Genomic alterations in ovarian endometriosis and subsequently diagnosed ovarian carcinoma

Endometriosis, a benign disease that affects 10–18% of young fertile women ( Giudice, 2010 ; Zondervan et al. , 2018 ), may transform into cancer in the ovaries later in life. There is substantial epidemiologic evidence supporting a link between ovarian endometriosis (OE) and epithelial ovarian carcinoma (OC). Women with endometriosis have a lifetime risk of OC of 2–3% compared to a risk of 1–1.5% in the overall female population ( Pearce et al. , 2012 ; Hermens et al. , 2020 ). There are no known predictive molecular markers or signs indicating which women with endometriosis are at risk of developing OC. If they were known, active surveillance or prophylactic surgery could be indicated to save lives. Endometrioid ovarian carcinoma (E-OC) and clear cell ovarian carcinoma (CC-OC) are the most common endometriosis-associated ovarian carcinomas (EAOCs). The genomic landscape of endometriosis and EAOC has been explored in recent years. Mechanisms involved in endometriosis development include epigenetic defects, chronic inflammation, and aberrations in the regulation of estrogen and progesterone receptor pathways, with complex interactions between stromal, inflammatory, and epithelial cells ( Bulun et al. , 2019 ; Mari-Alexandre et al. , 2019 ). Several findings of known cancer-driving mutations in deep infiltrating endometriosis (DIE), OE, and iatrogenic endometriosis with benign histology have been reported ( Anglesio et al. , 2015 , 2017 ; Lac et al. , 2018 ; Suda et al. , 2018 ). Frequently affected genes in endometriosis include ARID1A , PIK3CA , KRAS , PTEN , and TP53 , which are also frequently mutated in EAOC ( Kuo et al. , 2009 ; Wiegand et al. , 2010 ; Lu et al. , 2015 ; Murakami et al. , 2017 ). Endometriosis is defined as the presence of endometrial stroma and epithelium outside of the uterine corpus. The lesions are most typically located in the pelvis area, on the peritoneal surface or infiltrating nearby organs, but can be found throughout the abdominal cavity and at distant sites such as the lungs. OE, DIE, adenomyosis, and peritoneal endometriosis are suggested to be distinct subtypes. Endometriosis lesion in the vicinity of, or in continuity with, EAOC is a common histopathologic finding, in which shared somatic mutations have been demonstrated ( Sato et al. , 2000 ; Anglesio et al. , 2015 ; Er et al. , 2016 ; Wiegand et al. , 2010 ; Yamamoto et al. , 2011 ). Multiple mutations in cancer-associated genes in endometrial (eutopic) tissue samples and multifocal endometriosis (ectopic) lesions have previously been described, showing that KRAS and PIK3CA mutations are shared across multiple regions of endometriosis but not between eutopic and ectopic endometrial tissue of individual subjects ( Suda et al. , 2018 ). Conserved mutations in cancer-associated genes have also been shown between epithelial endometrial, endometriotic, and cancer cells derived from a single ancestral clone ( Suda et al. , 2020 ). Based on these findings, it is appealing to imagine a shared mutational lineage between endometrial tissue, endometriotic lesions, and possibly metachronous EAOC. Investigations on EAOC cases and endometriosis prior to cancer diagnosis have not yet been conducted and may shed light on causative mechanisms. In this study, we have explored the mutational and copy number landscapes in EAOC and preceding specimens of OE in paired samples from 11 patients using archival material, macro- and laser capture micro dissection, and whole-exome sequencing.

Eleven women with surgically removed OE who later developed OC were included in the present study. Seven developed E-OC and four developed CC-OC. The median age at diagnosis of OE and OC was 42 and 54 years, respectively ( Supplementary Table S2 ). There was no difference in age distribution between the patients at the time of OE (Mann–Whitney, P  = 0.315) or OC diagnosis (Mann–Whitney, P  = 0.649) considering the two histotypes. The median time between OE and E-OC was 10 years (range, 4–27) and CC-OC 9 years (range, 9–30). There was no correlation between age at OE and time to cancer diagnosis (Spearman’s rho –0.321, P  = 0.482). Four patients had bilateral OE, and of them, one patient developed E-OC in both ovaries. Three patients had OE and OC on opposite sides, of which two underwent unilateral salpingo-oophorectomy at the time of OE surgery (E-OC and CC-OC), and the third patient developed an ultramutated E-OC. Laterality of the OE was unknown for four patients ( Supplementary Table S2 ). Whole-exome sequencing of the 11 OE samples and their matched germline control was performed for the identification of endometriosis-associated somatic mutations. The median sequencing depth was 20 reads per identified variant (range, 5–726). A total of 4248 somatic mutations were detected with a median of 204 mutations per sample (range, 78–1453) or 22.6 (range, 5.5–281.8) somatic mutations per sequenced Mb (mut/mb) ( Supplementary Fig. S2 ). The median fraction of non-silent mutations per sample was 15% (3.5–26), while 15% (10–40) and 69% (33–87) were silent or located outside transcribed region, respectively ( Fig. 2A ). In all, 763 non-silent mutations were identified in 624 genes ( Supplementary Table S3 ). This corresponded to a median of 2.8 non-silent mut/Mb per