Compartment-specific inflammation-metabolism remodeling across eutopic and ectopic endometriosis tissues with cross-context evaluation in TCGA-UCEC

Juan Yang: Writing – original draft, Methodology, Data curation, Conceptualization. Shumin Wu: Resources, Formal analysis. Aimin Pu: Supervision, Methodology. Sujuan Xie: Resources. Mi Hu: Writing – review & editing, Supervision, Methodology.

This research received no external funding.

The prespecified single-cell cohort contained 14 identified donors and 24 primary-tissue libraries: Ctrl 3, EuE 9, EcP 8, and EcO 4. Post-QC gene detection is summarized by library in Fig. 1 A, and the UMI-gene relation is shown in Fig. 1B The mitochondrial fraction remained below the prespecified 20% library-level median threshold across the retained libraries ( Fig. 1C ). Harmony-integrated embeddings display tissue contexts in Fig. 1D and broad marker-score compartments in Fig. 1E . The included-library tally and balanced analyzed-cell counts are shown in Fig. 1F ; this panel records the actual balanced display cohort rather than implying equal raw-cell yields. Fig. 1 Quality control, integration, and included primary-tissue libraries. A-C, post-QC library complexity and mitochondrial fraction. D, Harmony-integrated embedding by tissue context. E, the same embedding by broad marker-score compartment. F, included library counts and balanced analyzed-cell counts. The single-cell analysis includes 24 primary-tissue libraries and 21,600 analyzed cells after QC and balanced sampling. Quality control, integration, and included primary-tissue libraries. A-C, post-QC library complexity and mitochondrial fraction. D, Harmony-integrated embedding by tissue context. E, the same embedding by broad marker-score compartment. F, included library counts and balanced analyzed-cell counts. The single-cell analysis includes 24 primary-tissue libraries and 21,600 analyzed cells after QC and balanced sampling. Unsupervised PC-space clusters are shown independently of the broad marker-score labels in Fig. 2A . Pooled retained-cell composition is shown in Fig. 2B Canonical marker-expression support is displayed in Fig. 2C , and library-level broad-compartment fractions are displayed in Fig. 2D Inflammation is mapped onto the embedding in Fig. 2E . The M2-like marker track is restricted to broad myeloid cells in Fig. 2F and is not treated as a validated macrophage-polarization classifier. Fig. 2 Broad compartments and inflammatory score localization. A, unsupervised PC-space clusters. B, pooled retained-cell composition. C, canonical marker-expression support. D, library-level compartment fractions. E, inflammation score on the Harmony-integrated embedding. F, M2-like marker-track distribution restricted to broad myeloid cells. Panels B, C, and E-F are descriptive cell-level displays; panel D uses libraries as the sampling unit. Broad compartments and inflammatory score localization. A, unsupervised PC-space clusters. B, pooled retained-cell composition. C, canonical marker-expression support. D, library-level compartment fractions. E, inflammation score on the Harmony-integrated embedding. F, M2-like marker-track distribution restricted to broad myeloid cells. Panels B, C, and E-F are descriptive cell-level displays; panel D uses libraries as the sampling unit. Cell-level inflammatory-program distributions are descriptive in Fig. 3A , while the context-by-compartment matrix in Fig. 3B shows heterogeneity across tissue contexts. Selected inflammatory transcript abundance is shown in Fig. 3C . The inflammation-chemokine relationship in Fig. 3D is a cell-level display and does not treat cells as independent biological replicates. Library-level interferon and NFkB summaries are shown in Fig. 3E . Broad-compartment mean inflammation is shown in Fig. 3F , supporting heterogeneous activity rather than a uniform lesion-wide increase. Fig. 3 Inflammatory-program displays across tissue contexts. A, cell-level inflammatory-program distribution. B, context-by-compartment inflammatory matrix. C, selected inflammatory transcript abundance. D, cell-level inflammation-chemokine relationship. E, library-level interferon and NFkB summaries. F, broad-compartment mean inflammation. Cell-level panels are descriptive and do not treat cells as independent patients. Inflammatory-program displays across tissue