{"paper_id":"2cba8f7c-7aef-499e-9caa-b3308e573722","body_text":"npj | women's health Article\nhttps://doi.org/10.1038/s44294-025-00099-3\nEnhanced analysis of endometriosis\npatients’ plasma using #Enzian annotation\nhighlights potential biomarkers for early-\nstages of disease\nCheck for updates\nDaniel Rodriguez Gutierrez1, Alina Astourian1, Marianne Spalinger1, Lucie Berclaz1, Monique Hartmann1,\nJulian Metzler2, Isabell Witzel2, Patrick Imesch2, Valentina Vongrad1 & Brigitte Leeners1\nEndometriosis is a chronic inﬂammatory condition marked by the presence of endometrial-like tissue\noutside the uterus, often co-occurring with leiomyoma and presenting a diagnostic challenge. We\nanalyzed 96 plasma cytokines and in ﬂammatory markers in 86 women undergoing surgery for\nsuspected endometriosis, using multiplex assays and unsupervised clustering methods. Patients\nwere classiﬁed using both rASRM and the more granular #Enzian system to assess lesion-speciﬁc and\nstage-speciﬁc biomarker patterns. We identi ﬁed ﬁve distinct patient clusters re ﬂecting disease\nheterogeneity, with improved resolution using the #Enzian classi ﬁcation. Notably, the presence of\nleiomyoma inﬂuenced cytokine pro ﬁles, potentially obscuring biomarker signals. Key biomarkers\nincluding perforin, TRAIL, sFasL, IL-17F, PDGF, VEGFA, and MCP-2 were associated with disease\npresence and stage. These ﬁndings highlight the value of advanced classi ﬁcation systems and\nemphasize the importance of accounting for comorbid conditions. Our results support the\ndevelopment of non-invasive biomarker panels for earlier and more accurate diagnosis of\nendometriosis.\nEndometriosis (EM) is a chronic gynecological disorder characterized by the\npresence of endometrium-like tissue at ectopic locations, including the\novaries, recto-sigmoid, fallopian tubes, and pelvic peritoneum1.A l t h o u g h\nthe exact incidence and prevalence of EM is dif ﬁcult to determine, it is\nestimated to affect 5 –10% of women of reproductive age worldwide,\nimposing a signiﬁcant burden on patients’ lives through chronic pelvic pain,\nfatigue, infertility, and diminished quality of life2,3. The most common types\nof pelvic EM are super ﬁcial peritoneal lesions (15–50% of EM patients),\ndeep-inﬁltrating lesions (DE, 20%), and endometriomas (2 –10% of EM\npatients and 50% of women treated for infertility)4.\nHistorically, researchers have viewed endometriotic lesions as a single,\nuniform disease phenotype5. However, extensive heterogeneity has been\ndemonstrated in endometriotic lesions, even among lesions from the same\npatient\n6. This diversity challenges traditional histological classi ﬁcations,\noften rendering them ineffective in precise diagnosis but also in guiding\ntreatment. It also highlights the n eed for more nuanced approaches to\ndiagnosis, classiﬁcation, and treatment of what accounts for the complex\nand variable nature of the disease. In clinical practice, better outcomes are\noften observed when endometriosis surgeries are performed by experienced\nteams in specialized centers, where comprehensive care and advanced\nsurgical expertise contribute to more effective symptom relief and reduced\nrecurrence\n7.\nIrrespective of its heterogeneity, EM is considered an in ﬂammatory\ndisease, marked by increased levels of activated macrophages and cytokines,\nsuch as interleukins, and tumor necrosis factor-alpha (TNF-α), in the\nperitoneal ﬂuid of affected women 8. Several in ﬂammation-related bio-\nmarkers show elevated levels in the plasma of women with EM, including\nannexin A2 (ANXA2), ﬁbronectin, collagen IV, C-reactive protein (CRP),\nand monocyte chemoattractant protein-1 (MCP-1)\n9–12.M o r e o v e r ,t h i sw a s\nalso the case for a wide range of other molecules, such as glycoproteins,\ngrowth factors, microRNAs (miRNAs), long non-coding RNAs (lncRNAs),\nand various proteins associated with angiogenesis and immune response\npathways\n13.\nWith the current lack of ef ﬁcient biomarkers, the gold-standard for\ndiagnosing EM is laparoscopic pathological biopsy when imaging studies\nare unremarkable14. Furthermore, there is currently no treatment available\n1Department of Reproductive Endocrinology, University Hospital Zurich, Zurich, Switzerland. 2Department of Gynecology, University Hospital Zurich,\nZurich, Switzerland. e-mail: Brigitte.Leeners@usz.ch\nnpj Women's Health |            (2025) 3:60 1\n1234567890():,;\n1234567890():,;\n\nto entirely cure EM. Hormonal therapies fail in up to 30% of cases 15,\nrecurrence is common after surgical resection (30% of cases)13, and, for 55%\nof EM patients, current medications offer only limited symptom relief in the\nlong term\n16,17, underscoring the urgent need for novel diagnostic and ther-\napeutic targets. Ideally, a noninvasive biomarker for diagnosing EM would\nbe derived from serum or plasma, as well as saliva, urine, or menstrual\nefﬂuent, offering a valuable tool for the differentiated diagnosis of different\nendometriotic lesions and therapeut ic monitoring of patients alongside\nstandard clinical evaluations.\nEmerging technologies such as proteomics, metabolomics, and geno-\nmics, allow exploration of extensive panels of molecules or gene pro ﬁles.\nThese technologies represent a signi ﬁcant advancement in the ﬁeld,\npotentially eliminating the need for invasive procedures like laparoscopies\nf o rd i a g n o s t i cp u r p o s e s .D e s p i t et h ep r o m i s eo ft h e s ea d v a n c e da p p r o a c h e s ,\nthe clinical translation of biomarkers for EM remains challenging. To date,\nno single biomarker or panel ha s been validated with the speci ﬁcity and\nsensitivity required for routine clinical diagnostic use\n9.\nAdditionally, comorbid conditions and medications can confound\np r o t e o m i ca n dm e t a b o l i cp r oﬁles, making it challenging to distinguish\ndisease-speciﬁc signals from those arising due to coexisting conditions.\nAdjusting models for comorbid conditions and medication has been shown\nto result in a notable reduction of signiﬁcant associations and unique pre-\ndictors, particularly in diseases with high comorbidity rates\n18–20.T h e s e\nﬁndings emphasize the necessity for research to incorporate both comor-\nbidity proﬁles and medication regimens as covariables to enhance the\nprecision of disease-speciﬁc risk factor identiﬁcation, ultimately aiding the\ndevelopment of more effective diagnostic and treatment strategies for\nchronic diseases.\nTherefore, the aim of the present study was to evaluate whether plasma\ncytokines and inﬂammatory markers could facilitate the diagnosis of EM\nbased on the most recent annotation systems rASRM and #Enzian. We also\ninvestigated whether speciﬁc biomarkers were associated with (i) disease\nstage, (ii) speciﬁc lesion types according to #Enzian annotation, and how\n(iii) potential confounders such as myoma or medication in ﬂuence bio-\nmarker ﬁndings.\nResults\nLack of distinct cytokine proﬁle differences between EM patients\nand controls without clustering\nThe comparison of 96 cytokine markers of women diagnosed with EM with\nthose of control women without considering the presence of myoma did not\nreveal any clear group separation nor did it identify signiﬁcantly regulated\nmarkers, even when all patients’ metadata were incorporated in the analysis\n(Fig. 1a and b, Supplementary Table 3). Although we observed signiﬁcant\ndifferences between women presenting with EM and control women with\nregard to age and parity (Table 1), these did not result in any differences\nbetween biomarker proﬁles.