Transcriptome differences in lesions by color, and cancer mutations in endometriosis

This work was supported by funding from Boston Center for Endometriosis.

With IRB approval (Mass General Brigham Protocol #2017P002765 and #2023P002237), informed consent for study participation and excess surgical tissue collection was obtained from patients scheduled for surgical management of endometriosis. Tissue samples were obtained from a single cohort of patients with surgically confirmed endometriosis or controls with surgically confirmed uterine fibroids without surgical evidence of endometriosis. All tissue samples were fresh or flash-frozen. The tissues from all 25 patients were obtained from and processed in the Departments of Obstetrics, Gynecology, and Reproductive Biology at the Center for Infertility and Reproductive Surgery at Brigham and Women’s Hospital. Inclusion criteria for the diseased cohort were pathologist-confirmed endometriotic superficial peritoneal lesions of a lesion size sufficient for excision surgery and microdissection. When possible, tissues were acquired from three separate sites on the same patient: endometrium, peritoneal tissue, and ectopic lesion. Ectopic lesions and their underlying peritoneal tissues were separated via dissection under a microscope, and tissue that microscopically appeared of mixed origin was excluded from analysis. Patient metadata were acquired by survey and consented under IRB guidelines. De-identified survey results from the WERF EPHect Endometriosis Questionnaire – Minimal (EPQ-M) ( Vitonis et al. 2014 ) were provided for analysis. A total of 200 patients were recruited for this study with IRB approval. No pediatric patients were involved in this study. The mean age of the cohorts was 30.25 years (the 20-woman endometriosis group) and 33 years (the 5-woman control group). Basic clinical characteristics and analysis inclusion details are displayed in Table 1 . Cohort description. Patient information and endometriosis characterization, grouped by extent of inclusion in sequencing and analysis. Eu, eutopic endometrial biopsy; Pt, peritoneal tissue; Lc, clear lesion; Lr, red lesion; Lb, black lesion; Lw, white lesion; Ct, endometrial biopsy from controls; DGE, differential gene expression; SVA, shared variant analysis. Total RNA was isolated by bead-based capture, purified, and confirmed as having an RNA integrity score of >6.5 before further processing (51/112 tissues). Whole transcriptome sequencing was performed on those tissues that passed internal quality controls for a total of 18 endometriotic lesions, 12 peritoneal tissues from below lesions in endometriosis patients, 17 endometrial biopsies from endometriosis patients, and 5 endometrial biopsies from control patients. Among endometriosis patients, 10 of 20 had high-quality tissue from eutopic endometrium, peritoneal tissue, and lesion acquired and sequenced and are thus classified as ‘triad’ patients. Paired-end libraries were generated using New England Biolab’s NEB II Ultra RNA Library Preparation Kit, and libraries were sequenced on Illumina’s NovaSeq 6000 platform using S2 and SP chips with a sample average of 47.2 million mapped paired-end reads and 2 × 100bp read length. RNA-seq short variant discovery workflow v4 (SNPs + Indels) in Broad Institute’s Genome Analysis Tool Kit (GATK) was used to call shared (‘germline’) variants in our cohort. To annotate discovered variants, we used Ensembl’s Variant Effect Predictor (VEP) ( McLaren et al. 2016 ). Output annotated variants were analyzed using a curated list of 299 cancer driver genes ( Bailey et al. 2018 ), derived from the National Cancer Institute’s The Cancer Genome Atlas database. Validation of called shared variants was performed utilizing the Broad Institute’s Integrated Genomics Viewer ( Robinson et al. 2011 ). We used GATK’s Somatic short variant pipeline with Mutect2 ( Benjamin et al. 2019 ) downstream of the RNA-seq short variant discovery workflow to call somatic mutations by comparing eutopic endometrium (normal) and lesion (‘tumor’) sequences. After conservative somatic variant calling, we filtered the calls to account for read orientation