Germline BRCA-Associated Endometrial Carcinoma Is a Distinct Clinicopathologic Entity

Inheritance of a pathogenic mutation in one allele of the breast cancer susceptibility genes, BRCA1 or BRCA2, results in the hereditary breast and ovarian cancer (HBOC) syndrome, characterized by severely increased lifetime risk to develop breast cancer and tubo-ovarian cancer (OC; refs. 1, 2). Other cancer types reported to be increased in patients with germline BRCA2 mutations (gBRCA) are pancreatic and prostate cancer (3, 4). Whether endometrial carcinoma (EC) should be considered part of gBRCA-associated HBOC syndrome is still under debate due to conflicting data (5–9). A number of studies have shown an increased risk to develop EC especially for gBRCA1 carriers, with highest risks observed for an aggressive subtype of EC—the serous-like ECs (5–7, 9–11). However, others did not observe this increased risk or attributed it to previous tamoxifen treatment rather than to the gBRCA mutation (8, 9, 11).

LOH of the wild-type BRCA1 or BRCA2 allele (gBRCA/LOHpos) is an important step in the carcinogenesis of gBRCA-associated tumors. This is supported by the observation that gBRCA/LOHpos breast cancers and OCs show significantly higher homologous recombination deficiency (HRD) scores compared with their gBRCA/LOHneg counterparts (12). The HRD score is based on the presence and quantification of “genomic scars” associated with BRCA deficiency, including the number of regions with LOH (13), large-scale state transitions (14), and telomeric allelic imbalances (15). Breast cancers and OCs arising in gBRCA carriers show variable LOH frequencies, with reported rates of 93% (gBRCA1) and 84% (gBRCA2) for OCs, and 90% (gBRCA1) and 54% (gBRCA2) for breast cancers (12). This signifies the relevance of LOH as a marker of causality and implies that gBRCA/LOHneg cancers are in fact sporadic tumors that develop independently of the gBRCA mutation and are not HRD.

The finding of recurrent clinicopathologic and molecular features in gBRCA-associated breast cancers and OCs has supported the concept that these cancers are distinct entities belonging to the gBRCA-associated HBOC syndrome. These features can also help identify tumors more likely to harbor BRCA1/2 mutations. For example, breast cancers arising in gBRCA1 carriers prototypically are of high grade and of the basal-like subtype with more frequent necrosis and lymphocytic infiltration (16, 17). BRCA1-associated high-grade serous tubo-ovarian carcinoma (HGSOC) shows more frequent (partial) Solid, pseudoEndometrioid, and/or Transitional morphology (SET morphology), which is distinctly different from the prototypical papillary and infiltrative growth generally encountered in sporadic HGSOC. Other features more frequently observed in BRCA1-associated HGSOC are necrosis, a higher mitotic index, and an increased number of tumor-infiltrating lymphocytes (TILs; refs. 18–20). On a molecular level, gBRCA-associated breast cancers and OCs share similar somatic copy-number profiles [somatic copy-number alteration (SCNA)-high] and frequent TP53 mutations (16–18, 20–22).

The Cancer Genome Atlas (TCGA) Research Network previously defined four distinct molecular subclasses with prognostic relevance in ECs (22). The “serous-like/SCNA-high” molecular subclass has poorest clinical outcome and interestingly displays molecular similarities to both basal-like breast cancer and HGSOC, including a high number of SCNAs and frequent TP53 mutations. Moreover, recent studies demonstrated that serous-like/SCNA-high ECs also frequently are HRD (22–24). This raises the question whether ECs occurring in gBRCA1/2 carriers might be enriched for certain features, but studies comprehensively evaluating this have not been performed to date.

We aimed to, for the first time, comprehensively describe the clinicopathologic and molecular features, stratified by LOH-status, of a large series of ECs that occurred in gBRCA carriers.

Patients with a history of EC and a pathogenic gBRCA1/2 mutation were identified from the “Hereditary Breast and Ovarian cancer study, the Netherlands (HEBON cohort study)” (25). The HEBON study is an ongoing nationwide study on families with HBOC for which all patients who undergo genetic testing for BRCA1/2 and CHEK2 mutations in one of the participating centers are eligible for inclusion [all eight university medical hospitals in the Netherlands and the Netherlands Cancer Institute (NKI)]. For participants, data on, among others, personal cancer history and therapeutic treatments are collected both retrospectively and prospectively through regular linkages with the Netherlands Cancer Registry. Data on prophylactic surgery are collected via the Dutch Pathology Registry (PALGA; ref. 26). All data are centrally collected and managed by trained data managers only. Women were eligible for inclusion when they had (i) a proven pathogenic germline BRCA1/2 (gBRCA1/2) mutation (PLON class 4 or 5; ref. 27), (ii) provided written informed consent for the HEBON study, and (iii) had a history of epithelial EC or developed an EC during follow-up, defined as a tumor with an International Classification of Diseases Oncology, Third Edition, First Revision (ICD-O-3.1; http://codes.iarc.fr/) topographical code of either C54 (Corpus Uteri) or C55 (Uterus, NOS).