sample (range, 0.2–73.9) with a median variant allele frequency (VAF) of 0.209 (range, 0.18–0.38). There was no correlation between age at OE diagnosis and the total mut/Mb (Spearman’s rho –0.409, P  = 0.212) or non-silent mut/Mb in cancer-associated genes (Spearman’s rho –0.482, P  = 0.133). Median VAF was ≥0.2 in 89.5% of the variants ( Fig. 2B ). For endometriosis that later developed into E-OC and CC-OC, the median VAF was 0.29 (range, 0.27–0.44) and 0.30 (range, 0.27–0.32), respectively. The median VAF showed a positive correlation with time to OC diagnosis (Spearman’s rho 0.615, P  < 0.05). The relationship was emphasized when including the VAFs for non-silent mutations only ( Fig. 2C , Spearman’s rho 0.853, P  < 0.001). The time between diagnoses showed a trend for positive correlation with cancer-associated mut/Mb that did not reach statistical significance (Spearman’s rho 0.550, P  = 0.079), and a corresponding sub-analysis of non-silent mutations showed a similar result (Spearman’s rho 0.495, P  = 0.121). There was a positive correlation between mut/Mb and number of individual biopsies with OE locus per specimen (Spearman’s rho 0.641, P  < 0.05), but not considering the number of spatially separate loci (Spearman’s rho 0.344, P  = 0.301). Hypermutated regions, kataegis, were observed in OE-03 and OE-05 ( Supplementary Fig. S3A and B ). They carried mutations possibly affecting apolipoprotein B mRNA editing enzyme catalytic polypeptide ( APOBEC ) genes, with an intronic variant in OE-03 ( APOBECC3F c.567-50G>A) and an intergenic variant in OE-05 ( RP11 -444J21.2- APOBEC1 n.7781293T>C). These two samples also demonstrated a significantly high median VAF (median 0.36 and 0.44) ( Fig. 2B ). Evaluation of the mutational landscape of ovarian endometriosis by whole-exome sequencing. ( A ) Distribution of SNVs per Mb, separated by impact. Data were dichotomized by median VAF (≤0.353). ( B ) All mutations detected per specimen, horizontal line marks median VAF. Symbols annotate the non-silent SNV in the cancer-associated gene with the highest VAF. The VAF of OE-03 was higher than that of all samples except OE-05, and OE-05 had the significantly highest VAF (Kruskal–Wallis, * P  < 0.05, ** P < 0.001). ( C ) Correlation between median VAF of non-silent mutations, and time between OE and OC diagnoses. Mb: megabase; OE: ovarian endometriosis; SNV: single-nucleotide variant; VAF: variant allele frequency; Y: year. Twenty-nine (29) genes were identified as affected by non-silent mutations in more than one patient ( Fig. 3A ). Eight of the samples displayed one or more non-silent mutations in a cancer-associated gene ( Fig. 3B ). In all, 136 mutations, corresponding to 37 cancer-associated genes were observed ( Supplementary Table S3 ). This included a recurrent mutation in KRAS , represented by two likely pathogenic mutations, c.35G>C and c.35G>A. Of the 37 mutated cancer-associated genes, 17 (46%) were associated with maintenance of genome integrity such as DNA damage, methylation and chromosomal rearrangement ( Fig. 3B ), and six samples displayed mutations likely to contribute to genomic instability ( Supplementary Table S4 ). Five samples harbored mutations affecting the MAPK signaling pathway, including mutations in MOS and CHUK (OE-03), FGFR4 (OE-05), KRAS (OE-06, OE-07), CACNA1B (OE-07), MAP3K13 , and PPP3CA (OE-09) ( Supplementary Table S3 ). A comparison with previously published exome-sequencing data of endometriosis, without any known association with cancer, showed that 52% of the mutated genes in the present study were mutated in OE ( Li et al. , 2014 ), while only three mutated genes were shared with endometriosis from other sites ( Anglesio et al. , 2017 ) ( Fig. 3C ). Of the cancer-associated genes identified in the current study, 68% (25/37) were also mutated in OE ( Fig. 3D ). A sub-analysis comparison to genes that were found mutated in >10% (2/17) of the samples in the external OE cohort showed that 47% (17/36) of the cancer-associated genes in our dataset also harbored a mutation in the dataset from OE in women without any known cancer association ( Li et al. , 2014 ). The corresponding comparison with data from non-OE showed that only two of the cancer-associated genes overlapped with OE: one from each data set. Non-silent mutations in ovarian endometriosis. ( A ) Genes harboring a mutation in more than one sample. Mutated genes assessed per sample and displayed with the histology of subsequent malignancy. In cases where the same basepair is altered in more than one specimen, this is marked in the figure. Additionally displayed is the reported mutation-frequency in COSMIC (left-hand side-panel). Color and density indicate the VAF of each somatic mutation. Reported spurious genes are excluded. ( B ) Mutated cancer-associated genes per sample. Patients ordered by histology of subsequent malignancy, OC-E (first seven columns) OC-CC (last four columns). Marked in the figure are the number of mutations according to low, medium, and high. Also shown is time between the two diagnoses and age at diagnosis. Overlapping functional annotations from GSEA are displayed (right-hand side-panel). ( C ) Venn diagram, comparison of the current dataset with previously published data. OE (Li et al. ) and endometriosis in other loci (Anglesio et al. ). ( D ) Equivalent comparison of cancer-associated genes. The