Introduction

Primary mucinous ovarian carcinomas are relatively infrequent tumors. They are most often low grade, indolent carcinomas, although it has been shown that prognosis of patients with mucinous ovarian carcinomas is strongly stage dependent [1]. In the now widely accepted, dualistic model of ovarian carcinogenesis, they are considered type I tumors, developing through a well-defined cystadenoma—atypical proliferative mucinous (borderline) tumor—carcinoma sequence. Although the cell of origin of primary mucinous tumors of the ovary is unknown, a growing body of literature suggests Brenner tumors and teratomas as possible sources [2].

Brenner tumors are ovarian tumors consisting of nests of transitional epithelium. The presence of mucinous tumors within Brenner tumors has been described with a frequency up to 33% [3,4,5]. Seidman et al. found specific types of calcifications occurring in both mucinous and Brenner tumors and complementary size distributions [4]. Immunohistochemical studies have shown that both the mucinous and the Brenner components have a non-Müllerian phenotype generally lacking expression of PAX2 and PAX8, while GATA3 expression in both components may be suggestive of a common source [3]. More recently, origin of mucinous tumors in Brenner tumors has also been supported by clonality studies using microsatellite genotyping [6]. MYC amplification found in a number of Brenner tumors and associated mucinous tumors is also suggestive of a clonal relationship [7].

Teratomas are thought to arise from oocytes following duplication errors in meiosis [8, 9]. Since all three germ layers are potentially allowed to develop, any type of tissue can be present in these tumors. Mucinous neoplasms coinciding with teratomas are commonly noted and express gastrointestinal markers, incompatible with a Müllerian phenotype [10,11,12]. Molecular studies using microsatellites and short tandem repeats have indicated a clonal relationship between teratomas and associated mucinous tumors [13,14,15].

Little is known about specific genetic alterations in these tumors such as copy number alterations or mutations and their role in progression from mucinous cystadenomas into atypical proliferative tumors and ultimately carcinomas. Ryland et al. have shown KRAS, TP53, and CDKN2A mutations in both mucinous cystadenomas, atypical proliferative tumors and carcinomas, whereas BRAF and NRAS mutations occurred solely in atypical proliferative tumors and carcinomas. ERBB2 amplification appears to be a late event, since no cystadenomas showed this [16]. MacKenzie et al. carried out next generation sequencing, finding comparable results with additionally PIK3CA mutations occurring in a number of atypical proliferative tumors and carcinomas [17].

The aim of this study was to examine a possible clonal relationship between mucinous ovarian tumors occurring with Brenner tumors or teratomas in the same ovary, based on morphological, immunophenotypical and molecular changes. The genome-wide copy number analysis and somatic gene panel analyses for this purpose also served to investigate the differences in the development of mucinous tumors arising from Brenner tumors and teratomas.

Materials and methods

Case selection

The nationwide network and registry of histo- and cytopathology in the Netherlands codes and saves pathology reports in the Netherlands from 1971 with nationwide coverage from 1991 [18].

A search for all patients with Brenner tumor-associated mucinous tumors and teratoma-associated mucinous tumors was performed in the National Pathology Database (registration number lzv1140A and B). For this main search, Brenner tumor-associated mucinous tumors and teratoma-associated mucinous tumors were defined as any type of mucinous ovarian tumor occurring with either any type of Brenner tumor or any type of teratoma in the same ovary. Other inclusion criteria were diagnosis on complete resection of the ovary from 2000 until 2014. Patients with any contralateral ovarian tumor other than a Brenner tumor or teratoma (either with or without a mucinous tumor) were excluded, as were patients with any other genital tract malignancy. Histories of ovarian and tubal tumors to rule out inclusion of recurrences and gastrointestinal tract tumors to rule out metastatic ovarian tumors were searched additionally. The archives of the Johns Hopkins Hospital were also searched for available cases.

Demographic, macroscopic and histologic features

Age, laterality, and size were extracted from pathology reports. Slides of a subset of cases could be requested from 12 collaborating hospitals and were reviewed. Diagnosis, distribution of Brenner nests and mucinous cysts (separate compartments or mixed), presence of mucinous metaplasia of Brenner nests, presence of mucinous stromal dissection and calcifications were recorded. If the Fallopian tube was resected, the presence of Walthard cell nests and transitional cell metaplasia, defined as occurring at the tubo-peritoneal junction, were recorded as well.

