Inflammatory and Molecular Mechanisms of Adenomyosis Associated Pain: Insights from Multiple Analytic Approaches

Adenomyosis is a common gynecologic condition, affecting approximately 30% of women of reproductive age. 1 It is generally classified as a benign, hormone-dependent disease of the uterus, primarily driven by estrogen. Clinically, affected women present with diverse symptoms, including chronic pelvic pain, dyspareunia, cyclic pain, abnormal uterine bleeding such as menorrhagia, and infertility. Traditionally, diagnosis was confirmed through histological identification of endometrial glands and stroma within the myometrium, either scattered or diffusely distributed. 2 However, advancements and availability in diagnostic imaging technologies, particularly ultrasound and magnetic resonance imaging (MRI), have shifted the diagnostic paradigm. These modalities allowed non-invasive detection of adenomyosis based on typical characteristics of uterus such as a globular and enlarged uterus, asymmetric myometrial thickening, a thickened or disrupted junctional zone, and focal areas of altered echogenicity or signal intensity within the myometrium. 3 The improved access to developed imaging tools has further contributed to earlier diagnosis, often identifying adenomyosis in asymptomatic women or those with mild symptoms. As a result, an increasing number of cases are detected before significant clinical manifestations develop and allow early medical interventions. In addition to advancements in diagnostic modalities, molecular studies have expanded our understanding of the pathogenesis of adenomyosis, which suggested possible therapeutic targets. Key molecular mechanisms implicated include fibrosis, decidualization, angiogenesis, and inflammation. These findings have been followingly integrated into two prevailing theories. The first theory attributes the condition to the penetration of the endometrium into the myometrium through a tissue injury and repair (TIAR) mechanism. The second suggests that adenomyosis originates from the metaplasia or differentiation of stem cells. Despite this progress, the connection between these molecular processes and various clinical presentations remains insufficiently understood. 2 Through understanding the different clinical symptoms and severity associated with adenomyosis from previous studies, it is well established that adenomyosis is at least partially involved in the innate immune system, as evidenced by altered expression of toll-like receptors and STING proteins. 4 , 5 Although these innate-immune alterations have been documented in adenomyotic tissue, they have largely been characterized at the level of disease presence rather than being explicitly linked to the pain phenotype, and the molecular basis of why some patients develop debilitating pain, while others remain asymptomatic is still poorly defined. Notably, both prevailing pathogenic theories converge on inflammation: the tissue injury and repair (TIAR) mechanism implies repeated micro-injury and a sustained wound-healing inflammatory milieu, whereas the stem-cell metaplasia model implies that aberrant progenitor populations contribute to a locally altered immune microenvironment. We therefore reasoned that pain-associated, rather than merely disease-associated, inflammatory programs might be identifiable at the cell-type level. This rationale is supported by other chronic pain disorders, including orthopedic conditions such as osteoarthritis, in which cytokines and chemokines (eg, TNF-α, IL-1, and CXC-family chemokines) produced within damaged tissue directly sensitize and activate innervating nociceptors. 6 Recognizing that inflammatory responses may be linked to the pathophysiology of pain in adenomyosis, this study aimed to identify the underlying gene regulatory mechanisms closely implicated in the inflammatory pathophysiology of pain in adenomyosis by comprehensively exploring direct and indirect inflammatory-associated gene expression profile patterns. To structure this investigation in line with the analyses performed, we first compared inflammatory activity between adenomyotic lesions and adjacent normal tissue (endometrium vs myometrium), and then examined molecular and cell-type-specific differences between adenomyosis patients with and without pain. To enhance the rigor and validity of this study, we employed a range of analytical approaches. These included enrichment analyses of inflammatory modules (ie, gene groups), exploration on specific inflammatory gene expression profiles, protein–protein interaction (PPI) network analysis of specific inflammation-associated cell-types, and cell–cell communication analysis of those certain cell-types. These analyses were conducted using bulk RNA-seq data from adenomyosis patients and single-cell RNA-seq data from adenomyosis patients with and without pain.