number of overlapping genes is displayed together with the percentage within parenthesis. Percentage was calculated from the cohort with the least mutated genes. Chr: chromosome; COSMIC: catalogue of somatic mutations in cancer; N: No; OC-CC: clear cell ovarian carcinoma; OC-E: endometrioid ovarian carcinoma; OE: ovarian endometriosis; VAF: variant allele frequency; Y: Yes; (Y): time in years. Whole-exome sequencing of the 11 OC samples and their matched germline control was performed for identification of cancer-associated somatic mutations. The median read depth was 21 (range, 5–633). A total of 33 232 somatic mutations, corresponding to 6909 affected genes were detected ( Supplementary Fig. S4 ). The median number of detected mutations per OC sample was 501 (range, 6–27 722). Two OCs were excluded from the statistical analysis and calculations; E-OC-10 had an excessive number of mutations, 27 722, of which 9666 were non-silent, and CC-OC-04 had an insufficient coverage of 1944 Mb at cutoff and only six mutations were identified, of which two were non-silent. After this exclusion, the range was 45–2097 mutations per sample, corresponding to a median of 32.9 mut/Mb (range, 3.95–328.9). Considering all mutations, 95.8% had a VAF ≥0.2, with a median VAF per sample of 0.294 (range, 0.22–0.44) ( Table 1 ) and a median VAF for non-silent mutations of 0.269 (range, 0.18–0.42). All OCs except E-OC-03 displayed an above-median VAF mutation in a cancer-associated gene ( Fig. 4A ). In all, 9996 non-silent mutations were identified in 6561 genes ( Supplementary Table S4 ). A median of 20% (range, 5–36%) of the mutations per sample were identified as non-silent ( Fig. 4B ), corresponding to a median number of non-silent mutations of 63 (range, 2–9666), or range of 3–295, after exclusion as above. The latter corresponded to a median of 4.8 mutations per Mb (range, 0.3–119.9). Characterization of mutations in endometriosis-associated ovarian carcinoma. ( A ) All mutations detected per specimen; the horizontal line marks the median VAF. Symbols annotate the non-silent SNV in the cancer-associated gene with the highest VAF. ( B ) Distribution of SNVs per Mb, separated by impact and location. Data were dichotomized by median VAF (≤0.402). ( C ) Heatmap of the 52 cancer-associated genes mutated in the 11 OC specimens. Color densities indicate the VAF of each somatic mutation. The hypermutated E-OC-10 was considered separately, and subsequently included for a complete data set. Horizontal demarcation indicates the six genes mutated in more than one sample, not including the contribution of E-OC-10. The asterisk (*) indicates genes expected to be affected by mutations in EAOC. EAOC: endometriosis-associated ovarian carcinomas; N: No; OC-CC: clear cell ovarian carcinoma; OC-E: endometrioid ovarian carcinoma; SNV: single-nucleotide variant; VAF: variant allele frequency; Y: Yes; (Y): Time in years. Data summary for all samples. CNA: copy number alterations; IHC: immunohistochemistry; IR: inconclusive results; MMRd: mismatch repair deficiency; MMRp: mismatch repair proficient; MSI: microsatellite instable; MSS: microsatellite stable; Mut/Mb: mutations per megabase; N: negative; P: positive; SNV: single-nucleotide variants; VAF: varian allele frequency. Subclonal MSH2/6. Negative MSH6. Low coverage. Hypermutated. There were 52 genes identified as affected by non-silent mutations in cancer-associated genes ( Fig. 4C ). The most frequently mutated cancer-associated gene was TP53 , which was mutated in five OC samples (three E-OC and two CC-OC). All TP53 mutations were predicted to be pathogenic. The genes KRAS , PIK3CA , AFF3 , and SPEN were mutated in three carcinomas. Two tumors, both of the CC-OC histotype, had deleterious mutations in ERBB3 . Two tumors, one of each OC histotype, had a predicted pathogenic mutation in PIK3CA hotspots (c.1624G>A and c.3140A>G). Three E-OC specimens harbored mutations in KRAS , of which two displayed a hotspot mutation in c.35G>A, while the hypermutated E-OC-10 harbored an atypical mutation, c.39C>T. Stop-gain mutations were detected in PTEN and ARID1A in one E-OC and one CC-OC, respectively. The ARID1A stop-gain mutation was confirmed with IHC, while the remaining specimens were positive for ARID1A ( Supplementary Fig. S5 ). The hypermutated E-OC-10 sample carried non-silent mutations in 331 cancer-associated genes, including a deleterious missense mutation in the exonuclease domain of POLE ; in all, it was analogous with the ultramutated genotype of endometrial carcinoma. This tumor also harbored mutations in several genes associated with maintenance of genome integrity ( Supplementary Table S5 ). Additionally, E-OC-10 showed evidence of kataegis ( Supplementary Fig. S3C ) and had multiple mutations in kataegis genes (seven in APOBEC genes and three in REV1 ), one of which was a missense variant ( APOBEC3B c.31C>T). A total of 308 shared mutations were detected in paired OE and OC samples corresponding to a median of 19 (range, 0–134) mutations per sample. All samples displayed at least one shared SNV between paired OE and OC, apart from Case 04 ( Supplementary Table S6 ). The median VAF per sample was 0.40 (range, 0.28–0.61) and 0.39 (range, 0.16–0.70) in the OE and corresponding OC. We detected 33 shared non-silent mutations in 31 genes ( Fig. 5 ). Eight cases shared 1–11 non-silent