\nComparison of biomarkers in women with and without EM and/or\nmyoma: Myoma obscures the differences between healthy sub-\njects and EM patients\nIn our cohort, among the 65 patients diagnosed with EM, 18 (27.7%) were\nalso positive for myoma. In contrast,within the control (no EM) group, 11\nout of 21 individuals (52.4%) presented with myoma (Fig.1c). To explore\nthe impact of myoma on plasma marker levels, we compared controls\nwithout myoma with those in whom myoma had been detected, as well as\nEM patients with and without myoma (Figure 1d–h, and Supplementary\nTables 1 and 4). In patients with myoma as a single condition, we observed a\nsigniﬁcant reduction in the plasma levels of perforin, IL-31, CXCL16, and\nENA-78 compared to controls (Fig. 1d). Perforin was also signi ﬁcantly\nreduced in EM patients without myoma (Fig.1e). In cases where EM and\nmyoma coexisted, a combined effect of both conditions was evident, with\nsigniﬁcant reductions in perforin, CXCL16, and TRAIL when compared\nwith controls without myoma (Fig.1f). The overlapping impact of these two\nconditions on plasma markers was further highlighted by the absence of\ndifferentially expressed markers when patients with EM alone were com-\npared with those with myoma alone (Fig.1g). These results may explain the\nmasking of potential EM markers by i ncluding patients diagnosed with\nmyoma among the control group. Use of hormonal medication did not\nsigniﬁcantly affect the plasma levels of the analyzed markers (Supplemen-\ntary Fig. 1).\nAssociation between biomarkers and disease stage: IL-17F,\nPDGF-AB/BB, VEGFA, MCP-2, and MPI-1β plasma levels were\nincreased in early stages of EM\nThe unsupervised clustering presented here successfully grouped patients\ninto manageable clusters based on their EM heterogeneity and severity\nwithout introducing experimenter bias and consequently generated a reli-\nable basis for the evaluation of stage-speciﬁcb i o m a r k e r s( F i g .2a–c, Sup-\nplementary Fig. 2). In contrast, rASRM groups often included patients with\nvery different degrees of heterogeneity and severity (Fig.2d, asterisks). For\nexample, the rASRM II group, comp osed mostly of patients with mild\nperitoneal lesions, also includes a\npatient with P2, T3, A2, B2, C2 lesions and\npatent tubes.\nWe then analyzed whether our #ENZIAN clustering approach out-\nperforms the traditional rASRM classi ﬁcation in highlighting potential\nbiomarkers for different stages of EM (Fig. 3). Using the rASRM classiﬁ-\ncation, we observed a signiﬁcant reduction of perforin plasma levels across\nvarious EM stages and in myoma patients compared to the control group\n(Fig. 3a). Other markers, such as sFasL, TRAIL, CXCL16, and PDGF-AB/\nBB, were differentially expressed in o ne or more stages, especially in the\nmore severe stages of the disease.\nUsing the #ENZIAN annotation, our clustering approach conﬁrmed\nthe reduction of perforin among cluster #II to #V as well as the reduction of\nCTACK in the most severe cluster, and sFasL in intermediate stages of the\ndisease (Fig.3b). Most importantly, this approach allowed the identiﬁcation\nof an initial stage cluster #I, where IL-17F, PDGF-AB/BB, VEGFA, MCP-2,\nand MPI-1β plasma levels were signiﬁcantly elevated. These elevations were\nunique to the earliest stages of EM and were not apparent with the rASRM\nclassiﬁcation.\nWe evaluated the discriminative power of the identiﬁed plasma bio-\nmarkers in classifying individuals intotheir respective groups based on their\nplasma marker proﬁles and predicted their optimal threshold in plasma\n(Fig. 4 and Supplementary Table 5). Our analysis revealed discriminative\nperformance of various of these plasma markers when comparisons of\ncontrol group with EM and/or myoma groups were made (Fig.4a). For the\nc o n t r o lg r o u pw i t h o u tE Mo rm y o m a ,b i o m a r k e r ss u c ha sp e r f o r i n\n(AUC = 0.82, predicted cutoff = >7.64 ng/ml), TRAIL (AUC = 0.75, pre-\ndicted cutoff = >68.73 pg/ml), and CXCL16 (AUC = 0.77, predicted\ncutoff = >802.25 pg/ml) demonstrated high discriminative ability, char-\nacterized by strong sensitivity and relatively low false positive rates. In the\nmyoma group, ENA-78 and IL-34 showed notable sensitivity values (1.00\nand 0.91, respectively) but were associated with higher false positive rates,\nindicating potential limitations in speciﬁcity. Focusing on the EM group, IL-\n31, GRO-alpha, and LIF exhibited moderate performance, with sensitivity\nvalues ranging from 0.70 to 0.91 and AUCs between 0.61 and 0.67. While\ntheir overall classiﬁcation ability was less robust, these biomarkers may still\nhave diagnostic relevance in speciﬁc contexts. The subgroup of individuals\nwith both EM and leiomyoma demonstrated moderate discriminative\nperformance for biomarkers like TARC (AUC = 0.68) and IL-17F\n(AUC = 0.67).\nThe biomarker analysis revealed notable differences in diagnostic\nperformance across the #Enzian clusters, with speciﬁc biomarkers showing\npromise in distinguishing early stages of EM (clusters #I and #II) from\ncontrols (Fig.4b, Supplementary Table 5). Regarding cluster #I, IL-17F and\nPDGF-AA demonstrated strong classiﬁcation performance, with AUCs of\n0.75 and 0.84, respectively, combined with balanced sensitivity (0.80 and\n0.80) and speciﬁcity (0.80 and 0.83). Similarly, VEGF-A exhibited robust\npredictive performance for cluster #I with an AUC of 0.83 and a sensitivity\nof 1.00, albeit poor speciﬁcity (0.58). In cluster #II, IL-31 exhibited medium\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 2\n\nperformance, with an AUC value of 0.66, while its sensitivity and speciﬁcity\nvalues were moderate (0.64 and 0.69, respectively). For prediction of later-\nstage clusters (#III, #IV, and #V), bio markers generally displayed lower\ndiscriminative power compared to those in early stages. Across all groups,\nbiomarkers such as CXCL16, TRAIL, and perforin demonstrated moderate-\nto-high diagnostic performance, withAUCs ranging from 0.77 to 0.82, and\nmoderate to high sensitivity and speci ﬁcity, underscoring their utility in\ndistinguishing healthy controls from EM cases. The diagnostic potential of\nbiomarkers such as TARC, GCP-2, and IL-17F in combined conditions (EM\n+ myoma) was mostly limited, with AUCs of 0.64–0.68.\nAdditionally, we explored the presence of trends among all markers\nanalyzed and the #Enzian clusters in order to highlight potentially\naffected mechanisms (Supplementary Fig. 3a). The expression trend of\nperforin, with reduced levels in myoma-only patients and a gradual\nreduction in EM stages, was unique and grouped individually. However,\nseveral markers followed the sFasL trend, with a reduced expression in\nthe more severe clusters and myoma. Among this group, we found\nGranzyme B, IL-7, IL-35, IL-16, HMGB1, and eotaxin 3. In contrast,\nCXCL16 and ENA-78 markers were signi ﬁcantly reduced in myoma\npatients but not in those with EM. Interestingly, we observed multiple\nmarkers speci ﬁcally elevated in #I cluster, including the signi ﬁcant\nVEGFA, MIP-β, MCP-2, PDGF-AB/BB, and IL-17F, as well as the non-\nsigniﬁcant INF-α2, TGF-α, CCL28, FGF-2, granzyme B, I-309, MCP-4,\nGM-CSF, and interleukins 1α, 17A, 34, 15, and 2. Additionally, CTACK,\nTRAIL, and GCP-2 were similarly decreased in EM and myoma patients,\namong eotaxins 1 and 2, and SCF markers.\nFig. 1 | Patient clustering and biomarker analysis in endometriosis. a PCA using\ncytokine data showing limited separation between endometriosis (EM) and control\n(CTRL) groups. b Volcano plot showing signi ﬁcantly altered markers in EM vs.\ncontrols; no markers were signi ﬁcantly upregulated. c Venn diagram showing\nleiomyoma (Myoma) co-occurrence: 18/65 EM patients and 11 controls had\nmyoma. d–g) Volcano plots comparing biomarker expression in patients with\nmyoma (d), EM ( e), and EM +myoma (f) against healthy controls without either\ncondition, and EM alone vs. myoma (g). h Boxplots showing normalized expression\nof selected markers; whiskers indicate 1.5× IQR. P-values (t-test or Mann–Whitney\nU test) are indicated above each box;*p < 0.05, **p < 0.01; non-signiﬁcant results are\nlabeled.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 3\n\nAssociation between biomarkers and lesion types: IL-17F and\nPDGF-AB/BB marker levels are associated with speciﬁc\n#Enzian-based lesion types\nThanks to the versatility of #Enzian classiﬁcation, we were also able to zoom\ninto different lesion types (Supplementary Figs. 3b and 4). Although we\ncould capture decline tendencies of perforin plasma levels in patients with\ntubule-ovarian lesions, as well as grade 3 ovarian lesions, those tendencies\ndid not reach signiﬁcance. However, IL-17F levels were signiﬁcantly elevated\nin EM patients without peritoneal lesions, with tubulo-ovarian lesions of\ngrade 2 and initial-stage lesions in therectum location. PDGF-AB/BB levels\nwere also signiﬁcantly higher in non- and grade 1 peritoneal, as well as grade\n3 ovarian lesions. Theseﬁndings suggest a lesion-type-dependent inﬂuence\non levels of certain cytokines and other plasma markers in EM patients.