artifacts with a genotype count greater than 10. We then called the filtered variant files in VEP. Visualizations were plotted using the GenVisR R package ( Skidmore et al. 2016 ). Human BAF (SWI/SNF) complex was computationally modeled in Schrodinger PyMOL ( PyMOL, version 2.0 ), using the experimental structure PDBID 6LTH ( He et al. 2020 ). ARID1A computational mutagenesis was executed in PyMOL to visualize resulting steric clashes with the native SWI/SNF complex conformation. To predict the impact on ARID1A function, SIFT ( Sim et al. 2012 ) and PolyPhen2 ( Adzhubei et al. 2010 ) algorithms were used. Read counts from STAR ( Dobin et al. 2013 ) aligner 2.7.9a were processed using the DESeq2 ( Love et al. 2014 ) package for differential gene expression analysis, and principal component analysis was conducted in R 4.1.1. Gene Ontology and KEGG pathway enrichment analyses were performed using functions in the ClusterProfiler R package ( Yu et al. 2012 ). Gene Set Enrichment Analysis ( Subramanian et al. 2005 ) using the Hallmark dataset was performed using the fGSEA ( Korotkevich et al. 2016 ) package on pre-ranked genes from DEseq2’s ‘stat’ variable. A ranked-order gene list for somatic mutations was created based on the net number of medium- and high-impact variants called in our ‘triad’ cohort across all 10 samples. An accompanying ranked-order gene list was created based on results from differential gene expression analysis (log2FoldChange > 1.5 and P -value adjusted <0.05) and ordered based on fold change. Network visualization was performed on GePHI ( Bastian et al. 2009 ) with annotated interactions derived from the STRING database ( Szklarczyk et al. 2017 ).

In line with the distinctive phenotypic differences between red and clear peritoneal lesions, we found 148 genes that were significantly differentially expressed ( Fig. 2A ) ( P adj. 1.5) between these two lesion types. Gene Ontology and pre-ranked GSEA analysis showed enrichment for terms related to ciliated motility in clear lesions, and terms related to lymphocyte and cytokine production in red lesions ( Fig. 2B ), suggesting an increased fraction of cell types expressing genes in these pathways in red lesions. This is supported by the statistically significant upregulation of TNF signaling and inflammatory response in red lesions ( Fig. 2C ). Transcriptional upregulation of immune processes in red endometriotic lesion tissue. (A) Volcano plot highlighting genes that are significantly differentially expressed between red and clear lesions. (B) Gene Ontology enrichment for biological processes in clear and red lesion comparison, highlighting terms such as ciliated motility, lymphocyte differentiation, and cytokine production. (C) Pre-ranked GSEA of clear and red lesions yielded statistically significant upregulation of TNF signaling and inflammation in red lesion tissue. Furthermore, our analysis reveals a number of genes with stark differences in expression (such as PARP15 , SLC1A3 , and NR4A2 in red lesions and PPM1E , FAM95C , and GGT2 in clear lesions), which may provide insights into distinct disease processes. We then investigated the transcriptional differences within eutopic and lesion tissue pairs in our 10-patient triad dataset. We found a substantial number of differentially expressed genes ( Fig. 3A ). Subsequent Gene Ontology analysis identified 290 enriched terms with top terms involving matrix, lymphocyte, and cell recognition ( Fig. 3B ). Pre-ranked GSEA demonstrated enrichment in Hallmark gene sets related to cell proliferation and estrogen response ( Fig. 3C ). Importantly, these differences reflect changes at the tissue level that may result both from differential regulation of gene expression within individual cell types and from differential cell type composition. It thus reflects changes in the local microenvironment caused by both types of changes. Transcriptional upregulation of cell proliferation and estrogen response gene sets in endometriotic lesion tissue, from data collected from patients in the triad group. (A) Volcano plot detailing differentially expressed genes between eutopic and lesion tissue, with teal dots passing the filter ( P adj 1.5). (B) Gene Ontology chart of differentially expressed genes. (C) Pre-ranked GSEA demonstrated enrichment in Hallmark gene sets related to cell proliferation and estrogen response in lesion tissue. n = 10. We performed a shared SNP/Indel variation analysis on all 51 tissues in our discovery cohort using the GATK RNA SNP/Indel variant calling pipeline, in order to identify variants that were shared across a patient’s tissue types. 