variants per case. Of the variants, 13 showed a higher VAF in OC versus OE, 13 had <0.05 difference in VAFs, and seven had a lower VAF in OC compared to OE. Except for one stop codon variant in NEB , all variants were of missense type. Of the common variants, 51% (17/33) were predicted tolerated and benign. Seven variants were assessed as being deleterious and possibly damaging, of these, two cases had higher VAF in OC than in OE: DLGAP3 and NUFIP1 (Case 01), and WI2-3308P17.2 (Case 09). Six cases (55%) shared mutations associated with effects on the immune response. Mutations in PIBF1 (Case 01), HLA-B (Case 02), and HLA-DQA2 (Case 08) were predicted as damaging or deleterious, while the remaining mutations were tolerable or benign, or no prediction annotation was made. The permutation test showed that the probability of observing one or more common missense mutations by chance was <5%. Shared non-silent mutations in paired endometriosis and carcinomas. Data are displayed by differences in variant allele frequency between paired samples of OE and OC. Data were trichotomized as increased VAF in OC (OE OC). Affected genes are displayed with base pair and amino acid alterations. Variants are predicted as tolerated (gray) or deleterious (black) in SIFT and as benign (pink) or possibly/probably damaging (purple) in PolyPhen. CC: clear cell ovarian carcinoma; EC: endometrioid ovarian carcinoma; OC: ovarian carcinoma; OE: ovarian endometriosis; VAF: variant allele frequency. Evaluation of the cosine similarity between the mutational profile of each sample and COSMIC SBS and DBS signatures suggested that DNA repair mechanisms were frequently represented by signatures with ≥0.5 similarity ( Fig. 6 ). Signatures attributed to mismatch repair deficiency (MMRd) (SBS6, 15, 26, 44, and DBS7), were observed in nine OE and seven OC samples, respectively. IHC staining for MMRd confirmed aberrations in two OCs (OC-E-03 and OC-E-10) ( Table 1 , Supplementary Fig. S6 ). IHC for OC-CC-11 was inconclusive due to fixation artefacts, while the remaining samples showed no MMR aberrations. Analysis of MSI suggested that OE-02 displayed high instability, while data from OE-04 and OE-05 were inconclusive, and the remaining OEs did not display MSI. Four OCs displayed MSI, two with high (OC-E-05 and OC-E-10) and two with low MSI (OC-CC-02 and OC-E-08). The signature associated with homologous repair deficiency (SBS3) was observed in eight OE and OC samples, respectively. The most frequent signature, suggestive of deficiency in transcription-coupled nucleotide excision repair (TC-NER) (SBS5) was seen in all samples, except OC-CC-04 and OC-E-10. Indel signatures (IDs), ID1 and ID2, suggestive of DNA slippage during replication, were the most frequently noted, in 77% (17/22) and 59% (13/22) of the samples. The OC-E-10, harboring missense mutations in POLE , POLD1 , MSH2 , MSH6 , and PMS2 , showed high similarity to additional signatures coupled to POLE or POLD1 mutations (SBS20 and 14) and MMRd (SBS44 and DBS10). Profile similarities of SBS, DBS, and ID COSMIC signatures for all samples. Mutation profiles with a minimum of one sample with a signature similarity of ≥0.5 are included in the figure. The color density indicates the level of similarity (cosine similarity range 0.5–0.95). CC: clear cell; COSMIC: catalogue of somatic mutations in cancer; DBS: doublet-base substitution; E: endometrioid; ID: indel; OC: ovarian carcinoma; OE: ovarian endometriosis; SBS: single-base substitution. When comparing paired samples, OC showed more CNAs than OE, with a median of 256 (range, 17–521) events per tumor and 135 (range, 62–379) events per OE specimen. All samples in the present cohort displayed CNAs affecting multiple genes with a median of 128 (range, 14–4740), and 5542 (range, 181–10 468) genes affected in OE and OC, respectively ( Table 1 , Supplementary Table S7 ). Overall, the OEs displayed more chromosomal loss than gain of genetic material ( Supplementary Fig. S7 ). Generally, chromosomal gains consisted of smaller chromosomal regions and partial gene sequences. Eight OEs displayed a CNA affecting one or more cancer-associated genes ( Table 1 ), median 49.5 (range, 0–165), where a predicted impact could be estimated (i.e. deletion or amplification of transcribed regions or whole gene/s, respectively). OE-01 displayed the highest number of CNAs, including loss of a large fraction of chromosome X (ChrXp22.33-q28, LogR –0.89) ( Fig. 7A ), affecting numerous cancer-associated genes (145) where the entire or majority of the gene displayed loss. Corresponding chromosomal loss was seen in the subsequent OC (ChrXp22.33-q28, LogR –0.44 to –1.16). The most telomeric region, with the most discrepant LogR, harbored two cancer-associated genes, FLNA and RPL10 . Overlapping loss of chromosome X was also seen in Case 05, where both OE and OC displayed loss of multiple cancer-associated genes including FLNA and RPL10 . Both patients developed E-OC. Two OEs that later developed into CC-OC displayed large gains on chromosome 16, 16p13.3-p13.13 and 16p13.3-p13.11 ( Fig. 7B ) in OE-07 and OE-11, respectively. This region contained six cancer-associated genes that were amplified in both specimens. For Case 07, overlap was found in OE and the subsequent CC-OC (16p13.3-p13.13), affecting eight cancer-associated genes. Data showed that, except for Cases 08 