contexts. A, cell-level inflammatory-program distribution. B, context-by-compartment inflammatory matrix. C, selected inflammatory transcript abundance. D, cell-level inflammation-chemokine relationship. E, library-level interferon and NFkB summaries. F, broad-compartment mean inflammation. Cell-level panels are descriptive and do not treat cells as independent patients. Cell-level glycolytic-program distributions are shown in Fig. 4A . The context-by-compartment metabolic matrix in Fig. 4B demonstrates that metabolic tracks vary with both tissue context and broad compartment. Iron-redox score localization is shown in Fig. 4C . Bulk program shifts are summarized independently at the biopsy-library level in Fig. 4D The glycolysis-OxPhos relationship in Fig. 4E is descriptive at cell level. The bulk fingerprint in Fig. 4F shows that control biopsies had the highest mean standardized glycolysis score (1.05), compared with −0.58 in EcP and −0.83 in EcO. Fig. 4 Metabolic-program displays and bulk context summaries. A, cell-level glycolytic-program distribution. B, context-by-compartment metabolic matrix. C, iron-redox program on the embedding. D, bulk program-score shifts. E, descriptive cell-level glycolysis-OxPhos relationship. F, bulk inflammation-metabolism fingerprint. Bulk panels use GSE179640 biopsies as the biological unit. Metabolic-program displays and bulk context summaries. A, cell-level glycolytic-program distribution. B, context-by-compartment metabolic matrix. C, iron-redox program on the embedding. D, bulk program-score shifts. E, descriptive cell-level glycolysis-OxPhos relationship. F, bulk inflammation-metabolism fingerprint. Bulk panels use GSE179640 biopsies as the biological unit. Within myeloid library summaries, inflammation and glycolysis correlated (rho = 0.52, P = 0.009; Fig. 5A ). The M2-like and iron-redox association did not meet the nominal threshold (rho = 0.40, P = 0.054; Fig. 5B ). The myeloid library program landscape is shown in Fig. 5C . The context-wise M2-like score summary is shown in Fig. 5D The myeloid iron-redox score summary is shown in Fig. 5E . In contrast, the bulk-wide inflammation-glycolysis association in Fig. 5F was weak (rho = 0.21, P = 0.320), supporting a compartment-specific rather than universal coupling interpretation. Fig. 5 Myeloid-restricted and bulk-wide inflammation-metabolism relationships. A, myeloid library inflammation-glycolysis relationship. B, myeloid library M2-like and iron-redox relationship. C, myeloid library program landscape. d -E, context-wise myeloid M2-like and iron-redox summaries. F, bulk-wide inflammation-glycolysis relationship. The M2-like track is a predefined marker score shown only in myeloid cells, not a validated polarization assignment. Myeloid-restricted and bulk-wide inflammation-metabolism relationships. A, myeloid library inflammation-glycolysis relationship. B, myeloid library M2-like and iron-redox relationship. C, myeloid library program landscape. d -E, context-wise myeloid M2-like and iron-redox summaries. F, bulk-wide inflammation-glycolysis relationship. The M2-like track is a predefined marker score shown only in myeloid cells, not a validated polarization assignment. The 24-library bulk transcriptome PCA is shown in Fig. 6A . DESeq2 identified 31 significant genes for EuE versus Ctrl in Fig. 6B ; this limited yield is reported directly and is not reframed as a validated disease shift. The peritoneal lesion contrast identified 4480 significant genes in Fig. 6C . The ovarian lesion contrast identified 4632 significant genes in Fig. 6D Leading contrast genes are displayed in Fig. 6E . Bulk program-score shifts are shown in Fig. 6F . Fig. 6 Count-aware bulk RNA-seq evidence. A, PCA of the 24 GSE179640 bulk libraries. B-D, DESeq2 volcano plots for EuE, EcP, and EcO versus Ctrl. E, leading contrast genes across the bulk cohort. F, bulk program-score shifts. Significant genes meet FDR < 0.05 and absolute log2 fold change ≥ 0.5. Count-aware bulk RNA-seq evidence. A, PCA of the 24 GSE179640 bulk libraries. B-D, DESeq2 volcano plots for EuE, EcP, and EcO versus Ctrl. E, leading contrast genes across the bulk cohort. F, bulk program-score shifts. Significant genes meet FDR < 0.05 and absolute log2 fold change ≥ 0.5. The TCGA-UCEC primary-tumor score space is displayed by reported histology in Fig. 7A . Paired tumor-normal program