Mutational landscape of mucinous ovarian carcinoma and its neoplastic precursors

Somatic mutation frequency and spectra

To profile the somatic mutation spectrum of mucinous tumors of the ovary, we performed
whole exome sequencing on 24 tumors including 5 benign cystadenomas, 8 borderline
tumors and 11 carcinomas (Table 1, Additional file 1: Table S1). A mean coverage depth of 144× was achieved in both neoplastic and non-cancerous
specimens (range 53-fold to 224-fold) and 91 % of the bases were covered by at least
20 uniquely mapping reads (Additional file 1: Table S2). Using stringent criteria, 1126 somatic coding and essential splice site
mutations were identified (1031 SNVs and 95 indels), of which 841 were predicted to
alter protein sequence (Additional file 1: Table S3). These included 44 (5.2 %) nonsense, 60 (7.1 %) frameshift indel, 16 (1.9
%) splice site, 27 (3.2 %) inframe indel and 694 (82.5 %) missense mutations. Benign
and borderline tumors had on average 25.4 (range 21–38) and 32.9 (range 2–76) coding
mutations per tumor, equating to a frequency of 0.8 mutations/Mb and 0.9 mutations/Mb
respectively. Although variable, this mutation burden did not differ between benign
and borderline tumors but was significantly lower when compared to the carcinomas
(average of 66.9 mutations per sample and 1.5 mutations/Mb) (P = 0.008 vs. benign and P = 0.047 vs. borderline) attributed mostly to an accumulation of missense mutations
in the carcinomas (Fig. 1a, Additional file 1: Table S5). There were no hyper-mutated cases (defined as 10 mutations/Mb) indicative
of a mutator phenotype such as mismatch repair deficiency. Relative to other cancer
types, MOTs showed a similar somatic mutation density to breast, serous ovarian and
pancreatic cancers, but a lower density than colorectal and stomach tumors 14], 15].

Table 1. Cohort summary

Fig. 1. Mutational landscape of MOTs identified by exome sequencing. Samples are grouped according
to pathological classification and ordered from lowest to highest mutation frequency.
a Somatic mutation frequency (left Y-axis) and number of coding mutations by consequence
(right Y-axis). b Relative frequency of somatic mutations according to base substitution type. Substitutions
were categorized by the six possible base-pair changes

The mutation spectrum was dominated by CT transitions, comprising 63.9 % of somatic
substitutions, and this was common to all three tumor subtypes (Fig. 1b). Mutations in this context demonstrated a marked preference for NpCpG trinucleotides
(Additional file 2, Figure S1a), the optimum motif for spontaneous 5-methylcytosine deamination 16]. An equivalent signature is frequently seen in other epithelial tumors of the gastrointestinal
tract, but is different to that observed in other cancers of the female reproductive
system including high-grade serous ovarian carcinoma (Additional file 2: Figure S1b). Taken together these findings are consistent with MOTs having a shared
lineage distinct from that of other ovarian epithelial tumors.

Profile of mutated genes in mucinous ovarian tumors

Protein-altering mutations were detected in 761 genes, of which 42 were mutated in
two or more of the 24 tumors. Among the most frequently mutated were known mucinous
ovarian cancer genes KRAS, BRAF and CDKN2A (Table 2, Fig. 2). Interestingly, TP53 was the second most frequently mutated gene, with seven mutations identified. Eight
genes were significantly mutated based on a statistically significant accumulation
of mutations by both MuSiC 9] and OncodriveFM 11] (Table 2). Other genes predicted by one algorithm were also notable; for example, ERBB3 (MuSiC) and GNAS and FBXW7 (OncodriveFM) (Table 2). Based on these predictions and observation in other cancer types, five novel candidate
drivers not previously reported in MOTs were selected for validation in an independent
cohort: TP53 (7/24), ELF3 (3/24), ERBB3 (2/24), GNAS (2/24) and KLF5 (2/24) (Table 2, Fig. 2, Additional file 1: Table S6). Our validation study of the tumor suppressor gene RNF43 has been published previously 8]. In addition, known cancer genes for this ovarian subtype were evaluated in parallel
to assess their relationship with new drivers including HER2 (by immunohistochemistry
and copy number analysis) and mutations in KRAS, BRAF, CDKN2A and other RAS pathway members NRAS and HRAS.