and 10, all samples shared common regions with CNAs, with a median of 10 affected genes per sample (0–540 and 0–2179 genes with gains and losses, respectively). Two pseudogenes, DUX4L1/3 , displayed gain in three samples, and in total 137 genes displayed gain in ≥2 cases, including five cancer-associated genes ( CAMTA1 , PRDM16 , RPL22 , SKI , TNFRSF1 ). Functional annotation suggested enrichment of tumor necrosis factor signaling (Cases 07 and 11). Case 07 also displayed enrichment of amplification of genes associated with negative regulation of HLA class I molecules. Shared chromosomal loss in ≥2 samples affected 683 genes, including 31 cancer-associated genes, where loss of BGN was found in three cases. Four OCs and two OEs displayed CNAs on Chromosome 6 affecting immune response genes. Loss of 6p21.33 ( Fig. 7C ) was seen in three samples (OE-03, OE-05, OC-E-06), where the smallest region of overlap contained four genes ( HLA-B , HLA-C , MICA , MICB ). OC-E-09 displayed chromosomal gain affecting an adjacent transcribed region of HLA-G (201 bp, LogR 1.01). Gain of 6p22.1-p21.33 was seen in OC-CC-02; this region contained the major HLA class I genes and two additional cancer-associated genes ( POU5F1 , TRIM27 ). OC-E-03 showed deletion of 6p21.32, containing a part of HLA-DPA1 (13 965 bp, LogR –0.82). Visualization of recurrent large copy number alterations. ( A ) Extensive loss of chr X in two OE and paired OC with endometrioid histotype. ( B ) Gain of chr16 in OE that later developed OC of clear cell histotype, a corresponding gain was detected in paired OC for one of the cases. ( C ) Two OEs and four OCs displayed CNAs affecting chr 6, including loss and gain affecting HLA genes of class I and II. The y -axis represents the normalized median logR. chr: chromosome; CNA: copy number alteration; Mb: megabase; OC-CC: clear cell ovarian carcinoma; OC-E: endometrioid ovarian carcinoma; OE: ovarian endometriosis.

Formalin-fixed and paraffin-embedded (FFPE) tissue specimens from 45 archival cases with EAOC and previous endometriosis, surgically removed prior to OC diagnosis, between 1998 and 2016 were identified at the Pathology Departments of Sahlgrenska University Hospital in Gothenburg (n = 13) and Skåne University Hospital in Lund (n = 32) ( Fig. 1 ). After pathology review (C.M./S.W.-F.), 25 cases were excluded due to reclassification into non-EAOC (n = 9) or because OE could not be confirmed (n = 16). Further cases (n = 9) had either insufficient endometriosis tissue or showed poor DNA quality in either normal or endometriosis tissue ( Supplementary Table S1 ). Eleven cases had satisfactory DNA from all three locations and were eligible for further analysis. FFPE sections rich in non-neoplastic epithelium from the fallopian tube or uterine cervix (n = 5) or blood samples collected at cancer diagnosis (n = 6) were used as the germline reference. Patient flow chart and inclusion. CC-OC: clear cell ovarian carcinoma; E: endometriosis; EAOC: endometriosis-associated ovarian carcinoma; E-OC: endometrioid ovarian carcinoma; HGSC: high-grade serous carcinoma; LCM: laser capture microdissection; LGSC: low-grade serous carcinoma; OE: ovarian endometriosis; SCC: squamous cell carcinoma; SUH: Sahlgrenska University Hospital; SUS: Skåne University Hospital. Created with BioRender.com. All tissue biopsies were assessed by a specialist in gynecologic pathology (C.M./S.W.-F.) on standard hematoxylin and eosin slides. OE epithelium was collected using the PALM MicroBeam system (Zeiss, Stockholm, Sweden) by Laser Capture Microdissection. Epithelium was collected from a median of one separate biopsy of OE tissue (range, 1–5) corresponding to a median of two (range, 1–5) disconnected OE loci per specimen ( Supplementary Fig. S1 ). DNA was extracted using the QIAamp DNA Micro kit (Qiagen) for Gothenburg cases and QIAamp FFPE kit (Qiagen) for Lund cases. Cell-rich carcinoma and normal tissue were macrodissected and DNA was extracted using the GeneRead DNA FFPE kit (Qiagen, Hilden, Germany) for Gothenburg cases and QIAamp FFPE kit (Qiagen, Hilden, Germany) for Lund cases. DNA from blood was extracted using QIAamp DNA Micro Kit according to the protocol (Qiagen). DNA concentration was assessed with the Qubit Fluorometer (ThermoFisher Scientific, MA, USA) and DNA quality was evaluated by TruSeq FFPE DNA Library Prep QC Kit where a ΔCt value <7.0 was required for continuation to exome sequencing. Sample and library preparation, enrichment, and elution were performed according to the manufacturer’s protocols (Human Comprehensive exome, Twist Bioscience, Carlsbad, CA). For library preparation, 50 ng of genomic DNA was fragmented by enzymatic digestion, followed by end repair and dA-tailing. The library was enriched by seven PCR cycles using twist unique dual index primers. Then, 200 ng of the library was hybridized to the Twist CoreExome EF Multiplex Complete hybridization probes in a six-sample multiplex reaction. The library was amplified with seven cycles, purified, quantified, and sequenced (Nextseq500, Illumina, San Diego, CA). Image analysis and base calling were performed (Illumina Real Time Analysis, Illumina) with default parameters. Read mapping and data processing were performed using the Sarek pipeline ( https://nf-co.re/sarek ). FreeBayes, Mutect2, and Strelka were run for calling single-nucleotide variants (SNVs). Variants were accepted if they passed all three callers, or if they passed FreeBayes and Strelka or Mutect2. For FreeBayes, a cutoff of QUAL >20 was used. Accepted mutations were filtered to remove those with an observed allele frequency of <0.5%, those with sequencing depth of <5 observed reads in samples, and those present in the normal reference sample. Variants were annotated with SnpEff, Variant Effect Predictor (VEP), and Catalogue of Somatic Mutations in Cancer (COSMIC) cancer census genes and counts (COSMIC, v95). Function and clinical implications were assessed based on Sift, PolyPhen, and ClinVar according to the VEP annotations. Previously suggested spurious genes ( Lawrence et al. , 2013 ; Tamborero et al. , 2013 ; Shyr et al. , 2014 ) were excluded unless identified as endometriosis-associated genes ( Joseph and Mahale, 2019 ). The Maftools R package was used for characterization and visualization of mutational patterns and kataegis analysis ( Mayakonda et al. , 2018 ). Gene ontology and gene set enrichment were assessed with GSEA ( Subramanian et al. , 2005 ; Liberzon et al. , 2015 ) and DAVID ( Dennis et al. , 2003 ). Mutational profiles were compared to the COSMIC single-base substitution (SBS), doublet-base substitution (DBS), and small indel (ID) mutational signatures ( Alexandrov et al. , 2013 ), by calculating cosine similarities (lsa package, version 0.73.3). Signatures with a similarity <0.5 to all samples were excluded. All samples and remaining signatures were grouped by similarity using hierarchical clustering and visualized as a heatmap (pheatmap package, version 1.0.12). Somatic copy number alteration (CNA) analysis was performed using SAMtools ( Li et al. , 2009 ) pileup of normal and tumor sample BAM files, followed by CNA calling with VarScan2 ( Koboldt et al. , 2012 ). Calls were normalized by the ratio of uniquely mapped reads in the normal sample to uniquely mapped reads in the tumor sample and adjusted for GC content. VarScan2 results were smoothed and segmented with DNAcopy package version 1.72.0 ( Seshan, 2022 ), and recentered genome-wide. CNA segments were merged by VarScan2’s mergeSegments.pl script using default parameters, and mapped to genomic annotations with the annotatr package 1.24.0 ( Cavalcante and Sartor, 2017 ). Immunohistochemical (IHC) staining was performed for MLH1 (FLEX Monoclonal mouse Anti-Human Clone ES05, Dako, Glostrup, Denmark), MSH2 (FLEX Monoclonal Mouse Anti-Human Clone FE11, Dako), MSH6 (FLEX Monoclonal Rabbit Anti-Human Clone EP49, Dako), and PMS2 (FLEX Monoclonal Rabbit Anti-Human 2 Clone EP51, Dako) using a Dako Autostainer (Autostainer Link 48) and EnVision FLEX, High pH (Link) kit. Loss of mismatch repair (MMR) protein expression was defined as granular or absence of nuclear staining for any of the MMR proteins in all epithelial cells or a distinct area of OE or tumor, with normal immunoreactivity in the internal positive control (i.e. lymphocytes and stromal cells). IHC for ARID1A (Monoclonal Rabbit Anti-Human clone EPR13501 , Abcam) was performed with Ventana Benchmark Ultra. Complete loss of ARID1A expression in all tumor or OE cells was considered negative and for a subclonal pattern, evidence of complete negativity in sections of the tumor or OE was required. IHC was assessed independently by two observers (C.M., S.W.-F.). One of the EAOC blocks with inadequate fixation was replaced by an omental metastasis from the same case. For microsatellite instability (MSI) analysis, remnant DNA from the exome sequencing was used when available (Lund) or freshly prepared (Gothenburg). Tissue samples were macrodissected and DNA was extracted from the FFPE (QIAamp DSP DNA FFPE Tissue Kit, Qiagen) and corresponding blood (QIAamp DNA Micro Kit). PCR reactions were set up with 2 ng DNA and prepared for electrophoresis (TrueMark™ MSI Assay, ThermoFisher Scientific). The fragment analysis reactions were diluted 1:100, or adjusted after initial assessment, before data collection (Applied Biosystems 3500 Genetic Analyzer). Data were analyzed with the TrueMark™ MSI Analysis Software (v1.0). Statistical analysis was performed using the IBM SPSS Statistics version 28. Groups were compared with the Mann–Whitney U or Kruskal–Wallis test for two or more groups, respectively. Correlation was evaluated with Spearman’s correlation. Two-sided tests were used and P  ≤ 0.05 was considered statistically significant. Two OC samples (CC-OC-04, E-OC-10) were excluded from statistics and calculations when the data were presented on OC-group level. Data in the text are presented as median with minimum and maximum values in parenthesis. A permutation test was performed to assess whether common missense mutations between paired samples were expected by chance. Permutations were performed for 1000 iterations, where mutations were randomly seeded in the exomes of 10 in silico -paired samples. The number of mutations per sample was assumed to follow a negative binomial distribution fitted to the mutation data in this study (except for E-OC-10 that harbored a much larger number of mutations). To make the permuted data comparable to the real data, mutations in each in silico sample were seeded in the exons affected by mutations in the corresponding sample. The study was approved by the ethics committee of the Sahlgrenska University Hospital (Dnr 201-15 and T522-17).