\nCombined effect of EM and concomitant myoma on plasma\nmarkers\nSince the presence of myoma was also associated with a downregulation of\nseveral of the analyzed markers, we tested whether this association was\nuniform among the deﬁned #Enzian clusters of EM patients. We observed\nthat the plasma levels of perforin ten dt ob el o w e rf o rp a t i e n t sw i t hc o n -\ncomitant EM and myoma compared with the levels of those presenting\nexclusively with EM (Fig. 5). This effect was also present in most of the\nclusters for sFasL, TRAIL, Granzyme B, and CTACK, albeit not signiﬁcant.\nIL-17F, PDGF-AB/BB, VEGFA, and MCP-2 elevation in #I cluster was\nsigniﬁcant for patients with EM only compared to levels for the control\ngroup. Additionally, within the same cluster, IL-17F levels were more\npronounced only in the presence of concomitant myoma. This effect was\nalso observed in MIP-1β levels in #I cluster.\nDiscussion\nEM and myoma are hormonally responsive conditions that often co-occur,\nposing signiﬁcant challenges for biomarker research. We identiﬁed potential\nkey immune markers, including perforin, TRAIL, and sFasL, whose levels\nwere reduced in both EM and myoma patients, suggesting impaired\napoptotic and cytotoxic responses that may contribute to lesion persistence.\nAdditionally, elevated IL-17F, PDGF-AB/BB, VEGFA, and MCP-2, levels in\nEM patients highlight their potential as early-stage biomarkers and ther-\napeutic targets.\nIn patients with myoma only, many markers were signiﬁcantly altered,\nshowing a reduction similar to that observed in EM patients. This overlap\ncomplicated the identiﬁcation of EM-speciﬁc biomarkers, as the presence of\nmyoma in the control group introduced substantial variability. In addition,\nuncontrolled factors such as hormonal treatment and menstrual cycle phase\nlikely contributed further to biomarker variability across groups. Our study\nunderscores the importance of precise classiﬁcation methodologies in dif-\nferentiating EM heterogeneity and the need for meticulous control group\nselection in biomarker analyses for their relevance in the identiﬁcation of\ntreatment targets for EM and myoma. While most studies rely on\nlaparoscopy-conﬁrmed EM-negative women\n21–25, this approach often\nintroduces bias, as the indication for surgery typically involves other\ngynecological disorders such as myoma. In our cohort, 69% of EM-negative\nwomen presented with myoma, reﬂecting its high global prevalence (77%)\namong adult women, with a substantial proportion remaining\nasymptomatic\n26. This observation aligns with recent proteomics studies27,\nwhich emphasize that differences between study groups and inclusion of\ngynecological comorbidities in control groups may substantially bias bio-\nmarker discovery.\nTable 1 | Descriptive statistics\nMissing Overall CTRL (no EM) EM P-Value\nn 86 21 65\nAge (years) 37.3 (6.3) 39.9 (5.2) 36.5 (6.5) 0.021 *\nBMI 8 24.7 (5.7) 25.3 (5.7) 24.6 (5.7) 0.699 ns\n# cycles during last year 4 10.2 (2.5) 10.6 (2.3) 10.1 (2.5) 0.487 ns\nEver pregnant, n (%) Yes 38 (44.2) 14 (66.7) 24 (36.9) 0.033 *\nNo 48 (55.8) 7 (33.3) 41 (63.1)\nMain symptom, n (%) Infertility 14 (16.3) 3 (14.3) 11 (16.9) 0.049 *\nPain 57 (66.3) 10 (47.6) 47 (72.3)\nBleeding dis. 7 (8.1) 3 (14.3) 4 (6.2)\nOther 4 (4.7) 2 (9.5) 2 (3.1)\nNo symptoms 2 (2.3) 1 (4.8) 1 (1.5)\nUnknown 2 (2.3) 2 (9.5)\nRelatives with endometriosis, n (%) Sister 2 (2.3) 0 (0.0) 2 (3.1) 0.705 ns\nMother 15 (17.4) 5 (23.8) 10 (15.4)\nGrandmother 1 (1.2) 0 (0.0) 1 (1.5)\nAunt 2 (2.3) 1 (4.8) 1 (1.5)\nCousin 2 (2.3) 1 (4.8) 1 (1.5)\nOther 2 (2.3) 0 (0.0) 2 (3.1)\nNone 62 (72.1) 14 (66.7) 48 (73.8)\nHormonal treatment, n (%) Combined Oral Contraceptives 4 (4.7) 1 (4.8) 3 (4.6) 0.831 ns\nProgesterone only 16 (18.6) 3 (14.3) 13 (20.0)\nGnRH agonist 1 (1.2) 1 (1.5)\nIUDs (copper) 2 (2.3) 2 (3.1)\nNone 63 (73.3) 17 (81.0) 46 (70.8)\nData are presented as mean (SD) or n (%) when indicated. For continuous variables, t-test was used to compare the means of the two groups. For categorical variables with two categories, t he two-\nproportion z-test was used to test for differences in proportions. For categorical variables with more than two categories, the chi-square test was u sed to evaluate associations between the\ngroups. *p < 0.05.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 4\n\nIn our study, we observed a signi ﬁcant decrease in plasma levels of\nperforin, TRAIL, and sFasL in patients with EM and/or myoma compared to\nhealthy controls. These molecules playcentral roles in immune surveillance\nthrough cytotoxic T cell and natural killer (NK) cell-mediated apoptosis.\nPerforin, a critical pore-forming protein released by cytotoxic lym-\nphocytes, plays a pivotal role in the cytotoxic activity of CD8+ Ta n dN K\ncells, facilitating granzyme entry into target cells to induce apoptosis28.T h e\nobserved decrease in perforin, together with lower granzyme B levels, sug-\ngests functional impairment of cytotoxic T lymphocytes and NK cells.\nPrevious studies have implicated cytotoxic T cells in the pathogenesis of EM,\nreporting a defective T cell response and reduced cytotoxicity toward\nautologous endometrial cells\n29.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 5\n\nBoth perforin and granzyme B have also been implicated in the\nshedding of human endometrium, contributing to endometrial\nmenstruation30,31.Ad e ﬁciency in perforin-med iated cytotoxicity may\ntherefore play a role in lesion persistence in EM. Further, Yang et al. 32\nproposed that interactions between endometriotic stromal cells and mac-\nrophages produce IL-10 and TGF- β, impairing NK and CD8 + T cell\ncytotoxicity in EM and promoting immune evasion.\nPerforin-mediated cytotoxicity is considered a rapid mechanism of\naction for cytotoxic T lymphocytes33. However, CD8+ T and NK cells can\nalso utilize a slower cytotoxic mechanism involving the Fas ligand (FasL)\npathway, which triggers programmed cell death in target cells through the\ninteraction of Fas expressed on the target cell surface with FasL on T cells.\nThe triggering of Fas by FasL results ininduction of programmed cell death\nin susceptible Fas-bearing cells\n34. In EM, a stage-dependent reduction in\nFasL expression has been observed in ectopic endometrial tissue and peri-\ntonealﬂuid, possibly mediated by macrophage-derived factors such as TGF-\nβ and PDGF35,36.\nIn addition, cytotoxic T and NK cells can kill target cells through the\nTNF-related apoptosis-inducing ligand (TRAIL) pathway37.T R A I Li sa\nTNF family member that induces apoptosis upon binding to its receptors.\nDecreased levels of granzyme B, perforin, and TRAIL have been reported in\nthe peritoneal ﬂuid of EM patients, indicating functionally defective NK\ncells\n38. Further, TRAIL antagonists such as osteoprotegerin (OPG) are ele-\nvated in EM peritonealﬂuid, interfering with TRAIL-induced apoptosis39.\nThe impairment of these apoptosis- inducing pathways collectively\nweakens the immune response of local NK and CD8+ T cells, potentially\nallowing ectopic lesions to survive and grow39,40. These mechanisms may be\npromoted by elevated TGF- β and estradiol levels41,w h i c ha r ek n o w nt o\nsuppress immune cell activity. In mouse and human studies, increasing\nTGF-β expression has been associated with decreased cytotoxic responses\nand enhanced lesion survival and invasion\n42.\nInterestingly, early-stage EM (cluster #I) displayed a distinct proﬁle,\nwith elevated levels of immune and angiogenic markers, including IL-17F,\nPDGF-AB/BB, VEGFA, MCP-2, and MIP-1 β. These elevations suggest\nearly in ﬂammatory and angiogenic res ponses supporting lesion\nestablishment.