EIF1AX and NRAS Mutations Co-occur and Cooperate in Low-Grade Serous Ovarian Carcinomas

Low-grade serous ovarian carcinomas (LGSC) differ from high-grade serous ovarian cancer (HGSC) by having RAS–MAPK pathway mutations (1), few DNA copy-number changes, a lower rate of somatic mutation, and wild-type TP53 (2). LGSC are associated with younger age of onset and, unlike HGSC that are typically sensitivity to primary platinum-based chemotherapy, patients with LGSC are often chemoresistant with a high risk of disease recurrence (3). Importantly, in LGSC patients with advanced-stage disease and macroscopic residual disease after primary treatment, outcomes are equivalent to HGSC (4, 5). Therefore, despite the term low-grade, LGSC is not an indolent disease unless it can be cleared surgically.

The small number of driver mutations identified in LGSC are generally stable, both spatially and temporally (6), and therefore targeting these events might be effective therapeutically. Potential novel treatments include RAS–MAPK pathway inhibitors, which have shown significant activity in other tumor types such as melanoma (7). Response to the BRAF inhibitor vemurafenib has been demonstrated in BRAF^V600E-mutated LGSC (8, 9) and we have observed substantial clinical response to dabrafenib in a young patient with recurrent LGSC with a BRAF^V600E mutation (Moujaber and colleagues, manuscript in preparation).

A previous exome sequencing study of eight LGSC found mutations in KRAS and BRAF but no other recurring mutations (10) and another study identified a case with a MAP2K1 (MEK1) mutation (1/18 cases; ref. 11). Patients with RAS pathway mutation negative tumors are associated with late disease stage (12) and poorer survival (13), and given few other recurrent mutations have been identified to date, further mapping of driver mutations is warranted. We therefore performed whole exome sequencing (WES) and limited whole genome sequencing (WGS) analysis to further characterize the mutational landscape of LGSC. In addition to expected activating mutations in RAS–MAPK pathway genes (BRAF, KRAS, and NRAS), we identified recurrent disruption of NF1, a negative regulator of the pathway. We also observed recurrent mutations in the protein translational regulator EIF1AX, which significantly co-occurred with mutations in NRAS. The functional significance of this association was demonstrated in a newly derived LGSC cell line, where coexpression of mutant NRAS and EIF1AX proteins resulted in increased proliferation and clonogenic survival.

The study population consisted of women diagnosed with epithelial ovarian, primary peritoneal, or fallopian tube cancer between 1992 and 2012. The women were treated at hospitals across Australia and were recruited through the Australian Ovarian Cancer Study (AOCS) or the Gynaecological Oncology Biobank at Westmead Hospital (NSW). All eligible cases underwent pathology review by expert gynecological pathologists on a complete set of hematoxylin and eosin (H&E)-stained diagnostic slides or 1 to 3 representative diagnostic block(s) selected by the original reporting pathologist. Our previous analysis has shown the two review methods are comparable (14). Representative H&E sections for sequenced tumors where formalin-fixed paraffin-embedded (FFPE) tissue was available are shown in Supplementary Fig. S1. We have incorporated the shift from a three-tier grading system to a two-tier grading system that has occurred over recent years, as recommended in the 2014 World Health Organization (WHO) classification of ovarian tumors (15). Briefly, Silverberg is a three-tier grading system, with grade 1 and some grade 2 cases segregating with LGSC (2). Grade 3 and most grade 2 cases segregate with HGSC with a defining characteristic being somatic mutations in TP53 (2). Our sequencing studies involved patients diagnosed with Silverberg grade 1 or 2 serous carcinoma of the ovary or peritoneum (16), with initial molecular characterization to exclude any grade 2 HGSC, based on TP53 mutation status and copy-number aberration (2). There was no record of previous serous borderline malignancy in our sequenced patient cohort except for patient 9849, who was initially diagnosed with a primary serous borderline tumor and invasive disease on recurrence (Supplementary Table S1). Of the 23 cases from the sequencing cohort, 21 had available FFPE tumor tissue blocks for Ki-67 staining. The average number of Ki-67–positive tumor cells was 16% (range, <5%–35%), consistent with the low proliferation rate seen in LGSC (Supplementary Table S2; ref. 17).