85% of our shared variant calls were previously annotated variants (Fig. S1A (see section on Supplementary materials given at the end of the article)) with the proportion of high-impact variants (nonsense, frameshift, and splice acceptor/donor variants) being 1.98% of all shared variants detected (Fig. S1B and C). We examined the variant profile of genes related to the well-investigated RAS and SWI/SNF pathways. In the transcriptome of our samples, we found a high prevalence of missense and nonsense variants in well-documented cancer genes such as ERBB2 , TP53 , EGFR , and ARID1A ( Fig. 4A ). We examined variant progression in ‘triad’ patients (Fig. S2A), and interestingly, there was evidence of high variant conservation across the three tissues in the same patient (Fig. S2B). Considering that germline mutations in these genes are associated with high incidence of cancer disease, and the lack of cancer syndrome in patients in our cohort, this observation is more likely due to the remnants of microscopic endometriotic lesion tissue within the sub-lesional peritoneal tissue samples, rather than being evidence of germline mutations. Analysis of shared cancer variants of tissue cohort (18 lesions, 12 peritoneal tissues, 17 endometrial biopsies, and 5 control endometrial biopsies, n = 51 from 25 patients) in RAS-related and SWI/SNF-related genes. (A) Shared cancer variants called in VEP variant pipeline as high or medium impact in all 51 tissues from 25 patients. (B) Principal component analysis of all high- or medium-impact shared variants in these tissues normalized in STAR counts. (C) Principal component analysis of normalized transcripts, showing separation between eutopic/control endometrium and peritoneal tissue/lesions. Tissue type: Ct (control), Eu (eutopic), L (lesion), and Pt (peritoneal). Looking only at the endometrium, we found that mutations in the three most frequently affected genes ( ERBB2 , TP53 , and PIK3R2 ) were significantly more common in the eutopic endometrium of patients vs unaffected controls (94 vs 67%, P = 0.015, Fisher’s exact test), in line with previous findings showing correlation between mutated genes in eutopic endometrium and endometrial cancer ( Yamaguchi et al. 2022 ). Thus, mutation in the eutopic endometrium might predispose patients to endometriosis. To glean insight into the effect of mutations on tissue transcriptome, we performed principal component analysis (PCA) on all VEP-annotated ‘medium’- and ‘high’-impact variants ( Fig. 4B ). Genes that are somatically hypermutated (IGLV|IGHV|IGKV|HLAs), as related to host immune response, were filtered prior to analysis. Genes that contributed most along principal component 1 primarily consisted of matrix components ( NID1 , LAMB2 , and LAMA5 ), mucins ( MUC5B and MUC16 ), and DNA-binding-related proteins (e.g. CHD3 ) as annotated by StringDB interactions (Fig. S4A). We confirmed that cycle-phase, which dramatically impacts transcriptional expression in hormonally responsive tissue such as eutopic and ectopic endometrium, did not distinguish genetic variants in PC1 and PC2 (Fig. S3B). In transcriptome PCA, we found a sizable separation between eutopic/control endometrium and peritoneal tissue/lesions ( Fig. 4C ), which is unsurprising considering the tissue composition differences. As expected, there was an observed cycle-dependency (Fig. S3A). Of the top 30 genes that contributed to principal components 1 and 2 in the shared variant PCA, only INMT , a methyltransferase, and PKD1L2 , a pseudogene, had intersections with the top 30 mutated genes in the transcriptional PC1 and PC2 (Fig S4B), suggesting that mutation is not a major driver of tissue transcript abundance in cis .