Additional database searches: comparison of tumor subtypes

For age comparison between ovarian tumor subtypes, a search for all ovarian tumors of endometrioid and clear cell histology from 2000 until 2010 and for all serous tumors in 2000, 2005, and 2010 was performed in the National Pathology Database (registration number lzv1140C). Of serous cystadenomas, serous borderline tumors, serous adenocarcinomas, endometrioid adenocarcinomas and clear cell adenocarcinomas, 200 cases each were included. Mixed tumors, bilateral tumors with different subtypes or differentiation, recurrences, tumors of uncertain origin and patients with a history of other malignancies were excluded. Endometrioid tumors were included if histology of the uterus was available and negative for endometrial endometrioid adenocarcinoma. Patients with other malignancies of the internal genital tract were excluded.

Tissue micro-array construction

Of a subset of requested cases found through the main search in the National Pathology Database, tissue blocks containing formalin-fixed paraffin embedded tissue were requested and 2 mm cores were taken for construction of tissue micro-arrays using the 3DHistech TMA Grand Master (Sysmex, Switzerland). Two cores were taken from each mucinous component and two cores of each Brenner component. Tissue micro-arrays were used for immunohistochemical staining and fluorescent in situ hybridization (FISH).

Immunohistochemical staining and FISH

Slides were stained for CK7, CK20, CDx2, ER, PR, WT1, PAX8, MUC4, CEA-m, Ki-67, p53, OC125, HNF4α, SATB2, p16, GATA3, Her2/neu, MSH2, MSH6, MLH1, PMS2. FISH was carried out for ERBB2 and MYC. Details are found in the Supplementary Methods.

Interpretation of staining

All stains were scored for both dominant intensity (0 = negative, 1 = weak, 2 = moderate, 3 = strong) and percentage of positive tumor cells unless indicated otherwise. These scores were multiplied and the scores of both tissue micro-array cores were averaged. A score 0–20 was considered negative, 21–100 weakly positive, 101–200 moderately positive and 201–300 strongly positive. For binary scoring, a score higher than 20 was considered positive. Mismatch repair protein staining was considered positive in case of any amount of at least weak nuclear staining. P53 was considered mutated in case of diffuse (>60%), strong staining or complete absence of staining and wild type in case of heterogeneous staining. Ki-67 was scored as percentage of positive cells. Her2/neu staining, ERBB2 FISH and ultimate Her2/neu status were determined according to the 2013 ASCO/CAP guidelines for breast cancer [19]. MYC FISH was scored as amplified when ratio MYC/cep8 ≥2.0 or nonamplified when the ratio <2.0.

DNA isolation

Areas of interest were marked on HE-stained slides and tumor cell percentage was scored (see Table S6 for estimated tumor cell percentages). Five to eight 30-µm-thick sections were cut from formalin-fixed paraffin embedded material, depending on the absolute amount of selected tumor cells. At the end a 4-µm-thick slide was HE-stained for comparison of the tumor containing area. After the sections were mounted on glass slides, the area of interest was manually microdissected and put into a 1.5 mL Eppendorf tube. Depending on the amount of tissue, 100–200 µL of 5% Chelex-100 in TET lysis buffer was added. Of this amount, 10% of proteinase K (20 mg/mL) was added. Samples were incubated in thermomix (Thermomixer comfort, Eppendorf) for at least 16 h at 56 °C, 350 rounds per minute (rpm). When all tissue was dissolved, samples were incubated for exactly 10 min at 95 °C, 350 rpm and centrifuged for 10 min at room temperature, 14,000 rpm. The supernatant containing DNA was transferred to a second 1.5 mL Eppendorf tube, separating it from the Chelex-100 and coagulated paraffin. DNA concentrations were measured using Qubit dsDNA BR Kit (Thermofisher). DNA from normal tissue was isolated if available from the same or another tissue block.

Genome-wide DNA copy number analysis

Shallow genome-wide sequencing for copy number analysis was performed at The Tumor Genome Analysis Core, department of Pathology, VU Cancer Center Amsterdam. In short, library preparation was carried out, comprising fragmentation of DNA by sonication, repair of DNA ends, 3′ adenylation, adapter ligation and purification and PCR amplification of the fragments with adapters and again purification. The Illumina Next Generation Sequencing platform was used for shallow whole-genome sequencing. The shallow whole-genome sequencing data were analyzed using the Bioconductor package QDNAseq [20]. Reads per 100 kbp bin were counted to make a genome-wide profile. Correction for sequence mappability and GC content was carried out, as well as filtering of problematic genomic regions. A subset of normal samples was analyzed and profiles were used for dewaving tumor genome-wide copy number profiles using the R-package ‘NoWaves’ [21]. Details of this method are described elsewhere [20].