For comparison of patient demographics and clinical outcome with LGSC, HGSC from AOCS were chosen that had undergone pathology review and were confirmed to be serous, grade 3 cases according to standardized criteria (16). Neoadjuvant chemotherapy may interfere with accurate tumor grading and therefore only patients with tissue collected prior to chemotherapy were included in this study. Progression-free survival (PFS) was defined as the time interval between histological or cytological diagnosis and disease progression based on GCIG (Gynaecological Cancer Inter Group) CA125 criteria (18), imaging or clinical evaluation. Overall survival (OS) was defined as the interval between histological or cytological diagnosis and death from any cause. Clinical characteristics of the patient cohort are summarized in Supplementary Table S1. Ethics board approval was obtained at all institutions for patient recruitment, sample collection, and research studies. Written informed consent was obtained from all participants.

Normal DNA was isolated from peripheral lymphocytes as previously described (14). Tumor DNA was isolated from frozen tissue cryosections with a section for H&E staining taken before and after sections for DNA, to assess tumor content. For samples containing >70% tumor, 4 × 50 μm sections were used for DNA extractions. For samples containing <70% tumor, needle dissection of tumor cells was performed on up to fifty 10-μm sections. Extractions were performed using the DNeasy blood and tissue kit (QIAGEN). DNA was quantified using the Qubit dsDNA BR assay.

DNA (1 μg) was fragmented to approximately 200 bp using a focal acoustic device (Covaris) and used to prepare libraries and hybridization capture following the SureSelect v4 or v5 (Agilent Technologies) recommended protocols. Exons 1 and 7 of EIF1AX were not included in the Agilent v4 exome capture and were therefore screened by targeted amplicon sequencing (see below). Three indexed samples were pooled in a single lane of an Illumina HiSeq2500 flow cell (paired-end 100 bp) according to the standard protocol (Illumina). Read data were processed and analyzed using an in-house bioinformatics pipeline as described in Supplementary Methods. Sequence data for an FFPE sample (65855) was equivalent to fresh-frozen samples (Supplementary Table S3). Coding genome for 10693 (WGS) was calculated as described in Patch and colleagues (19). WGS data have been deposited in the European Genome-phenome Archive (EGA) repository under the accession code EGAD00001000877.

Mutation verification using DNA or cDNA was performed by Sanger sequencing, using previously described methods (2, 20) or by targeted amplicon sequencing (19). Primer sets for complete sequencing of exons 1 to 6 and exon 7 CDS of EIF1AX, TP53 (exons 2–11), and mutation hotspots for KRAS (exons 2 and 3), NRAS (exons 2 and 3), and BRAF (exons 11 and 15) were designed using Primer3 (21). Primer sequences are given in Supplementary Table S4. For targeted sequencing, primer sets were tagged with CS1 and CS2 (common sequence) universal tags (Fluidigm). The first PCR utilized the CS-gene-specific primers and Phusion High-Fidelity DNA Polymerase (Thermo Scientific), purified product used as a template in a second PCR utilizing PE-CS primers with a unique barcode for each sample as described previously (19). DNA from each PCR product was pooled, purified using AMPure XP Beads and analyzed on Agilent BioAnalyzer High Sensitivity DNA chip. Libraries were then used to generate 150-bp paired-end reads on MiSeq (Illumina). Sequence analysis is described in Supplementary Methods.

To establish the AOCS2 cell line from patient 65549, ascites fluid (2 mL) collected at recurrence was centrifuged at 1,500 rpm for 5 minutes to create a cell pellet. Supernatant was removed and the cell pellet was resuspended in 10 mL of complete RPMI media (RPMI 1640, 10% FBS, 50 U/mL penicillin, and 50 mg/mL streptomycin) and transferred into a standard humidified incubator (37°C, 5% CO₂). Media were replaced after 48 hours, and then once every 2 to 4 days, until an adherent cell line was established. Cells were passaged 10 times from the time of collection and stocks cryopreserved. The AOCS2 cell line was authenticated against the patient germline DNA using STR profiling (GenePrint 10 System, Promega) and shown to be free of Mycoplasma (MycoAlert, Mycoplasma Detection Kit, Lonza) in December 2013 and at each revival. Revived cell stocks were passaged three times prior to performing experiments. The AOCS2 cell line has a doubling time of approximately 2.1 days. Cell pellets were fixed in 10% neutral-buffered formalin (NBF), set in 4% agarose, and embedded in paraffin prior to immunohistochemical staining of PAX8, WT1, p53, and Ki-67 according to standard protocols. SNP microarray copy-number analysis of cells from ascites and cells grown in culture were performed on Illumina OmniExpress arrays as previously described (2).