Table 2. Candidate driver genes with significantly recurrent somatic mutations in mucinous
ovarian tumors

Fig. 2. Candidate driver genes in MOTs. Significantly mutated genes identified by OncodriveFM
and MuSiC analyses are arranged vertically by their frequency of mutated samples in
the whole exome sequencing data. Color indicates mutation consequence. Selected genes were also investigated in a validation
cohort of mucinous tumors. Each column denotes an individual tumor (ordered as listed in Additional file 1: Table S1), which have been arranged to emphasize mutational groups. Genomic aberrations
in other MAPK pathway genes were also screened for mutations. LOH loss of heterozygosity

Prevalence of mutations in known mucinous ovarian cancer genes

We and others have previously described the importance of the RAS pathway and p16
in MOTs 1]–3]. Here we extended this analysis and found 56 cases with mutations in KRAS, BRAF and NRAS (68.3 %). BRAF mutations were significantly more prevalent in the carcinomas (7/31, 22.6 %) than
in the borderline (3/29, 10.3 %) or benign tumors (0/22, P = 0.036 Fisher’s exact test), suggesting an association with a more aggressive phenotype.
An alternative mechanism for activation of the MAPK pathway was identified through
mutation of the ras-like gene RRAS2 in one benign tumor that was KRAS/BRAF wild type (Fig. 2). We previously reported RRAS2 gene amplification in this sample 2]; consistent with this, the Sanger sequencing validation confirmed homozygous amplification
of the mutant allele (Additional file 2: Figure S2). This 9 bp duplication, resulting in reiteration of Gly-Gly-Gly (codons
22–24), occurs in the region of RRAS2 that is complementary to codons 11–13 within the G1 phosphate-binding loop of conventional
ras proteins (P-loop, amino acids 10–17). Interestingly, rare reports of comparable
events appear in the literature. Huang et al.17] described a three amino acid RRAS2 duplication (Gly24_26dup) in the human uterine leiomyosarcoma cell line ST-UT-1;
this mutation resulted in enhanced GTP-binding and conferred transforming activity
in vitro. Similarly, in KRAS, 9 bp and 12 bp tandem repeats of codons 10–12 and 10–13 respectively were identified
as an alternative mechanism for KRAS oncogenic activation in 2 of 18 chemically induced rat renal mesenchymal tumors 18]. Triple residue insertions in the P-loop of HRAS also demonstrated increased preference for GTP-binding and increased interactions
with downstream Raf kinase compared to wild type 19]. Taken together, these observations indicate that although the RRAS2 duplication described in this study is an unconventional mutation for ras proto-oncogene activation, it is predicted to result in up-regulated MAPK pathway
activity.

A previous study found KRAS mutation and HER2 amplification to be almost mutually exclusive 1]. Although the number of cases we studied was smaller, we did not see this exclusivity:
2/6 HER2+ borderline and 3/6 HER2+ carcinomas also carried KRAS mutations. One caveat to this observation is that HER2 status in this study was based
on IHC and/or high-level amplification (SNP array analysis) rather than a combined
score including chromogenic in situ hybridization.

Candidate mucinous ovarian cancer genes

In addition to the seven somatic TP53 mutations identified by exome sequencing, Sanger sequencing of the DNA binding domain
(exons 4–9) in the validation cohort identified a further 15 mutations at an overall
frequency of 22/82 (26.8 %) MOTs, of which 21 were missense mutations (Table 2, Fig. 2). All 22 mutations have been previously reported in a somatic context (IARC TP53 mutation database release 17). There was a significant difference in TP53 mutation frequency among the three tumor subtypes (P = 0.003, Chi-square test). While there was a similar frequency of TP53 mutations in benign (2/22, 9.1 %) and borderline (4/29, 13.8 %) tumors, 16/31 (51.6
%) of carcinomas harbored a TP53 mutation (P = 0.002 and P = 0.002 compared to benign and borderline tumors respectively, Fisher’s exact test),
suggesting that aberrant p53 contributes to the invasive phenotype in a proportion
of these ovarian cancers. Both low-grade and high-grade carcinomas harbored mutations,
which trended towards increasing frequency with grade (45.5 %, 53.8 % and 66.7 % in
Grades 1, 2 and 3 respectively), and with an overall frequency similar to that of
gastrointestinal mucinous carcinomas (Additional file 2: Figure S3). While it is well accepted that TP53 mutation is an obligatory event in the genesis of high-grade serous ovarian carcinoma,
we show by direct sequencing that mutant p53 is also common in mucinous-type ovarian
carcinomas, but is a late event in their molecular progression. Interestingly, this
group does not share the widespread genomic instability that typifies high-grade serous
carcinomas that is contributed to, at least in part, by mutant TP53, suggesting different p53 activity in these two contexts.