In the present study, we assessed the genomic landscapes of paired OE and EAOC; 11 samples from women with OE who later developed either endometrioid or clear cell OC were analyzed. To our knowledge, this is the first study that has explored the mutation and copy number profiles of endometriosis with a confirmed subsequent malignant association. We here identified conserved mutations between OE and paired OC, indicating a potential common ancestry. However, although mutations in cancer-driver genes have been suggested as the origin of malignant transformation of OE ( Suda et al. , 2020 ), no shared cancer-associated mutations were observed. Thus, although mutations in conventional cancer-driver genes were confirmed to occur frequently in OE, our original hypothesis could not be supported. Noteworthy, the median VAFs of the shared mutations were higher than the overall median VAFs, suggesting a positive selection for clones harboring these mutations ( Suda et al. , 2018 ). Conserved mutations in genes associated with an immune response were the most prevalent, with affected genes in 5 of the 11 cases. Additionally, CNA analyses showed that aberrations involving immunological pathways were conserved. In total, seven cases exhibited genetic aberrations suggestive of an altered immune response. The single most frequently mutated type of genes was human leukocyte antigen (HLA) genes; four cases harbored a shared non-silent mutation in an HLA gene, three in class II, and one in HLA class I genes. Furthermore, the present data showed CNAs in four carcinomas and two endometriomas, mainly affecting HLA class I genes, emphasizing the importance of these mechanisms in the development of endometriosis and EAOC. In line, chromosomal instability in the HLA super-locus on 6p21 has previously been described as a common alteration in EAOC ( Micci et al. , 2014 ), and aberrant HLA expression has been described for both endometriosis ( Semino et al. , 1995 ; Matalliotakis et al. , 2001 ) and OC ( Hazini et al. , 2021 ). Additionally, the most frequently observed aberration in paired samples was chromosomal loss of BGN (encoding biglycan). Low BGN expression has been associated with decreased immunogenicity and poor patient outcome in KRAS-driven malignancies ( Subbarayan et al. , 2022 ). Furthermore, a deleterious mutation in PIBF1 (encoding progesterone immunomodulatory binding Factor 1) was seen in paired samples. This protein regulates the immune response during normal pregnancy, and has been suggested to facilitate tumor growth by suppressing the local antitumor response ( Szekeres-Bartho and Polgar, 2010 ). Taken together, these genetic aberrations suggests that adaption to inflammation is an early and crucial event, in line with known mechanisms by which tumors evade the immune system ( Jhunjhunwala et al. , 2021 ). Furthermore, this finding suggests that immune-associated treatments may be beneficial not only for the treatment of endometriosis but may also have cancer-preventive effects. Considering the chronic inflammation accompanying endometriosis ( Zondervan et al. , 2018 ), and the well-described association between inflammation and the genomic instability associated with tumor development ( Coussens and Werb, 2002 ), it is reasonable to assume that the inflammatory environment contributes to the increased incidence of EAOC. Based on other types of cancer and their association with inflammation ( Balkwill and Mantovani, 2001 ), it could be expected that chronic inflammation would increase the incidence of cancer to an even higher extent than what is seen in association with endometriosis ( Pearce, et al. , 2012 ; Hermens et al. , 2020 ). Furthermore, chronic inflammation is not exclusive to OE, but universal to endometriosis, and mutations in cancer-driver genes have been shown for different types of endometriosis, although no association with EAOC has been described ( Anglesio et al. , 2015 , 2017 ). Interestingly, the present data show a strong positive correlation between VAF and time between the two diagnoses. This implies that a dominant OE may delay tumor establishment; in other words, protect against an emerging malignancy. This is in line with the previously shown cancer-preventive properties of nonmalignant epithelium in association with an accumulation of cancer-associated mutations ( Colom et al. , 2021 ). Moreover, the data show that there was no correlation between age at OE diagnosis and mutational burden, suggesting that mutations are selected, unlike mutations in normal endometrium that passively accumulate over time ( Moore et al. , 2020 ). This is in line with clonal evolution of cancer ( Greaves and Maley, 2012 ), and supports previous observations that clonal expansion of OE involves the accumulation of mutations in cancer-driver genes ( Suda et al. , 2018 ). However, the absence of conserved cancer-associated mutations between OE and OC indicates that the mutations associated with the malignant transformation are later events that occur in an originally less-potent precursor. There is sparse data on the genomic configuration of multiple and/or bilateral endometriotic lesions in the same individual. One case report described multiple shared mutations in cancer-associated genes in synchronous samples from the endometrium, different endometriosis lesions, and CC-OC ( Suda et al. , 2020 ) and in another cohort, the same group found shared KRAS or PIK3CA mutations between multiple unilateral OE lesions in six individuals, and between bilateral OE lesions in a single case ( Suda et al. , 2018 ). These findings suggest an inherent potential for progression to cancer in all OE lesions