Next generation sequencing using single-molecule molecular inversion probes

Details of this method are extensively described elsewhere [22]. Samples were selected based on a tumor cell percentage of at least 20%. For preparation of the single-molecule molecular inversion probe pool for targeted enrichment, molecular inversion probes were designed for 56 cancer related genes using the procedure described elsewhere [23] (see Table S1). The single-molecule molecular inversion probes were divided over two pools. The molecular ratio between gDNA and single-molecule molecular inversion probes was set to 1:800 for every individual single-molecule molecular inversion probe. Per sample, four PCRs were performed and the resulting PCR products were pooled. Per run 47 up to 77 libraries were sequenced, depending on the presence of normal tissue samples for which lower read depth was accepted.

Variant selection

The commercial analysis software SeqNext (Sequence Pilot, JSI medical systems, Ettenheim, Germany) was used for read mapping and variant calling. Settings for generating consensus reads and variant calling are described elsewhere [22]. During coverage calculations read pairs are counted as two reads. Minimal absolute coverage was set at ≥20 combined reads. Variants were only called if the variant percentage was ≥5% and ≥5 combined reads present showing the variant. Variants identified in normal samples were filtered from the matching tumor samples. Intronic variants (outside the canonical splice sites) and synonymous variants were discarded. Variants with at least a three-fold difference in the number of forward and reverse reads, and SNPs with a population allele frequency in ExAC [24] ≥0.01 were removed.

Remaining variants with at least ≥10% allele frequency ≥10 combined reads and detection in at least two molecular inversion probes targeting opposite strands were selected. Variants were excluded if SeqNext hinted at cytosine deamination and allele frequency or number of combined reads were <10% or <10%, respectively. If these were ≥10% or ≥ 10%, variants were manually curated, as were those with mutant allele frequency between 5% and 10%. Exclusion criteria during manual curation were the following (a) variants within homo-polymers; (b) variants within reads with many Ns; (c) variants only present directly at the start or end of reads and absent in the overlapping molecular inversion probe; (d) variants in poorly aligned reads or reads representing pseudogenes; (e) suspected cytosine deaminations.

Concerning clonality, loci of variants detected in either a mucinous tumor or Brenner tumor or teratoma, were specifically reassessed in the complementary tumor (i.e., the aligned reads were evaluated) and coverage of the specific locus was determined.

Coverage per unique region was calculated using the aligned reads based on the output tables from SeqNext.

Statistical analysis

For comparison of means of two populations, independent t-tests were used, for distributions in two or more populations, Pearson’s Chi-squared test was used (IBM SPSS Statistics v20.0). The statistical programming language R was used for copy number data analyses. To test for clonality between tumor pairs the R-package ‘Clonality’ was used [25]. First, the chromosomes were split into p and q arms, and the dewaved, median-normalized log2-transformed read counts per 100 kbp bin were averaged into blocks of three consecutive markers. Subsequently, clonality analysis was performed. This methodology uses an algorithm to identify the most significant change in each chromosome arm, then calls gains or losses and looks at matching patterns of gains or losses for the matching sample pairs and sample combinations that are independent by default using a log likelihood ratio measure. A p-value of < 0.05 was considered significant. In addition, the dewaved, median-normalized log2-transformed read counts per 100 kbp bin were segmented using R-package ‘DNAcopy’ [26] to determine the Pearson correlation between samples.

To identify differences in copy number alterations between Brenner tumor-associated and teratoma-associated mucinous tumor samples, these were called into five categories (deletion, loss, normal, gain or amplification) with the R-package ‘CGHcall’ [27] and a dimension reduction was performed with the R-package ‘CGHregions’ [28]. Subsequently, the Wilcoxon rank-sum test was performed for each chromosomal sub-region. A p-value of < 0.05 was considered significant.

Results

Age in relation to Brenner tumor-associated and teratoma-associated mucinous tumors

The main search identified 531 cases, of which 516 through the National Pathology Database and 15 through the Johns Hopkins Hospital. Of these, 352 were Brenner tumor-associated mucinous tumors, 178 were teratoma-associated mucinous tumors and one case had both a Brenner tumor and teratoma with a mucinous cystadenoma in one ovary. Mean age of all patients was 55 years. Mean age of patients with Brenner tumor-associated mucinous tumors was 61 and of patients with teratoma-associated mucinous tumors 43 years (p < 0.001), for carcinomas only, this was 60 and 62 years (p = 0.84).