\nAmong the immune markers, IL-17F plays a signiﬁcant role. IL-17F\nshares strong homology with IL- 17A and is produced mainly by Th17\ncells\n43. Sisnett et al. 44 hypothesized that IL-17 produced by Th17 cells\nexacerbates EM by recruiting immune cells to lesion sites and enhancing\nlesion establishment. Elevated IL-17 levels have been consistently observed\nin EM, especially in its early stages\n45–47. Moreover, increased RNA expression\nof IL-17A, IL-17F, IL-12B, and TGF-β1 in ectopic tissues suggests coordi-\nnated upregulation of in ﬂammatory and immune-modulatory\npathways44,48.\nThe macrophage inﬂammatory protein MIP-1β (CCL3), expressed in\nthe endometrium, is correlated with NK cell recruitment to the endometrial\nzone, as indicated by a strong correlation between the endometrial MIP-1β\nconcentration and the number of endometrial NK cells49. In vitro studies\ndemonstrated higher secretion of MIP-1 β by lymphocytes from women\nwith EM, indicating a role in the altered immune environment50.O u rd a t a\nshow that MIP-1β is elevated particularly in patients with both myoma and\nEM at early stages, suggesting that myoma may amplify in ﬂammatory\nchemokine responses. MCP-2, a C-C chemokine subfamily member, acti-\nvates basophils, mast cells, and NK cells. Its increased expression in endo-\nmetriotic lesions further suggests a role in chronic inﬂammation51.\nAngiogenesis is another critical mechanism in the establishment of\nearly EM lesions, with VEGFA and PDGF playing central roles. VEGFA\npromotes vascularization and is upregulated in the peritonealﬂuid of EM\npatients, with enhanced regulation during menstruation\n52,53. In vitro studies,\ninhibition of VEGFA reduced ectopic endometrial mesenchymal stem cell\nproliferation, motility, and angiogenesis\n54,55. According to our results,\nincreased levels of VEGFA might also be involved in the stimulation of\nproliferation, motility, and angiogenesis in lesions at early stages of the\ndisease.\nPlatelet-derived growth factor isoforms (PDGF-AB/BB) exhibit\nangiogenic effects by stimulating endo metrial stromal cell proliferation,\nmigration, and invasion\n56. Both VEGFA and PDGFs not only contribute to\nneovascularization but also modulate immune responses, potentially\ncreating a permissive environment for early lesion growth.\nAltogether, our results suggest that early EM development is char-\nacterized by a dual imbalance: weakened immune clearance through\nimpaired cytotoxicity and enhanced in ﬂammatory-angiogenic signaling.\nDefective apoptosis pathways, reﬂected by reduced plasma levels of perforin,\nTRAIL, and sFasL, may allow survival and implantation of ectopic endo-\nmetrial fragments, while elevated in ﬂammatory and angiogenic markers\nsuch as IL-17F, MCP-2, MIP-1 β,V E G F A ,a n dP D G F - A B / B Bp r o m o t e\nlesion establishment, vascularization, and persistence. This coordinated\ndisruption of immune surveillance and tissue remodeling highlights the\ncomplex pathophysiology underlying EM and may explain its frequent\nprogression from early, asymptomatic stages to more severe, chronic dis-\nease. Understanding this provides a mechanistic framework for identifying\nreliable non-invasive biomarkers, guiding the development of diagnostic\npanels, and identifying new therap eutic targets focused on restoring\nimmune function and limiting angiogenesis in EM patients. In line with this,\nSchoeman et al. results\n27 further support the need to integrate immunolo-\ngical, metabolic, and angiogenic axes into future biomarker research and\nexplore the potential of these biomarkers for stratifying patients based on\ndisease stage, lesion subtype, and comorbidities to enhance clinical applic-\nability and relevance.\nOur analysis revealed the potential of several biomarkers, such as\nperforin, IL-17F, and PDGFs, for application in the early diagnosis of EM\nand related conditions. A decline in perforin plasma levels below 7.64 ng/ml\nmay re ﬂect cytotoxic dysfunction, consi stent with the immune evasion\nmechanisms described earlier. Conversely, elevated levels of IL-17F\n(>40.09 pg/mL) and PDGF-AA ( > 1.78 ng/mL) may reﬂect active inﬂam-\nmatory and angiogenic signaling duri ng early lesion establishment. The\nidentiﬁcation of these patterns in mild or early-stage EM highlights the\npotential for plasma biomarker-based tools to aid in early, non-invasive\ndiagnosis, a major unmet clinical need.N o n e t h e l e s s ,f u r t h e rv a l i d a t i o ni s\nrequired in larger, prospective cohorts with careful stratiﬁcation by disease\nstage, hormonal treatment, and comorbidities such as myoma.\nThe diversity of EM lesions complicates patient classiﬁcation, diagnosis,\nand treatment monitoring. Understanding the clinical impact of individual\nFig. 2 | EM patient clustering based on the #Enzian classi ﬁcation system. EM\npatient clustering according to #Enzian classiﬁcation. a Schematic representation of\nthe #Enzian classi ﬁcation coding system. Possible values and lesion types for dif-\nferent anatomical regions. Example annotations are provided to illustrate how the\ncoding is applied from surgeon ’s annotation, decomposed and translated into a\nseverity score (SC) derived from the lesion types and their respective severity levels\n(0-3). The dimensionality reduction model (Kernel-PCA) is presented in 2D and 3D.\nEM patients (orange) and controls (blue) are plotted. Unsupervised clustering (K-\nmeans) was used for classifying patients. Clusters were named according to the mean\nEnzian Severity Index (mESI) of each group as control (Ctrl, 0.00), #I (0.02), #II\n(0.12), #III (0.31), #IV (0.48), and #V (0.89) respectively. b Principal Component\nAnalysis (PCA) showing the distribution of each lesion classi ﬁer from #Enzian\namong the different clusters. c Parallel categories plot representing the #Enzian\nclassiﬁcation across different patient clusters. This plot visualizes the distribution of\nlesion types and severities. The categories show the relationship between the extent\nof disease and patient groups, providing a comprehensive overview of the classi ﬁ-\ncation. d PCA showing the #Enzian cluster, rASRM group, and ESI generated for\neach patient. Asterisks: patients showing dissociation between rASRM classiﬁcation\nand ESI. A: retrovaginal space, B: sacrouterine ligaments, C: rectum, CTRL: control,\nEM: endometriosis, FA: adenomyosis, FI: intestinum, Fother: other, FU: ureter, O:\novarian, P: peritoneal, T: tubo-ovarian, Tpt: patency test.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 6\n\nFig. 3 | Plasma biomarker analysis by #Enzian cluster and myoma status.\na Comparison of revised American Society for Reproductive Medicine (rASRM)\nclassiﬁcation (groups I –IV) and #Enzian-based classi ﬁcation (groups #I –#V). Vol-\ncano plots compare each cluster to the control (Ctrl) group without leiomyoma\n(myoma). Labels for signi ﬁcantly reduced (blue) and signi ﬁcantly elevated (red)\nmarkers are shown. Markers for which the p-value was < 0.05, but the fold change\nwas lower than 2, are labelled in gray. Boxplots display the normalized expression of\nsigniﬁcantly expressed markers across different groups. The central box shows the\ninterquartile range (IQR) and median (horizontal line inside the box). The “whis-\nkers” extend to the smallest and largest values within 1.5 times the IQR, respectively.\nb Expression of signiﬁcantly differenciated markers among the clustering groups. P-\nvalues from the statistical tests (t-test or Mann-Whitney U test) comparing each\ngroup with the control group are shown on top of each group ’s box. Signi ﬁcant p-\nvalues are indicated by asterisks ( *p < 0.05, **p < 0.01, ***p < 0.001), while non-\nsigniﬁcant p-values are represented with their exact values.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 7\n\nlesions is essential to address these challenges. The #Enzian system provides a\ncomprehensive user-friendly mapping of EM, accounting for anatomic\nlocation, lesion size, adhesions, and the degree of involvement of adjacent\norgans\n57,58 with excellent inter- and intra-observer agreement in MRI-based\ndiagnosis of EM59. Using this classiﬁcation system, we generated coherent\npatient clusters with similar EM heterogeneity and severity, revealing bio-\nmarkers, such as IL-17F, that may remain undetected when using other\nclassiﬁcation systems, particularly in the early stages of the disease.