Prior to analyzing cellular and molecular inflammation patterns in the endometrium associated with adenomyosis, we first explored how inflammatory responses of endometrium are different from those of myometrium. This examination utilized a core set of inflammatory gene list obtained from Topper et al to evaluate the activation levels of key inflammatory modules (gene groups) as shown in Figure 1 . 20 From the primary immune system (ie, Innate Immune, Extracellular Immunity, and Mitochondrial Innate Immune), the activation levels of inflammatory modules generally exhibited opposing trends between the endometrium and myometrium, excluding mitochondrial DNA (mtDNA) and double-stranded RNA (dsRNA). Specifically, even within canonical and non-canonical modules, which typically operate as the first line of defense or finetuning its inflammatory response, the opposing trends observed across the two tissues may suggest significant differences between them. Figure 1 Lollipop plots showing activation levels of inflammatory modules from the endometrium and myometrium (P-value < 0.5). Positive NES values indicate activation in adenomyotic tissues. Six lollipop plots comparing inflammatory module activation between endometrium and myometrium. Innate Immune. X-axis: module (Antigen Presentation, Canonical, Inflammation, Non-canonical). Y-axis: NES (no unit), about minus 1 to plus 1. Antigen Presentation shows only a myometrium normal marker near 0. Canonical shows adenomyosis endometrium near plus 1 and normal endometrium near minus 1. Inflammation shows adenomyosis endometrium near plus 1.2 and adenomyosis myometrium near plus 0.8. Non-canonical shows adenomyosis endometrium near plus 0.9 and normal endometrium near minus 1. Extracellular Immunity. X-axis: Antigen Presentation, Cytokines, Interleukins, Surface Receptor Signaling. Y-axis: NES, about minus 1 to plus 1. Antigen Presentation shows adenomyosis myometrium near plus 0.8. Cytokines and Interleukins show adenomyosis endometrium near plus 1 and normal endometrium near minus 1. Surface Receptor Signaling shows adenomyosis endometrium near plus 0.8 and normal endometrium near minus 1. Mitochondrial Innate Immune. X-axis: mtDNA, mtRNA and ssRNA, mtdsRNA. Y-axis: NES, about minus 1 to plus 1.5. mtDNA shows adenomyosis endometrium near plus 1.3 and adenomyosis myometrium near plus 1.1. mtRNA and ssRNA shows adenomyosis endometrium near plus 1.4 and adenomyosis myometrium near plus 0.6. mtdsRNA shows adenomyosis myometrium near plus 1.0 and normal endometrium near minus 1. Unfolded Protein Response. X-axis: Endoplasmic Reticulum, Mitochondrial. Y-axis: NES, about minus 0.5 to plus 1. Endoplasmic Reticulum shows adenomyosis endometrium near plus 0.9 and adenomyosis myometrium near plus 1.0. Mitochondrial shows adenomyosis myometrium near plus 0.8 and normal endometrium near minus 0.5. Integrated Stress Response. X-axis: AA-Uptake and Biosynthesis, Anti-Oxidant, Autophagy, Cytokines and Chemokines, DNA Damage, Growth Factors, ISR Activation, ISR Inhibition, Metabolism, Survival Factors. Y-axis: NES, about minus 1 to plus 1.2. AA-Uptake and Biosynthesis shows normal endometrium near minus 1 and normal myometrium near minus 1. Anti-Oxidant shows adenomyosis endometrium near plus 1. Autophagy shows normal endometrium near minus 0.6 and adenomyosis myometrium near plus 1. Cytokines and Chemokines shows adenomyosis endometrium near plus 1.2 and adenomyosis myometrium near plus 1.0. DNA Damage shows adenomyosis endometrium near plus 1.1 and adenomyosis myometrium near plus 0.9. Growth Factors shows normal endometrium near minus 0.6. ISR Activation shows normal endometrium near minus 0.7 and normal myometrium near minus 0.6. ISR Inhibition shows normal endometrium near minus 0.7 and normal myometrium near minus 0.6. Metabolism shows adenomyosis endometrium near plus 1.0 and adenomyosis myometrium near plus 0.9. Survival Factors shows adenomyosis endometrium near plus 1.1 and adenomyosis myometrium near plus 1.0. Renin-Angiotensin-Aldosterone System. X-axis: AGT Activation, Bradykinin Degradation, Complement Activation, Fibrin Deposition, Hyaluronan Accumulation, NADPH Oxidase, PANoptosis, Syndecans. Y-axis: NES, about minus 1 to plus 1. AGT Activation shows adenomyosis endometrium near plus 1 and normal myometrium near minus 1. Bradykinin Degradation shows normal endometrium near minus 1 and normal myometrium near minus 0.8. Complement Activation shows adenomyosis endometrium near plus 0.9 and normal myometrium near minus 1. Fibrin Deposition shows adenomyosis endometrium near plus 0.9 and normal myometrium near minus 1. Hyaluronan Accumulation shows normal endometrium near minus 0.8 and normal myometrium near minus 0.7. NADPH Oxidase shows normal endometrium near minus 0.8 and normal myometrium near minus 0.7. PANoptosis shows adenomyosis endometrium near plus 0.8 and normal myometrium near minus 1. Syndecans shows adenomyosis endometrium near plus 1 and adenomyosis myometrium near plus 0.8. Legend: Tissue endometrium square, myometrium triangle; Condition adenomyosis and normal; grey indicates p-value greater than 0.5. Lollipop plots showing activation levels of inflammatory modules from the endometrium and myometrium (P-value < 0.5). Positive NES values indicate activation in adenomyotic tissues. The endometrium and myometrium also revealed distinct patterns in the indirectly associated systems (ie, Unfolded Protein Response, Integrated Stress Response, and Renin–Angiotensin–Aldosterone System), from the inflammatory module activation perspective. The Unfolded Protein Response, which is partially involved in regulating immune responses, disclosed comparable trends in the Endoplasmic Reticulum, but demonstrated divergent characteristics in the Mitochondrial module. Another stress response system, the Integrated Stress Response, which comprises nine inflammatory modules, displayed distinct activation patterns in three modules such as Anti-Oxidant, Autophagy, and Sensor/Initiator. The other system indirectly regulating immune responses, the Renin–Angiotensin–Aldosterone System, also unveiled three modules (ie, AGT Regulator Axis, Complement Activation/Fibrin Deposition, and PANoptosis) with different activation patterns between the two tissues. This disparity may suggest a potential shift in the immune landscape associated with adenomyosis, driven by tissue-specific immune regulation in its pathophysiology. Since the primary objective of this study was to investigate