As shared variants were conserved across intra-patient tissue comparisons, we employed the Mutect2 pipeline to systematically call somatic variants evident primarily in lesion tissue by using patient eutopic tissue as ‘normal’ and paired endometriotic lesion tissue as ‘tumor’ ( Fig. 5 ). Of the 162 cancer driver genes detected with ‘medium’- or ‘high’-impact variants, 16 had variants in more than half of our ‘triad’ cohort ( Fig. 5A ) that may have arisen in lesions or been selected for during their generation. These include previously well-known SWI/SNF-related gene ARID1A , RTK/Ras-related gene EGFR , WNT-related gene CHD8 , Notch-related genes EP300 and NCOR1 , Hippo-related gene LATS1 , and several cancer driver genes less studied in the endometriosis context: MACF1 , NIBPL , ARHGAP35 , HUWE1 , KMT2C , KMT2D , MED12 , POLRMT , and RBM10 . Network analysis of these genes through StringDB showed significant enrichment of interaction ( Fig. 5B ). Furthermore, Gene Ontology analysis identified enrichment of genes related to histone modification, chromatin modification, and regulation of intracellular steroid hormone receptor signaling ( P adj < 0.05) (Fig. S6). Somatic mutational profiling in RAS-related pathway and SWI/SNF-related genes, comparing eutopic endometrium and lesion tissue ( n = 10). (A) Waterfall plot detailing somatic mutations in cancer-driver genes in lesion tissues from patients in the triad group. (B) Network plot of tumor suppressor genes in which somatic mutations were identified, showing their co-expression. (C) Waterfall plot of somatic mutations in RAS-related pathway and SWI/SNF-related genes in lesion tissues from patients in the triad group. (D, E, F, G) Lollipop plot detailing somatic mutations along the ARHGAP35, ARID1A, PBRM1, and PIK3CA genes. Previous studies emphasized the role that variants in oncogenes and tumor suppressor genes involved in the RAS pathway play in endometriosis pathophysiology ( Orr et al. 2023 ), and we found 30 RTK/Ras-related genes somatically mutated in our study ( Fig. 5C ). These included well-studied genes ARID1A and PIK3CA and several tumor suppressor genes that are infrequently linked with endometriosis, including Rho GTPase-activating protein 35 ( ARHGAP35 ) and TSC complex subunit 1 ( TSC1 ), and oncogenes EFT , MET , and SOS1 , which had deleterious variants in more than half the cohort. To further explore the downstream effects of missense mutations in one of the genes of interest ( ARID1A ), we computationally modeled the SWI/SNF complex using Schrodinger PyMOL. Of the somatic variants observed in this study, three encode missense mutations in ARID1A that have been previously reported. One of the three (rs1410136655) encodes a G-to-V mutation at position 1848, which is close to the interface between ARID1A, SMARCB1, and SMARCC2. Computational mutagenesis showed significant steric clashes resulting from this mutation, suggesting a disruption of protein folding/interaction ( Fig. 6 ). To validate this, SIFT (sequence-based) and PolyPhen2 (structure-based) analysis was conducted, yielding ‘deleterious’ and ‘probably damaging’ effects on protein function, respectively. Notably, this variant has been identified once previously, in a case of primary cutaneous CD8+ aggressive epidermotropic cytotoxic T cell lymphoma ( Bastidas Torres et al. 2022 ). The other two missense mutations, R1528Q and P550H are not located centrally in the SWI/SNF complex but are still predicted to lead to loss of function in ARID1A using SIFT and PolyPhen2. In silico modeling of somatic ARID1A missense mutations using the SWI/SNF chromatin remodeling complex (PDBID 6LTH). Of the observed ARID1A somatic variants, three missense mutations were modeled. One of the three (rs1410136655) encodes a G-to-V mutation at position 1848, which is close to the interface between ARID1A, SMARCB1, and SMARCC2. Computational mutagenesis showed steric clashes (in red) resulting from the mutation, suggesting a disruption of protein folding/interaction. To understand how somatic variants in cancer driver genes may be associated with lesion transcript expression, we took a ranked-order list of ‘medium’- and ‘high’-impact somatic variants and compared them to a ranked-order list of differentially expressed genes. We annotated based on a previous list of oncogenes ( Fig. 7A ) and tumor suppressor genes, and the interaction strengths were annotated by StringDB. Interactions between genes noted in the 10 canonical cancer-related pathways are highly prevalent in our network analysis. Most notably, the analysis highlights the widespread impact that mutations in PIK3CA , EGFR , and EP300 have on the lesion transcriptome. Network visualization of somatic mutation prevalence and differentially expressed genes in the triad cohort. (A) Somatically mutated oncogenes (blue dots) and differentially expressed genes (gray dots) in network visualization. (B, C, D) PI3K, NOTCH, and RAS pathway somatically mutated genes (clear) and StringDB-related annotated differentially expressed genes (green-red) based on log2FoldChange.