In total, 3,726 gBRCA carriers provided informed consent between 1999 and 2014, of which the majority was provided in 2012 and 2013 (62.5%). Of these women, 41 (1.1%) developed an EC. We were able to retrieve 39 of 41 tumors from pathology laboratories across the Netherlands. One tumor was a sarcoma and was therefore excluded. Of these 38 HEBON-ECs, 21 ECs occurred preceding to study enrollment (mean: 4.7 years, SD: 2.79) and 16 ECs occurred after study enrollment (mean: 4.5 years, SD 3.52). For one case, the date of study enrollment was not available. The HEBON-ECs were supplemented with four ECs from known gBRCA1/2 carriers previously diagnosed in the Leiden University Medical Center (LUMC).

For all ECs, hematoxylin and eosin (H&E)-stained slides, anonymized pathology reports, and at least one representative formalin-fixed, paraffin embedded (FFPE)-tumor block were collected through the Dutch Pathology registry (PALGA; ref. 26) from pathology laboratories across the Netherlands. If applicable, material from the (salpingo-)oophorectomy or OC specimen was also requested. The HEBON study is approved by the medical ethical committees of all participating centers, and all participants gave written informed consent to participate in the study. The HEBON study is performed in accordance with the Declaration of Helsinki. Our study was performed after the study protocol was approved by the HEBON steering committee (date: November 30, 2017) and by the Institutional Review Board of the Netherlands Cancer Institute; IRBd18086. All specimens were handled in compliance with the Code of Conduct for dealing responsibly with human tissue in the context of health research (2011) drawn up by the Federation of Dutch Medical Scientific Societies.

All cases were independently reviewed by three expert gynecologic pathologists (V.T.H.B.M. Smit, T. Bosse, and B.E. Howitt). They were aware that the ECs occurred in gBRCA carriers; however, they were blinded for LOH status. The World Health Organization (2014) criteria were used for histologic subtype diagnosis. Reviewers were not allowed to use immunostains to aid classification and diagnoses were solely based on H&E stains. Cases were classified ambiguous when overlapping features of both high-grade endometrioid and serous carcinomas were present in the tumor and when tumors failed to show prototypic features of a certain subtype. Discordant cases were discussed during a consensus meeting to assign final histologic subtype. ECs with ambiguous morphology were considered nonendometrioid for statistical analyses. After final histologic subtype was assigned, histologic subgroups were made. For ambiguous cases, TP53 mutation status was used to assign histologic subgroup. Cases were categorized as follows: “Endometrioid” for Endometrioid, mucinous and TP53-wild-type ambiguous carcinomas, “serous-like” for uterine serous carcinomas (USCs), uterine carcinosarcomas (UCSs) and TP53-mutant ambiguous carcinomas, or “clear cell” for clear cell carcinomas. Review of adnexa, depth of myometrial invasion, cervical involvement, lymph node status, and presence of lymphovascular space invasion was performed by one expert gynecologic pathologist (T. Bosse) on which FIGO-2009 stage was based upon. When slides were not available, these data were retrieved from the original pathology reports.

Nineteen morphologic characteristics were assessed by one expert gynecologic pathologist (B.E. Howitt) on all available tumor slides, blinded for LOH status. For additional details on this, see Supplementary Materials and Methods.

Cases were stained for p53 (clone DO-7, 1:2,000, DAKO), Wilms tumor 1 (WT-1, clone 6F-H1, 1:3,200, Invitrogen), estrogen receptor (ER, Clone EP1, 1:200, DAKO), progesterone receptor (PR, Clone Pgr636, 1:400, DAKO), and CD8 (Clone 4B11, 1:2,000, Novocastra). Procedures and scoring methods are described in the Supplementary Materials and Methods.

Tumor DNA was isolated from FFPE-tissue blocks either by using three 0.6-mm tumor cores (n = 16) or by using microdissected tissue from 5 to 10 tissue sections (10 μm; n = 26). DNA isolation was performed fully automated using the Tissue Preparation System (Siemens Healthcare Diagnostics) as described previously (28). The median tumor cell percentage of the isolated areas was 80% (range, 25%–90%).