Patients with Brenner tumor-associated mucinous tumors showed no age difference with patients with endometrioid (p = 0.25) and serous adenocarcinomas (p = 0.85). They were older than patients with clear cell carcinomas (mean age 59 years, p = 0.02) and patients with all serous tumors together (mean age 56 years, p < 0.01). Patients with teratoma-associated mucinous tumors were younger compared to patients with endometrioid (60 years, p < 0.01), clear cell (59 years, p < 0.01) and serous adenocarcinomas (61 years, p < 0.01) and all types of serous tumors together (56 years, p < 0.01) (Fig. 1 and Table S2).

Fig. 1
figure 1

Age distribution of patients with endometrioid, serous and clear cell adenocarcinomas and Brenner tumor-associated mucinous tumors and teratoma-associated mucinous tumors

Histology of Brenner tumor-associated and teratoma-associated mucinous tumors

Revised cases included 104 Brenner tumor-associated mucinous tumors (80 cystadenomas, 20 atypical proliferative tumors, and 4 carcinomas) and 58 teratoma-associated mucinous tumors (40 cystadenomas, 14 atypical proliferative tumors and 4 carcinomas) (Fig. 2a, b). One case (1%) was excluded after alteration of the diagnosis to serous cystadenoma, in 14 cases (9%) diagnosis was alters in means of cystadenoma, atypical proliferative tumor or carcinoma. Mean size, tumor laterality and frequency of salpingectomy did not differ significantly between the two groups. In case of Brenner tumor-associated mucinous tumors only, mucinous metaplasia was seen significantly more often in cases with mixed distribution of Brenner nests and mucinous cysts than in cases with compartmental distribution (63% vs. 25%, p < 0.01, Fig. 3a–c). Calcifications were significantly more often seen in Brenner tumor-associated mucinous tumors than in teratoma-associated mucinous tumors (56% vs. 22%, p < 0.01, Fig. 3d). Mucinous stromal dissection was significantly more often seen in teratoma-associated mucinous tumors (13% vs. 43%, p < 0.01, Fig. 2e). When the Fallopian tube was available for review, tubal Walthard cell nests were seen more often in Brenner tumor-associated mucinous tumors than teratoma-associated mucinous tumors (42% vs. 15%, p < 0.01), whereas for transitional cell metaplasia no difference was observed (Fig. 3f–g).

Fig. 2
figure 2

a, b Differences in frequencies of histological characteristics and protein expression patterns in Brenner tumor-associated and teratoma-associated mucinous tumors, respectively. c Differences in frequencies of histological characteristics and protein expression patterns in Brenner tumor-associated and teratoma-associated mucinous tumors. Brenner tumor-associated mucinous tumors are indicated in blue, teratoma-associated mucinous tumors in red from light to dark for cystadenomas, atypical proliferative mucinous (borderline) tumors and carcinomas, respectively

Fig. 3
figure 3

a, b Histology of mixed and compartmental distribution of Brenner tumor and mucinous components. c Mucinous metaplasia in Brenner tumor. d Calcifications. e Mucinous stromal dissection. f Transitional cell metaplasia. g Walthard cells nests. h–n Protein expression of CK7, CK20, CDx2, CEA, HNF4α, SATB2 and PAX8, respectively

Protein expression in Brenner tumor-associated and teratoma-associated mucinous tumors

Thirty-eight Brenner tumor-associated mucinous tumors (17 cystadenomas, 17 atypical proliferative tumors, and 4 carcinomas) and 27 teratoma-associated mucinous tumors (12 cystadenomas, 11 atypical proliferative tumors and 4 carcinomas) were randomly selected for immunohistochemical analyses (Figs. 2c, 3h–n). Brenner tumor-associated mucinous tumors were significantly more often positive for CK7 (p = 0.02) and negative for both CDx2 (p = 0.04), MUC4 (p = 0.02) and SATB2 (p < 0.01). Associated Brenner tumors showed comparable patterns. No differences were observed between Brenner tumor-associated and teratoma-associated mucinous tumors for gastrointestinal markers such as CK20, HNF4α, and CEA-m and gynecological markers such as OC-125, ER, PR, PAX8, and WT1.