effects in 23 patients are summarized in Fig. 7B , including glycolysis (1.92, BH q = 5.77e-5) and OxPhos (0.63, BH q = 1.75e-6). Paired differential expression is shown in Fig. 7C . Glycolysis differs across the reported grade groups descriptively in Fig. 7D (Kruskal-Wallis P = 8.52e-5). The exploratory composite does not separate overall survival in Fig. 7E (log-rank P = 0.969). Its univariable Cox estimate in Fig. 7F is null (HR = 1.00, 95% CI 0.82–1.22, P = 0.968). Fig. 7 Independent TCGA-UCEC cancer-context analysis. A, primary-tumor score space. B, paired tumor-normal program effects in 23 patients; red bars meet BH q < 0.05. C, paired tumor-normal DESeq2 volcano. D, descriptive glycolysis score by reported grade. E, exploratory composite survival curves for 541 patients with 90 deaths. F, univariable Cox estimate, which is null. TCGA-UCEC is not a longitudinal or mechanistic validation cohort for endometriosis. Independent TCGA-UCEC cancer-context analysis. A, primary-tumor score space. B, paired tumor-normal program effects in 23 patients; red bars meet BH q < 0.05. C, paired tumor-normal DESeq2 volcano. D, descriptive glycolysis score by reported grade. E, exploratory composite survival curves for 541 patients with 90 deaths. F, univariable Cox estimate, which is null. TCGA-UCEC is not a longitudinal or mechanistic validation cohort for endometriosis. Standardized program effects are compared without claiming cohort equivalence in Fig. 8A . Count-aware DEG yields are reported in Fig. 8B The observed GSE179640 bulk gene-correlation network is shown in Fig. 8C . Leave-one-library-out glycolysis effects are shown in Fig. 8D Within-donor paired effects for the five comparison programs are shown descriptively in Fig. 8E (EcP-EuE n = 3, EcO-EuE n = 4, and EcO-EcP n = 5); these small paired sets are not used for confirmatory claims. The five-program comparison in Fig. 8F was not correlated across endometriosis lesion-control and TCGA tumor-normal contrasts (n = 5, rho = −0.30, P = 0.624; direction concordant 2/5), reinforcing that the cancer cohort should not be presented as a direct validation of an endometriosis mechanism. Fig. 8 Cross-context audit and sensitivity analysis. A, standardized program-effect audit across GSE179640 and TCGA-UCEC. B, count-aware DEG yields. C, observed GSE179640 bulk gene-correlation network. D, leave-one-library-out lesion-control glycolysis effects. E, within-donor paired program effects in GSE179640 bulk biopsies (EcP-EuE n = 3, EcO-EuE n = 4, and EcO-EcP n = 5); these small paired sets are descriptive. F, five-program effect comparison across independent cohorts. Panels A and F are descriptive cross-context comparisons and do not imply causal continuity. Cross-context audit and sensitivity analysis. A, standardized program-effect audit across GSE179640 and TCGA-UCEC. B, count-aware DEG yields. C, observed GSE179640 bulk gene-correlation network. D, leave-one-library-out lesion-control glycolysis effects. E, within-donor paired program effects in GSE179640 bulk biopsies (EcP-EuE n = 3, EcO-EuE n = 4, and EcO-EcP n = 5); these small paired sets are descriptive. F, five-program effect comparison across independent cohorts. Panels A and F are descriptive cross-context comparisons and do not imply causal continuity. To validate key stem cell markers identified through scRNA-seq, flow cytometry was performed on endometriotic tissue samples using a sequential gating strategy ( Figure 9 ). Lymphocytes comprised 89.1% of total events ( Figure 9A ), with 99.6% confirmed as singlets to ensure single-cell accuracy ( Figure 9B ). Viability assessment revealed 86.4% viable cells ( Figure 9C ), and CD45 expression analysis identified 42.0% immune cells within the viable population, consistent with the inflammatory nature of endometriosis ( Figure 9D ). Most importantly, 4.73% of CD45⁻ non-immune cells expressed CD24 at the protein level, directly validating its role as a stem cell marker identified in scRNA-seq ( Figure 9E ). Across independent biological samples, the proportion of CD24⁺ cells within the CD45⁻ compartment was higher in endometriotic tissue (n = 3, 4.73 ± 0.85%, mean ± SD) than in control endometrium (n = 3, 1.92 ± 0.42%), with a statistically significant between-group difference (P < 0.01, unpaired two-tailed Student’s t -test), indicating expansion of the stem cell