whether the inflammatory patterns in patients with adenomyosis experiencing pain are associated with specific cell types, we analyzed single-cell RNA seq data of patients with pain from the inflammatory response perspective. In this analysis, the activation levels of inflammatory modules across all cell types were compared with those observed in bulk RNA-seq data considered to be the control as depicted in Figure 2 . Overall, the inflammatory activation levels of most cell types in patients with pain revealed similar patterns to those in the bulk data. This not only indirectly confirmed that both datasets originated from the endometrium tissue but also suggested the potential association of inflammatory patterns in multiple cell types with pain. Figure 2 Lollipop plots comparing activation levels of endometrium bulk RNA-seq and individual cell types of scRNA-seq data (P-value < 0.5). Positive NES values of cell types indicate activation in pain tissues. Six lollipop plots comparing normalized enrichment score across immune and stress modules by cell type and bulk. Innate Immune; Extracellular Immunity; Mitochondrial Innate Immune; Unfolded Protein Response; Integrated Stress Response; Renin-Angiotensin-Aldosterone System. A lollipop-plot grid showing normalized enrichment score comparisons for bulk and single-cell pseudobulk across modules and cell types. Each subplot uses x-axis module categories (no unit) and y-axis labeled NES (no unit). Innate Immune: y-axis from minus 2 to plus 2. Categories: Antigen Presentation, Canonical, Inflammation, Non-canonical. Most points cluster near 0; several positive lollipops around plus 1 to plus 2 in Canonical, Inflammation and Non-canonical. Extracellular Immunity: y-axis from minus 1 to plus 2. Categories: Antigen Presentation, Cytokines, Interleukins, Surface Receptor Signaling. Positive lollipops mainly between plus 1 and plus 2 for Cytokines, Interleukins and Surface Receptor Signaling; Antigen Presentation near 0. Mitochondrial Innate Immune: y-axis from minus 2 to plus 2. Categories: mtDNA, mtDNA dsRNA, mtRNA. mtDNA and mtDNA dsRNA show positive lollipops near plus 1 to plus 2; mtRNA shows negative lollipops around minus 1 to minus 2. Unfolded Protein Response: y-axis from minus 1.5 to plus 1.0. Categories: Endoplasmic Reticulum, Mitochondrial. Endoplasmic Reticulum shows positive lollipops near plus 1; Mitochondrial shows negative lollipops near minus 1 to minus 1.5. Integrated Stress Response: y-axis from minus 1 to plus 2. Categories: AA Uptake Biosynthesis, Antioxidant, Autophagy, Cytokines Chemokines, Death Factors, ISR Activator, ISR Inhibitor, Inflammation Initiator, Survival Factors. Mixed positive and negative lollipops; several positives near plus 1 to plus 2 in AA Uptake Biosynthesis, Cytokines Chemokines, Death Factors and Survival Factors; negatives near minus 1 in Antioxidant, Autophagy, ISR Activator and ISR Inhibitor. Renin-Angiotensin-Aldosterone System: y-axis from minus 1 to plus 2. Categories: AGT Receptor, Bradykinin Degradation, Complement Activation, Complement Fibrin Deposition, Hyaluronan Accumulation, NADPH Oxidases, PANoptosis, Syndecans. Multiple positives near plus 1 to plus 2 in AGT Receptor and Syndecans; several negatives near minus 1 in Bradykinin Degradation, Hyaluronan Accumulation and NADPH Oxidases. Legends: Analysis includes bulk and sc-pseudobulk. Condition includes adenomyosis, normal and p-value greater than 0.5. Single-Cell Celltypes list includes endothelial cell, mesenchymal cell, red blood cell, multipotent stem cell, myofibroblast, NKT cell, other, epithelial cell, myeloid cell and opposite patterns. Lollipop plots comparing activation levels of endometrium bulk RNA-seq and individual cell types of scRNA-seq data (P-value < 0.5). Positive NES values of cell types indicate activation in pain tissues. Our focus in this analysis was, however, on exploring specific cell types within inflammatory modules having higher activation levels compared to bulk levels. This focus stems from the hypothesis that such cases may be potentially associated with pain in adenomyosis. Among the nine cell types, excluding a cell group named “other,” the most considerable difference was detected in multipotent stem cell of Anti-Oxidant of Integrated Stress Response (ISR), with an increase of 1.37 times compared to the control. Myofibroblast cell of Endoplasmic Reticulum of Unfolded Protein Response exhibited the second highest level of inflammatory activation relative to the control as 1.3 times elevation among all the cell types. Other cell types, such as the myeloid cell of Cytokines/Chemokines of ISR also demonstrated relatively increased activation levels, with at least a 1.2-fold increase. These cell types, which revealed pronounced inflammatory activation specifically in adenomyosis patients experiencing pain, may suggest a potential association with pain. Having investigated how inflammatory-associated genes collectively revealed their coordinated activation levels across different cell-types, we explored the alterations in the expression of these genes in all cell types in adenomyosis compared to the control as shown in Figure 3 . Comparison of the main immune systems (ie, Innate Immune, Extracellular Immunity, and Mitochondrial Innate Immune) between adenomyosis and the control revealed that only a few genes, such as CGAS and TNF, unveiled significant changes (ie, P-value < 0.05). Previous studies have identified genes such as CGAS as being implicated in the pathophysiology of adenomyosis. 