The human EIF1AX Precision LentiORF cDNA (pLOC-WT-EIF1AX) construct was purchased from GE Dharmacon (ThermoFisher Scientific) and EIF1AX c.5C>T (p.P2L), EIF1AX c.22G>A (p.G8R) and EIF1AX c.23G>A (p.G8E) mutant constructs were generated from pLOC-WT-EIF1AX by Mutagenex Inc. based on mutations observed in this and previous studies (see below). DNA constructs were entirely sequenced. The human WT-NRAS and NRAS^Q61K lenti-vector expression constructs have been previously described (22). Lentiviral particles for EIF1AX and NRAS constructs were produced using Lenti-X (Clontech) packaging mix according to the manufacturer's protocol. AOCS2 cell lines were transduced with EIF1AX and NRAS virus particles individually or in combination. NRAS construct–positive cells were isolated by FACS sorting for GFP-positive cells and EIF1AX construct–positive cells were selected in blastcidine (15 mg/mL) over a week. Overexpression of EIF1AX and NRAS were verified by RT-PCR and Western blot. A lentiviral constitutive shRNA system (mir30 pGIPZ) was used to knock down endogenous EIF1AX expression in AOCS2 cells that were transduced with an NRAS^Q61K overexpression construct. Five unique EIF1AX-specific shRNA were transduced and knock down validated by Western blot. We subsequently selected the two most effective shRNA (EIF1AX-sh2 and EIF1AX-sh3) for functional characterization. Reverse transcription and quantitative-PCR was performed as described previously (23) with primer sequences given in Supplementary Table S3.

Further experimental methods are given in Supplementary Methods.

To characterize the patient cohort, we first compared the survival characteristics of 684 patients with advanced-stage serous cancer, either LGSC or HGSC. Survival analysis showed a trend toward improved progression-free and overall survival in LGSC (grade 1) patients debulked to no macroscopic residual disease. However, survival was equivalent in patients with grade 1 (LGSC) and grade 3 (HGSC) tumors, if macroscopic residual disease remained after debulking surgery (median overall survival 38.9 and 34.6 months, respectively, long-rank P value < 0.673). This result confirms the similarly poor outcome reported for LGSC and HGSC patients who were suboptimally debulked (Fig. 1; Supplementary Table S1A; ref. 5). We observed a shift in age distribution (t test P < 0.01; Supplementary Fig. S2A) in patients with LGSC (grade 1) toward a greater proportion of younger patients (≤40) compared with HGSC (grade 3) patient groups (Fisher test P < 0.0001 when patients stratified by age ≤ or >40; Supplementary Fig. S2B). Young age at diagnosis has been associated with worse outcomes in LGSC patients (24); however, we did not observe a significant association with survival in our cohort, which may in part be explained by limited sample size.

We next selected a cohort for WES (n = 22) and WGS (n = 1) analysis based on histology and molecular characteristics consistent with LGSC (TP53 wild-type and low copy-number variation as described in our previous study (Supplementary Table S1B; ref. 2). We achieved a mean target base coverage of 136× for WES of both tumor and normal samples and for the WGS sample 78× and 55× average fold coverage in the tumor and normal sample, respectively (Supplementary Table S3). We identified an average of 50 (range, 19–114) small mutations, including single-nucleotide variants (SNVs) and insertions or deletions (indels) per tumor sample (Supplementary Tables S5 and S6). These included an average of 15 (range, 5–34) deleterious variants, including missense, nonsense, frameshift, or splice-site mutations, and one (range, 0–4) cancer driver gene (25) mutation per sample. The mutation rate of LGSC was 0.75 coding SNV mutations per Mb (Fig. 2A), which is approximately half that of primary HGSC (1.4 mutations per Mb; ref. 19). The most common base substitutions were C>T transitions in an NpCpG context (Fig. 2B), which is a mutational signature previously associated with patient age at diagnosis (26). Accordingly, we identified an association between patient age and the total number of somatic variants (Fig. 2C; Spearman P < 0.002).