Following extraction, DNA was quantified using the Qubit fluorometric assay (Thermo Fisher Scientific) and further assessed for quantity and quality using a quantitative PCR assay (hgDNA Quantitation and QC kit, KAPA Biosystems). Library preparation and sequencing were performed as previously described for the UCM-OncoPlus Assay (29). Briefly, approximately 100-ng DNA was fragmented using the Covaris S2 (Covaris). The fragmented DNA was amplified using the KAPA HTP Library Preparation Kit (Kapa Biosystems) along with a set of patient-specific indexes (Roche). The pooled library was captured using a custom SeqCap EZ capture panel (Roche) featuring a collection xGen LockdownProbes (IDT) for 1,213 genes. The pooled captured library was sequenced on the Illumina HiSeq 2500 system (Illumina) in rapid run mode (2 × 101 bp paired end sequencing). Somatic mutation and copy number calling were performed across all 1,213 genes using a custom in-house bioinformatics pipeline previously described (29). The five-tier pathogenicity classification described by Plon and colleagues, 2008, was used to categorize variants (27). Only class 4 (likely pathogenic) and 5 (pathogenic) mutations are reported in the manuscript.

Known gBRCA1/2 mutations were assessed for LOH of the wild-type allele by evaluating the following parameters: estimated tumor cell purity, BRCA1/2 mutation variant allele frequency (VAF), local copy number status, and adjacent SNP VAF, using a similar approach to what has been described by Khiabanian and colleagues, 2018 (30). For LOH analyses, we applied the following model, taking into account the chromosomal copy number at the BRCA locus; VAF = [(1 − p) +cmut × p]/[2 × (1 − p) + Y × p], with p being the tumor purity, cMut being the mutation's chromosomal copy number, and Y being the ploidy of the tumor cells. LOH events occur when cMut = 1 and Y = 1 or cMut > 1 and Y > 1. Because all BRCA1/2 mutations were germline mutations, the expected VAF in the absence of LOH was 1/2 (50%) for all cases. LOH of the gBRCA1/2 wild-type allele was considered to be present if (i) cMut = 1 and Y = 1 or cMut > 1 and Y > 1, (ii) the observed gBRCA1/2 mutation VAF was similar to the expected VAF according to the formula, (iii) adjacent observed SNP VAF (if present) supported the findings, and (iv) sequencing quality was sufficient. Mutations that were considered to have an LOH event were classified as either copy-neutral (no evidence of local copy-number change) or copy-number loss. gBRCA/LOHpos ECs were defined as gBRCA-associated, gBRCA/LOHneg ECs as “sporadic.”

For the copy-number calling, we used a clinically validated bioinformatic tool that has previously been detailed and published (29). Briefly, copy-number analysis involved evaluation of average exon interval depths recorded via the Genome Analysis Toolkit DepthofCoverage module. A historical normalized baseline for each interval in the panel was generated using 24 nonmalignant clinical samples. Test sample data were subjected to a normalization algorithm to control for individual gene profile run-specific variability. To detect the potential copy-number regions, fold change and Z-scores were calculated for each interval, and thresholds were set at >200% (gain) or <66% (loss) with Z-score >3 or ≤2, respectively. Genes with more than half the intervals showing copy-number changes in the same direction were then identified. Overall copy-number status was assessed manually by assessing the copy-number plots across the entire territory and determining how many large-scale (arm or subarm-level changes) copy-number alterations were present in each case. Cases considered to be “low” copy number had 0 large-scale copy-number alterations, “borderline” had 1 to 2 large-scale copy number alterations, and those considered “high” had >2 large-scale copy-number changes.

For MSI testing, a metric similar to that proposed by Kautto and colleagues 2017 (31) was employed to quantify the stability of a homopolymer locus. For each locus, distribution over different homopolymer lengths (normalized to a fraction of total depth at the locus) was generated. Then, absolute value of the stepwise difference between that sample distribution and normal distribution was calculated as a distance score (d). The baseline distribution was generated using average values across 23 nonmalignant spleen samples. Thresholds for assignment of “stable” or “instable” status for a locus involved using training sets of MSI-stable and MSI-high samples, tested previously by PCR assay or IHC staining. Samples with unstable loci <9% were classified as microsatellite stable, 9% to 15% were classified as indeterminate, and >15% were classified as microsatellite instable (MSI).

Tumor mutational burden (TMB) was quantified as mutations/Mb using a 1,132-gene territory from the UCM-OncoPlus assay. Variants that met any of the following criteria were excluded from the calculation: <10% VAF, synonymous, variants present in either 1,000 genomes or ExAC population databases. In addition, variants were rescued if there were >10 entries in COSMIC database with an ExAC frequency of <0.001.

