The Driver Mutational Landscape of Ovarian Squamous Cell Carcinomas Arising in Mature Cystic Teratoma

Mature cystic teratoma (MCT) of the ovary (also known as dermoid cyst and benign cystic teratoma) is a common benign gynecological tumor, usually arising before the menopause (1). The origin of MCT remains uncertain. Karyotype analyses suggested that there may be five separate types arising from different replication errors during meiosis (2) and that MCT may represent primary oocytes that have escaped from meiotic arrest (3).

Because MCT contain all three germ cell layers, they often display multiple differentiated tissue types, including teeth, bone, and hair. However, secondary transformation into invasive malignancy can also occur, at reported rates of 0.1% to 1% (4, 5). Up to 80% of these transformed teratomas contain squamous cell carcinoma (SCC), with the remaining 20% containing adenocarcinoma, thyroid carcinoma, or carcinoid tumors (6).

Due to their rarity, there are few large published series and no prospective clinical trials in MCT-derived SCC, and the recent Gynecologic Cancer Intergroup (GCIG) consensus review concluded that there were insufficient data to provide clear guidance on optimal treatment (7). However, the prognosis for these SCC is poor, especially for stage II–IV disease (summarized in ref. 6). As a consequence, there is a need to characterize the genomic features of these tumors, to identify opportunities for patients to enroll in clinical trials.

Here we present analysis of the mutational state, copy number, and zygosity of 151 cancer genes, as well as HPV status, in 25 cases of MCT with SCC from four large UK gynecological cancer centers.

All samples were acquired and utilized under the auspices of the NHS Greater Glasgow and Clyde Biorepository following approval by West of Scotland Research Ethics Committee 4 (Reference 10/S0704/60). Overall survival was calculated from the date of diagnosis to the date of death or the last clinical assessment where known. Overall survival was calculated by log-rank test (Mantel–Cox) using Prism v6.0 (GraphPad, San Diego, CA), whereas multivariate analysis was calculated using a Cox Proportional Hazard model in SPSS; P < 0.05 was considered to be significant.

Formalin-fixed paraffin-embedded specimens were identified from the pathology archives of participating centers. Following review by an expert gynecological pathologist, areas of SCC, mature cystic teratoma (MCT), and normal tissue were marked for macro- or microdissection. Laser capture microdissection (LCM) was performed on consecutive cresyl violet-stained sections using a Leica CTR6500 microscope (Leica Microsystems, Milton Keynes, UK). Sections were collected onto LCM-compatible polyethylene terephthalate (PET) frame slides (Leica Microsystems).

DNA was extracted from 10 × 10 μm sections using QIAmp DNA FFPE Tissue Kit (Qiagen) according to manufacturer's protocol. For macrodissected samples, paraffin was removed by the xylene/ethanol method. For microdissected samples, paraffin was removed by the heptane method. Briefly, 500 μL heptane was added to the sample, followed by 10 minutes incubation at room temperature. Twenty-five microliters of methanol was added, followed by centrifugation and aspiration. One milliliter of absolute ethanol was then added to the pellet, followed by centrifugation and aspiration. The pellet was then air dried. From here the standard QIAmp extraction method was followed as per manufacturer's protocol. DNA was quantified on a Qubit 2.0 Fluorometer (Life Technologies, Paisley, UK).

High-/intermediate-risk HPV serotypes (16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 68) were assessed in 50 ng SCC DNA using real-time PCR (HPV High Risk Taqman PCR Kit; Norgen BioTek, Thorold, ON, Canada) using a CFX96 Touch Real-Time PCR System (Bio-Rad, Watford, UK).

Fifty to 200 ng DNA was sheared with a Covaris LE220 focused-ultrasonicator (Covaris) to produce 100 to 200 bp fragments. Libraries were generated using SureSelect XT standard protocol (Agilent Technologies) for low-input and FFPE samples. Pre-capture libraries were quantified as above, then assessed for library size and impurities with a High Sensitivity DNA BioAnalyzer Chip (Agilent Technologies). Libraries were captured with 120nt biotinylated RNA baits designed for the ClearSeq Comprehensive Cancer Panel (Agilent Technologies), covering 151 cancer-associated genes (Supplementary Table S1). Hybridization for capture libraries less than 3Mb was performed according to manufacturer's protocol. Post-capture PCR incorporated primers with unique 8-bp indexes (Agilent Technologies) to facilitate multiplexing. Final captured-libraries were quantified with a Qubit Fluorometer High Sensitivity DNA Assay (Invitrogen) and assessed for size distribution and quality on a High Sensitivity DNA BioAnalyzer Chip (Agilent Technologies). Pooled capture-libraries were sequenced on a NextSeq 500 instrument (Illumina) or MiSeq instrument (Illumina) with 2 × 76-bp, paired-end reads according to manufacturer's instructions.