5 In contrast, considerably differentiated genes were captured when cell subpopulations were compared based on pain status. Generally, except for NK/T cells, significant differentially expressed genes (DEGs) were detected across all other cell types, with each cell-type having at least one DEG. Although red blood cells and the undefined cell clusters labeled as “other” displayed differential expression of six and nine genes, respectively, they were excluded from further analysis due to their minimal association with inflammation. In the primary immune systems of adenomyosis (compared to the control), twenty-five genes were identified as significant DEGs, excluding one gene (TNFRSF4) in the red blood cells. From the Innate Immune, five genes (CCL5, CCL2, PYCARD, TNFAIP6, TNF) displayed upregulation, whereas eleven genes (OAS3, ISG15, IFITM1, IFIT2, IFIT1, IFI27, ZC3HAV1, SERPING1, PLVAP, HLA-A, TNFSF10) were downregulated. In the Extracellular Immunity, seven genes (CXCL12, CSF3, CCL19, CCL8, CCL5, CCL2, CD74) demonstrated upregulation; however, exceptions to downregulation were observed for CXCL13 and BCL6. Within the Mitochondrial Innate Immune, three genes (NFKBIA, IL1R1, PYCARD) were positively differentially expressed, while IL1R1 demonstrated downregulation in epithelial cells. Figure 3 Heatmaps displaying Wald test statistics of inflammatory-associated genes in three primary immune systems: Innate Immune, Extracellular Immunity, Mitochondrial Innate Immune. Heatmaps show gene expression in immune systems with Wald test stats, cell types and gene regulation levels. Three heatmaps present Wald test statistics for inflammatory-associated genes across Innate Immune, Extracellular Immunity and Mitochondrial Innate Immune systems. The horizontal axis lists genes, while the vertical axis includes cell types such as endothelial, mesenchymal, stem, myofibroblast, NK/T, epithelial, myeloid, red blood and others. The statistic scale ranges from negative 4.5 to positive 4.5, indicating gene regulation levels. Gray cells indicate missing data. Annotation bars compare adenomyosis versus normal and pain versus no-pain conditions. Subcategory bars label gene groups like antigen presentation, inflammation, cytokines, interleukins and mitochondrial signaling. In the Innate Immune panel, genes like CCL5 and TNF show upregulation in specific cells, while ISG15 and IFIT1 are downregulated. In the Extracellular Immunity panel, genes such as CXCL12 and CCL2 are upregulated across cell types. In the Mitochondrial Innate Immune panel, NFKBIA and PYCARD are upregulated, whereas IL1R1 is downregulated in epithelial cells. Heatmaps displaying Wald test statistics of inflammatory-associated genes in three primary immune systems: Innate Immune, Extracellular Immunity, Mitochondrial Innate Immune. We continuously examined gene expression profile changes of inflammatory modules of indirectly associated immune systems such as Unfolded Protein Response, Integrated Stress Response, and Renin–Angiotensin–Aldosterone System. As illustrated in Figure 4 , only a few genes, specifically TNF and ANGPT1, were detected as differentially expressed genes (DEGs) between adenomyosis and the control, however, thirty-one DEGs were captured from the comparison based on pain status. In the Unfolded Protein Response, substantial expression alterations were observed, with DNAJB9 and DERL3 being considerably upregulated and SOD2 downregulated. Of particular interest, DNAJB9 demonstrated a cell-type-specific expression pattern, characterized by upregulation in mesenchymal cells, multipotent stem cells, and myofibroblasts, while being downregulated in epithelial cells. Within the Integrated Stress Response, ten DEGs were identified (ie, GDF15, TNFSF10, TNF, CXCL8, CXCL3, CCL2, IL23A, BID, GSTA, TXN), where most of these genes were upregulated, except for TNFSF10. In the Renin–Angiotensin–Aldosterone System, contrasting expression trends were observed, as eight genes (ie, KLK6, KLK4, KLK2, SDC1, F13A1, C5AR1, PYCARD, AGT) disclosed increased expression, whereas ten others (KLK13, KLK10, HMMR, THBD, PLVAP, C1R, SERPING1, MLKL, ANGPT1, CTSA) demonstrated reduced expression. Figure 4 Heatmaps displaying Wald test statistics of inflammatory-associated genes in three secondary immune systems: Unfolded Protein Response, Integrated Stress Response, and Renin–Angiotensin–Aldosterone System. Wald test heatmaps for UPR, ISR, RAAS genes. Three heatmaps titled Unfolded Protein Response, Integrated Stress Response and Renin-Angiotensin-Aldosterone System display gene columns by row groups with a category strip. Gene symbols label columns. A shared legend shows a numeric scale from -4.5 to 4.5, with NA as gray, representing Wald test statistics for Adenomyosis vs. Normal and Pain vs. No-pain comparisons, noting endometrium under the first. A cell-type legend includes endothelial, mesenchymal, red blood, multipotent stem, myofibroblast, NKT, other, epithelial and myeloid cells, matching row-group strip colors. No axis titles or units are present. Most cells cluster near 0, with some at higher positive and lower negative values. The Renin-Angiotensin-Aldosterone System heatmap has several NA cell blocks across genes, while the other heatmaps show fewer NA blocks. Heatmaps displaying Wald test statistics of inflammatory-associated genes in three secondary immune systems: Unfolded Protein Response, Integrated Stress Response, and Renin–Angiotensin–Aldosterone System. More importantly, having recognized that myofibroblasts, multipotent stem cells, and myeloid cells demonstrated higher activation levels in certain inflammatory modules in this study, we further examined the representative genes associated with these three modules. The upregulation of DNAJB9 in myofibroblasts within the Unfolded Protein Response and GSTA1 in multipotent stem cells within the Integrated Stress Response contributed to the observed heightened activation. Additionally, myeloid cells in the Integrated Stress Response exhibited increased levels of CXCL8 and CXCL3, underscoring their potential involvement in pain-specific inflammatory regulation in adenomyosis. Given that certain cell types, such as the multipotent stem cells and NK/T cells, may be associated with adenomyosis-associated pain, we performed an analysis from a different perspective to search for potential functional clusters with respect to associations between proteins encoded by differentially expressed genes (DEGs). This analysis was carried out based on the Protein–Protein Interaction (PPI) network by leveraging high-confidence interaction data sourced from the STRING database as described in Figure 5 . Through our analysis of single-cell RNA-seq data from adenomyosis patients with and without pain, a multipotent stem cell population was detected predominantly in the pain group; given the small number of patients and the low number of captured cells in the no-pain samples; however, we cannot exclude that this reflects limited cell capture or stochastic sampling rather than a strictly pain-exclusive population. Also, recent studies have reported that SFRP4+IGFBP5hi NK/T cells can promote the differentiation of progenitor cells into neurogenic cells, contributing to pain in adenomyosis patients. 