Three mutations in the epithelial-specific ETS transcription factor E74-like factor 3 (ELF3) were detected in three tumors by exome sequencing. ELF3 was significantly mutated above background (MuSiC) and had an excess of likely deleterious
mutations (OncodriveFM) including two frameshift insertions (p.Val345Glyfs*126, p.Asp239Glyfs*62)
and a missense substitution (p.Met324Val) (Table 2, Figs. 2 and 3). Sequencing of the coding regions in the expanded cohort identified an additional
splice site mutation in a borderline tumor (c.1001 + 1_1001 + 2insGG). Although ELF3 is thus infrequently mutated (6.9 % borderline tumors and 6.5 % carcinomas), the
shared characteristics of the four heterozygous mutations is indicative of a pathogenic
role. Three of the mutations are overtly deleterious, including two frameshift indels
and a canonical splice site mutation, while the missense mutation is predicted to
be deleterious by computational analyses 20]–22]. We further investigated the exon 8 splice donor site mutation by cDNA sequencing,
which confirmed the use of an alternative donor splice sequence in the mutant allele
(Additional file 2: Figure S4) consistent with the in silico prediction 23]. This mutation would result in out-of-frame, continued translation into the 3?-untranslated
region (p.Tyr335Glyfs*113). cDNA sequencing of this and the two other truncating mutations
found that all three mutations were readily detected in the tumor RNA, indicating
that these mutations are not the subject of strong nonsense-mediated decay (Additional
file 2: Figure S4). Truncating mutations in this epithelial-specific transcription factor
have recently been reported in other cancers, including cancer of the cervix, stomach
and bladder 24]–26]. Interestingly, ELF3-mutated cervical carcinomas express ELF3 at a higher level compared to wild-type
tumors 24]. This result may suggest that both copies of this gene are required and mutation
of one allele results in up-regulation of the gene in an attempt to compensate. Alternatively,
ELF3 mutations may only have a selective advantage in tumors highly expressing ELF3.

Fig. 3. Distribution of somatic mutations identified in novel significantly mutated genes.
ELF3, KLF5, GNAS and ERBB3 are shown in the context of protein domains as predicted by UniProt, with somatic
mutations identified in the exome (closed circle) and validation (open circle) cohorts mapped to each gene. I-IV extracellular domains I, II, III and IV, AT hook NLS AT-hook domain and nuclear localization signal, C2H2 zinc-finger C2H2 domain, ETS DNA binding domain, GTP GTP nucleotide binding region, PNT pointed domain, SAR serine-rich and aspartic acid-rich domain, TAD transactivation domain, TKD tyrosine kinase domain

ELF3 has previously been identified as a candidate cancer gene; however, its role appears
to be context dependent, in keeping with the tissue-specific nature of its transcriptional
target genes. An oncogenic role has been suggested for breast cancer, with the gene
being amplified and overexpressed 27], 28]. A positive feedback loop between ELF3 and HER2 exists in breast cancer, where ELF3
is both a downstream mediator and activator of HER2 signaling 29]. Of note in this study, two of two mutated carcinomas were HER2+, while the two mutated
borderline tumors were HER2-. However, in a gastrointestinal tissue context, ELF3
may act as a tumor suppressor, because it is involved in positively transcriptionally
regulating TGFBR2, facilitating the growth inhibitory consequences of TGF-? signaling 30], 31]. ELF3 was identified as a cancer gene in a recent pan-cancer study, with enrichment for
mutations in bladder and colorectal cancer 32]. The frequency of mutations in MOTs suggests that ELF3 is indeed a cancer gene in this tumor type, but its exact role is unclear from the
mutational profile—while the mutations are detrimental in nature, the retention of
the wild-type allele argues against a classical tumor suppressor gene functional mechanism.

Another transcription factor with a proclivity for truncating mutations was KLF5, which encodes a zinc finger transcriptional activator (Table 2, Figs. 2 and 3). Exome sequencing identified two heterozygous frameshift mutations (p.Phe123Leufs*3
and p.Asp238Argfs*16); however, sequencing the coding region in a validation cohort
of carcinomas failed to identify additional changes. Collectively, KLF5 is mutated in 6.7 % (2/30) of mucinous ovarian carcinomas. Like ELF3, it has been identified as a pan-cancer gene but enriched for mutations in bladder,
colorectal, and head and neck squamous carcinoma 32], and has been variously described as both an oncogene 33] and a tumor suppressor gene 34].

Considering exome and validation cohorts, five constitutively activating mutations
at arginine codon 201 of the oncogene GNAS were identified, including 2/22 (9.1 %) benign cystadenomas, 2/29 (6.9 %) borderline
tumors and 1/30 (3.3 %) carcinomas (Table 2, Figs. 2 and 3). Hotspot mutations in this guanine nucleotide-binding protein alpha subunit have
recently been identified in other pre-malignant or non-aggressive mucinous-type tumors
of gastrointestinal origin, albeit at a higher frequency (Additional file 2: Figure S3), including intraductal papillary mucinous neoplasm of the pancreas and
bile duct 35], 36]; appendiceal mucinous neoplasms (and its associated pseudomyxoma peritonei) 37], 38]; and adenoma of the colon/rectum, stomach and small intestine 39], 40]. In this context, constitutive activation of GNAS through codon 201 mutation has been shown to increase levels of cAMP, resulting in
prominent mucin production but not cell growth 38]. Consistent with previous reports, simultaneous KRAS mutations were present in four MOTs, although this association was not statistically
significant. Thus, unlike gastrointestinal mucinous-type tumors, GNAS activation occurs only rarely in those involving the ovary.

