Mutation and methylation profiles of ectopic and eutopic endometrial tissues

This study included a cohort of 25 women with adenomyosis and/or endometriosis who had hysterectomies at the Johns Hopkins Hospital between 2014–2018. The clinicopathological features of all cases are summarized in Table 1 . Of the 25 women included in the study, 16 (64%) had concomitant eutopic endometrium, adenomyosis and endometriosis while 9 (36%) had eutopic endometrium and adenomyosis. The women included in this study had a median age of 43 years (range: 36 to 61 years; mean ± SD: 43 ± 5.1 years). After examining the hysterectomy specimens, adenomyosis was diffuse in 20 (80%) cases and the remaining 5 had adenomyomas, representative gross and microscopic images of diffuse adenomyosis is shown in supplementary material , Figure S1 . The distribution of peritoneal endometriosis included Fallopian tube/paratubal soft tissue, ovary, pelvic surface, uterine serosa, omentum, and appendix. All 20 cases (patients) studied for WES had normal-appearing endometrium and adenomyosis, and among them, 16 also had peritoneal endometriosis. Four remaining cases, 3, 17, 18, and 19 did not have endometriosis lesions available for study. Glandular epithelium and stroma within each sample were enriched by LCM separately ( Figure 1A ), and all microdissected specimens were visualized under the LCM microscope for quality control to ensure complete separation ( Figure 1B ). In all, we analyzed 111 LCM samples for WES. The average distinct exome coverage was reported in supplementary material , Table S1 . Within this study cohort, we identified a total of 214 somatic mutations (200 SNVs, 12 deletions, and 2 insertions). We found that all groups except normal-appearing endometrial epithelium had a high mutant allele frequency ( Figure 1C ). All mutations are presented in supplementary material , Table S2 and are schematically listed according to specimens in supplementary material , Figure S2 . The mutant allele frequency in normal-appearing endometrial epithelium was significantly lower than any other groups (p<0.0001, ANOVA). All 20 cases except one (case 17) had somatic mutations, and 62 (56%) of 111 LCM samples examined harbored somatic mutations including 11 canonical cancer driver gene mutations: 6 in adenomyosis epithelium, 3 in endometriosis epithelium, and 2 in endometrial epithelium ( Table 2 ). None of the 55 LCM stroma samples contained driver mutations. Thus, we detected mutations in cancer driver genes only in epithelium. Based on the inference of somatic mutations for clonal development, we found that the many of LCM samples from normal-appearing endometrium, adenomyosis, and endometriosis were clonal, especially in epithelial components of individual tissue types ( supplementary material , Table S3 ). Somatic mutations were seen in 146 unique genes including 8 different cancer driver genes and 138 different passenger genes ( supplementary material , Table S2 ). The 8 mutated cancer driver genes ( FBXW7 , CDH4 , ERBB3 , EYS , BRCA1 , CTNND1 , ARHGAP35 , and PIK3R1 ) were exclusively detected in epithelium but not stroma in cases 6, 8, 9, 10, 11, 13, and 20 ( supplementary material , Table S2 ). Only two recurrent mutated cancer driver genes were detected; FBXW7 was present in adenomyosis epithelium in case 6 and endometriosis epithelium in case 20, and ARHGAP35 was detected in normal-appearing epithelium in both case 6 and case 13. The presence of cancer driver mutations was not associated with any clinicopathological features on the basis of a small sample size ( Table 1 ). The percentage of somatic mutations among the 20 analyzed cases was higher in epithelium than in stroma. In any given case, if stroma was positive for mutation, its corresponding epithelium was also positive for mutation, but not vice versa, except in normal-appearing endometrial stroma in case 9 ( supplementary material , Table S3 ). Next, we inferred the clonal relationships among endometrium, adenomyosis, and peritoneal endometriosis based on the presence or absence of shared somatic mutations. We separated glandular epithelium and stroma in individual cases for phylogenetic analysis. Among the 20 cases for which WES was performed, there were a total of 15 informative cases in which somatic mutations were found in at least two of the LCM samples from the same epithelial or stromal components of endometrium, adenomyosis, and