For carcinomas only, there were no statistical differences, although all Brenner tumor-associated carcinomas were negative for SATB2. Mismatch repair proteins were expressed in all carcinomas. Her2/neu status and MYC amplification was not different between Brenner tumor-associated and teratoma-associated mucinous tumors.

Expression of CK20, CDx2, CEA-m and HNF4α increased significantly when Brenner tumor-associated mucinous tumors progressed (cystadenoma—atypical proliferative tumor—carcinoma) (p < 0.01, p < 0.01, p = 0.02 and p = 0.03, respectively), whereas CK7 expression decreased in this group (p < 0.01). Teratoma-associated mucinous tumors showed a comparable trend without reaching statistical significance.

One Brenner tumor-associated carcinoma and one teratoma-associated atypical proliferative tumor were positive for Her2/neu. No MYC amplification was noted in any tumor. See Table S3.

Copy number analysis in Brenner tumor-associated and teratoma-associated mucinous tumors

Copy number analysis was performed for 24 Brenner tumor-associated and 22 teratoma-associated mucinous tumor pairs that were randomly selected. Twelve random normal samples (6 from Brenner tumor-associated and 6 from teratoma-associated mucinous tumor patients) were used for normalization of tumor profiles. Brenner tumor-associated mucinous tumors showed a mean of 7% aberrant bins compared to 19% in teratoma-associated mucinous tumors (p = 0.04). For carcinomas only, the mean percentage of aberrant bins was 14% for Brenner tumor-associated mucinous tumors and 34% for teratoma-associated mucinous tumors (p = 0.24). Brenner tumors had 3% aberrant bins compared to 2% in teratomas (p = 0.63). Frequency plots of copy number alterations in Brenner tumor-associated and teratoma-associated mucinous tumors are found in Figure S1. The ten lowest p-values were selected/ (range 0.02–0.07). Statistically significant were alterations of 5q11.1–5q13.2 (p = 0.02), 5q13.2–5q13.3 (p = 0.02), 5q13.3 (p = 0.04), 5q13.3-5q35.3 (p = 0.02), 9p21.3a (p = 0.03) and 11p15.5–11q25 (p = 0.03). Regions 9p21.3b, 4p16.3–4q35.2, 9q22.32 and 13q11–13q34, showed trends but no statistical significance. Some of the regions concerned large parts of chromosomal arms (see Fig. S2 and Table S4). The region 9p21.3b overlaps with the genes CDKN2A/CDKN2B, and 9p21.3a is positioned just 100 kbp upstream. The gene MTAP is localized in between 9p21.3a and 9p21.3b. Although not statistically significant, deletion or loss of CDKN2A/CDKN2B was detected in eight Brenner tumor-associated mucinous tumors (33%), compared to only two deletions in teratoma-associated mucinous tumors (9%). These occurred only in atypical proliferative tumors and carcinomas. Co-deletion or loss of 9p.21.3a and MTAP was found in eight out of ten cases. Three Brenner tumors showed an identical gain of 5q13.2–5q13.3 as the associated mucinous tumor. No other identical copy number alterations were observed for the regions mentioned. Details of copy number alterations of these regions per mucinous tumor are displayed in Fig. 4a.

Fig. 4
figure 4

Copy number alterations and mutations specifically depicted per tumor. a The heatmap showing amplifications, deletions, copy number gains and losses in Brenner tumor-associated and teratoma-associated mucinous tumors. Regions 9p21.3a and b coincide with MTAP and CDKN2A/CDKN2B. *p < 0.05. b The heatmap showing mutations per gene for Brenner tumor-associated and teratoma-associated mucinous tumors. Brenner tumor-associated mucinous tumors are indicated in blue, teratoma-associated mucinous tumors in red from light to dark for cystadenomas, atypical proliferative mucinous (borderline) tumors and carcinomas, respectively

Based on copy number alterations, one cystadenoma and Brenner tumor were likely clonal. At first, clonality analysis showed three tumor pairs with a high likelihood of clonality (Fig. 5a). However, two cases showed 0% and around 1% copy number alterations, while one case, a Brenner tumor and cystadenoma, showed a high frequency of copy number alterations in both samples (Fig. 5b). This case also showed an identical gain of 5q13.2–5q13.3 in the Brenner tumor and the associated mucinous tumor. Normal tissue of this patient did not contain the same copy number alteration, excluding germ line variations.