A full description of mapping, QC, point mutation and indel calling, copy number calling (Supplementary Table S2), and zygosity analysis is provided in Supplementary Methods.

Thirty-one cases were originally identified. Samples from three patients were missing. There were low DNA yields (<50 ng) in two cases, and sequencing failed in one SCC case, leaving 77 samples from 25 cases that were successfully sequenced (Fig. 1; see Supplementary Table S3 for QC). Twenty cases had sample trios (normal, MCT, SCC; Table 1), five of which had two or more regions of SCC and one of which had two MCT samples. Four cases had normal–SCC pairs and one case had MCT–SCC with no matching normal. In the sequenced cases, median age at diagnosis was 51.0 years (range 25–86) and 56% (14/25) had stage I disease (Table 1).

Figure 2 shows a summary of key mutation and copy number alterations in both SCC and MCT—full details are given in Supplementary Tables S4 and S5. MCT were genomically quiet, with few mutations per sample (median 0, mean 1, range 0–7), and only one sample showed high-level copy number alterations—amplification of MYC and EGFR in SC025 (Supplementary Fig. S1A), the latter of which was shared with the arising SCC. Two further MCTs (SC006 and SC013) showed evidence of low-level copy number gains (Supplementary Fig. S1B), whereas the remaining 18 had normal copy number profiles (Supplementary Fig. S1C).

SCC samples contained a median of 8 mutations per case (range 0–39), with an overall median mutational burden of 10.2 mutations per MB of sequenced DNA (range 0–49). This is similar to that of lung SCC (8.4 per MB; refs. 8, 9), and higher than head and neck SCC (3.2–5.0/MB; refs. 10, 11), but lower than skin SCC, which has one of the highest mutation rates of any malignancy (at least 50 mutations/MB; ref. 12). All cases were negative for 13 high-/intermediate-risk HPV genotypes (Supplementary Table S5).

The most commonly altered genes were TP53 (20/25 cases, 80%; Fig. 2B), including 4/20 (20%) cases with mutations in codon 285 (p.E285L/Q), PIK3CA (13/25 cases, 52%), and CDKN2A, both by homozygous deletion (Supplementary Table S6) and by point mutation (11/25 cases, 44%), a pattern similar to lung (8) and skin SCC (12). Many SCCs showed evidence of highly rearranged genomes, with multiple low-level gains and losses (Supplementary Fig. S2A). Amplification of MYC was called in two cases (SC005 and SC021; Supplementary Table S6) and gain of MYC, falling just below the threshold for amplification calling, was seen in a further three cases (SC011, SC020, and SC025), making the potential prevalence as high as 20% (5/25; Supplementary Fig. S2B). The only other recurrent amplification was JAK2 in two cases (SC020 and SC025; Supplementary Table S6).

In five cases, there were two or more separate SCC samples. One of these (SC030) contained no detectable mutations, but in the remaining four cases, the TP53 mutation was an early and shared event, even in samples that showed some heterogeneity between samples (Supplementary Table S6). Two multisample cases (SC022, SC031) showed identical mutation profiles across samples, whereas the remaining two (SC014, SC025) showed some overlap: SC014 shared two mutations (TP53, GNAQ) with PIK3CA and CDKN2A changes in one sample only. SC025 had four mutations in one sample and three in the second, but with only the TP53 variant being shared, and therefore the only one occurring prior to divergence of the two clones. Where there were multiple mutations in one gene (e.g., DDX3X in SC022; APC in SC031), there was complete concordance between samples.