Due to the difficulty of diagnosis and extensive variation in phenotype, the prevalence and impact of endometriosis are likely underestimated. Additionally, the recurrence of lesions is typical within a patient’s lifetime, as evidenced by the fact that many who elect for surgical resection require repeat surgery ( Zanelotti & DeCherney 2017 ). Our mutation data suggesting the presence of endometriosis-derived cells in ‘unaffected’ peritoneum may help explain both diagnostic challenges and the need for repeat surgery. If occult lesion cells are present in visually normal peritoneum, then complete excision may not be possible. Furthermore, if these cells induce inflammatory changes, they may cause symptoms without generating observable lesions, thereby accounting for diagnostic challenges. Our findings are consistent with studies showing that even apparently normal peritoneal tissue can contain microscopic endometriosis epithelium, i.e. the dissemination of endometriosis tissue extends beyond frank lesions ( Gubbels et al. 2020 ). Clinical observations suggest that endometriosis is a heterogeneous disease in terms of both lesion phenotype and patient symptomatology ( Koninckx et al. 2020 ). As such, we hypothesize that genetic variations between phenotypically different endometriosis lesions may underlie individualized pathophysiology. We analyzed the mutational and transcriptional landscapes of superficial peritoneal endometriotic lesion tissue and their potential interactions. To our knowledge, this is the first systematic approach to study the transcriptome and mutational landscape of triad tissue (from peritoneum, endometrium, and lesion) in patients with endometriosis. Our reporting of cancer driver variants in superficial peritoneal lesions gives new insight into the mutational profiles of superficial peritoneal lesions and accompanying tissues thought to be involved in disease progression. It is hypothesized that the primary source of endometriotic lesions, particularly those that reside in the peritoneal cavity, is endometrium from retrograde menstrual flow ( Ólafsson et al. 2026 , preprint). In our study, lesion shared variant profiles of high and moderate impact showed clustering with peritoneal tissue, rather than eutopic or control endometrium. As outlined above, this may be a result of more widespread dissemination of endometrium-derived cells than is typically appreciated. However, it could also be a result of deleterious mutations in extracellular matrix-related proteins, such as mucins and laminin, which prime the squamous epithelial cells of the peritoneal wall for invasion, aligning with the coelomic metaplasia hypothesis for endometriosis development ( Seli et al. 2003 ). If shared variant PCA is used in our triad comparison (Fig. S5B), we find lesions aggregate in between eutopic and peritoneal tissue, implying that lesions may be comprised of a combination of invasive endometrial epithelial lining and primed peritoneal lining or that ectopic endometrium is transformed in the peritoneal environment. In previous studies, somatic mutations in ARID1A and PIK3CA have been demonstrated to mediate epithelial invasiveness in ovarian clear-cell carcinomas ( Chandler et al. 2015 ). In addition, inhibition of P300 (encoded by EP300 ) in ARID1A -mutant cells inhibits SERPINE1-mediated invasion in vitro ( Wilson et al. 2020 ). We confirm the prevalence of ARID1A in both frequency and distribution of mutations in our study of superficial peritoneal lesions with accompanying mutations in SWI/SNF-related genes PIK3CA and PBRM1 . However, coexistent deleterious EP300 mutations (4/5) were detected in endometriotic lesions with ARID1A missense mutations (Fig. S7). Thus, the therapeutic potential of P300 inhibition in preventing lesion formation must be further explored. To predict the effects of the observed missense mutations on the ARID1A structure within the chromatin remodeling complex, computational protein modeling was used. This in silico analysis predicted a novel structural basis for the loss of function that occurs in the G1848V variant. Given that this mutant has been previously documented in a rare form of T cell lymphoma, the overlap between cancer and endometriosis pathophysiology is again clear. We encourage researchers to explore the effects of this mutation on the ARID1A structure in vitro . In addition, we discovered multiple potentially deleterious mutations in a host of chromatin modification-, segregation-, and