9 These findings provide a compelling rationale for our focus on exploring protein associations involving NK/T cells and multipotent stem cells. Figure 5 Protein–protein interaction (PPI) networks of multipotent stem cells (p 0.5) and NK/T cells (p < 0.1). Ten most enriched biological pathways for each cell type, with each pathway featuring a false discovery rate below 0.001 and including at least three counts in the network. Nodes represent proteins encoded by differentially expressed genes, and edges represent high-confidence STRING interactions; each distinct node color denotes membership in one of the enriched biological pathways listed in the legend, and proteins participating in more than one enriched pathway are shown with multiple colors. Gray edges indicate protein–protein associations, with edge thickness scaled to the STRING combined-interaction confidence score. Protein-protein interaction networks for multipotent stem cells and NK/T cells, highlighting enriched pathways. The image shows protein-protein interaction networks for multipotent stem cells and NK/T cells. The multipotent stem cell network includes nodes representing proteins encoded by differentially expressed genes, with edges indicating high-confidence STRING interactions. Pathways include nitric oxide transport, chronic inflammatory response and others. The NK/T cell network features pathways such as positive regulation of chemokine production and leukocyte cell-cell adhesion. A map indicates pain and no-pain groups, with distinct colors for each. The legend lists pathways associated with each cell type, with nodes colored according to pathway membership. Proteins participating in multiple pathways are shown with multiple colors. Gray edges represent protein associations, with edge thickness scaled to the STRING confidence score. Protein–protein interaction (PPI) networks of multipotent stem cells (p 0.5) and NK/T cells (p < 0.1). Ten most enriched biological pathways for each cell type, with each pathway featuring a false discovery rate below 0.001 and including at least three counts in the network. Nodes represent proteins encoded by differentially expressed genes, and edges represent high-confidence STRING interactions; each distinct node color denotes membership in one of the enriched biological pathways listed in the legend, and proteins participating in more than one enriched pathway are shown with multiple colors. Gray edges indicate protein–protein associations, with edge thickness scaled to the STRING combined-interaction confidence score. From the multipotent stem cells, some functional pathways (ie, chronic inflammatory response, mammary gland morphogenesis, glomerulus development, regulation of alpha-beta T cell differentiation, response to estrogen, and positive regulation of smooth muscle cell proliferation) were considerably enriched while substantial enrichment on the NK/T cells was involved in other pathways (ie, leukocyte cell–cell adhesion, positive regulation of interferon-gamma production, response to nutrient, and myeloid leukocyte differentiation). Among these protein networks, specifically, key chemokines such as CCL11 and CXCL13 were detected, where chemokines are well-established mediators in the pathogenesis of chronic pain. 24 Blockade of CCR3, the receptor for CCL11, has demonstrated potential in alleviating neuropathic pain in preclinical models by interacting with certain chemokines such as CCL5, CCL7, and CCL11, which are important for nociception. 25–30 Additionally, CXCL13, acting through its receptor CXCR5, is implicated in pain modulation. 31 These findings suggest that chemokines secreted by multipotent stem cells, which are exclusively present in the pain group, may play a crucial role in the nociceptive processes observed in adenomyosis patients. Building upon our protein-interaction network analysis of multipotent stem cells and NK/T cells, we extended our exploration to encompass the broader pathophysiology and pain-associated mechanisms at the cellular level by cell-type-specific communication analysis in both the pain and no-pain groups. While the majority of communication patterns were consistent across the two groups, we identified particular differences in three critical signaling pathways: Interleukin-1 (IL-1), Estradiol, and 2-Arachidonoylglycerol (2-AG) as shown in Figure 6 . In the patient group with pain, the IL-1 signaling pathway was predominantly directed towards red blood cells and epithelial cells. This is in contrast to the no-pain group, where mesenchymal cells emerged as the principal signaling target. Estradiol signaling in the pain group was substantially enhanced in epithelial and multipotent stem cells, while it was attenuated and primarily targeted endothelial cells in the no-pain group. 2-AG signaling appeared to be significantly diminished or absent in the pain group. Through additional examination on ligand–receptor interactions, we captured that the IL1B-(IL1R1+IL1RAP) and IL1A-(IL1R1+IL1RAP) pairs in the pain group were ranked higher than their rankings in the no-pain group (as fourth and seventh, respectively), and the IL18-(IL18R1+IL18RAP) pair was exclusively active in the no-pain group. Figure 6 Comparison of cell type-specific interaction patterns between pain and no-pain group. Ten most significant signaling pathways and the strengths of each cell in interactions as sender/receiver. Charts compare cell signaling in pain vs no-pain groups, highlighting key pathways. The images show heatmaps and bar charts comparing cell signaling patterns between pain and no-pain groups. In both images, the x-axis represents cell types: endothelial, mesenchymal, red blood, multipotent stem, myofibroblast, NK T, other, epithelial, myeloid. The y-axis shows signaling pathways: ADGRE, CSF, IL1, GALECTIN, ANGPTL, CCL, CX3C, PARs, KLK, Estradiol. The heatmaps display outgoing and incoming signaling patterns with a strength scale from 0 to 1. In the pain group, IL1 and Estradiol pathways are prominent, targeting epithelial and red blood cells. In the no-pain group, mesenchymal cells are key targets, with IL18 and 2-AG pathways more active. Bar charts show the relative contribution of ligand-receptor pairs. In pain, top pairs include THY1-ADGRE2 and IL1B-IL1R1+IL1RAP. In no-pain, THY1-ADGRE2 and IL34-CSF1R are significant. The charts use a relative contribution scale without specific units, highlighting differences in signaling dynamics between conditions. Comparison of cell type-specific interaction patterns between pain and no-pain group. Ten most significant signaling pathways and the strengths of each cell in interactions as sender/receiver.