within these individuals. Hence, despite the discrepancy in laterality between OE and OC in some of the cases in our study, the pair-wise comparisons made are relevant and interesting in relation to clonal evolution in cancer progression. KRAS was the most frequently mutated cancer-associated gene in the current study. All affected specimens displayed the same hotspot mutation (c.35G) except the hypermutated OC-10, which carried a rarer mutation in an adjacent codon. It has been suggested that selection for specific c.35G mutations drives different tumor types ( Ostrow et al. , 2016 ). The most well-described KRAS mutation in OE (p.G12V, c. G35T) ( Suda et al. , 2018 ) was observed in one of the OE samples. This specific mutation has previously been linked to inflammation in OE ( Yachida et al. , 2021 a). Of note, expected cancer-associated genes such as PIK3CA , ARID1A , PPP2R1A , and FBXW7 ( Anglesio et al. , 2015 , 2017 ; Suda et al. , 2018 ) did not harbor any non-silent mutations in OE. Moreover, the number of OE specimens with observed cancer-associated mutations in the present data was higher than what has previously been reported ( Anglesio et al. , 2017 ; Suda et al. , 2018 ). Comparative analyses suggested that OEs share several commonly affected genes ( Li et al. , 2014 ), including a large fraction (68%) of cancer-associated genes, compared to only 1% in data from endometriosis at other locations ( Anglesio et al. , 2017 ), suggesting that this genetic profile is associated with the OE subtype. A limitation of the present study was the small cohort, thus further studies are warranted to determine whether a specific genetic profile of OE contributes to the development of EAOC. Considering OC, the mutation frequency of the conventional EAOC-associated genes differed from what has previously been described. The most frequently reported mutated gene in EAOC, ARID1A , described in 50% and 30% of CC-OC and E-OC tumors ( Yachida et al. , 2021 ), was only found to be mutated in one CC-OC. The effect of the mutation was confirmed by IHC as the single case displaying aberrant ARID1A protein expression. Thus, the frequency of aberrant ARID1A protein expression was also considerably lower than expected ( Heinze et al. , 2022 ). PIK3CA , with a described mutation frequency of 41% and 53% in E-OC and CC-OC ( Yachida et al. , 2021 b), respectively, was found mutated in E-OC only. In contrast, the most frequently mutated gene in the tumor specimens was TP53 : mutated in three E-OC and two CC-OC specimens, respectively. This suggests that in the present cohort TP53 mutations were overrepresented, particularly in CC-OC. The data showed that both OE and OC displayed mutational profiles associated with deficiency in genes controlling genomic integrity. Analysis of cosine similarity to mutational signatures suggested that most of the specimens displayed a profile associated with deficiency in homologous repair as well as defective DNA mismatch repair. Although these findings could only be confirmed for two OCs using IHC, the MSI results suggested that at least one OE and four OCs displayed instabilities consistent with MMR deficiency. Furthermore, seven of the OEs and four of the OCs displayed genetic aberrations such as non-silent mutations or CNAs in genes associated with maintenance of genomic integrity or repair mechanisms. This included aberrations shared between paired samples such as mutations in the homologous repair gene GEN1 ( Lee et al. , 2015 ) and deletion of FLNA , associated with the BRCA damage response pathway ( Velkova et al. , 2010 ). Additionally, the ultramutated OC and two of the OEs displayed signs of cancer-like kataegis ( Alexandrov et al. , 2013 ). However, only the OC specimen displayed the nucleotide substitution patterns typically associated with APOBEC-related kataegis. The OEs, on the other hand, displayed significantly high VAFs and prolonged time between diagnoses, suggesting an exceedingly strong clone. Taken together, these data suggest that an innate defective DNA damage response and overall genomic instability, while central to the malignant transformation, also account for the resilience of OE. The study has several limitations. The aim was to study possible causality in surgically resected OE prior to OC diagnosis in the same patient. We acknowledge that even though this is the first study of its kind, the sample size was small, the sequenced endometriosis was resected, sometimes on the contralateral side of subsequent OC, and no in-between surgery data was known, i.e. recurrence of endometrioma. However, 8 of the 11 OCs were at stages I and II at the time of diagnosis and second surgery, which might suggest that these patients were more closely surveilled since OCs are commonly diagnosed at stages III and IV. Taken together, the present study suggests that, as a robust endometriosis lesion prevails, the accumulation of specific cancer-driver mutations may ultimately lead to the development of a malignancy. The time required, and probability, for this to occur are associated with the genotype of the original endometriotic lesion and the environment where the precursor is located. The positive correlation between the number of distinct OE subpopulations and mutational burden, and the overall low number of recurrent mutations in the cohort, suggest intra- and inter-patient heterogeneity. This study provides a novel comprehensive understanding of the genetic diversity of OE associated with metachronous carcinoma. The complexity presented here prompts further investigative studies for an in-depth understanding of the risk of malignant transformation associated with OE.