endometriosis. The remaining 5 cases (5, 8, 9, 17, and 18) contained either no mutations or mutations only in endometrium, adenomyosis, or endometriosis, and were, therefore, not informative ( supplementary material , Table S3 ). None of the 15 informative cases had the same set of mutations in any of the two LCM samples. Thus, phylogenetic analysis was completed for these 15 cases ( Figure 2 ). Ten of the 15 cases demonstrated at least one shared mutation in epithelium from normal-appearing endometrium, adenomyosis, and endometriosis, suggesting that they were derived from the same progenitor cell, while private mutations were evident in individual samples. In the remaining 5 cases (cases 1, 10, 12, 13, and 20), each showed a distinct mutation pattern without any shared mutations in epithelium from normal-appearing endometrium, adenomyosis, and endometriosis. As there were fewer somatic mutations in the stroma, there were only 8 informative cases. We performed MethylationEPIC analysis on DNA extracted from each of the 66 FFPE tissue samples including epithelial and stromal components from 15 pairs of adenomyosis and normal-appearing endometrium together with 3 endometriosis lesions in women from whom sufficient DNA was available. In the remaining cases, the amount of DNA remaining after WES was insufficient for methylation analysis. In the 15 cases analyzed, β-value distributions for all probes indicated that the methylation data was of suitable quality for downstream analyses ( Figure 3A ). Unsupervised analysis identified the most significant difference in comparison between epithelium and stroma ( supplementary material , Figure S3A ). The principal component analysis also showed a complete separation of epithelium and stroma ( Figure 3B ). The clean separation excludes the possibility of significant cross contamination of epithelial and stromal cells during LCM or DNA contamination during extraction. If there was a significant mixture of epithelial and stromal component, then the heatmap would not be as distinct between epithelial and stromal profiles. Significant differences were observed among individuals ( Figure 3C , supplementary material , Figure S3B ). In cases 12, 13 and 16, in which methylation data from adenomyosis, endometriosis, and endometrium were all available for analysis, they showed case-specific co-clustering. Epithelial or stromal components from adenomyosis, endometriosis, and normal-appearing endometrium from one patient were much more closely related than in comparison to other patients ( Figure 3C ). Next, we compared different sample locations including after correcting for tissue type (epithelium vs. stroma) and individual differences. We found significant differential methylation between adenomyosis and normal-appearing endometrium, in both epithelium and stroma ( Figure 3D , E ). Epithelial samples divided cleanly into normal-appearing endometrium versus lesions, though the distinctions were not as evident as in epithelium versus stroma ( Figure 3D and supplementary material , Figure S3C ). Stromal samples exhibited clearer differences between conditions, though some cases clustered with normal-appearing endometrium ( Figure 3E and supplementary material , Figure S3D ). Moreover, we found that the methylation pattern in epithelial and, to a lesser degree, stromal of adenomyosis was more heterogeneous and less defined as one group than that in normal-appearing endometrium epithelium (p< 0.05) ( Figure 3F , G ). We observed no correlation among mutation, methylation patterns and clinico-pathological features. In the absence of independent test samples, we employed a leave-one-out cross-validation procedure to verify the finding. The methylation profile of each left-out epithelial sample was compared to all other epithelial samples with the expectation that it would most closely resemble those of the same type. As shown in Figure 4 , each of the four rows correspond to the tissue type of the test sample, and columns, consisting of two adjacent box plots, correspond to cases (patients). Each pair of boxes shows the similarity measured by Spearman, rank-based correlation between the left-out sample and all other adenomyosis (red) and normal-appearing endometrium (blue) of the same class. In the majority of cases, left-out adenomyotic tissue matched the other adenomyotic tissues (red) better than it matched normal-appearing tissues.