Public transcriptome data from the Gene Expression Omnibus (GEO) repository were utilized in this study. Specifically, three bulk RNA-seq datasets (accession number; GSE157718 , GSE185392 , GSE190580 ) and one scRNA-seq dataset (accession number; GSE218044 ) were obtained. 7–10 The bulk RNA-seq data included endometrial tissues of adenomyosis patients ( GSE157718 , n = 3; GSE185392 , n = 10; GSE190580 , n = 6) and healthy controls ( GSE157718 , n = 3; GSE185392 , n = 10; GSE190580 , n = 5). Additionally, myometrial tissues of the same participants were sequenced in accession number GSE190580 . By comparison, the scRNA-seq data comprised endometrial samples from adenomyosis patients (n = 4) categorized based on the presence of dysmenorrhea (with pain, n = 2; without pain, n = 2), without including healthy controls. After the requisite data were downloaded, cell cluster identification was conducted for the scRNA-seq dataset using the Seurat (v5.1.0) R package. 11 We applied identical marker genes to ensure consistency with the original paper; consequently, nine cell types were distinguished with ambiguous clusters labeled as “other.” Following annotation, gene counts were summed within each sample to form pseudobulk samples, including whole-cell and cell type-specific aggregates. The cell type-specific pseudobulk samples were then utilized for differential analyses of “pain vs no-pain” in each cell type. Pseudobulk aggregation was chosen because it accounts for biological replicate structure and has been shown to control false discoveries more reliably than single-cell-level differential testing. 12 We note, however, that the single-cell cohort comprised only two patients per group, so all cell-type-specific results derived from it are exploratory and hypothesis-generating; this constraint is considered explicitly in the Discussion. For the bulk RNA-seq datasets, given that GSE157718 provided only sequenced reads, we aligned these reads to the reference genome (GRCh38) using STAR (v2.7.10b) and quantified with HTseq-count (v2.0.2). 13 , 14 The resulting read counts were combined with those from GSE185392 and GSE190580 to create a gene count matrix, which also included whole-cell pseudobulk samples as four adenomyosis samples. The ComBat_seq model from the sva (v3.46.0) R package was then applied to the matrix to minimize batch effects from dataset variations and the inclusion of pseudobulk samples. 15 , 16 ComBat_seq models batch as a known covariate using a negative-binomial framework while preserving the integer count structure required by DESeq2; nonetheless, jointly correcting bulk and pseudobulk libraries within a single matrix assumes that the dominant unwanted variation is captured by the specified batch factor, and residual platform-related effects cannot be fully excluded. 17 Because of this, the integrated bulk–pseudobulk comparisons were treated as exploratory and were interpreted alongside the within-platform analyses rather than in isolation. For the myometrial data, raw gene counts from GSE190580 were used directly as the count matrix. Following preprocessing, differential analyses were performed according to the DESeq2 (v1.44.0) protocol for two comparisons: “adenomyosis vs normal” in both endometrial and myometrial tissues, and “pain vs no pain” within each annotated cell type. 18 Subsequently, we implemented fast gene set enrichment analyses (fGSEA), deploying customized gene sets defined by Topper and Guarnieri et al (2023) to identify upregulation in core inflammatory modules. 19 , 20 For fGSEA input, DESeq2 results were filtered with a p-value threshold of 0.5. This threshold was deliberately permissive and was not used as a significance cutoff for individual genes; rather, fGSEA operates on a continuously ranked gene list, and the filter served only to remove genes carrying essentially no signal while retaining a sufficiently large ranked input for stable enrichment-score estimation. Module-level significance was subsequently assessed using the enrichment statistics and their associated p-values, so this lenient pre-filtering step does not by itself inflate the number of genes called significant at the gene level. Guided by the outcomes of differential and enrichment analyses, supplementary analyses were conducted at three distinct levels to provide further insights. First, differential gene expression patterns in the endometrium and its cell subpopulations were examined through heatmap visualizations. We repeatedly utilized a gene set customized by Topper et al to focus on inflammatory genes. 20 Second, we constructed a protein–protein interaction (PPI) network using STRING (v12.0) to investigate the proteins encoded by differentially expressed genes (DEGs) and their interactions. 21 Unlike the previous reference, to target inflammatory DEGs, we utilized a customized gene set compiled from human collections of the Molecular Signatures Database (MSigDB) using keyword “inflammation” for the search. Lastly, cell–cell interaction analysis was conducted using CellChat (v2.1.2) to explore intercellular communication within the tissue microenvironment, with a specific emphasis on inflammatory-related signaling pathways by restricting the search to receptor keywords associated with inflammatory responses. 22 , 23

Our study demonstrates that adenomyosis is characterized by distinct inflammatory profiles, with a pronounced upregulation of canonical and non-canonical inflammatory modules in the endometrium and significant tissue-specific differences compared to the myometrium. Key findings include the activation of interferon-sensitive genes and the involvement of the innate immune system via the cGAS-STING pathway, underscoring mitochondrial dysfunction as a potential trigger of immune activation. Furthermore, our cell type-specific analyses suggested that differential regulation of IL-1, estradiol, and 2-AG signaling pathways may underpin the heterogeneous nature of adenomyosis-associated pain, with genes such as CGAS, TNF, DNAJB9, GSTA1, CXCL8, and CXCL3 emerging as candidate mediators that warrant further validation. By identifying these candidate inflammatory pathways and signaling alterations, our study generates testable hypotheses regarding the pain-specific inflammatory mechanisms of adenomyosis. Rather than establishing definitive therapeutic targets, these findings advance the current understanding of adenomyosis pathophysiology and provide a prioritized framework for future research, which will require validation in larger, well-characterized cohorts and functional experiments before clinical translation can be considered.