Fig. 5
figure 5

a Scatter plot showing likelihood of clonality based on copy number analysis. The x-axis shows the Pearson’s correlation, the y-axis shows clonality analysis based on copy number alterations of chromosomal arms. Thresholds were set at 90% of the data. The three cases appearing in the upper right dark green square have a high likelihood of clonality. b The percentage of copy number alterations per tumor pair in numerical order (mucinous tumor followed by precursor), with the clonally related cases highlighted  by rectangular boxes. The percentage of copy number alterations in case 30 is 0.0%

Mutations in Brenner tumor-associated and teratoma-associated mucinous tumors

Single-molecule molecular inversion probe-based next generation sequencing mutation analysis was performed for 13 Brenner tumor-associated mucinous tumors and 17 teratoma-associated mucinous tumors, their associated Brenner tumor or teratoma and normal tissue. One Brenner tumor-associated mucinous tumor pair was excluded due to the low coverage. Details of coverage per target region per tumor are found in Table S5.

Mucinous tumors showed mutations in KRAS (37%), TP53 (17%), RNF43 (13%), ELF3 (10%), GNAS (7%), CDKN2A (7%), BRAF (3%), PIK3CA (3%), ERBB2 (3%) and ACVR1B (3%). KRAS, TP53, CDKN2A and ELF3 mutations were observed in both Brenner tumor-associated (31%, 15%, 8% and 15%) and teratoma-associated mucinous tumors (41%, 18%, 6%, and 6%). KRAS and TP53 mutations were found in cystadenomas, atypical proliferative tumors and carcinomas. BRAF and PIK3CA mutations occurred only in Brenner tumor-associated mucinous tumors (both 8%) and mutations in RNF43, GNAS, ERBB2 and ACVR1B occurred only in teratoma-associated mucinous tumors (24%, 12%, 6%, and 6%, respectively). RNF43 mutations did not occur in cystadenomas. One atypical proliferative mucinous tumor and associated Brenner tumor showed an identical deletion in CDKN2A that was not present in normal tissue. Mutations per mucinous tumor are depicted in Fig. 4b. Details on mutations per gene per tumor are found in Table S6.

In the two tumors containing CDKN2A mutations we further investigated the presence of loss of heterozygosity (LOH). In case 12, the variant allele frequency of CDKN2A was 51% with an estimated tumor cell percentage of 40%. This meant that the tumor cell percentage per definition represents an underestimation making it probable that no or very few tumor cells contained a wild type CDKN2A allele, although this was not confirmed with copy number analysis. In case 32, the variant allele frequency was 24% with a tumor cell percentage of 30%. In a nearby region one informative SNP was found with a variant allele frequency of 34% in the tumor sample and 54% in the normal tissue. In case of a tumor cell percentage of 30%, this indicates LOH of CDKN2A. In this case, loss of CDKN2A was also confirmed with copy number analysis.

Discussion

In this study we collected a large cohort of Brenner tumor-associated and teratoma-associated mucinous tumors and their associated Brenner tumor and teratoma, respectively. For two Brenner tumor-associated mucinous tumor pairs we could substantiate a clonal relationship. Several remarkable differences between Brenner tumor-associated and teratoma-associated mucinous tumors were that Brenner tumor-associated mucinous tumors frequently showed loss of CDKN2A and MTAP. In contrast, patients with teratoma-associated mucinous tumors were significantly younger and their tumors expressed SATB2 and showed mutations in RNF43. These findings support the thought that mucinous tumors of the ovary can originate from both Brenner tumors and teratomas, and that these are distinct entities.