\nThe #Enzian annotation enables us to categorize patients based on a\nspeciﬁc lesion type, facilitating the identiﬁcation of trends in plasma bio-\nmarkers associated with that type. H owever, this approach cannot fully\nisolate the effects of the selected lesion type from the in ﬂuence of other\ncoexisting lesions. Consequently, the observed trends may be masked or\ndiluted by the“noise” generated by these additional lesions. For instance, the\ntrends of IL-22 and IL-24 in patients with peritoneal lesions, as well as IL-3,\nI-Tac, APRIL, and TPO in patients with tubo-ovarian lesions, underscore\nthis limitation. These insights hi ghlight the need for larger and more\ngranular studies of lesion-speci ﬁc biomarkers. Such efforts could sig-\nniﬁcantly enhance our understanding of EM biology and pave the way for\nmore effective, tailored therapeutic approaches.\nOne of the main limitations of our study is the small number of patients\nin the control group, which is largely due to the stringent criteria required for\naccurate diagnosis. Identifying control patients who are deﬁnitively negative\nfor EM necessitates invasive laparoscopy, and further distinguishing those\nFig. 4 | Evaluations of biomarkers as diagnostic predictors of disease presence\nand cluster-based classiﬁcation. a Receiver Operating Characteristic (ROC) curves\nand permutation test results for the top 5 biomarkers in distinguishing control\n(CTRL) groups from various disease conditions, including leiomyoma (Myoma),\nendometriosis (EM), and EM + Myoma. b Analysis of biomarker performance\nwithin clusters generated based on #Enzian annotations (groups #I –#V), reﬂecting\ndisease subtypes. The ROC curves illustrate the diagnostic accuracy of the top ﬁve\nbiomarkers for each cluster/condition, with Area Under the Curve (AUC) values\nindicating their predictive power. Permutation test results are shown as −log10(p-\nvalue), with Bonferroni correction applied for multiple testing. The signi ﬁcance\nthreshold after Bonferroni correction is marked with a red dotted line.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 8\n\nwithout myoma adds an additional layer of complexity. This constraint\nresulted in a limited sample size that may impact the generalizability of our\nﬁndings. The signiﬁcant age difference between EM patients and women\nwithout EM (p = 0.021) might have inﬂuenced ﬁndings. Additionally, the\nmenstrual cycle phase was not included as a variable in the analysis due to\nthe inconsistent availability of this i nformation in routine clinical doc-\numentation. While all patients were recruited during standard preoperative\nassessments, cycle phase data were not uniformly recorded across sites. As\nsuch, we cannot exclude the possibility that physiological variation across\nt h ec y c l em a yh a v ec o n t r i b u t e dto biomarker variability.\nOur ﬁndings revealed a signiﬁcant reduction in key immune markers,\nincluding perforin, TRAIL, and sFasL, in both EM and myoma patients,\nindicating impairments in apoptoticand cytotoxic immune responses that\nmay contribute to lesion persisten ce. Patient clustering using #Enzian\nannotation identiﬁed elevated levels of IL-17F, PDGF-AB/BB, VEGFA, and\nMCP-2 in early-stage EM, highlighting their potential as early biomarkers\nand therapeutic targets. The interplay between immune suppression and\nangiogenesis in early-stage EM suggestsp r o m i s i n ga v e n u e sf o rc o m b i n a t i o n\ntherapies targeting these pathways. Advanced classi ﬁcation systems and\nprecise biomarker analyses are essential for gaining deeper insights into EM\npathophysiology. At the same time, identifying early-stage biomarkers offers\nhope for more accurate and timely diagnosis, enabling earlier interventions.\nThese ﬁndings could drive the development of personalized and effective\ntreatments, ultimately enhancing the quality of life for individuals affected by\nthis challenging condition.\nMethods\nStudy design\nPatient recruitment and enrollment through senior gynecologists, sup-\nported by a study nurse, took place between 2021 and 2023 at the Depart-\nments of Reproductive Endocrinology and Gynecology at the University\nHospital Zurich, Switzerland. The study was approved by the national\nmedical ethics committee (BASEC 2020-02117), and all participants pro-\nvided written informed consent prior to inclusion. All procedures involving\nhuman participants were conducted in accordance with the ethical stan-\ndards of the institutional and national research committee and with the 1964\nDeclaration of Helsinki and its later amendments.\nInclusion criteria for the study comprised 18- to 50-year-old women\nundergoing surgery (including hysterectomy) for EM, as well as women\nwithout EM undergoing surgical management of uterine leiomyoma, or ster-\nilization. Exclusion criteria for the clinical study were current pregnancy,\nbreastfeeding, a positive diagnosis of immunodeﬁciency or autoimmune dis-\neases, or increased risk of bleeding. Participants were interviewed regarding\ntheir lifestyle (i.e., alcohol consumption, smoking status, sports activity) and\nmedical history, especially with regard topain perception, as well as medication\nintake during the week prior to surgery. A meticulous investigation of the\npatients’ history, plus general and gynecological exams served to identify any\nadditional pathology before the surgical procedure. All baseline demographic\nand clinical data were collected as metadata information. Myomas were\nassessed by transvaginal ultrasound and palpation, and further classi ﬁed\naccording to the FIGO (International Federation of Gynecology and Obste-\ntrics) uterineﬁbroid classiﬁcation system, including size, type, and location. The\nreference test for the diagnosis of EM was laparoscopy with visualization of\ntypical lesions and histological evaluation, performed by expert surgeons with\nat least ten years of experience. EM was classiﬁed according to the revised\nAmerican Society for Reproductive Medicine (rASRM) and the #Enzian\nclassiﬁcation by the surgeon immediately after surgery.\nPatients were included in further analysis if they had clear and com-\nprehensive information regarding EM and/or uterine leiomyoma, including\ndetailed rASRM and #Enzian annotations, and sufﬁcient plasma for bio-\nmarker analysis. Of 89 initially selected patients, only three were excluded\nbased on these criteria. Patients diagnosed with EM were assigned to the EM\ngroup, while those without EM were included in the control (no EM) group,\nregardless of the presence of leiomyoma.\nPlasma sample collection\nEDTA-blood was collected from patients on the day of the surgery\naccording to standard operating procedure. Blood samples of a minimum\n9 ml were taken into EDTA-coated tubes for plasma collection.\nAll blood specimens were collected immediately upon transfer from\nthe ward, and prior to the induction of anesthesia, the induction of anes-\nthesia was initiated only after blood collection to avoid confounding effects.\nSamples were kept at 4 °C until processing. HAV IgG/IgM Combo and\nHBsAg /HCV /HIV /Syphilis Combo Rapid Test Cassettes (CiTest Diag-\nnostics, Canada) were used to measure major infections. Within one hour\nafter collection, the samples were centrifuged at 2000 G at 4 °C for 10 min.\nThe plasma was aspirated, aliquoted into 500 μl volumes, and stored at\n−80 °C until analysis.\nFig. 5 | Combined effect of EM and myoma on signiﬁcant markers in the previous\nanalyses. Normalized expression of these markers among the clustering groups,\ndiscriminating by the presence (grey) or absence (white) of myoma. Data are\nrepresented as the mean ± SD. Asterisks ( *) represent statistical signi ﬁcance for\ncomparisons of clusters generated based on #Enzian annotations (groups #I –#V) vs\nits corresponding control (Ctrl). Daggers ( †) represent statistical signi ﬁcance for\ncomparisons between no myoma and myoma conditions within each cluster. A\np-value < 0.05 was considered signi ﬁcant.