This retrospective study was approved by the Institutional Review Boards of Johns Hopkins Medical Institutions (Baltimore, MD, USA). Formalin-fixed paraffin-embedded (FFPE) tissue blocks were selected from patients undergoing hysterectomy between 2014 and 2018. The case selection was based on the availability of eutopic endometrium, adenomyosis with or without endometriosis from the same women without gynecologic neoplastic or precancerous diseases. As a result, 25 hysterectomy specimens were retrieved and the diagnoses were validated by independent gynecologic pathologists using previously published criteria [ 23 , 24 ]. Adenomyosis was diagnosed on the basis of the distance between the endo-myometrium junction and the adenomyosis, exceeding the diameter of a 100x microscopic field (about 2 mm). To determine somatic sequence mutations, we also sampled and analyzed the matched normal smooth muscle, Fallopian tube, or colonic epithelium. The samples from cases 1–20 were submitted for whole exome sequencing (WES). Among these, cases 1–3, 6, 9, 12, 13, 16, 17, and 19, together with additional cases 21–25 were submitted for global DNA methylation profiling. Laser capture microdissection (LCM), DNA extraction, and WES were performed, and these methods have been described previously [ 25 ]. The epithelium and stroma were dissected separately from normal-appearing endometrium, adenomyosis, and endometriosis. We sampled the basal layer in the eutopic endometrium and the functional layer, if needed. The histopathological features and LCM in a representative case are illustrated in Figure 1A . Control tissues were also laser capture microdissected. The control tissue is the normal tissue to help rule out germline mutations. Only the somatic variant was called a somatic mutation due to absence of a nonreference allele in the normal sample. We performed WES on genomic DNA obtained from LCM samples including adenomyosis epithelium, adenomyosis stroma, endometriosis epithelium, endometriosis stroma, endometrium epithelium, and endometrium stroma from cases 1–20, except for cases 3, 17, 18, and 19, which had paired eutopic endometrium and adenomyosis but without concurrent endometriosis by pathology. Thus, 132 DNA samples were available for assessing somatic mutations. Whole genome libraries were generated from these samples using the ThruPLEX® Tag-seq 48S Kit (Takara Bio USA Inc., Mountain View, CA, USA). Exome regions were enriched with the Agilent SureSelectXT Human All Exon V6 (Agilent Technologies Inc., Santa Clara, CA, USA). We failed to prepare a library for the endometriosis stroma of case 14. A total of 131 samples were submitted for WES using the Illumina HiSeq 4000 sequencer (Illumina Inc., San Diego, CA, USA) with 150PE. Details for phylogenetic analysis are provided in Supplementary materials and methods . The paired-end sequencing raw data (FASTQ format) were aligned to the human reference genome (hg19) using the Burrows-Wheeler aligner software (BWA, v0.7.15). Candidate somatic mutations were detected by comparing sequencing data from lesions with matched eutopic tissues using the MuTect1 and Strelka for single-nucleotide variants (SNV) and InDels, respectively. All candidate somatic mutations were validated by visual inspection using the Integrated Genome Viewer (IGV) [ 26 ]. Somatic mutations in the test samples were defined according to the following criteria: (1) the mutation identified in six or more distinct reads; the mutation frequency more than 10%, and the mutation depth greater than 15 reads, (2) the mutation frequency in the control sample is less than 2% and the mutation not present in any of the reads in the matched control sample or in the other 19 control samples analyzed in this study to rule out germline mutations, and (3) the mutation uncommon in dbSNPs (frequency < 1%). Both non-synonymous and synonymous mutations were reported. Putative driver mutations were determined based on previously listed cancer driver genes [ 27 ]. We analyzed global DNA methylation in 15 cases including cases 1–3, 6, 9, 12, 13, 16, 17, 19, and 21–25. Since some of these turned out to have insufficient amounts of DNA for the assay after WES, we identified 66 LCM DNA samples qualified for DNA methylation analysis using the MethylationEPI BeadChip platform. The EZ DNA methylation kit (Zymo Research Corp., Irvine, CA, USA) was used to bisulfite-convert 50 ng of DNA according to the manufacturer’s protocol. Infinium FFPE DNA restoration kit (Illumina Inc.) was used to restore the EPIC assay Bisulfite-treated (BST) DNA to amplifiable length according to the manufacturer’s manual. After removing nonhybridized and nonspecifically-hybridized DNA, the captured DNA served as a template for single-base extension on the BeadChip incorporating fluorescence-labeled dNTPs. Additional details for methylation analysis are provided in Supplementary materials and methods .