Clonality analysis of 46 tumor pairs based on the copy number analysis showed three pairs of both Brenner tumor-associated and teratoma-associated mucinous tumors which were likely clonal. However, since two of these patterns showed the virtually complete absence of copy number alterations, these calls are probably false positive, resulting in only one case of copy number alteration-based clonality. With mutation analysis, we found one identical deletion in CDKN2A in a Brenner tumor and its associated atypical proliferative mucinous (borderline) tumor. Although deletions in this genomic region have been reported before, the absence of this specific deletion in COSMIC suggests it is uncommon and therefore a strong indication for a clonal relationship. Clonality between Brenner tumor-associated and teratoma-associated mucinous tumors and their precursor lesions has been suggested earlier using repetitive DNA sequences [6, 13,14,15]. Our findings further support an origin of primary mucinous ovarian tumors in either Brenner tumors or teratomas. The fact that we could confirm the clonal relationship in only two tumors pairs, may be explained by the limited information obtained by our assays due to the relatively low number of copy number alterations in mucinous ovarian carcinomas, especially compared to highly genetically unstable carcinomas such as high-grade serous carcinomas [29,30,31]. The included Brenner tumors and teratomas also contained very few copy number alterations, meaning that during (pre)malignant transformation tumors become more genetically unstable, developing a more complex copy number profile than their original precursor and thus showing fewer similarities. In short, copy number analysis seems suboptimal for clonality analysis. For mutation analysis, low coverage may have impeded detection of (identical) variants. DNA quality may have influenced our results since we used formalin-fixed paraffin embedded material, sometimes over 15 years old. However, by using single unique molecule identifiers and targeting both strands for next generation sequencing the accuracy increased and formalin induced artifacts could be recognized [22]. Especially for cystadenomas, the amount of tumor cells in a selected slide may be low and difficult to macrodissect manually. Also, the specific subclones in the Brenner tumors and teratomas from which the mucinous component arose may be completely overgrown. Furthermore, we cannot certainly exclude collision tumors with another origin than assumed. However, this is unlikely since the combination of mucinous tumors with Brenner tumors and teratomas is rare.

We found changes that were shared between Brenner tumor-associated and teratoma-associated mucinous tumors. Firstly, frequently mutated genes in both tumors were KRAS and TP53, occurring in cystadenomas, atypical proliferative tumors and carcinomas. KRAS encodes a GTPase protein in the tyrosine kinase induced RAS-RAF-MEK-ERK pathway leading to cell growth, proliferation and survival and is frequently mutated in colorectal cancer [32]. Mucinous cystadenomas, atypical proliferative mucinous tumors and mucinous carcinomas of the ovary have been shown to frequently contain KRAS mutations [16, 17, 33, 34]. This seems to be an early event in tumorigenesis, where the lack of KRAS mutations in precursor Brenner tumors and teratomas suggests a role in development of mucinous phenotype. We found TP53 mutations in cystadenomas, atypical proliferative tumors and carcinomas, consistent with earlier studies [16, 17, 33, 34]. TP53 mutations, which are mainly associated with high-grade serous carcinomas and serve as a hallmark for type II ovarian carcinomas, are generally associated with poor prognosis. These findings suggest that mutations in TP53 are indicative of biological behavior of tumors rather than a feature of specific tumor types and that the assumption of mucinous ovarian carcinomas being solely type I carcinomas is unjust. Although mucinous ovarian carcinomas are generally regarded low grade, indolent tumors, a minority presents as advanced stage disease with extremely poor prognosis [1]. It is possible that the origin of the carcinoma is the underlying cause of this prognostic difference, but this remains to be investigated.

Secondly, Brenner tumor-associated and teratoma-associated mucinous tumors both show a gastrointestinal immunophenotype during transformation towards malignancy. Brenner tumor -associated mucinous tumors showed no expression of typical Müllerian markers such as PAX8 as reported before [3, 10, 12]. This further supports that Brenner tumor-associated and teratoma-associated mucinous tumors follow a comparable carcinogenetic pathway culminating in carcinoma, which is characterized by a gastrointestinal mucinous phenotype. This might offer new possibilities for tailored treatment for this subgroup, along the lines of systemic treatment for colorectal cancer.

Interestingly, there also were remarkable differences between Brenner tumor-associated and teratoma-associated mucinous tumors. Patients with teratoma-associated mucinous tumors were significantly younger than patients with Brenner tumor-associated mucinous tumors and patients with serous, clear cell, and endometrioid adenocarcinomas. There is a bias comparing age data with other carcinoma subtypes, since the majority of teratoma-associated mucinous tumors are premalignant. However, the same holds true for Brenner tumor-associated mucinous tumors, therefore the age difference between Brenner tumor-associated and teratoma-associated mucinous tumors is not affected by this bias. The age difference between Brenner tumor-associated and teratoma-associated mucinous tumors is only present in the premalignant stages, while carcinomas (albeit small in number) are diagnosed at a similar age.

The most striking difference in protein expression between Brenner tumor-associated and teratoma-associated mucinous tumors was SATB2 expression, being (weakly) positive in only one Brenner tumor-associated mucinous tumor and strongly expressed in almost half of teratoma-associated mucinous tumors. SATB2 is a protein involved in transcription regulation and chromatin remodeling, which is thought to be useful in the distinction of primary mucinous carcinomas and metastases of gastrointestinal origin, the latter being generally positive [35, 36]. However, our observations suggest that SATB2 is also positive in teratoma-associated mucinous tumors.