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 9\n\nDimensionality reduction and unsupervised clustering\nThe actual rASRM classi ﬁcation score alone does not provide a detailed\ndescription of the heterogeneity or the full extent of multiple lesions and\ncannot map DE\n57,60,61. Thus, we also used the #Enzian annotation, which\nallows a comprehensive mapping of EM, including anatomical location,\nlesion size, adhesions, and involvement of adjacent organs, using picto-\ngrams to increase reliability and convenience of scoring\n57. The #Enzian\nannotation was decomposed to single lesion type variables (P, O left and\nright, T left and right, A, B left and right, C, Fa, Fb, Fi, Fu, Fother, and\npatency test) (Fig. 1), or combined to obtain an #Enzian Severity Index\n(ESI) by averaging the scores from all #Enzian variables. An ESI score can\nbe used to indicate the overall stage of the disease (Fig. 1a). The dimen-\nsionality reduction of these 15 variables distributed the patients according\nto the combined weight of every lesion-type severity, using a kernel PCA\nalgorithm to visualize the distribution of patients according to those\nvariables in 3D and 2D space\n62. For control patients, we assigned an\n#Enzian annotation with a 0 value on every variable. Theoretically, ESI\nvalues can range from 0 (no EM) to 2.2 (assuming the maximum value in\nall #Enzian variables). To identify the optimal number of clusters, we\nperformed silhouette analysis, which evaluates the consistency within\nclusters and the separation between them (Supplementary Fig. 2). Based\non the silhouette scores, we selectedﬁve as the optimal number of clusters\nfor K-means unsupervised clustering of the resulting distribution. This\napproach generated 5 distinct clusters. We named each cluster after its\nmean ESI (mESI), using similar terminology to that of the rASRM clas-\nsiﬁcation and adding a # to indicate the #Enzian origin (Fig. 1a and c).\nThus, we distinguished between #I (mESI = 0.02), #II (mESI = 0.12), #III\n(mESI = 0.31), #IV (mESI = 0.43), and #V (mESI = 0.89). We manually\nseparated the control group (mESI = 0.00) from the #I cluster. Cluster #I\n(n = 5) was composed of patients with one unique lesion of grade 1 or 2\nand no peritoneal or tubule-ovarian lesions. Cluster #II ( n = 28) was\nmainly composed of grade 1 and 2 peritoneal lesions and few low-grade\novarian and deep lesions at the sacrouterine ligaments. #III cluster\n(n = 14) comprised mainly patients with peritoneal and deep lesions. #IV\ncluster ( n = 11) contained patients with multiple lesions of medium to\nhigh grade, including peritoneal, ovarian, and deep lesions from multiple\nlocations. #V (n = 7) was considered the most severe cluster, composed of\npatients with grade 3 tubulo-ovarian lesions coexisting with peritoneal,\novarian, and deep lesions.\nMultiplex analysis of biomarker measurements\nAll methods were carried out in accordance with the relevant guidelines and\nregulations. A total of 500 µl of EDTA-plasma samples were aliquoted and\nsent to Eve Technologies Corp. (Calgary, Alberta, Canada). Multiplexing\nanalysis was performed using the Luminex™ 200 system (Luminex, Austin,\nTX, USA). Ninety-six markers were simultaneously measured in the sam-\nples using Eve Technologies’ Human Cytokine 96-Plex Discovery Assay®,\nwhich consists of two separate kits, the Panel A 48-plex and the Panel B 48-\nplex (MilliporeSigma, Burlington, Massachusetts, USA). The assay was run\naccording to the manufacturer’s protocol. The Panel A 48-plex consisted of\ns C D 4 0 L ,E G F ,e o t a x i n ,F G F - 2 ,F L T - 3l i g a n d ,f r a c t a l k i n e ,G - C S F ,G M - C S F ,\nGROα,I F N -α2, IFN-γ,I L - 1α,I L - 1β, IL-1RA, IL-2, IL-3, IL-4, IL-5, IL-6, IL-\n7, IL-8, IL-9, IL-10, IL-12(p40), IL-12(p70), IL-13, IL-15, IL-17A, IL-17E/IL-\n25, IL-17F, IL-18, IL-22, IL-27, IP-10, MCP-1, MCP-3, M-CSF, MDC, MIG/\nCXCL9, MIP-1α,M I P - 1β, PDGF-AA, PDGF-AB/BB, RANTES, TGF α,\nTNF-α,T N F -β, and VEGF-A. The Panel B 48-plex consisted of 6CKine,\nAPRIL, BAFF, BCA-1, CCL28, CTAC K, CXCL16, ENA-78, eotaxin-2,\neotaxin-3, GCP-2, granzyme A, granzyme B, HMGB1, I-309, I-TAC, IFNβ,\nIFNω, IL-11, IL-16, IL-20, IL-21, IL-23,IL-24, IL-28A, IL-29, IL-31, IL-33,\nIL-34, IL-35, LIF, lymphotactin, MCP-2, MCP-4, MIP-1δ,M I P - 3α,M I P - 3β,\nMPIF-1, perforin, sCD137, SCF, SDF-1, sFAS, sFASL, TARC, TPO, TRAIL,\nand TSLP. All shared sample information was fully anonymized, and the\nEve Technologies Corp. personnel conducting the assays were blinded to\npatient-identifying information and diagnoses. Estimated concentrations\nwere presented in pg/ml.\nConfounders\nWhen using plasma as a source of biomarkers for a speci ﬁcd i s e a s e ,i ti s\ncrucial to highlight that analyses represent bulk measurements, inﬂuenced\nby all the conditions of the patient. In EM, most patients present with\nmultiple lesions that often coexist with other conditions, such as myoma.\nWe hypothesized that the presence of additional conditions could obscure\nthe detection of differences between control subjects and those with EM.\nMyoma was of particular interest, given its high prevalence in affecting more\nthan 70% of reproductive-aged women worldwide, its signiﬁcant overlap\nwith EM, and the role of in ﬂammation in the pathogenesis of both\nconditions\n26,63. Furthermore, concomitant medication may inﬂuence ﬁnd-\nings in biomarkers64. Therefore, we controlled ourﬁndings for the presence\nof myoma as well as the potential inﬂuence of current treatments, including\ncombined oral contraceptives, progesterone therapy, GnRH agonists, and\ncopper IUDs, on inﬂammatory biomarkers levels.\nStatistics\nAltogether, data from 65 women with endometriosis and 21 controls were\navailable for analysis. The data were processed using Python program-\nming language with open-source packages such as pandas, scikit-learn,\nscipy, seaborn, and matplotlib\n65–68. Results of the descriptive analysis (i.e.\npatient’s clinical data) were presented as mean ± standard deviation (SD)\nwhile the concentrations of the measured proteins were presented as\nmean ± SD when variables were normally distributed and as mean and\nquartiles (Q1, Q3) when non-normally distributed (Table 1). Values that\nexceeded the mean ± 3xSD threshold were considered outliers and\nremoved from the analysis\n69. Fisher’s exact and Chi-square tests were used\nfor comparison of categorical variables. For continuous variables, t-test\nwas used to compare the means of two groups and ANOVA for multiple\ngroups. For categorical variables with two categories, the two-proportion\nz-test was used to test for differences in proportions. For categorical\nvariables with more than two categories, the Chi-square test was used to\nevaluate associations between the groups. The normal distribution of\nevery marker was tested using the Shapiro-Wilk test. For markers with a\nnormal distribution, ANOVA was used to compare the means between\ngroups. For markers that did not follow a normal distribution, the\nKruskal-Wallis test was applied. Apart from single proteins, additional\nvariables were constructed representing ratios of the proteins ’ con-\ncentrations. The expression data were normalized using mean-centering\nand standard deviation normalization. In brief, for each cytokine marker,\nthe mean expression level across all samples was calculated. This mean\nwas then subtracted from each individual sample ’s expression level,\nresulting in a distribution centered around zero. After mean-centering,\neach marker’s expression values were divided by the standard deviation of\nthe expression levels across all samples, thus scaling the data such that\neach marker has a standard deviation of one, and ensuring comparability\nacross different markers. Corrected P-values of <0.05 were considered\nsigniﬁcant.\nAfter a ﬁrst comparison of biomarker proﬁl e si nw o m e nd i a g n o s e d\nwith EM compared with control women, we added the presence of myoma\nas well as concomitant medication as additional factors in our analysis.