Whole exome sequencing and genome-wide methylation analyses [ 28 ] have become effective tools to study pathogenesis, especially when only formalin-fixed and paraffin-embedded tissues are available. In this study, we employed both methods to address one of the most fundamental questions pertaining to the pathogenesis of adenomyosis and peritoneal endometriosis, i.e., whether both lesions are clonal in origin, and if so, how are they related to the corresponding eutopic endometrium. These seemingly straightforward questions are difficult to answer without separately analyzing the glandular epithelium and stromal cells which constitute eutopic endometrium, adenomyosis, and endometriosis. Analyzing both epithelial and stromal components together, as has been a common practice, may not be informative because the data obtained are the average of molecular changes in epithelium and stroma, whose proportions vary in individual samples. This limitation prompted us to embark on this study. First, we demonstrated for the first time, like epithelial cells, stromal cells surrounding glands in adenomyosis and peritoneal endometriosis can be of clonal origin because of somatic mutations detected. Likewise, as for eutopic endometrium, which has recently been reported to carry cancer-driving mutations in individual glands [ 29 – 34 ], we found somatic mutations in stroma alongside glandular epithelium. The lack of mutations detected in the stroma from some of the cases may have two explanations. One is the limitation of exome sequencing, rather than whole genome sequencing, employed in this study. The other is since unlike glandular epithelial cells that form structurally well-defined individual glands, stromal cells lack architectural compartmentation, we may have microdissected stromal areas spanning multiple clonal patches, and therefore individual mutant allele frequencies were “diluted” to a low level beyond detection. The patches have not been well defined but may be from the perivascular regions [ 35 ]. This study provides the first unbiased analysis demonstrating that cancer-driver mutations only occur in epithelium, a result suggesting that epithelium as compared to stroma is prone to cancer-driver gene mutations. This view resonates with the fact that endometrial carcinomas [ 36 ] but not sarcoma are characterized by those mutations. The relatively high mutant allele frequency in both epithelium and stroma in adenomyosis and peritoneal endometriosis strongly suggests that the same mutations involve almost all the glandular epithelial or stromal cells rather than a small subpopulation, suggesting that the mutations occurred very early during lesion development, i.e., clonal expansion of a stem cell with mutations that establishes the lesion, so the progeny cells harbor the same mutations. In contrast, the mutant allele frequency in epithelium from eutopic endometrium was significantly lower than in adenomyosis and peritoneal endometriosis ( Figure 1C ). This is likely because we microdissected and analyzed aggregates of multiple glands, and these adjacent individual glands may be clonally distinct [ 30 , 31 ]. Accordingly, the endometrial glands we analyzed might represent a composite genomic landscape attributable to clonally distinct individual endometrial glands. From this perspective, both adenomyosis and peritoneal endometriosis exhibit relative genetic homogeneity. Surprisingly, the stromal component is also clonally derived and appears to coevolve with the glandular epithelium. Also, the relatively high MAF (mutant allele frequency) of detected mutations could help us exclude substantial tissue type contamination since the MAF in some of the mutations is high in epithelial cells but is almost zero in stroma. Theoretically, the immune cells and endothelial cells in stroma may “dilute” the MAF of stromal cell mutations to a level below our detection cutoff but we do detect somatic mutations in the stromal component, ruling out a significant contamination by immune and endothelial cells. To infer the clonal development history, we applied phylogenetic analysis based on selection-neutral mutations including synonymous and missense passenger mutations that are unlikely to result in changes of fitness. Among informative cases, many adenomyosis and peritoneal endometriosis lesions likely originate from the same founder epithelial or stromal progenitor cells as eutopic endometrium, on a basis that they shared the same unselected mutations. Intriguingly, we noted several occurrences of concordant mutations in both epithelium and stroma in 7 adenomyosis samples, 4 peritoneal endometriosis samples, and 8 eutopic endometrial samples. This is unlikely coincidental as the probability of the same somatic mutation (non-hot spot mutation) occurring simultaneously in epithelial and stromal components of a single tissue is estimated only 3×10 −7 when considering 30 million base pairs in human exomes. It can be argued that the mutations detected in both epithelium and stroma are an artifact from cross contamination during LCM or sample preparation. However, a perfect separation of epithelium and stroma in our unsupervised methylation profiling did not support this argument ( Figure 3B and supplementary material , Figure S3A ). We cannot exclude the possibility that the shared mutations are in fact germline variations which were technically undetected in control tissues. If we can exclude that potential artifact, one plausible explanation remained is that both epithelial and stromal progenitor cells descended from the same multipotent Müllerian stem cell before their differentiation into endometrial epithelial and stromal lineages. After leaving the endometrium, the theoretical stem cells develop to epithelial and mesenchymal lineage, and the lineage diversion may be driven by the epigenetic and transcriptional reprogramming. Then these progenitor cells develop into adenomyosis in myometrium and peritoneal endometriosis since they acquire their own private mutations; therefore, their clonal trajectories diverge from eutopic endometrium early on [ 32 ]. This theory of stem cell dissemination may help explain reported cases of fetal and prepubescent endometriosis and adenomyosis [ 37 – 39 ]. Our data demonstrate that adenomyosis, can be a genetically and epigenetically distinct entity, especially for those lesions located deep into the myometrium. The mechanism by which clonally defined glands and associated stroma from the basal layer of endometrium grow together into myometrium remains speculative. As compared to other adenomyosis cases, case 14 is unique, as adenomyosis and eutopic endometrium, either epithelium or stroma, shared almost all truncal mutations without branches, meaning that they were more related than seen in other cases. Histologic examination of this case showed a rather superficial adenomyosis within 20% of myometrial thickness, likely extending directly from the overlying basal endometrium where the LCM was performed. This finding supports the view that direct ingrowth from the horizontally inter-connected endometrial glands and associated stroma in basalis [ 40 ] results in early adenomyosis [ 41 ]. Those epithelial and stromal progenitor cells in adenomyosis then develop independently from the adjacent endometrium basalis once they reside in the myometrium, similar to the development of peritoneal endometriosis, as proposed in a previous study [ 42 ]. In a recent report [ 43 ], Filby et al . found the endometrial epithelial progenitors and perivascular mesenchymal stem cells residing in the basalis are highly clonogenic, and have multilineage (mesodermal and decidual) differentiation potential. These findings support our view. While complete separation between epithelium and stroma is expected, the difference in methylation patterns among tissue types (adenomyosis and endometriosis versus eutopic endometrium) was less apparent than the differences among individual cases. There are two implications. First, the global methylation pattern of either adenomyosis or endometriosis resembles its matched normal-appearing eutopic endometrium despite their aberrant locations, a finding consistent with a previous report [ 44 ]; and second, individual women have an idiosyncratic methylation signature in endometrium under the same hormonal exposure, regardless of its status being eutopic or ectopic. More importantly, despite the genome-wide similarity between a given lesion and its corresponding eutopic endometrium, after separating epithelium and stroma, and adjusting for individual differences, we were still able to detect significant focal differences in methylation patterns between adenomyosis/endometriosis and eutopic endometrium. This observation is consistent with a previous study [ 45 ] demonstrating difference between eutopic and endometriotic endometrium stroma, suggesting that adenomyosis and endometriosis are epigenetically different from eutopic endometrium. More interestingly, the methylation pattern of adenomyosis epithelium is more heterogeneous than other tissue groups except endometriosis of which the sample size is too small to compare. In this study, we did not analyze eutopic endometrium in the absence of adenomyosis and a large case control study is needed to determine if there is a molecular difference between eutopic endometrial tissues from patients with adenomyosis and those without [ 46 ]. Another limitation is that we did not map the whole eutopic endometrium given technical and financial challenges. To do that, we have to submit at least 20 sections that comprehensively represent the entire endometrium and perform a relatively large number of individual molecular analyses for only one uterus. Likewise, a large study is required to determine the association between molecular alterations (somatic mutations and methylation patterns) and clinicopathological features. Although we applied a leave-one-out cross-validation procedure for validation since this method is widely used when an independent validation cohort is not available, a future independent study is required to validate our methylation profiles. Nevertheless, this study provides compelling evidence of clonal derivation of adenomyosis and peritoneal endometriosis from normal-appearing eutopic endometrium, and provides new insight into the role of endometrial stromal cells in the pathobiology of endometrium-related disorders.