The following surrogate markers were used to classify ECs in the four molecular subgroups defined by the TCGA (22, 32, 33); POLE exonuclease domain mutations for the POLE/ultramutated group, MSI-high profile for MSI-high/hypermutated group, TP53 mutations for SCNA-high/serous-like group, and the absence of surrogate markers for no surrogate marker profile (NSMP)/SCNA-low group (22, 32, 33). When two molecular classifiers were present, subgroups were assigned in line with what has previously been published by the TCGA (22); POLE&MSI-high or POLE&TP53 as POLE and MSI-high&TP53 as MSI-high.

Associations between categorical variables were tested using a two-sided Fisher exact test or χ² statistics when more than two variables were compared. Associations between continuous variables were tested using the Mann–Whitney U test. Overall survival was calculated using the Kaplan–Meier Method with log-rank test and was calculated from the date of EC diagnosis to the date of death while patients who were alive were censored at the date of last follow-up. For HEBON cases, the date of last linkage with the Dutch Municipal Personal Record Database was used as last date of follow-up (April 11, 2019, for all except for case 2; December 23, 2016). P values <0.05 were considered significant. Statistical analysis was performed using IBM SPSS version 23.0 (SPSS, Inc.) and GraphPad Prism (GraphPad Software Inc.).

In total, 42 ECs that occurred in gBRCA1/2 carriers were analyzed (32 gBRCA1, 10 gBRCA2). Clinicopathologic characteristics of the complete cohort are described in Supplementary Table S1. The cohort comprised 26 endometrioid ECs (61.9%), of which 17 (40.5%) were grade 1, three (7.1%) were grade 2, five (11.9%) were grade 3, and one (2.4%) was a mucinous carcinoma. Sixteen ECs were classified as nonendometrioid (38.1%), of which seven (16.7%) were USC, four (9.5%) were UCS, and five (11.9%) were classified as high-grade ambiguous.

Molecular analysis was conducted to stratify for LOH of the gBRCA1/2-wild-type allele, which succeeded for all but two cases (n = 40, 95.2%), which were excluded from final analyses (one USC and one EEC grade 1, Supplementary Table S2). The known gBRCA1/2 mutation was confirmed in all 40 cases included in final analyses. Overall, 60% (24/40) of ECs were gBRCA/LOHpos. When stratified for gBRCA1 and gBRCA2 mutations, 66.7% (n = 20/30) and 40% (n = 4/10) showed LOH, respectively (P = 0.159; Fig. 1; Supplementary Table S2). Plotting the position of the gBRCA mutations across the coding DNA sequence for BRCA1 and BRCA2 did not show enrichment of mutations in a specific region of the gene [www.cbioportal.org/visualize (34, 35); Supplementary Fig. S1].

Clinicopathologic characteristics stratified by LOH status are summarized in Table 1 and Fig. 1. Compared with gBRCA/LOHneg ECs, gBRCA/LOHpos ECs were significantly more often FIGO grade 3 (6.3% vs. 79.2%, P < 0.001) with nonendometrioid and serous-like histology (both 6.3% vs. 58.3%, P = 0.001) and more often presented with lymphovascular space invasion (41.7% vs. 0%, P = 0.003). The 5-year overall survival rate of gBRCA/LOHpos ECs was lower (81.3%) compared with gBRCA/LOHneg ECs (93.3%, P = 0.084; Supplementary Fig. S2).

In total, morphologic characteristics were informative for 39 cases (one case was excluded because of neoadjuvant therapy). A higher frequency of “SET features” in gBRCA/LOHpos ECs was observed compared with gBRCA/LOHneg ECs (52.2% vs. 0%, P < 0.001; Fig. 2). Other histologic features that were significantly more often observed in gBRCA/LOHpos ECs were destructive type of invasion, desmoplastic stromal reaction, nonglandular dominant growth pattern, geographic necrosis, trabecular growth pattern, slit-like spaces, high nuclear grade, tumor giant cells, and a higher median mitotic index (Table 2; Supplementary Fig. S3). We did not find a significant difference for intraepithelial TILs or peritumoral lymphocytes assessed on H&E, nor for CD8-positive T cells (Supplementary Fig. S4). gBRCA/LOHpos ECs were more often estrogen receptor negative (45.5% vs. 6.8%, P = 0.012) and progesterone receptor negative (79.2% vs. 12.5%, P < 0.001) compared with gBRCA/LOHneg ECs.