To strengthen the assumption that teratoma-associated mucinous tumors present with a gastrointestinal phenotype, we found RNF43 mutations exclusively in teratoma-associated mucinous tumors, limited to the more advanced stages (atypical proliferative tumors and carcinomas). RNF43 is a tumor suppressor gene negatively regulating Wnt signaling and is also frequently mutated in colorectal carcinomas [37]. Earlier studies also reported mutations in RNF43 in atypical proliferative mucinous tumors (6–9%) and carcinomas (19–21%), but not in cystadenomas [16, 38]. The nearly exclusive occurrence of SATB2 positivity and RNF43 mutations in teratoma-associated mucinous tumors, is an indication that these tumors more closely resemble gastrointestinal tumors than Brenner tumor-associated mucinous tumors, especially in earlier phases.

Copy number alterations deductable to specific cancer related genes were limited, but remarkable was deletion of 9p21 mainly in Brenner tumor-associated mucinous tumors, and exclusively in atypical proliferative tumors and carcinomas. This region contains CDKN2A/CDKN2B and MTAP, and is frequently deleted in a multitude of tumors. CDKN2A and CDKN2B code for the p16INK4A and p14ARF, and the p15INK4B proteins respectively, which have tumor suppressive properties involved in cell cycle regulation. Genetic imbalance of chromosome 9p encompassing CDKN2A has been more often detected in mucinous cystadenomas (60%), atypical proliferative mucinous tumors (77%) and mucinous carcinomas (67–75%) [29, 34, 39]. Another study found LOH and homozygous deletions of CDKN2A in 55%, 59%, and 52% of mucinous cystadenomas, atypical proliferative mucinous tumors and mucinous carcinomas, respectively [16]. We found co-deletion or -loss of MTAP in 80% of cases, which is described in various premalignancies and cancer types in which it also predicts poor outcome [40,41,42]. MTAP codes for a key enzyme in polyamine salvage metabolism. This gene may act as a tumor suppressor, independent of CDKN2A alterations and directly related to its enzymatic activity [43]. Deletion of this gene offers possibilities for targeted therapies attaining purine deprivation or inhibition of purine synthesis [44].

In daily practice, the majority of mucinous ovarian carcinomas will present without an evident precursor. Part of these tumors will concern Brenner tumor-associated and teratoma-associated mucinous carcinomas. In the latter case, characterized by SATB2 expression and mucinous stromal dissection, there is a difficult differential diagnosis with metastatic colorectal carcinomas. In addition, a number of mucinous ovarian carcinomas is PAX8 positive [45, 46], whereas in our study cohort, PAX8 expression was practically absent. This suggests that there might be a yet unclarified pathway leading to development of PAX8 positive mucinous ovarian carcinomas. Validation of the described features on mucinous carcinomas without precursors is needed to further unravel different carcinogenetic pathways, determine necessary pathological workup and examine clinical implications of different subtypes of mucinous ovarian carcinomas.

Concluding, we further substantiated that at least part of primary mucinous ovarian carcinomas originates from Brenner tumors or teratomas. Our data suggest that in both Brenner tumors and teratomas early events such as KRAS mutations are responsible for acquisition of a mucinous phenotype. Subsequently, Brenner tumor-associated and teratoma-associated mucinous tumors follow a comparable carcinogenetic pathway to carcinoma, as shown by closing of the age gap and acquisition of a gastrointestinal immunophenotype. However, specific molecular changes in these pathways, such as deletion or loss of CDKN2A/CDKN2B and MTAP mainly in Brenner tumor-associated mucinous tumors and SATB2 expression and RNF43 mutations nearly exclusively in teratoma-associated mucinous tumors support that Brenner tumor-associated mucinous tumors and teratoma-associated mucinous tumors remain entities with specific characteristics, with teratoma-associated mucinous tumors resembling gastrointestinal tumors most closely. Although we did not include survival or therapy related data, it is worth consideration that either teratoma-associated mucinous tumors or Brenner tumor-associated mucinous tumors might develop into mucinous ovarian carcinomas associated with poor prognosis, more often presenting in advanced stage. Also, specific molecular changes such as KRAS mutation, ERBB2 amplification and deletion of MTAP may have implications for future options for targeted therapy.