Genes significantly mutated over the background mutation rate were identified using MuSiC (Supplementary Table S7 and Supplementary Methods). Recurrent significantly mutated genes (LRT, likelihood ratio test P < 0.05 and FDR-correct P <0.2) are shown in Fig. 2A. Consistent with previous studies, mutations in KRAS, NRAS, or BRAF were common (2, 10, 27) and observed collectively in 57% (13/23) of cases. Additionally, two samples had disruptive frameshift indels in NF1, a negative regulator of the RAS pathway (Fig. 2A). All RAS pathway mutations were mutually exclusive. Other recurrently mutated genes are shown in Supplementary Table S8 and included missense mutations in EIF1AX and FFAR1 and disruption of USP9X (Fig. 2A), also described by Hunter and colleagues (27). All samples were shown to be of sufficiently high tumor purity to detect high confidence variants (Supplementary Fig. S3A). A complex structural rearrangement comprised of an inversion and two translocations was identified in the WGS sample and was predicted to disrupt the USP9X locus through the homozygous deletion of exons 6 to 13 (Fig. 2D and Supplementary Table S5). USP9X is a deubiquitinase, tumor suppressor with a role in genome stability and apoptosis (27–29).

EIF1AX encodes X-linked eukaryotic translation initiation factor 1A commonly mutated in uveal melanoma (30) and thyroid cancer (31, 32). We screened an independent set of 55 serous ovarian tumors (25 grade 1 and 30 grade 2) for mutations in EIF1AX (all exons), and mutation hotspots in NRAS (exons 2 and 3), KRAS (exons 2 and 3) and BRAF (exons 11 and 15). We identified a further 14 cases with missense mutations in KRAS (n = 10), BRAF (n = 3), or NRAS (n = 1). The NRAS mutation co-occurred with a mutation in EIF1AX (Supplementary Table S9; Supplementary Fig. S3B). Grade 2 cases with no identified mutation (n = 28) were sequenced for TP53 (exons 2–11) and 26 cases with deleterious mutations were identified and excluded from further analysis as these were presumed to be HGSC (Supplementary Table S10). EIF1AX mutations identified in the WES (n = 3) and validation cohorts (n = 1) combined (4/51, 8%; Fig. 3A and Supplementary Fig. S3B) co-occurred with NRAS mutations (Fisher test P < 0.001). This is consistent with anaplastic and poorly differentiated thyroid cancers, where EIF1AX mutations co-occurred with mutations in the RAS pathway (31, 33).

The amino acid sequence of eIF1A is highly conserved between human and yeast (34, 35). Moreover, yeast eIF1A can replace human factor in an in vitro translation assay (36), suggesting that mutations found in yeast eIF1A may have similar effects on protein synthesis in human tumors with analogous mutations in EIF1AX. Studies in yeast and mammals showed that EIF1AX acts synergistically with eIF1 and is essential for the scanning ability of the 48S ribosomal preinitiation complex (PIC) and for selection of the cognate AUG initiation codon (37). The structure of the human EIF1AX comprises a central domain with an oligonucleotide/oligosaccharide-binding (OB) fold, two extensions (N- and C-termini) and a small helical subdomain, which are absent in its bacterial homologue IF1 (34). N- and C-terminal residues of eIF-1A have opposing effects on the fidelity of start codon selection (38).

We observed that missense mutations in LGSC appear to be clustered in exons 1 to 2 encoding the unstructured N-terminus of EIF1AX (Supplementary Tables S8 and S9). We expanded our analysis to mutations reported across cancer types (27, 30, 31, 33), including 51 unique mutations in 81 patients from various studies including ours (Fig. 3B and Supplementary Table S11), and superimposed these on a three-dimensional structure of EIF1AX. We observed that mutations in EIF1AX across cancer types (Supplementary Table S11) clustered within the modeled structure (Fig. 3C) and that a majority of mutations were located at the unstructured N-terminus. Biochemical studies in yeast showed that N-terminal tail (NTT) of eIF1A inhibits scanning-competent conformation of the PIC (37). Structural data suggest that NTT would stabilize the locked conformation of initiator transfer RNA (tRNAi) in the P site of the 40S subunit by interacting with the ribosomal protein eS31, mRNA, rRNA, and the anticodon stem loop (ASL) of tRNAi (39–41).

A second group of EIF1AX mutations was found in the OB domain, which binds in the A site of the 40S subunit and interacts with the 18S rRNA and ribosomal proteins uS12 (S23) and eS30 (S30; refs. 39, 40, 42). Four mutations occurred in the helical subdomain (E99, A113, N116, and E117), while one (R24T) was in a region of the helical domain juxtaposed with the N-terminal region. R24 likely forms an arginine–carboxyl pair with E99 in wild-type eIF1A, connecting the helical and N-terminal subdomains and stabilizing the interface between the helical subdomain and the head domain of the 40S subunit. Both the helical subdomain and the N-terminus interact with helix h30 of the 18S rRNA. One mutation, D127A, was located at the beginning of an unstructured C-terminal tail (Fig. 3C). Amino acid residue D127 in yeast eIF1A is a part of the Scanning Enhancer (SE) 1, which together with the SE2, stimulates recruitment of initiator tRNA–eIF2–GTP ternary complex and also prevents initiation at UUG codons (43).