Although human epidermal growth factor receptors have been implicated in MOT progression
through amplification and overexpression of ERBB2 (HER2), activating mutations in
other family members have not been previously described. Exome sequencing identified
three ERBB3 (HER3) mutations (a borderline tumor with concurrent mutations, and a carcinoma)
(Table 2, Figs. 2 and 3), including two in the extracellular domain (p.Met91Ile and p.Glu332Lys) and one
in the kinase domain (p.Glu925Lys). No additional mutations were identified in a validation
screen of 19 carcinomas, giving a final frequency of 4.7 % in MOTs. Frequent ERBB3 mutations have recently been reported in other cancer types, including those of the
colon, gallbladder and stomach 41], 42]. Although ERBB3 contains an impaired kinase domain, it is capable of ligand binding and preferentially
hetrodimerizes with ERBB2 to potently activate cellular signaling pathways 43]. Thus the ERBB3 mutations described here are predicted to cooperate with ERBB2 to promote ligand-independent oncogenic transformation, as functionally demonstrated
for other kinase and extracellular mutations in this gene 41]. Of note, the ERBB3 mutant carcinoma was also HER2+. We also identified a single somatic extracellular
domain mutation in another ErbB receptor, ERBB4 (p.Glu57Asp).

Additional mutated candidate genes

We also identified somatic mutations in epigenetic regulatory genes, including the
chromatin-remodeling factors ARID1A (1/5 benign MOTs and 1/11 carcinomas; a predicted significantly mutated gene) and
ARID2 (1/11 carcinomas) (Table 2, Fig. 2). Both genes are recognized suppressors of tumorigenesis in multiple cancer types
44], 45]. Consistent with this, the two ARID1A mutations result in protein truncation (p.Gln1894Profs*7 and p.Arg2116Thrfs*33).
A further missense mutation was found in the Polycomb-group protein member ASXL1.

In addition to ELF3 and KLF5, other genes implicated in the control of gene expression were collectively mutated
in multiple samples. Three transcriptional co-regulatory proteins contained somatic
mutations, including the consensus driver gene BCL-6 corepressor (BCOR; 2/11 carcinomas including splice donor and missense mutations) 45], and proposed pan-cancer drivers NCOR2 (inframe indel in a benign tumor) and ARHGAP35 (1/11 carcinomas) 46], 47]. Other genes involved in transcription also featured, such as a single mutation in
TAF1 that forms the large subunit of the transcription factor II D complex and facilitates
the initiation of transcription by RNA polymerase II, and a missense mutation at the
serine 34 hotspot of pre-mRNA splicing factor U2AF1, which has been shown to alter the cancer transcriptome 48]. The GATA3 transcription factor was also mutated in one carcinoma.

One other important group of genes mutated in MOTs included those associated with
ubiquitin-mediated protein degradation. As well as frequent deleterious mutations
in the E3 ubiquitin ligase RNF438], the consensus cancer gene and tumor suppressor FBXW7 is noteworthy 45], encoding for the substrate recognition component of SCF (complex of SKP1, CUL1 and
F-box protein)-type ubiquitin ligases. Recurrent heterozygous mutations (1/8 borderline
tumors and 1/11 carcinomas) (Table 2, Fig. 2) are predicted to result in proteins with impaired (p.Asp560Asn) or absent (p.Arg278*)
substrate binding capability that dominantly interfere with wildtype protein through
the intact dimerization domain. Interestingly, the mutant borderline tumor also harbored
bi-allelic mutations in another SCF complex gene CUL1. We also identified a missense mutation in the E3 ubiquitin-protein ligase and cancer
gene UBR546].

Other genes identified based on significance prediction and mutated in 2/24 MOTs by
exome sequencing included leucine-rich repeat kinase 2 (LRRK2), the ribosomal gene transcription termination factor TTF1, and LPHN3, which encodes a member of the latrophilin subfamily of G protein-coupled receptors
(Table 2, Fig. 2). DCLK1, a newly identified marker of transformed stem cells in the gut 49], was also mutated in 8.3 % of cases (Table 2, Fig. 2), and is a recurrent target for mutation in neoplasms of the stomach 25], appendix 37] and skin 50]. Clonal heterogeneity may be a feature of DCLK1, as we and others 25], 50] observed mutations at low allelic fractions. Validation in a larger cohort of samples
is needed to interpret the role of these genes in mucinous ovarian tumorigenesis.