compartment in disease. These findings provide independent protein-level confirmation of our transcriptomic results, supporting CD24 as a clinically relevant stem cell biomarker for endometriosis. Fig. 9 Flow cytometry validation of stem cell markers. A, FSC-A vs SSC-A gating identifying lymphocyte population (89.1%). B, FSC-A vs FSC—H doublet discrimination showing singlets (99.6%). C, Zombie BV510 viability gating showing viable cells (86.4%). D, CD45 expression analysis identifying immune cells (42.0% CD45⁺). E, CD24 expression within CD45⁻ cells showing stem cell population (4.73% CD24⁺). F, Alternative visualization of CD24 expression distribution. Flow cytometry validation of stem cell markers. A, FSC-A vs SSC-A gating identifying lymphocyte population (89.1%). B, FSC-A vs FSC—H doublet discrimination showing singlets (99.6%). C, Zombie BV510 viability gating showing viable cells (86.4%). D, CD45 expression analysis identifying immune cells (42.0% CD45⁺). E, CD24 expression within CD45⁻ cells showing stem cell population (4.73% CD24⁺). F, Alternative visualization of CD24 expression distribution.

The single-cell and bulk endometriosis data were obtained from GSE179640 . The single-cell primary-tissue analysis included Ctrl, EuE, EcP, and EcO matrices. Patient-derived organoids (EOR) and ectopic-adjacent tissues (EcPA/EcPA2) were excluded from the prespecified primary-tissue comparison. The bulk analysis included all 24 available Ctrl, EuE, EcP, and EcO biopsies. TCGA-UCEC count and clinical data were used as a separate cancer context with one primary tumor retained per patient, a matched tumor-normal subset, and a patient-unique survival subset. Quality control required 1000–6000 detected genes, 5000–40,000 UMI counts, and mitochondrial fraction <20%, applied before balanced sampling. Each library was capped at 900 retained cells only after filtering, preventing a high-yield library from dominating visual displays. Highly variable genes were used for principal components and Harmony on library identity. Broad compartments were assigned by predefined marker-score maxima and are reported as descriptive marker-score compartments, not as reference-validated cell identities. Each predefined program score was the mean log1p normalized expression across its listed genes. M2-like scoring was displayed and interpreted only within the broad myeloid compartment. For single-cell between-context checks, library summaries rather than individual cells were used as the independent unit. Bulk differential expression was performed with DESeq2 from raw featureCounts, with significance defined as FDR < 0.05 and absolute log2 fold change ≥ 0.5. Program comparisons are exploratory and no clinical threshold was trained. For the matched subset, tumor-normal differences were tested within 23 patient pairs. The cancer grade panel is descriptive. Survival analyses used 541 patient-unique records and 90 deaths; the high/low split is an exploratory median split of a composite expression score. A log-rank test and an univariable Cox model are displayed so that null outcome evidence is visible rather than omitted. Protein-level validation was performed using flow cytometry on freshly prepared single-cell suspensions. Cells were stained with fluorochrome-conjugated antibodies: Zombie BV510 (viability), CD45-FITC (immune marker), and CD24-PE (stem cell marker). Data acquisition used BD FACSAria III flow cytometer with 405 nm, 488 nm, and 640 nm lasers, collecting ≥100,000 events per sample. Sequential gating strategies identified: (1) lymphocyte populations (FSC-A vs SSC-A), (2) singlets (FSC-A vs FSC—H doublet discrimination), (3) viable cells (Zombie BV510-negative), (4) CD45+ immune and CD45- non-immune populations, and (5) CD24+ stem cells within CD45- fraction. FlowJo software (v10.8.1) analyzed data with FMO controls establishing positive gating thresholds. Flow cytometry was performed using independent tissue specimens from n = 3 patients with endometriosis and n = 3 control participants. Each patient contributed one biologically independent sample, and technical acquisitions from the same specimen were not treated as independent replicates. Data are presented as mean ± SD. Comparisons between the two groups were performed using an unpaired two-tailed Student’s t -test, with P < 0.05 considered statistically significant.