In this study, we explored the cellular and molecular mechanisms underlying adenomyosis, specifically focusing on the role of inflammatory responses in driving pain, which is the condition’s most debilitating symptom. Our hypothesis posited that gene regulatory mechanisms associated with inflammation might play a central role in the pathophysiology of pain associated with adenomyosis. To investigate the underlying mechanisms, we employed multiple analytical approaches, including pathway enrichment analysis of core inflammatory genes, expression profile pattern analysis of these genes, PPI network analysis, and cell–cell communication analysis. Collectively, these analyses provided a comprehensive framework for understanding how specific core inflammatory-associated molecular activities, through genes, modules, and their interactions, may contribute to pain in adenomyosis, offering potential avenues for targeted therapeutic interventions. Our findings indicated a predominant upregulation of both canonical and non-canonical inflammatory modules in the endometrium, featuring the significant role of interferon-driven and interferon-sensitive genes (ISGs) in the inflammatory pathophysiology of adenomyosis ( Figure 1 ). 4 , 5 These two inflammatory modules, primarily composed of upregulated ISGs, demonstrated strong activity in the endometrial tissue, providing compelling evidence for the involvement of inflammation in adenomyosis. In contrast, the divergent patterns observed in the myometrium suggest substantial compartmentalization of the inflammatory response between tissues. This discrepancy may reveal differences in the microenvironmental impact of adenomyosis across tissue types or variations in disease progression and immune activity specific to each compartment. The innate immune system, particularly systems triggered by cytosolic mtDNA and mtdsRNA release due to mitochondrial dysfunction, appeared to play a critical role. Dysregulated mitochondrial integrity, likely activating pathways such as cGAS-STING signaling, has been implicated in adenomyosis progression. 32 Our analysis supports this finding, as the upregulation of core innate immune modules in the endometrium suggests that mitochondrial mtDNA may serve as a key driver of immune activation in adenomyosis. Additionally, extracellular immune signaling, mediated by cytokines and chemokines, displayed considerable alterations in adenomyotic tissues. 33 This inflammatory system has been known to directly or indirectly be associated with the inflammatory milieu of adenomyosis, further reinforcing the disease’s immune-mediated nature. Secondary inflammatory systems, including the Unfolded Protein Response and Integrated Stress Response, have been previously reported to unveil deviations among different adenomyosis patients. 34 , 35 Similarly, the Renin–Angiotensin–Aldosterone System, which has been indirectly linked to adenomyosis, displayed variations between the two tissues (endometrium and myometrium), but no clear trends were observed in each tissue within our dataset. 36 These findings suggest that secondary systems may play a context-specific role in the pathophysiology of adenomyosis, underscoring the complexity of the inflammatory response by emphasizing key immune-associated pathways and highlighting the need for further investigation into subpopulation-specific immune dynamics. Based on these findings, we shifted our focus to the inflammatory module activation of individual cell-types in pain-based comparisons, where compartmentalized patterns (hyperactivity and co-activation) may provide further insights into the heterogeneous landscape of adenomyosis. While similar activation patterns between the bulk endometrium and most cell types were captured in the three primary immune systems, endothelial cells, along with various other cell types that varied by module, disclosed activation patterns that consistently aligned with the bulk across all modules. These indicate that the three major immune systems may collectively contribute to pain through the involvement of diverse cell types. Through the bulk endometrium analysis, a small number of genes reached statistical significance with relatively large Wald test statistics. These genes appeared to be implicated in modulating inflammatory pathways according to enrichment analysis, suggesting a potential role in adenomyosis pathogenesis and warranting further study. CGAS and TNF could be particularly important due to their established links with adenomyosis. (i) CGAS, a key component of the cGAS-STING pathway, was upregulated in adenomyosis patients, consistent with the activation of this pathway, as evidenced by the co-upregulation of genes like STING and TBK. 32 , 37 (ii) TNF, a crucial mediator in various inflammatory and immune pathways, including the cGAS-STING pathway, was also upregulated in these patients, with expression levels correlating positively with dysmenorrhea. 32 Moreover, TNF-α has been reported to enhance NF-κB binding activities and elevate the protein levels of COX-2, VEGF, and TF, all associated with adenomyosis. 38 Cellular stress-associated systems, although displaying similar patterns, are likely driven by hyperactive pathways in specific cell types. The “pain vs no pain” comparison analysis revealed significant alterations in gene expression within distinct cell types, indicating a localized or cell type-specific inflammatory response. The genes identified—DNAJB9, GSTA1, CXCL8, and CXCL3—emerge as potential candidates for mediating pain-specific inflammatory responses in adenomyosis. The molecular chaperone DNAJB9, involved in the function of myofibroblasts, plays a role in the Unfolded Protein Response. However, GSTA1, abundantly expressed in multipotent stem cells, is integral to the Integrated Stress Response through its capacity to mitigate oxidative stress. While the direct contributions of DNAJB9 and GSTA1 to pain modulation have not been extensively studied, their roles in autophagy and cellular stress responses underscore a promising avenue for further investigation. The chemokines CXCL8 and CXCL3, prominently elevated in myeloid cells, play crucial roles in leukocyte recruitment and the activation of proinflammatory pathways. CXCL3, particularly noted for its interaction with the CXCR2 receptor, has been well-documented for its significant contribution to the modulation of neuropathic pain. 39 , 40 In parallel, CXCL8 (IL-8), a key proinflammatory mediator, has been extensively studied for its involvement in inflammatory pain and its potential as a therapeutic target. 