histone accessibility-related genes, such as BCOR , CREBBP , EP300 , LATS1 , KMT2A/B/C/D , NIPBL , SETD2 , and SMC1A . Furthermore, genes related to B cell, stem cell, and hematopoietic cell differentiation, such as CDK12 , NOTCH1 , and PTPRC , were enriched in our somatic mutation analysis, implying that lesion formation may partially result from dysregulation of stem cell differentiation because of injury to the peritoneal wall. Importantly, the recruitment of cytokines and leukocytes to superficial peritoneal lesions has been well documented ( Oală et al. 2024 ), and here, we concur with previous studies examining the peritoneal lesion microenvironment. Differential expression of T cell activation, cytokines, and leukocyte-related gene ontology terms suggests that in phenotypic red lesions, there is increased capacity for the recruitment of T cells, monocytes, and macrophages. GSEA Hallmark terms related to the inflammatory response and TNF-a/NFkB signaling pathways further suggest changes in cell composition and gene expression consistent with increased inflammation in red lesions. PARP15 and ERMN, suggested as prognostic markers in endometrial cancer ( Wang et al. 2020 ), are differentially expressed in red lesions compared with clear lesions. Specifically, PARP15 is significantly upregulated ( P adj = 0.00924), suggesting that it may be a therapeutic target in endometriosis patients with red lesions, but not in those with clear lesions. This proposition of course needs further interrogation, as well as functional validation. In our analysis protocol, we utilized best-practice filtering pipelines to ensure that common variants called in both our shared variant and somatic mutation pipelines were discarded. Furthermore, we highlighted exonic regions of known endometriosis-related genes and their prevalence. However, the data include the entire transcriptome that allows for detection of multiple forms of noncoding RNA and both known and unknown features. Uniquely, a significant subset of our cohort was paired, allowing for simultaneous transcriptional and mutational profiling. As a resource, these data and pipelines provide researchers with new avenues of understanding ‘tumor’ progression in endometriotic lesions and the mutational profile of this tissue. Our triad dataset structure enables several comparisons using a similar method. We performed somatic mutation calling by three different comparisons, using eutopic (normal) vs lesion (‘tumor’), peritoneum (normal) vs lesion (‘tumor’), and eutopic (normal) vs peritoneum (‘tumor’). As expected, the first comparison yielded the most pathologically relevant information about oncogene mutations in endometriosis. Our study has a few limitations. Because we study transcription at the tissue level, we can make only limited inference about intracellular signaling. We also cannot attribute the observed sequence variants to specific cell types (epithelial, stromal, inflammatory, etc.). However, this approach allows us to more comprehensively speak to the signaling microenvironment in the tissues because the contributions of all local cells (or cell processes in the case of neurons) are represented in our data. Furthermore, functional validation of our findings, including any implications of the somatic mutations in cancer-associated genes on protein expression, remains to be elucidated and is a subject of interest for future work. Our study is unique in that it not only confirms existing narratives about the significance of somatic mutations in chromatin remodeling complexes and chromatin accessibility genes in endometriotic lesion pathophysiology but also sheds light on the role that mutations and aberrant transcriptional expression may play in priming peritoneal tissue for the attachment and proliferation of endometrial cells. Considering that retrograde menstruation occurs in up to 90% of women ( Halme et al. 1984 ) but only approximately 10% of women develop endometriosis, the role of the peritoneum is a critical avenue of research. Furthermore, we demonstrate transcriptional differences that may accelerate the development of therapeutics that target an individual patient’s phenotypic lesion presentation (in this case based on color). Taken together, our data make clear the heterogeneous nature of endometriosis – recognizing this nature and working toward the development of patient-specific diagnostic – and treatment options will greatly enhance medical care for endometriosis patients.