TP53 mutations showed a striking pattern with 8/20 (40%) mutated cases containing two or more variants (Fig. 2A). Several pieces of evidence suggested strongly that these abnormalities are bi-allelic rather than resulting from the presence of two or more discrete clonal populations or the existence of kataegis-like clusters. First, in SC006, the mutations were sufficiently close to be phased, but were not detected in the same sequencing read pair and thus must lie on separate alleles (Supplementary Fig. S3A). Although the allele frequencies (18.4% and 13.1%) and copy number state (neutral, log₂ ratio shift of 0.05) do not preclude the possibility that these mutations are in separate clones, the heterogeneity analysis of the multiregion sampled SCCs shows that even where subclones are present, they share the same TP53 mutation (Supplementary Table S7). Second, high TP53 mutant allele frequencies in two cases can only be reconciled if the mutations are in the same cells. SC013 has TP53 mutation frequencies of 47.2% and 55.8%. If these mutations are in trans, they must account for 100% of TP53 alleles and thus cannot be in separate cells unless those cells are either homozygously mutated (and therefore bi-allelic for the same mutation) or have loss of the second copy to remove all wild-type alleles. However, the copy number state for TP53 in SC013 is a log₂ ratio shift of −0.35, which, at most, would indicate single-copy loss of TP53 in a fraction of cells. The possibility that these mutations are in cis cannot be excluded here. The same logic holds for SC021, which has TP53 mutation frequencies of 36.4% and 46.3% and a TP53 copy number log₂ ratio shift of 0.11, indicating no copy number alteration. Third, two separate mutations were identified in SC005 at the same allele frequency at the same position (R280K at 21.4%, R280T at 22.9%), along with a third mutation (E285K) 14bp away that is in phase with the R280K mutation but always on a different allele to the R280T (Supplementary Fig. S3). There is a fourth, splice region TP53 mutation in this case that was too far away to phase with the other mutations. The copy number state for TP53 in this case again indicated no copy number alteration (log₂ ratio shift of 0.21). Possible explanations are therefore that there are two identically sized subclones, one of which has either one or two TP53 mutations and the other has either two or three, depending on how the distant mutation phases (although it is important to note that the three subclone explanation is not possible in the absence of copy number alteration). Alternatively, all TP53 mutations exist in the same cell with the splice mutation on the same allele as either the E285K/R280K or the R280T mutations. The only explanation common to all three of the scenarios described above is that the TP53 mutations in these cases are bi-allelic.

There were other examples, albeit rare, of genes with multiple mutations, in a pattern consistent with an APOBEC signature (9). Two samples (SC001 and SC005) had more than two TP53 mutations, SC022 had three mutations in DDX3X, SC002 had three PIK3CA mutations, and SC009 had four mutations in NOTCH1. These clusters of mutations were mainly C>T (12/19) and C>G (4/19) but, in a TpC context rather than the CpG context that would indicate the ubiquitous cytosine deamination signature. However, numbers are small and sequencing of a larger genomic footprint would be required to capture enough mutations to carry out a formal signature analysis.

The PIK3CA alterations were canonical activating mutations, with the two most frequent mutations (p.E545K and p.E542K) corresponding to the two most frequent hotspots in Catalogue of Somatic Mutations in Cancer (COSMIC). CDKN2A was inactivated by a variety of mechanisms, the commonest being a large-scale deletion (eight cases), with nonsense mutation occurring in a single case and missense mutations in three cases. The missense mutations all occurred at mutation peaks in COSMIC, consistent with these events being under positive selective pressure and therefore inactivating.

Other recurrent changes included mutation of SMARCA4 and KMT2A (four cases each), gain of MYC (three cases), and gain of JAK2 (two cases). Driver point mutations, as opposed to structural changes, in KMT2A are relatively rare, as evidenced by its point mutation profile in COSMIC, which shows a very low level of mutation spread evenly across the whole length of the gene. KMT2A has a relatively long-coding sequence, making it more likely that mutations occur in it by chance. Given the size of our dataset we are unable to determine whether these KMT2A mutations are part of the so-called “long tail” of driver mutations seen in many cancer types or if they represent the background mutation rate. The SMARCA4 mutations are likely passengers as none of them matched the five most recurrent variants in COSMIC. Mutations in BRCA1 and BRCA2 were identified in two cases each; all mutations were somatic missense alterations (Supplementary Table S4).

Follow up data were available on 23 cases. Twelve patients died and median overall survival was 20.1 months (Fig. 3A), with no deaths occurring more than 24 months following diagnosis. The median follow-up for living patients was 55.8 months (range 6.8–172.6 months) with 91% followed up for at least 41.6 months. As previously described, survival was significantly better for stage I compared to stage II–IV disease (Fig. 3B; HR = 0.301; P = 0.021). There was no difference in survival based on age at diagnosis (Fig. 3C), PIK3CA or CDKN2A mutation status (Supplementary Fig. S4A), but there was a statistically significant association with TP53 mutation (HR = 0.178; P = 0.002; Fig. 3D). Exploratory analysis suggested that cases with bi-allelic TP53 mutations had significantly better overall survival than those with mono-allelic mutations (HR = 0.140; P = 0.029; Supplementary Fig. S4B). The association of survival with stage and TP53 mutation status remained statistically significant (P = 0.047 and 0.011, respectively) in multivariate analysis (Table 2).