41–43 Together, these insights advance our understanding of the mechanistic pathways implicated in pain pathogenesis and their potential association with adenomyotic pain. Recognizing that adenomyosis-associated pain may be driven by distinct cellular subpopulations within the endometrium that actively participate in inflammatory mechanisms through key candidate genes, our investigation also revealed critical differences in cell–cell signaling dynamics between adenomyosis patients with and without pain, underscoring the involvement of three pivotal pathways: IL-1, estradiol, and 2-AG signaling. The observed alterations in IL-1 signaling dynamics between the pain and no-pain groups may indicate a significant shift in the inflammatory landscape. In the pain group, red blood cells (RBCs) and epithelial cells emerged as the primary recipients of IL-1 signals, in contrast to mesenchymal cells, which were predominant in the no-pain group. This shift may imply a reorganization of immune signaling pathways, potentially enhancing inflammation at the epithelial interface and contributing to nociceptive sensitization. The prominence of the IL1B-(IL1R1+ILRAP) ligand-receptor pair in the pain group further underscores IL-1β’s well-established role as a driver of proinflammatory pathways associated with pain. 44 Previous studies have reported that IL-1β can activate neurogenic inflammation through the JNK pathway, necessitating the upregulation of neurotrophins and their receptors in endometrial stromal cells. 45 , 46 These findings may suggest the presence of a localized inflammatory environment conducive to pain sensitization in adenomyosis. In individuals experiencing pain, estradiol signaling was substantially increased in epithelial and multipotent stem cells, demonstrating the hormonal dependency typical of adenomyosis. However, those without pain exhibited diminished estradiol signaling, predominantly affecting endothelial cells. This variation in signaling patterns may indicate that estradiol could potentiate inflammatory and neurogenic pathways in pain-associated adenomyosis, potentially via enhanced interactions between epithelial cells and stem cells. The specific estradiol signaling detected in epithelial cells associated with pain may imply that estradiol modulates local inflammation and nociception, likely through receptor-mediated signaling cascades that enhance pain perception. The lack of 2-AG signaling in the pain group contrasts sharply with its presence in the no-pain group, where myeloid cells predominantly served as senders and mast/B cells as the primary receivers. Given 2-AG’s established role as an endocannabinoid with anti-inflammatory and analgesic functions, its absence in the pain group may suggest a disruption of a potential pain-modulating mechanism. This absence may indicate a dysregulated endogenous system that enhances nociceptive signaling in adenomyosis. Therefore, restoring 2-AG signaling could represent a potential therapeutic target for mitigating pain associated with this condition. These findings underscore the essential role of specific ligand–receptor interactions in modulating the immune and endocrine signaling networks involved in adenomyosis-associated pain. The increased activity of IL-1 and estradiol pathways, alongside the absence of 2-AG signaling, may indicate a multifaceted disruption of regulatory mechanisms that typically mitigate inflammation and nociception. Should these patterns be confirmed in larger cohorts, they would suggest the value of comprehensive therapeutic strategies that target inflammatory mediators such as IL-1β, hormonal modulators like estradiol, and interventions aimed at restoring endocannabinoid signaling. Accordingly, future studies could explore the therapeutic potential of inhibiting IL-1β signaling, modulating estradiol activity, and enhancing 2-AG pathways for the management of adenomyosis-associated pain. Moreover, validating these findings in larger cohorts and integrating additional omics data from other system levels, such as proteomics or metabolomics, would offer more comprehensive insight into the molecular mechanisms underlying this condition. Several limitations should be considered when interpreting these results. First, and most importantly, the single-cell analyses were based on a small cohort of four patients (two with pain and two without). With only two biological replicates per group, cell-type-specific differential expression and cell–cell communication findings have limited statistical power and may be influenced by inter-individual variability; the observation that a multipotent stem cell population was detected predominantly in the pain group, in particular, may partly reflect limited cell capture in the no-pain samples rather than a strictly pain-exclusive biology. To mitigate spurious gene-level calls, we used pseudobulk aggregation, which respects biological replicate structure, but minimal biological replication remains an intrinsic constraint. 12 Second, potential confounders that strongly influence endometrial transcriptomes—including menstrual cycle phase, hormonal medication use, and lesion location or severity—were not annotated in the available datasets and could not be adjusted for. Third, the integration of bulk and pseudobulk libraries within a single matrix, although corrected with ComBat_seq, may introduce residual platform-related effects; we therefore treated these integrated comparisons as exploratory and interpreted them alongside within-platform analyses. 17 Fourth, the study is entirely computational and retrospective, based on static transcriptomic snapshots; it cannot establish causality or disease progression, and none of the candidate genes or pathways were experimentally validated. For these reasons, the findings that are most robust—namely the tissue-level differences in inflammatory module activation between endometrium and myometrium, supported by larger bulk datasets—should be distinguished from the cell-type-specific and interaction-level observations, which are exploratory and require confirmation in larger, well-characterized cohorts and in functional experiments.