All ECs were classified into one of the four molecular subgroups previously defined by the TCGA (Fig. 1). All but two gBRCA/LOHpos ECs were classified in the TP53-mutated subgroup, compared with only one of the gBRCA/LOHneg ECs (91.7% vs. 6.3%, P < 0.001). In line with this, gBRCA/LOHpos ECs more often had a CN-high profile compared with gBRCA/LOHneg ECs (95.5% vs. 0%, P < 0.001; Fig. 3). Compared with gBRCA/LOHneg ECs, gBRCA/LOHpos ECs had significantly more mutations in TP53 (95.8% vs. 12.5%, P < 0.001) and fewer mutations in PTEN (16.7% vs. 93.8%, P < 0.001), PIK3CA (16.7% vs. 56.3%, P = 0.015), PIK3R1 (4.2% vs. 43.8%, P = 0.004), ARID1A (4.2% vs. 43.8%, P = 0.004), and CTNNB1 (0% vs. 37.5%, P = 0.002; Fig. 3). In total, gBRCA/LOHpos ECs harbored significantly fewer class 4 or 5 mutations (other than the gBRCA mutation) compared with gBRCA/LOHneg ECs; no statistically significant difference was observed for TMB (Supplementary Fig. S5A and S5B).

To ensure that the ECs did not represent misclassified OCs, salpingo-oophorectomy specimens were rereviewed to detect (pre)malignant lesions. Of the 40 cases included in our final cohort, 39 (97.5%) cases underwent salpingo-oophorectomy either prior to or at the time of hysterectomy. Women who developed gBRCA/LOHpos ECs more often previously underwent a risk-reducing salpingo-oophorectomy (RRSO) compared with women with gBRCA/LOHneg ECs (75% vs. 31.3%, P = 0.009), and the time interval between the RRSO and EC diagnosis was significantly longer; 73.2 months (range, 35.7–187) versus 12.2 months (range, 4.9–82.9, P = 0.037). Because this is a historical cohort, sectioning and extensively examining the fimbriated end was not routinely performed. In total, 36 of 39 (92%) adnexal specimens were available for rereview, of which the fimbriae could be (partially) examined for 16 of 22 (72.7%) of gBRCA/LOHpos ECs and seven of 14 (50%) of gBRCA/LOHneg ECs. None of the ECs showed adnexal involvement and none of the RRSO specimens showed a serous tubal intraepithelial carcinoma (STIC). In two cases, tubal lesions were detected at the time of hysterectomy; one TP53 signature (case 6, USC) and one serous tubal intraepithelial lesion (STIL, case 35, EEC grade 1). In addition, according the pathology report of case 31 (EEC grade 1, adnexa not available for review), the tubal lining showed focal epithelial “atypia and p53 positivity,” which could indicate the presence of a p53 signature, STIL, or STIC. Case 31 presented with a simultaneous EEC and endometrioid ovarian cancer, which were considered to be synchronous primary tumors and not to be secondary adnexal involvement of the EC.

A minority of cases displayed WT-1 positivity (n = 7, 17.5%), of which three (7.5%) displayed heterogenous staining; two USCs, one UCS, and four (10%) displayed diffuse staining; one USC, one UCS, and two ambiguous cases (Table 2). Six of seven women with a WT-1-positive EC had a history of RRSO, none of which showed a (pre)malignant lesion upon rereview. For all but one (case 5), slides available for rereview included sections through the fimbriae. For case 5 (EC diagnosis 2015), the fimbriae could not be examined because of scarring of the fimbriae as a result of a previous bilateral oophorectomy (1995) performed prior to the salpingectomy (2005), as the complete tubes were submitted for histology review. For the one WT-1-positive EC that did not have a history of RRSO (case 6), both adnexa were removed during therapeutic hysterectomy and a p53 signature was detected in one fallopian tube. When excluding all ECs that displayed WT-1 staining, nonendometrioid and serous-like histology remained significantly more common in gBRCA/LOHpos ECs than in gBRCA/LOHneg ECs (both n = 7/17, 41.2% vs. n = 1/16, 6.3%, P = 0.039).

In total, 19 women had a history of breast cancer, which was not significantly different for women with gBRCA/LOHpos ECs compared with gBRCA/LOHneg ECs (54.2% vs. 37.5%, P = 0.349). Although women with gBRCA/LOHpos ECs more frequently had a history of tamoxifen use (including one case for which the type of hormone treatment was not specified), this difference was not significant (n = 6, 25% vs. n = 1, 6.3%, P = 0.210; Table 1; Fig. 1). When excluding all tamoxifen-treated individuals, nonendometrioid and serous-like histology remained significantly more common in gBRCA/LOHpos ECs than in gBRCA/LOHneg ECs (both n = 8/18, 44.4% vs. n = 1/15, 6.7%, P = 0.021). Across the entire cohort (both gBRCA/LOHpos and gBRCA/LOHneg), a history of tamoxifen use was significantly associated with serous-like histology (n = 6/15, 40% vs. n = 1/25, 4.0%, P = 0.007). When only including women who received tamoxifen for 2 or more years (excluding the patient for which hormone treatment duration was unknown), this association was not observed anymore (n = 3/14, 21.4% vs. n = 1/25, 4%, P = 0.123).