To determine the stage of oogenesis that gave rise to the MCT, we analyzed SNPs in the 151 sequenced genes for heterozygosity, and for a change in zygosity between normal tissue and MCT. All cases had some homozygous SNPs, but analysis of nonhomozygous SNPs revealed four groups (Table 3 and Supplementary Fig. S5). Group A (10/21, 48%) showed the same levels of heterozygosity in normal and MCT, suggesting that they arise from cells prior to crossover (e.g., primordial germ cells or oogonia), whereas group B (3/21 cases, 14%) showed substantially fewer heterozygous positions in the MCT compared to normal. This is consistent with development after prophase I in meiosis I, when chromosomal crossover occurs. Group C (3/21,14%) had SNPs at the expected 0:1 and 0.5:0.5 ratios for diploid cells, but also had SNPs with other allelic ratios. SC011 and SC002, for example, showed SNPs at an allelic ratio of approximately 0.3:0.7. Given the absence of any copy number alterations in these samples, this would indicate a tetraploid MCT with some regions of the genome in a 2:2 maternal:paternal ratio and some in a 1:3 ratio. SC013 showed an additional allelic ratio of approximately 0.2:0.8, consistent either with some regions being at 6N in a 1:5 ratio between the maternal and paternal chromosomes or contamination of a group B-type profile with normal cells. SC013 did show some regions of copy number change, making it difficult to eliminate this as an explanation of the altered SNP allele frequencies. However, in the other two MCTs with copy number alterations, there were insufficient SNPs within the limited regions of copy number change to impact on the peaks of SNP allele frequencies (Supplementary Fig. S1B and Supplementary Fig. S6, and Table 3). Group D (6/21, 29%) did not have any SNPs in a 0.5:0.5 ratio but instead each showed a single allelic ratio consistent with a polyploid genome with unequal numbers of maternal and paternal chromosomes. For example, SC001, SC006, and SC030 all appeared to be pentaploid in a 2:3 ratio, whereas SC022, SC007, and SC029 were triploid (1:2 ratio), tetraploid (1:3 ratio), and hexaploid (1:5 ratio), respectively.

We believe this to be the first description of the mutational landscape of squamous cell carcinoma (SCC) arising within mature cystic teratoma of the ovary (MCT, also known as dermoid cyst and benign cystic teratoma). The data presented here show that MCT are genomically bland, but the associated SCC have a high mutation burden characterized by a high frequency of mutations in TP53, CDKN2A, and PIK3CA. TP53 mutation is an early event, and likely bi-allelic TP53 mutations are seen frequently. Although previous individual case reports describe high-risk HPV in an ovarian SCC (13), all 25 SCC analyzed here were negative for high-/intermediate-risk HPV genomes. There are few data on recurrent genomic abnormalities in malignant ovarian germ cell tumors (GCT) but SCC arising in MCT differ greatly from malignant testicular GCT, which have low mutational burden (0.9/MB) and are universally TP53 wild type (14).

Although we found mutations in genes frequently mutated in both lung and skin SCC (TP53 and CDKN2A), these two tumor types also contain frequent mutations in NOTCH1, 2, and 3 (15, 16). We only identified NOTCH1 mutations in two of our 25 cases (8%), one of which had only a synonymous change. However, our sequencing panel did not include other NOTCH genes. The rates of both PIK3CA and CDKN2A abnormality in our samples are potentially higher than in lung and skin SCC, but, overall SCC arising in MCT has features in common with other non-HPV SCC. A larger sample size would be required to make more definite statements about relative mutational frequency. The overall mutational burden estimate for the SCC was high at 10.2 mutations per MB of sequenced DNA. Assessment of mutation rates derived from cancer gene panels may overestimate the genome-wide rate as these panels include only genes that are known to be under selective pressure. Chalmers and colleagues recently compared mutational burden estimates from whole exome sequencing with targeted panels, demonstrating close correlation between the two (10). They also suggested that sequencing of approximately 1.1Mb of coding genome could accurately estimate overall mutational burden, with significant variance only below 0.5 Mb. Our data were derived from approximately 0.8Mb, and thus may be an overestimate, but are largely in line for data from other SCC types, including NSCLC.

Bi-allelic TP53 mutations have been identified in skin SCC (17). Exome sequencing of eight skin SCC identified TP53 abnormalities in seven. Four of these showed LOH at chromosome 17p, but the three remaining cases lacked LOH and contained two or more distinct TP53 mutations, presumed to be biallelic. We are not aware of other descriptions of bi-allelic alterations in other SCC, nor of any previous potential correlation between bi-allelic TP53 mutations and improved outcome.

There remains considerable debate as to the origin of MCT. Originally thought to represent parthenogenic activation of oocytes (i.e., embryological development in the absence of a male gamete) at the end of meiosis I (18), cytogenetic analyses later suggested the existence of five separate MCT groups, arising from distinct stages of meiosis (2). More recent short tandem repeat analyses generated conflicting data, supporting origins in oogonia (19) or primary oocytes that have escaped meiotic arrest (3).