The presence of clustered missense mutations is consistent with a gain-of-function mutation (44), although given EIF1AX is on the X-chromosome it remained possible that EIF1AX mutation may represent loss of tumor suppressor function in conjunction with X-inactivation of the wild-type allele. We therefore analyzed EIF1AX expression where LGSC RNA was available for analysis. Consistent with a previous report of escape from X-inactivation (30), we observed expression of both the wild-type and mutant alleles of EIF1AX (Supplementary Fig. S4). To rule out a contribution of wild-type gene transcripts from normal cell contamination, we performed immunohistochemical analysis and observed cytoplasmic expression of EIF1AX protein in the epithelial fraction of all tumor samples examined (n = 18; Fig. 4). Low EIF1AX expression was seen in normal ovarian surface epithelium (Fig. 4A). Higher expression was seen in normal fallopian tube epithelium, with more intense staining seen in secretory cells compared with ciliated cells (Fig. 4B). The cellular origin of LGSC is unresolved (45); however, it is noteworthy that among these tissues the highest expression of EIF1AX was observed in fallopian tube secretory cells, the presumed precursor of HGSC and possibly LGSC (45, 46). EIF1AX staining intensity varied between cases (Fig. 4C–I), but there was no relationship with the presence or absence of EIF1AX, KRAS, BRAF, or NRAS mutation (Table 1).

To test whether EIF1AX and NRAS mutations cooperate functionally, we expressed wild-type and mutant forms of EIF1AX and NRAS, individually and in combination in a newly established LGSC, RAS-WT cell line, AOCS2. Cells grown in culture were positive for PAX8 and WT1, consistent with epithelial serous ovarian cancer (Supplementary Fig. S5A). Variable TP53 expression was observed (wild-type pattern) and amplicon sequencing confirmed wild-type TP53 status. A relatively stable copy-number profile with few, large chromosomal gains and losses in tumor tissue, cells from ascites and cell lines grown in culture was consistent with TP53 wild-type LGSC (2). Variation in the copy-number pattern between the solid tumor tissue and ascites samples compared with the cell line presumably reflects clonal diversity and in vitro selection of a subpopulation of cells (Supplementary Fig. S5B).

Overexpression of NRAS and EIF1AX was validated at the transcript (Supplementary Fig. S6A–S6C) and protein levels (Fig. 5A). We performed proliferation and clonogenic assays across cell lines transduced with control vectors (EV, empty vector; RFP, red fluorescent protein), WT-EIF1AX or EIF1AX^G8E individually, and in a WT-NRAS or NRAS^Q61K background. Expression of NRAS^Q61K alone led to a significant increase in cell proliferation (Fig. 5B) and clonogenic survival (Fig. 5C) compared with controls; however, this was not observed for WT-NRAS, WT-EIF1AX, or EIF1AX^G8E. Importantly, overexpression of EIF1AX^G8E on an NRAS^Q61K background led to significantly increased cell proliferation and clonogenic survival compared with cells expressing WT-NRAS and WT-EIF1AX alone or in combination. These results demonstrate a functional cooperation between NRAS and EIF1AX mutations in LGSC. Overexpression of EIF1AX^G8R and EIF1AX^P2L mutants in combination with NRAS^Q61K had a limited effect, suggesting that different mutations in EIF1AX may not be equivalent (Supplementary Fig. S6D and S6E).

Loss-of-function experiments achieved knockdown of WT-EIF1AX protein using two unique EIF1AX shRNAs in AOCS2 cells transduced with NRAS^Q61K overexpression constructs (Fig. 5D and Supplementary Fig. S6F). In contrast to overexpression studies, suppression of EIF1AX on the background of NRAS^Q61K had no effect on cell proliferation (Fig. 5E) or clonogenic survival (Fig. 5F).

EIF1AX and NRAS function in the mTOR and RAS/ERK signaling pathways to regulate protein translation, cell proliferation, and cell survival (47–49). To understand how overexpression of mutant EIF1AX and NRAS proteins may affect intracellular signaling, we evaluated pathway activation by Western blot. Although mutant and wild-type EIF1AX and NRAS proteins were successfully overexpressed, we were unable to identify differential upregulation of mTOR or RAS/ERK pathway members in the AOCS2 cells cotransduced with mutant versus wild-type alleles (Supplementary Fig. S7). Our additional functional experiments were constrained by the lack of LGSC and borderline cell lines with appropriate background mutations. Derivation of additional LGSC cell lines, such as AOCS2, would benefit further characterization of the role of EIF1AX.