By integrating primary-tissue single-cell and bulk transcriptomic analyses with an independent TCGA-UCEC comparison and flow-cytometry validation, this study demonstrates that inflammation-metabolism remodeling in endometriosis is compartment and tissue-context specific. The myeloid inflammation-glycolysis association was not reproduced as a bulk-wide relationship, and ovarian and peritoneal lesions showed substantially larger count-aware transcriptomic differences from control than eutopic endometrium. TCGA-UCEC displayed distinct tumor-normal metabolic effects but did not provide cross-context or survival evidence that validates an endometriosis progression mechanism. These results favor context-aware, sample-level interpretation of inflammatory and metabolic programs and define testable hypotheses for future functional and independent-cohort studies.

This reanalysis shows that inflammatory and metabolic remodeling in endometriosis is strongly dependent on both tissue context and broad cellular compartment. The most reproducible coupling observed in the single-cell data was a myeloid-restricted association between inflammation and glycolysis, whereas the same relationship was weak across bulk biopsies. Bulk DESeq2 further showed markedly larger transcriptomic shifts in peritoneal and ovarian lesions than in eutopic endometrium. Importantly, the independent TCGA-UCEC analyses did not reproduce a coherent five-program endometriosis pattern and provided null exploratory survival evidence. Together, these findings support compartment-specific remodeling rather than a single lesion-wide inflammation-metabolism axis or an endometriosis-to-cancer molecular continuum. The single-cell results indicate that inflammatory-program activity differs across tissue contexts and broad cellular compartments rather than increasing uniformly in ectopic lesions. This interpretation is consistent with the cellular diversity described in the GSE179640 atlas and with broader evidence that endometriosis-associated inflammation reflects interactions among epithelial, stromal, vascular, and immune populations [ 4 – 6 ]. Accordingly, the cell-level distributions in this study are treated as descriptive, while between-context inference relies on library-level summaries rather than individual cells as independent biological replicates. The myeloid findings should also be interpreted at the level at which they were defined. The M2-like track is a predefined marker score restricted to the broad myeloid compartment and is not a validated macrophage-polarization classifier. Macrophage phenotypes in endometriosis are heterogeneous and can depend on cellular origin and tissue environment [ 7 ]. Thus, the present data support context-dependent myeloid remodeling but do not justify a binary or universal M1/M2 assignment across lesions. Metabolic program scores also varied by tissue context and compartment. The positive library-level association between myeloid inflammation and glycolysis (rho = 0.52, P = 0.009) supports a compartment-specific inflammation-metabolism relationship. In contrast, the M2-like/iron-redox association did not meet the nominal threshold (rho = 0.40, P = 0.054), and the bulk-wide inflammation-glycolysis relationship was weak (rho = 0.21, P = 0.320). These contrasts argue against treating inflammation and glycolysis as a universally coupled lesion-wide axis. The bulk fingerprint further showed that control biopsies had the highest mean standardized glycolysis score, whereas EcP and EcO were lower. This result should not be interpreted as direct evidence of metabolic flux: transcript-based gene-set scores summarize expression programs and may be influenced by tissue composition, cell-state abundance, and pathway regulation. Prior work supports altered glycolytic signaling and iron-associated redox stress in the endometriosis microenvironment [ 8 , 9 ], but functional assays of substrate use, lactate production, mitochondrial respiration, and redox state will be needed to establish the biochemical consequences of the transcriptomic patterns observed here. Count-aware bulk analysis revealed a limited EuE-versus-Ctrl differential-expression yield (31 significant genes) but much larger lesion-control contrasts for EcP (4480 genes) and EcO (4632 genes). These differences reinforce the importance of anatomical context and support reporting eutopic and ectopic tissues separately rather than treating all endometriosis-associated specimens as one molecular state. TCGA-UCEC was included only as an independent cancer context [ 12 ]. Paired tumor-normal comparisons showed significant glycolysis and OxPhos effects, and glycolysis differed across reported grade groups descriptively. However, the five-program effects were not correlated across endometriosis lesion-control and TCGA tumor-normal contrasts (rho = −0.30, P = 0.624; direction concordant 2/5), and the exploratory survival composite was null (log-rank P = 0.969; HR = 1.00, 95% CI 0.82–1.22, P = 0.968). These results do not support presenting TCGA-UCEC as validation of an endometriosis mechanism, a molecular progression sequence, or a prognostic extension of the endometriosis analysis. Flow cytometry identified a higher CD24-positive fraction within the CD45-negative compartment in endometriotic tissue than in control endometrium (4.73% versus 1.92%, P < 0.01), providing protein-level confirmation of differential CD24 surface expression. The biological interpretation should nevertheless remain specific. Endometrial stem/progenitor populations are heterogeneous and require careful phenotypic and functional definition [ 13 ]. CD24 itself is expressed in multiple epithelial, hematopoietic, and other cellular contexts and has biological functions extending well beyond stemness [ 14 ]. Therefore, the present flow-cytometry result validates a CD24-positive CD45-negative population; it should not by itself be equated with a universally defined stem-cell fraction. The finalized analysis is tissue based and does not establish a diagnostic classifier or a minimally invasive clinical test. No blood-, urine-, or menstrual-effluent assay was evaluated, and no externally validated AUC is reported in the current Results. Consequently, potential biomarker applications should be viewed as future hypotheses rather than demonstrated clinical utility. Any translation toward diagnosis will require independent prospective cohorts, prespecified thresholds, appropriate comparators, and evaluation against contemporary guideline-based diagnostic pathways [ 15 ]. Prior biomarker reviews likewise emphasize the need for rigorous validation before candidate molecular markers can be used clinically [ 16 ]. Several limitations define the scope of inference. First, the primary single-cell analysis is based on a modest number of donors and unequal numbers of libraries across tissue contexts. Harmony integration on library identity addresses an important technical source of variation, but residual biological and technical heterogeneity cannot be excluded. Second, broad compartments are marker-score assignments rather than reference-validated cell identities, and the M2-like track is descriptive rather than a validated polarization classifier. Third, program scores are exploratory transcriptional summaries and do not directly measure metabolic flux or establish clinical thresholds. Fourth, many cell-level panels are descriptive; the library, not the individual cell, is the appropriate independent unit for between-context comparisons. These constraints are consistent with broader endometriosis reviews emphasizing mechanistic and phenotypic heterogeneity [ 17 ]. Endometriosis can also have systemic manifestations beyond pelvic lesions, which further cautions against extrapolating a tissue-specific transcriptional signal to a universal disease state [ 18 ]. Established pathophysiological frameworks likewise support multiple interacting estrogen-dependent, inflammatory, and tissue-remodeling mechanisms rather than a single linear pathway [ 19 ]. For future scRNA-seq studies, best-practice workflows recommend explicit quality control, normalization, data integration, annotation, and replicate-aware inference [ 20 ]. Differential-expression analyses that ignore biological replicate structure can generate false discoveries [ 21 ], while multi-sample differential-state or pseudobulk frameworks provide a principled route for sample-level inference [ 22 ]. Benchmarking studies further show that batch-correction method choice can materially influence the balance between removing technical variation and preserving biological structure [ 23 ]. Additional limitations concern cross-cohort interpretation and validation. The within-donor paired bulk subsets are small and are therefore descriptive. TCGA-UCEC represents a biologically distinct cancer cohort and cannot establish longitudinal progression from endometriosis, shared susceptibility, or causal continuity. The flow-cytometry panel is limited to viability, CD45, and CD24 and therefore cannot resolve the full phenotype or function of the CD24-positive population. Future work should prioritize independent endometriosis cohorts and spatially resolved profiling; spatial transcriptomics can retain tissue architecture while localizing gene-expression programs [ 24 ]. Orthogonal metabolomics and stable-isotope tracing can determine whether transcript-based pathway scores correspond to altered metabolite abundance or biochemical flux [ 25 ]. Translational biomarker development should remain cautious because systematic-review evidence has historically found no peripheral marker sufficiently validated for routine clinical use [ 26 ]. Finally, studies identifying endometrial stem/progenitor populations in menstrual blood and peritoneal fluid reinforce the need for phenotype-specific and functional validation when interpreting marker-positive populations [ 27 ].