Endometriosis is characterized by the growth of endometrial stromal and glandular tissue outside of the uterine cavity ( Sampson 1927 ). Clinically, patients present with chronic pelvic pain, dysmenorrhea, dyspareunia, gastrointestinal symptoms, and infertility, leading to diminished quality of life ( Hadfield et al. 1996 ). Treatment regimens comprise primarily laparoscopic surgery, hormonal therapy, and pain management. While endometriosis is estimated to affect approximately 10% of reproductive-age women globally ( Giudice 2010 ), the condition is significantly underdiagnosed as a result of many factors, including a lack of validated biomarkers, confusion with primary dysmenorrhea, and the invasiveness of surgical identification ( Bulletti et al. 2010 ). As a result, there is an average delay of seven to nine years between the onset of symptoms and diagnosis ( Agarwal et al. 2019 ). The disease is most commonly classified according to lesion location/organization (deep infiltrating endometriosis, endometriomas, and superficial peritoneal disease) or lesion color (clear, red, blue/black, and white). Lesions may also be staged according to the rASRM score or Enzian classification, which are prone to interobserver variability ( Buchweitz et al. 2005 , Grasso et al. 2010 , Haas et al. 2013 , Johnson et al. 2017 ). The heritability of endometriosis is approximately 50% ( Treloar et al. 1999 , Saha et al. 2015 ), and some of the genes and polymorphisms of interest have been identified by using genome-wide association studies in patients with endometriosis ( Sapkota et al. 2017 , Rahmioglu et al. 2023 ). Encouragingly, these studies have led to new non-hormonal treatment targets ( Tapmeier et al. 2021 ). The peritoneal microenvironment ( Beste et al. 2014 , Jørgensen et al. 2017 ), blood-based metabolomics, and external macro-environment have been characterized through metadata acquisition of several long-term patient survey datasets ( Vitonis et al. 2014 ). Furthermore, whole exome studies on the epithelium of lesions and eutopic endometrium from endometriosis cohorts have identified somatic mutations in cancer-related genes ( Anglesio et al. 2017 ). Among the cancer-related genes mutated in endometriosis are AT-rich interaction domain 1A ( ARID1A ), a key component of the switch/sucrose non-fermentable SWI/SNF family of chromatin remodeling complexes and phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha ( PIK3CA ) ( Anglesio et al. 2017 ). Somatic mutations in these two genes have been found in various cancer diagnoses, including gynecological cancers ( Alqahtani et al. 2019 , Mullen et al. 2021 ), raising significant interest in the relationship between dysregulation in chromatin remodeling complexes and the pathogenic proliferation of endometriotic tissue. An integrated approach toward SNP/Indel analysis with accompanying transcriptomic analysis has yet to be applied to investigate endometriosis. RNA sequencing has been instrumental in revealing new disease biology, biomarkers, and pathways ( Ozsolak & Milos 2011 ), particularly when combined with existing genomics, proteomics, metabolomics, and phenomics data; such an approach has proved to be incredibly useful in cancer models ( Vasaikar et al. 2018 ). We hypothesized that pairing transcriptome mutation profiling with differential transcriptional expression analysis may give new insights into the etiology and development of superficial peritoneal endometriotic lesions in addition to more associations with the biological pathways involved ( Fig. 1 ). Through bulk RNA sequencing of tissue from 20 endometriosis patients and 5 control patients, we were able to identify somatic mutations in cancer-driver genes, such as ARID1A, KRAS, PIK3CA, ARHGAP35, and PBRM1 , that may be critical to endometriosis pathophysiology. Our observed conservation of shared cancer gene variants in the peritoneum and lesion tissue of endometriosis patients without cancer suggests that endometriotic tissue may spread beyond visible lesions themselves. Finally, novel transcriptional differences between endometriotic lesions were discovered based on lesion color. To synthesize somatic variant data with differential expression, network analysis was conducted, which highlighted the influence that mutations in oncogenes PIK3CA , EGFR , and EP300 have on lesion transcriptional expression. Summary of sample collection, sequencing, and analysis pipelines. All 51 samples from 25 patients (20 endometriosis and 5 control) were used for shared variant calling, while only 10 triad patients (each with eutopic endometrium, lesion, and peritoneum tissue) were used for somatic variant calling and differential gene expression analysis. Created using BioRender.

The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

JYH wrote the manuscript, generated figures, performed data analysis, and was involved in conception and design of the study. SR and RSM were involved in writing and editing the manuscript and generating the figures. LL conceptualized the manuscript, performed the data analysis, prepared the figures, and drafted the manuscript. AM contributed to the study design, obtaining informed consent from patients, and sample collection and processing. INW consented patients, surgically excised the tissue, and isolated tissue for further evaluation. MLS was involved in patient recruitment and consenting, sample processing, database management, and patient demographic data collection. AL contributed to data collection and tissue sample collection, preparation, processing, and categorization. ERD, RMG, and ASM performed experiments, data analysis and manuscript editing. JC contributed to computational analysis and data interpretation. ANY contributed to study design and manuscript editing. BJQ, MRL, MDH, and MSR critically revised and edited the manuscript and contributed to the interpretation of the results. GMC and RMA were involved in the conception and design of the study, critically revising and editing the manuscript, and the interpretation of the results.