This is the first study to describe gBRCA-associated EC as a distinct entity enriched for high-grade, nonendometrioid tumors with frequent TP53 mutations and recurring morphologic features. LOH of the wild-type gBRCA allele was present in 60% of ECs diagnosed in gBRCA carriers, and therefore these should be regarded as “gBRCA-associated ECs.” Importantly, the remaining 40% did not show LOH and therefore are “sporadic ECs” despite the presence of a gBRCA mutation. gBRCA-associated ECs were histologically high-grade in 79%, which is much more frequent than the 21% to 28% of ECs that would be expected on the basis of population frequencies (36, 37). We have shown that these tumors are not misclassified OCs, nor exclusively the result of previous tamoxifen treatment. In summary, our findings strongly support that EC is part of the gBRCA-associated HBOC syndrome.

There are no strict criteria to which a tumor type should adhere to be considered part of a hereditary cancer syndrome. It is generally accepted, however, that tumors that are part of a cancer syndrome should occur more frequently and develop at a younger age compared with what would be expected in the general population. A distinct phenotype of tumors in a cancer syndrome is considered to be in support of a causal relationship. Although previous studies show contradictory results about excess risk of EC (all histotypes) for gBRCA carriers (6–11, 38), most recent studies did find increased risks to develop serous-like ECs, with reported standardized incidence ratios (SIR) ranging from 14.29 to 32.2 (6, 7, 10). These SIRs are comparable with the reported relative risk increase for prostate cancer (up to 20-fold) and pancreatic cancer (up to 10-fold) for gBRCA2 carriers (1). The gBRCA-associated ECs in our study were diagnosed at a median age of 60.5 years (range, 33–74 years). Because these tumors were enriched for EC histotypes that generally occur at an older age (e.g., USC, UCS, EEC grade 3; refs. 36, 37), our data are suggestive that gBRCA-associated ECs indeed occur at a younger age compared with their sporadic counterparts, although no definitive conclusions can be drawn without a proper control group. The combination of the excess risk reported in literature and the phenotype of gBRCA-associated EC described here strongly support adding (TP53-mutated/serous-like) EC to the HBOC syndrome.

Our observation that gBRCA-associated (gBRCA/LOHpos) EC and sporadic (gBRCA/LOHneg) EC show marked histologic and molecular differences supports previous findings that tumors arising in gBRCA carriers are not necessarily causally related to the gBRCA1/2 mutation (12). ECs arising in gBRCA carriers showed LOH relatively infrequently (67.7% of gBRCA1 and 40% gBRCA2) compared with OCs and breast cancers in gBRCA1 carriers (93% and 90%) and OCs in gBRCA2 carriers (84%) but with similar rates to what has been found for breast cancers in gBRCA2 carriers (54%; ref. 12). This is an important finding, as it emphasizes that tumors that develop in gBRCA carriers are not HRD per default and thereby may not respond to treatments targeting this DNA repair defect. This concept impacts the interpretation of clinical trials assessing efficacy of PARP inhibitors in tumors with BRCA1/2 mutations that show LOH relatively infrequently and suggests that LOH should be included in stratification algorithms for studies assessing therapy efficacy in tumors from gBRCA carriers (39–42). In fact, LOH status may explain the less pronounced efficacy of olaparib (PARP-inhibitor) for gBRCA2 carriers with HER2-negative metastatic breast cancer compared with gBRCA1 carriers as observed in the OlympiAD-trial (42).

Our observation should increase awareness of the association between gBRCA and high-grade EC and may have clinical implications in selecting patients with EC and their families for gBRCA testing. Previous studies testing gBRCA mutations in unselected EC cohorts resulted in relatively low incidences (0.5% and 0.6%), with only minor increase (1.1% and 3%) when limited to USC and UCS (43, 44). The morphologic clues described in our study, however, may serve to enrich for gBRCA carriers and therefore facilitate cost-effective gBRCA testing in patients with EC and their families, a concept that merits further study. Currently, one might consider gBRCA testing in patients with high-grade EC with a previous history of breast cancer or a positive family history for gBRCA-associated malignancies. Although our study was not aimed to determine the excess risk in women with gBRCA1/2 mutations to develop EC compared with the general population, our study supports to at least inform gBRCA carriers about the association with EC, as the ECs arising in this background are of an unfavorable subtype.