Adenomyosis and endometriosis are two major gynecologic disorders causing chronic pelvic pain, dysmenorrhea, dyspareunia and infertility [ 1 – 4 ]. Worldwide, adenomyosis and endometriosis are estimated to affect, respectively, 10–80% and 10% of premenopausal women [ 1 ], and their prevalence is higher in women with symptoms such as infertility and chronic pelvic pain [ 5 – 8 ]. Both disorders are sources of significant socio-economic burden not only in the expense of medical care but in the loss of work productivity [ 9 , 10 ]. Adenomyosis is defined as the presence of endometrial tissue within the myometrium whereas endometriosis is the presence of endometrial tissue residing outside the uterus [ 11 , 12 ]. According to the anatomic locations, endometriosis can be further classified into superficial endometriosis on the peritoneal surface (“peritoneal endometriosis”), deep infiltrating endometriosis penetrating into the musculature of pelvic organs, and ovarian endometriotic cyst. Despite various types, these ectopic endometrial tissues are histologically difficult to distinguish from eutopic endometrium where endometrial glands are surrounded by a variable amount of stromal, endothelium, and immune cells. Biologically, both diseases represent an unusual pathologic condition in humans. Their aberrant location prevents them to be shed during menstruation and is thought to cause the disease phenotypes. From this perspective, the two estrogen-dependent diseases, adenomyosis and endometriosis, appear to be closely related [ 13 , 14 ]. Adenomyosis co-exists in 20–50% of patient with endometriosis [ 15 – 17 ], and disease co-occurrence is more commonly seen in patients with severe endometriosis (American Society for Reproductive Medicine stage IV) [ 18 ]. Despite their similarities, certain clinical characteristics set these two disease entities apart. Endometriosis is associated with younger age, nulliparity, and having a mother or sister with endometriosis. On the other hand, adenomyosis is usually diagnosed later in life, occasionally in early life [ 19 ], probably due to the fact that hysterectomy is usually not performed in child-bearing ages and hysterectomy is the standard to confirm the diagnosis. Despite being two of the most common gynecologic disorders, only a few molecular genetic studies have analyzed either endometriosis or adenomyosis at a genome-wide scale [ 20 , 21 ]. These studies focused either on studying lesions as a whole (combining epithelium and stroma) or only on analyzing the epithelium [ 22 ]. Given the importance of stromal components in co-developing adenomyosis and endometriosis with glandular epithelium, we have carefully separated epithelium and stroma using laser capture microdissection and applied whole exome sequencing to determine whether the stroma in both eutopic and ectopic endometrium, like its epithelial counterpart, is of clonal origin. Based on somatic mutations, we inferred the phylogenetic development of adenomyosis and endometriosis from eutopic endometrium from both epithelial and stromal perspectives. We further employed global methylation profiling to compare the epigenetic landscapes of epithelial and stromal components in adenomyosis and corresponding eutopic endometrium by separating glandular epithelium and stroma. Our results provide new insight into complex evolutionary trajectories in the origin of adenomyosis and peritoneal endometriosis.

Figure S1. The gross and microscopic feature of a diffuse adenomyosis involving a uterus Figure S2. List of somatic mutations in all LCM samples Figure S3. Global methylation profiles in different groups of samples analyzed Table S1: Sequence Analysis Summary of Whole-Exome Sequencing Table S2. Somatic mutations in 20 cases Table S3. Clonal status in different lesions from all cases