Endometriosis is a chronic, estrogen-responsive inflammatory disorder characterized by endometrial-like tissue outside the uterus and substantial heterogeneity in symptoms and lesion location [ 1 ]. Contemporary reviews emphasize that the disease is biologically heterogeneous and involves dynamic interactions among endocrine, immune, stromal, epithelial, vascular, and neural processes [ 2 ]. Molecular and endocrine studies further support estrogen-dependent inflammatory signaling and lesion-specific cellular programs [ 3 ]. The GSE179640 single-cell atlas demonstrated marked differences among control endometrium, eutopic endometrium, peritoneal lesions, and ovarian lesions, providing the principal transcriptomic resource reanalyzed here [ 4 ]. Immune dysregulation is a major component of the endometriosis microenvironment [ 5 ]. Detailed immunological reviews further describe altered macrophage, lymphocyte, and innate immune functions across lesions and peritoneal compartments [ 6 ]. Experimental work demonstrates that macrophage effects can differ according to cellular origin, underscoring the limitations of a simple uniform polarization model [ 7 ]. In parallel, metabolic adaptation has been linked to endometriosis-associated peritoneal remodeling, including TGF-beta1/ID2-related Warburg-like glycolytic changes [ 8 ]. Repeated hemorrhage and iron accumulation provide an additional source of oxidative and redox stress in ectopic sites [ 9 ]. These observations motivate testing whether inflammation-metabolism coupling is lesion-wide or restricted to particular cellular compartments. Single-cell RNA sequencing provides a means to localize transcriptional programs to broad cell compartments, whereas bulk RNA sequencing retains biopsy-level statistical power and facilitates count-aware differential expression. For multi-library scRNA-seq datasets, batch-aware integration is an important analytical safeguard; Harmony provides an established framework for reducing dataset-associated technical variation in a shared embedding [ 10 ]. At the biopsy level, count-aware RNA-seq differential expression can be performed with DESeq2, which models dispersion and fold-change estimates using biological replicates [ 11 ]. Together, single-cell localization and bulk sample-level inference permit separation of descriptive cellular patterns from biopsy-level statistical evidence. Endometrial carcinoma was not treated as a longitudinal extension of endometriosis in the present study. Instead, TCGA-UCEC was used as an independent cancer context to ask whether selected inflammatory and metabolic program directions observed in endometriosis were also present in tumor-normal comparisons. The integrated genomic characterization of endometrial carcinoma provides a well-defined external tumor context for this purpose [ 12 ]. Because the cohorts differ in disease biology, sampling, and study design, cross-context similarity cannot establish progression, common cellular ancestry, or mechanistic validation. Accordingly, we reanalyzed primary-tissue scRNA-seq and bulk RNA-seq data from GSE179640 to define broad cellular compartments, map inflammatory and metabolic programs, test myeloid-restricted inflammation-metabolism associations at the library level, and quantify count-aware lesion-control differential expression. We then evaluated the same program framework in TCGA-UCEC as a separate cancer context and used flow cytometry to assess CD45 and CD24 protein expression in endometriotic tissue. The analysis was designed to preserve the boundary between descriptive cellular patterns, biopsy-level statistical inference, and cross-cohort comparison.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

GSE179640 is publicly available from the NCBI Gene Expression Omnibus (GEO) and was used for the endometriosis single-cell and bulk RNA-seq analyses. TCGA-UCEC RNA-count and clinical data were obtained through the Genomic Data Commons (GDC). Processed analysis outputs and code used for the present reanalysis are available from the corresponding author upon reasonable request.