Our study was not designed to identify cell of origin of MCT, but to analyze whether there were obvious SCC precursor mutations present in MCT. These we did not find; indeed, the MCT had few mutations or copy number alterations. However, we did identify four separate zygosity states in the 22 MCT samples from 21 cases. Groups A and B appear to have arisen primordial germ cells/oogonia and secondary oocytes, respectively. Group C contained some SNPs with altered allelic ratios. In the absence of any detectable copy number alterations, one potential mechanism for generating such a state would be fusion of a cell that has undergone meiosis I but not completed meiosis II with a somatic cell, such that regions that had crossed over would end up in a 1:3 ratio, whereas the rest of the genome would be 2:2. Alternatively, this profile would be consistent with a mixture of a group B profile and contaminating normal cells. Group D, which lacked any SNPs in a 0.5:0.5 ratio but instead in each case showed a single allelic ratio consistent with a polyploid genome with unequal numbers of maternal and paternal chromosomes, may arise due to errors in meiosis I, when sets of homologous chromosomes should be separated. Given that only a single allelic ratio is present in these cells it seems unlikely that crossover has occurred.

Analysis of any rare cancer is logistically problematic, and we acknowledge that there are potential shortcomings in our study. Identifying rare cancer samples is challenging, and we acquired the original 31 samples only by interrogating the archives of four large UK Gynaecological Cancer Centres, with limited clinical data available. However, the clinical outcomes we describe are in line with a large systematic review (6), which suggests that our patients are broadly representative of this patient population. SCC arising within MCT has the added disadvantage that it is almost invariably an unexpected pathological diagnosis in an apparently benign tumor—there are no proven imaging characteristics to differentiate a benign MCT from one containing SCC (20, 21). We therefore had to work with archival formalin-fixed paraffin-embedded material up to 17 years old. Finally, our sequencing panel, although containing 151 genes, did not cover the whole exome, and so there may be critical mutations that we have missed.

Despite these issues, this first systematic genomic evaluation of ovarian SCC arising in MCT shows clearly that they are similar to other non-HPV SCC, especially NSCLC, but with distinct features, including bi-allelic TP53 mutations. Further studies will be required to address the question of MCT cell of origin and to understand what triggers transformation of the MCT into SCC. However, our data suggest that patients with MCT–SCC could be included in trials of SCC-specific therapy and may, like lung SCC (22), derive benefit from immune checkpoint inhibition.

P. Beer is an employee of Karus Therapeutics and Leeds Hospitals NHS Trust. A.V. Biankin reports receiving speakers bureau honoraria from AstraZeneca, Celgene, and Clovis Oncology, is a consultant/advisory board member for Cure Forward, and reports receiving commercial research grants from Celgene. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D. Ennis, R.M. Glasspool, D. Millan, A.V. Biankin, I.A. McNeish

Development of methodology: S.L. Cooke, L. Evers, D. Millan, A.V. Biankin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Evers, S. Dowson, M.Y. Chan, L. Hirschowitz, R.M. Glasspool, N. Singh, S. Bell, E. Day, A. Kochman, N. Wilkinson, D. Millan, I.A. McNeish

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.L. Cooke, L. Evers, S. Dowson, P. Beer, D. Millan, I.A. McNeish

Writing, review, and/or revision of the manuscript: S.L. Cooke, L. Evers, S. Dowson, M.Y. Chan, J. Paul, L. Hirschowitz, R.M. Glasspool, N. Singh, S. Bell, A. Kochman, N. Wilkinson, P. Beer, A.V. Biankin, I.A. McNeish

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Evers, S. Martin, D. Millan

Study supervision: L. Evers, J. Paul, A.V. Biankin, I.A. McNeish

Other (pathology review): S. Bell

Other (selection and identification of relevant tissues): D. Millan

This Scottish Genomes Partnership is funded by the Chief Scientist Office of the Scottish Government Health Directorates (grant reference SGP/1) and The Medical Research Council Whole Genome Sequencing for Health and Wealth Initiative. Additional funding was provided by the Medical Research Council (the Glasgow Molecular Pathology Node, grant reference MR/N005813/1), Cancer Research UK [grant references A15973 (IMcN) and A17263 (to A.V. Biankin)], the Wellcome Trust [grant reference 103721/Z/14/Z (to A.V. Biankin)], and the Beatson Cancer Charity [grant reference 15-16-051 (IMcN)]. We thank Craig Nourse for technical assistance, and Kymab for licensing GeneCN to us.

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.