In this study, it was relevant to ascertain that all included carcinomas were of endometrioid and not of tubo-ovarian origin. To exclude misclassification of secondary involvement of the endometrium by HGSOC as EC, we rereviewed all available salpingo-oophorectomy slides with emphasis on putative precursor lesions in the distal fallopian tube. None of the serous-like ECs showed adnexal involvement, supporting the endometrium as primary origin. In addition, we stained all ECs for WT-1, a marker that assists in distinguishing between USC and HGSOC, with reported nuclear positivity rates ranging from 0% to 44% for USCs and 95% to 100% for serous OCs (45–48). Although cutoff values for WT-1 positivity are unclear, “diffuse WT-1” is generally accepted to be uncommon in EC. WT-1 positivity was observed in seven of 40 ECs (17.5%), of which four (10%) showed diffuse WT-1 positivity. There was no macro- and microscopic indication for a tubo-ovarian carcinoma in the WT-1 positive ECs; nevertheless, we cannot completely rule out the theoretical possibility of a “drop metastasis” from the fallopian tube. The large time interval between the RRSO and EC diagnosis (median 5.7 years, range, 4.0–9.4 years) that was previously performed in six of seven cases, in combination with the absence of any tubal involvement upon rereview, favors primary endometrial origin. For the remainder WT-1-positive EC (case 6), both adnexa were removed during therapeutic hysterectomy, in which a p53 signature was detected unrelated to the EC. We therefore conclude that all cancers in this study, including those that showed WT-1 positivity, are most likely of primary endometrial origin.

Another relevant aspect is a history of tamoxifen treatment, as 2 or more years of tamoxifen treatment has been associated with a two- to sevenfold increased risk to develop ECs (49–52). ECs of tamoxifen-treated individuals are enriched for less favorable histologic subtypes compared with nontreated individuals, especially carcinosarcomas and sarcomas (10.6%–13.8% vs. 2.9%–8.7%, respectively), and for ECs with abnormal p53 expression (49, 53, 54). Tamoxifen is thought to have a stimulatory effect on the endometrium and uterine body while having an antiestrogenic effect in breast tissue (49, 55). This stimulatory effect on the endometrium is unlikely the responsible mechanism for the observed association with serous-like ECs as these ECs are mostly hormone independent (49). A more plausible, alternative hypothesis for this association may be the DNA-damaging effect of tamoxifen. It has been suggested that tamoxifen induces the generation of reactive oxygen species (ROS; ref. 56). ROS can cause DNA damage resulting in replicative stress and DNA double-stranded break formation (1, 57). Previous literature showing the association between tamoxifen use and EC risk did not take gBRCA status into account. In our study cohort of gBRCA carriers, we found an enrichment for serous-like histology in women previously treated with tamoxifen. We recently showed that BRCA1/2-mediated HR is commonly abrogated in TP53-mutated serous-like ECs (24). Cells that are HRD are more prone to DNA damage due to the error-prone repair of the DNA double-strand breaks caused by ROS and estrogen metabolites (58). Thereby, we hypothesize that tamoxifen might facilitate (but not initiate) early carcinogenesis of serous-like precursors in gBRCA1/2 carriers, as these women are already more prone to develop these tumors. This hypothesis should be further studied, as it may alter the balance between advantages and disadvantages of tamoxifen treatment in gBRCA carriers.

This study has some limitations. First, we did not include a matched control group of ECs from non-gBRCA1/2 carriers. Therefore, we are unable to assign sensitivity and specificity of the morphologic features described. Second, we have defined gBRCA-associated EC based on LOH status alone and did not interrogate the presence of BRCA-related genomic scars to support our definition of gBRCA-associated EC. Third, the study design, in which women were included only after providing informed consent and in which ECs were collected both retrospectively (period before providing informed consent) and prospectively (period after providing informed consent), may have led our study cohort to be enriched for ECs with more favorable histotype and survival.

In conclusion, we provide novel evidence that EC is part of the gBRCA-related tumor spectrum, with enrichment for EC subtypes associated with unfavorable clinical outcome and distinct histopathologic and molecular features. We also show that tumors with and without LOH of the gBRCA1/2 wild-type allele are clearly different, thereby providing evidence that establishing LOH status is critical when assessing treatment efficacy of drugs targeting HRD in BRCA1/2-mutated tumors.