\nBiomarker evaluation was performed using a logistic regression model for\neach biomarker independently, following a one-vs-rest strategy for multi-\nclass classiﬁcation. Each biomarker’s discriminative ability was quantiﬁed\nusing the Area Under the Receiver Operating Characteristic Curve (AUC),\ncalculated via 5-fold strati ﬁed cross-validation to ensure robust and\nunbiased performance estimates. For each cluster/condition, biomarkers\nwere ranked by their mean AUC scores, and the topﬁve biomarkers with the\nhighest AUCs were selected for further analysis. To assess the statistical\nsigniﬁcance of the observed AUC values, permutation testing was con-\nducted by randomly shufﬂing the class labels 1000 times to generate a null\ndistribution of AUCs under thehypothesis of no association.P-values were\ncalculated as the proportion of permuted AUCs that exceeded the observed\nAUC. To correct for multiple compa risons, Bonferroni correction was\napplied to the permutation-derived p-values, with statistical signiﬁcance\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 10\n\ndeﬁned as an adjusted p-value < 0.05. The results were presented as\n−log10(p-value). Youden’s J statistic was used to identify the optimal cutoff\nfor the prediction of either EM and/or myoma or a speciﬁc# E n z i a nc a t e g o r y\nby the selected biomarkers. These cutoffs represent the biomarker plasma\nconcentration at which the balan ce between sensitivity and speci ﬁcity is\nmaximized, ensuring robust discrimination between the studied classes.\nData availability\nThe data underlying this article are available in the article and in its online\nsupplementary material.\nReceived: 6 February 2025;Accepted: 27 August 2025;\nReferences\n1. Johnson, N. P. et al. World Endometriosis Society consensus on the\nclassiﬁcation of endometriosis. Hum. Reprod. 32, 315–324 (2017).\n2. Richter, M. et al. From donor to the lab: a fascinating journey of primary\ncell lines. Front Cell Dev. Biol. 9, 711381 (2021).\n3. Ramin-Wright, A. et al. Fatigue - a symptom in endometriosis. Hum.\nReprod. 33, 1459–1465 (2018).\n4 . I m p e r i a l e ,L . ,N i s o l l e ,M . ,N o ë l ,J .C .&F a s t r e z ,M .T h r e eT y p e s\nof Endometriosis: Pathogenesis, Diagnosis and Treatment. State\nof the Art. J. Clin. Med. 12. https://doi.org/10. 3390/jcm12030994\n(2023).\n5. Tan, Y. et al. Single-cell analysis of endometriosis reveals a\ncoordinated transcriptional programme driving immunotolerance and\nangiogenesis across eutopic and ectopic tissues (vol 24, pg 1306,\n2022). Nat. Cell Biol. 24, 1679–1679 (2022).\n6. Colgrave, E. M. et al. Extensive heterogeneity in the expression of\nsteroid receptors in super ﬁcial peritoneal endometriotic lesions.\nReprod. Biomed. Online 48, 103409 (2023).\n7. Bougie, O. et al. Impact of surgeon characteristics on endometriosis\nsurgery outcomes. J. Minim. Invasive Gynecol. 32, 709–717.e706\n(2025).\n8. Mu, F. et al. A Prospective study of in ﬂammatory markers and risk of\nEndometriosis. Am. J. Epidemiol. 187, 515–522 (2018).\n9. Anastasiu, C. V. et al. Biomarkers for the Noninvasive Diagnosis of\nEndometriosis: State of the Art and Future Perspectives. Int. J. Mol.\nSci. 21. https://doi.org/10.3390/ijms21051750 (2020).\n10. Vodolazkaia, A. et al. A high sensitivity assay is more accurate than a\nclassical assay for the measurement of plasma CRP levels in\nendometriosis. Reprod. Biol. Endocrinol. 9. https://doi.org/10.1186/\n1477-7827-9-113 (2011).\n11. Warzecha, D. et al. Plasma and Peritoneal Fluid Fibronectin and\nCollagen IV Levels as Potential Biomarkers of Endometriosis. Int. J.\nMol. Sci. 23, https://doi.org/10.3390/ijms232415669 (2022).\n12. Zhang, T. et al. MDSCs drive the process of endometriosis by\nenhancing angiogenesis and are a new potential therapeutic target.\nEur. J. Immunol. 48, 1059–1073 (2018).\n13. Ghafouri-Fard, S., Shoorei, H. & Taheri, M. Role of non-coding RNAs\nin the Pathogenesis of Endometriosis. Front. Oncol. 10, 1370 (2020).\n14. Gratton, S. M. et al. Diagnosis of Endometriosis at Laparoscopy: A\nvalidation study comparing surgeon visualization with histologic\nﬁndings. J. Obstet. Gynaecol. Ca 44, 135–141 (2022).\n15. Wang, Y. N. et al. Distinct subtypes of endometriosis identi ﬁed based\non stromal-immune microenvironment and gene expression:\nimplications for hormone therapy.Front. Immunol. 14. https://doi.org/\n10.3389/ﬁmmu.2023.1133672 (2023).\n16. Zakhari, A. et al. Endometriosis recurrence following post-operative\nhormonal suppression: a systematic review and meta-analysis. Hum.\nReprod. Update 27,9 6–107 (2021).\n17. Drechsel-Grau, A. et al. Long-term follow-up regarding pain relief,\nfertility, and re-operation after surgery for deep Endometriosis.J. Clin.\nMed. 13, 5039 (2024).\n18. Jaagura, M. et al. Comorbidities confound metabolomics studies of\nhuman disease. Sci. Rep. 14, 24810 (2024).\n19. Pusic, M. et al. Antibody arrays identi ﬁed cycle-dependent plasma\nbiomarker candidates of peritoneal Endometriosis. J. Pers. Med 12,\n852 (2022).\n20. Rocha, A. L., Vieira, E. L., Maia, L. M., Teixeira, A. L. & Reis, F. M.\nProspective evaluation of a panel of plasma cytokines and\nchemokines as potential markers of pelvic Endometriosis in\nsymptomatic women. Gynecol. Obstet. Invest. 81, 512–517 (2016).\n21. Mihalyi, A. et al. Non-invasive diagnosis of endometriosis based on a\ncombined analysis of six plasma biomarkers. Hum. Reprod. 25,\n654–664 (2010).\n22. Starodubtseva, N. et al. Identi ﬁcation of potential endometriosis\nbiomarkers in peritoneal ﬂuid and blood plasma via shotgun\nlipidomics. Clin. Mass Spectrom. 13,2 1–26 (2019).\n23. Vanhie, A. et al. Plasma miRNAs as biomarkers for endometriosis.\nHum. Reprod. 34, 1650–1660 (2019).\n24. Waelkens, E. et al. Discovery of Endometriosis biomarkers in\nmenstrual plasma samples using a proteomics approach.Fertil. Steril.\n106, E275–E275 (2016).\n25. Zafari, N. et al. A Panel of Plasma miRNAs 199b-3p, 224-5p, and Let-\n7d-3p as non-invasive diagnostic biomarkers for Endometriosis (Jan,\n10.1007/s43032-020-00415-z, 2021). Reprod. Sci. 28, 1000–1000\n(2021).\n26. Giuliani, E., As-Sanie, S. & Marsh, E. E. Epidemiology and\nmanagement of uterine ﬁbroids. Int. J. Gynaecol. Obstet. 149\n,3 –9\n(2020).\n27. Schoeman, E. M. et al. Identi ﬁcation of plasma protein biomarkers for\nendometriosis and the development of statistical models for disease\ndiagnosis. Hum. Reprod. 40, 270–279 (2025).\n28. Osinska, I., Popko, K. & Demkow, U. Perforin: an important player in\nimmune response. Cent. Eur. J. Immunol. 39, 109–115 (2014).\n29. Osuga, Y. et al. Lymphocytes in Endometriosis. Am. J. Reprod.\nImmunol. 65,1 –10 (2011).\n30. Male, V. et al. Immature NK cells, capable of producing IL-22, are\npresent in human uterine Mucosa. J. Immunol. 185, 3913–3918\n(2010).\n31. Konno, R. et al. Apoptosis of human endometrium mediated by\nperforin and granzyme B of NK cells and cytotoxic T lymphocytes.\nTohoku J. Exp. Med. 187, 149–155 (1999).\n32. Yang, H. L. et al. The crosstalk between endometrial stromal cells and\nmacrophages impairs cytotoxicity of NK cells in endometriosis by\nsecreting IL-10 and TGF- β. Reproduction 154, 815–825 (2017).\n33. Hassin, D., Garber, O. G., Meiraz, A., Schiffenbauer, Y. S. & Berke, G.\nCytotoxic T lymphocyte perforin and Fas ligand working in concert\neven when Fas ligand lytic action is still not detectable. Immunology\n133, 190–196 (2011).\n34. Gogacz, M. et al. Fas-related apoptosis of peritoneal ﬂuid\nmacrophages in Endometriosis patients: understanding the disease.\nJ. Immunol. Res. 2017. https://doi.org/10.1155/2017/3175394\n(2017).\n35. Garcia-Velasco, J. A., Arici, A., Zreik, T., Naftolin, F. & Mor, G.\nMacrophage derived growth factors modulate Fas ligand expression\nin cultured endometrial stromal cells: a role in endometriosis. Mol.\nHum. Reprod. 5, 642–650 (1999).\n36. Sturlese, E. et al. Dysregulation of the Fas/FasL system in\nmononuclear cells recovered from peritoneal ﬂuid of women with\nendometriosis. J. Reprod. Immunol. 92,7 4–81 (2011).\n37. Morales-Kastresana, A. et al. Essential complicity of perforin-\ngranzyme and FAS-L mechanisms to achieve tumor rejection\nfollowing treatment with anti-CD137 mAb.J. Immunother. Cancer 1,3\n(2013).\n38. Jeung, I., Cheon, K. & Kim, M. R. Decreased Cytotoxicity of peripheral\nand peritoneal natural killer cell in Endometriosis.Biomed. Res. Int.-Uk\n2016. https://doi.org/10.1155/2016/2916070 (2016).