L.L. Ritterhouse reports receiving speakers bureau honoraria from Bristol-Myers Squibb, Abbvie, Loxo Oncology, and Personal Genome Diagnostics. J.P. Segal is an unpaid consultant/advisory board member for Novartis, Bristol-Myers Squibb, and AstraZeneca, and reports receiving commercial research support from Abbvie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.M. de Jonge, C.D. de Kroon, M.P.G. Vreeswijk, C.J. van Asperen, V.T.H.B.M. Smit, T. Bosse

Development of methodology: M.M. de Jonge, J.P. Segal, M.A. Rookus, B.E. Howitt, T. Bosse

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.M. de Jonge, L.L. Ritterhouse, C.D. de Kroon, J.P. Segal, HEBON Group, H. Hollema, M.A. Rookus, C.J. van Asperen, F.E. van Leeuwen, V.T.H.B.M. Smit, B.E. Howitt, T. Bosse

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.M. de Jonge, L.L. Ritterhouse, C.D. de Kroon, M.P.G. Vreeswijk, J.P. Segal, M.A. Rookus, F.E. van Leeuwen, V.T.H.B.M. Smit, B.E. Howitt, T. Bosse

Writing, review, and/or revision of the manuscript: M.M. de Jonge, L.L. Ritterhouse, C.D. de Kroon, M.P.G. Vreeswijk, J.P. Segal, R. Puranik, H. Hollema, M.A. Rookus, C.J. van Asperen, F.E. van Leeuwen, V.T.H.B.M. Smit, B.E. Howitt, T. Bosse

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.M. de Jonge, L.L. Ritterhouse, R. Puranik, T. Bosse

Study supervision: C.J. van Asperen, V.T.H.B.M. Smit, T. Bosse

Other (responsible for quality evaluation, processing, and troubleshooting of samples on next-generation sequencing platform in Clinical Laboratory): R. Puranik

The Hereditary Breast and Ovarian Cancer Research Group Netherlands (HEBON) consists of the following Collaborating Centers: Netherlands Cancer Institute (coordinating center), Amsterdam, the Netherlands: M.A. Rookus, F.B.L. Hogervorst, F.E. van Leeuwen, M.A. Adank, M.K. Schmidt, and D.J. Jenner; Erasmus Medical Center, Rotterdam, the Netherlands: J.M. Collée, A.M.W. van den Ouweland, M.J. Hooning, and I.A. Boere; Leiden University Medical Center, Leiden, the Netherlands: C.J. van Asperen, P. Devilee, R.B. van der Luijt, and T.C.T.E.F. van Cronenburg; Radboud University Nijmegen Medical Center, Nijmegen, the Netherlands: M.R. Wevers and A.R. Mensenkamp; University Medical Center Utrecht, Utrecht, the Netherlands: M.G.E.M. Ausems and M.J. Koudijs; Amsterdam Medical Center, Amsterdam, the Netherlands: H.E.J. Meijers-Heijboer and T.A.M. van Os; VU University Medical Center, Amsterdam, the Netherlands: K. van Engelen and J.J.P. Gille; Maastricht University Medical Center, Maastricht, the Netherlands: E.B. Gómez-Garcia, M.J. Blok, and M. de Boer; University of Groningen, Groningen, the Netherlands: J.C. Oosterwijk, A.H. van der Hout, M.J. Mourits, and G.H. de Bock; The Netherlands Comprehensive Cancer Organisation (IKNL), Utrecht, the Netherlands: S. Siesling and J.Verloop; and the nationwide network and registry of histo- and cytopathology in the Netherlands (PALGA): E.C. van den Broek. HEBON thanks the study participants and the registration teams of IKNL and PALGA for part of the data collection. The authors thank E.J. Dreef and N.T. ter Haar (LUMC) for their technical assistance, Sabah Kadri and Sushant Patil (UC) for their bioinformatics assistance, and Wilbert Zwart (NKI). They thank all pathology departments from the hospitals that have send pathology material for study purposes, including the NKI-AVL Biobank. M.M. de Jonge thanks the “Leids Universiteits Fonds/Fonds Van Trigt” and the “René Vogels Foundation” for their financial support in the form of an International Travel Grant. The HEBON study is supported by the Dutch Cancer Society grants NKI1998-1854, NKI2004-3088, and NKI2007-3756, the Netherlands Organisation for Scientific Research grant NWO 91109024, the Pink Ribbon grants 110005 and 2014-187.WO76, the BBMRI grant NWO 184.021.007/CP46, and the Transcan grant JTC 2012 Cancer 12-054. This work was supported by internal departmental funds [Stanford University School of Medicine (to B.E. Howitt) and Leiden University Medical Center (to T. Bosse)]. M.M. de Jonge received an International Travel Grant from the “Leids Universiteits Fonds/Fonds Van Trigt” and the “René Vogels Foundation.”

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.