\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 11\n\n39. Harada, M. et al. Concentration of osteoprotegerin (OPG) in peritoneal\nﬂuid is increased in women with endometriosis. Hum. Reprod. 19,\n2188–2191 (2004).\n40. Bersinger, N. A. et al. PAPP-A and osteoprotegerin, together with\ninterleukin-8 and RANTES, are elevated in the peritoneal ﬂuid of\nwomen with endometriosis. Am. J. Obstet. Gynecol. 195, 103–108\n(2006).\n41. Shen, Z., Rodriguez-Garcia, M., Patel, M. V. & Wira, C. R. Direct and\nIndirect endocrine-mediated suppression of human endometrial\nCD8+T cell cytotoxicity. Sci. Rep. 11, 1773 (2021).\n42. Young, V. J., Ahmad, S. F., Duncan, W. C. & Horne, A. W. The role of\nTGF-β in the pathophysiology of peritoneal endometriosis. Hum.\nReprod. Update 23, 548–559 (2017).\n43. Chang, S. H. & Dong, C. IL-17F: Regulation, signaling and function in\ninﬂammation. Cytokine 46,7 –11 (2009).\n44. Sisnett, D. J. et al. The Dysregulated IL-23/TH17 Axis in Endometriosis\nPathophysiology. J. Immunol. 212, 1428–1441 (2024).\n45. Shi, J. L. et al. IL-17: an important pathogenic factor in endometriosis.\nInt. J. Med. Sci. 19, 769–778 (2022).\n46. Miller, J. E. et al. IL-17A modulates peritoneal macrophage\nrecruitment and M2 Polarization in Endometriosis.Front. Immunol. 11.\nhttps://doi.org/10.3389/ﬁmmu.2020.00108 (2020).\n47. Sikora, J., Smycz-Kubanska, M., Mielczarek-Palacz, A., Bednarek, I.\n& Kondera-Anasz, Z. The involvement of multifunctional TGF- β and\nrelated cytokines in pathogenesis of endometriosis. Immunol. Lett.\n201,3 1–37 (2018).\n48. Soni, U. K. et al. A high level of TGF-B1 promotes endometriosis\ndevelopment via cell migration, adhesiveness, colonization, and\ninvasiveness. Biol. Reprod. 100, 917–938 (2019).\n49. Kitaya, K. et al. Expression of macrophage in ﬂammatory protein-1β in\nhuman endometrium:: Its role in endometrial recruitment of natural\nkiller cells. J. Clin. Endocrinol. Metab. 88, 1809–1814 (2003).\n50. Sadlocha, M., Toczek, J., Czuba, Z. & Stojko, R. Effects of\nmacrophage inﬂammatory Protein 1 Alpha and Beta Chemokine\nconcentrations in the progression of Endometriosis.Clin. Exp. Obstet.\nGyn. 49. https://doi.org/10.31083/j.ceog4907153 (2022).\n51. Aksak, T., Gümürdülü, D., Çetin, M. T. & Polat, S. Expression of\nmonocyte chemotactic protein 2 and tumor necrosis factor alpha in\nhuman normal endometrium and endometriotic tissues. J. Gynecol.\nObstet. Hum. 50, 101971 (2021).\n5 2 . Y o u n g ,V .J . ,A h m a d ,S .F . ,B r o w n ,J .K . ,D u n c a n ,W .C .&H o r n e ,A .\nW. Peritoneal VEGF-A expression is regulated by TGF- β1 through\nan ID1 pathway in women with endometriosis. Sci. Rep. 5,1 6 8 5 9\n(2015).\n53. Danastas, K., Miller, E. J., Hey-Cunningham, A. J., Murphy, C. R. &\nLindsay, L. A. Expression of vascular endothelial growth factor A\nisoforms is dysregulated in women with endometriosis. Reprod. Fert.\nDev. 30, 651–657 (2018).\n54. Li, H. et al. FGA controls VEGFA secretion to promote Angiogenesis\nby activating the VEGFR2-FAK signalling pathway. Front. Endocrinol.\n13. https://doi.org/10.3389/fendo.2022.791860 (2022).\n55. Zubrzycka, A., Migdalska-Sek, M., Jedrzejczyk, S. & Brzezianska-\nLasota, E. Circulating miRNAs related to Epithelial-Mesenchymal\nTransitions (EMT) as the new molecular markers in Endometriosis.\nCurr. Issues Mol. Biol. 43, 900–916 (2021).\n56. Matsumoto, H. et al. Regulation of proliferation, motility, and\ncontractility of human endometrial stromal cells by platelet-derived\ngrowth factor. J. Clin. Endocrinol. Metab. 90, 3560–3567 (2005).\n57. Keckstein, J. et al. The #Enzian classi ﬁcation: A comprehensive non-\ninvasive and surgical description system for endometriosis. Acta\nObstet. Gynecol. Scand. 100, 1165–1175 (2021).\n58. Forschung, S. E. Vol. IOS, Android (ed Stiftung Endometriose\nForschung) (App Store).\n59. Harth, S. et al. Application of the #Enzian classi ﬁcation for\nendometriosis on MRI: prospective evaluation of inter- and\nintraobserver agreement. Front Med.10, 1303593 (2023).\n60. Vercellini, P. et al. Reproductive performance, pain recurrence and\ndisease relapse after conservative surgical treatment for\nendometriosis: the predictive value of the current classi ﬁcation\nsystem. Hum. Reprod. 21, 2679–2685 (2006).\n61. Vercellini, P. et al. Endometriosis and pelvic pain: relation to disease\nstage and localization. Fertil. Steril. 65, 299–304 (1996).\n62. Colgan, R. E., Gutierrez, D. E., Sundram, J., Bhaskar, T. G. Analysis of\nmedical data using dimensionality reduction techniques.Proceedings\nof the Conference: AMALTHEA–2013. Melbourne, https://doi.org/10.\n13140/2.1.2270.1762 (2013).\n63. Lou, Z. et al. Global, regional, and national time trends in incidence,\nprevalence, years lived with disability for uterine ﬁ\nbroids, 1990-2019:\nan age-period-cohort analysis for the global burden of disease\n2019 study. BMC Public Health 23, 916 (2023).\n64. Wehbe, E. et al. Therapeutic Drug Monitoring and Biomarkers:\ntowards Better Dosing of Antimicrobial Therapy. Pharmaceutics 16.\nhttps://doi.org/10.3390/pharmaceutics16050677 (2024).\n65. McKinney, W. Data Structures for Statistical Computing in Python.\nProceedings of the Python in Science Conference,5 6–61. https://doi.\norg/10.25080/majora-92bf1922-00a (2010).\n66. Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach.\nLearn. Res. 12, 2825–2830 (2011).\n67. Virtanen, P. et al. SciPy 1.0: fundamental algorithms for scienti ﬁc\ncomputing in Python. Nat. Methods 17, 261–272 (2020).\n68. mwaskom/seaborn: v0.8.1 (2017).\n69. Kwak, S. K. & Kim, J. H. Statistical data preparation: management\nof missing values and outliers. Korean J. Anesthesiol. 70,4 0 7–411\n(2017).\nAcknowledgements\nThis work was supported by an Innosuisse grant (grant number 44311.1 IP-\nLS). The funder had no role in study design, data collection, data analysis,\ndata interpretation, or writing of the manuscript. We would like to thank the\npatients and all clinical personnel of the University Hospital Zurich involved in\nthis research.\nAuthor contributions\nD.R.G., B.L., and V.V. were involved in the study design and\nconceptualization. D.R.G., M.S., A.A., and L.B. performed the statistical\nanalysis. All authors were involved in data interpretation. M.H., I.W., P.I., and\nJ.M. collected human samples and clinical data. D.R.G. drafted the original\nmanuscript. All authors contributed to the writing of the manuscript, made\ncritical comments, and approved the ﬁnal version.\nCompeting interests\nThe authors declare no competing interests.\nAdditional information\nSupplementary informationThe online version contains\nsupplementary material available at\nhttps://doi.org/10.1038/s44294-025-00099-3\n.\nCorrespondenceand requests for materials should be addressed to\nBrigitte Leeners.\nReprints and permissions informationis available at\nhttp://www.nature.com/reprints\nPublisher’s note Springer Nature remains neutral with regard to\njurisdictional claims in published maps and institutional afﬁliations.\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 12\n\nOpen Access This article is licensed under a Creative Commons\nAttribution-NonCommercial-NoDerivatives 4.0 International License,\nwhich permits any non-commercial use, sharing, distribution and\nreproduction in any medium or format, as long as you give appropriate\ncredit to the original author(s) and the source, provide a link to the Creative\nCommons licence, and indicate if you modi ﬁed the licensed material. You\ndo not have permission under this licence to share adapted material\nderived from this article or parts of it. The images or other third party\nmaterial in this article are included in the article ’s Creative Commons\nlicence, unless indicated otherwise in a credit line to the material. If material\nis not included in the article’s Creative Commons licence and your intended\nuse is not permitted by statutory regulation or exceeds the permitted use,\nyou will need to obtain permission directly from the copyright holder. To\nview a copy of this licence, visit http://creativecommons.org/licenses/by-\nnc-nd/4.0/\n.\n© The Author(s) 2025\nhttps://doi.org/10.1038/s44294-025-00099-3 Article\nnpj Women's Health |            (2025) 3:60 13","source_license":"CC0","license_restricted":false}