Subtypes of ovarian cancer differ in biological behavior and underlying molecular features. Most comprehensive genomic studies to date have focused on the more common histotype, HGSC (19, 32). Here, we performed WES and WGS of LGSC, a molecularly distinct and relatively understudied subtype of ovarian cancer that occurs in younger women and typically does not respond to standard chemotherapy. Somatic RAS–MAPK pathway mutations have been described in LGSC and are seen in about half of LGSC cases. In addition to mutations in BRAF, KRAS, and NRAS, we show that frameshift mutations in the negative regulator of RAS activity, NF1, may further contribute to deregulation of the RAS–MAPK pathway in LGSC. We have also shown an association between the presence of NRAS mutations and EIF1AX. The co-occurrence of mutations in NRAS and EIF1AX found in LGSC mirrors the association found in other cancer types, including uveal melanoma (30) and poorly differentiated thyroid cancer (31–33).

Translation initiation factor eIF1A, encoded by the EIF1AX gene, is universally conserved across all kingdoms of life (34, 35). It is essential for initiation of protein synthesis, particularly for the scanning ability of the 48S ribosomal PIC. It acts synergistically with eIF1 to promote assembly of the 48S PIC at the AUG initiation codon and destroy aberrant ribosomal complexes on the mRNA (37, 50). Solution structure of human eIF1A revealed a central domain with an oligonucleotide/oligosaccharide-binding fold, small helical subdomain that is absent in IF1, bacterial homologue of eIF1A, and charged basic and acidic unstructured N- and C-termini, respectively (34). Our analysis across cancer types shows that mutations in EIF1AX cluster into four main groups, all of which are expected to weaken AUG selection. For example, mutations in the N-terminus of eIF1A are predicted to weaken its ability to stabilize tRNAi in the base-paired state, resulting in reduced start codon recognition and a “leaky scanning” phenotype (43). Subsequently, translation from the second AUG may increase, altering the ratio of different isoforms of some proteins and translation regulation by short upstream reading frames. In addition, N-terminal mutations are expected to decrease alternative translation from noncanonical UUG start sites. Mutations may therefore act through deregulation of global translation, by altering the ratio of isoforms of specific proteins and/or destabilize translational regulation by short upstream reading frames (51). Increased translation of the shorter isoform of C/EBPβ has been associated with cancer (52). Translation from noncanonical start sites has been observed in data derived from mouse embryonic stem cells (53) but its extent in cancer is less well defined.

Consistent with the co-occurrence of EIF1AX and NRAS mutations, our functional experiments demonstrated that overexpression of both mutant proteins together led to increased cell proliferation and clonogenic survival, compared with cells overexpressing the wild-type proteins. How such effects relate to altered intracellular signaling pathways is unclear, as we were unable to detect differential modulation of mTOR or ERK/RAS pathways in mutant EIF1AX and NRAS double transfectants. Broader and more quantitative proteomic measures, such as with reverse phase protein arrays (54), may provide additional insight. Suppression of EIF1AX had no effect on cell proliferation or clonogenic survival, supporting our conclusion that EIF1AX mutations are likely to be activating and consistent with predictions based on analysis of EIF1AX protein structure. These findings suggest functional cooperation in malignant progression through a novel oncogenic mechanism involving relaxed translational initiation and mutant NRAS expression. This first example of a cooperative activating mutational event in ovarian cancer may represent a new treatment target for patients with NRAS-mutated LGSC.

No potential conflicts of interest were disclosed.

Conception and design: D. Etemadmoghadam, W.J. Azar, Australian Ovarian Cancer Study Group, R. Sharma, D.D.L. Bowtell, A. deFazio

Development of methodology: D. Etemadmoghadam, W.J. Azar, Y. Lei, S. Fereday, Australian Ovarian Cancer Study Group, R. Sharma, G.M. Arnau, T. Semple, S.M. Grimmond, H. Rizos

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Etemadmoghadam, W.J. Azar, Y. Lei, T. Moujaber, C.J. Kennedy, S. Fereday, C. Mitchell, Y.-E. Chiew, J. Hendley, Australian Ovarian Cancer Study Group, P.R. Harnett, G. Au-Yeung, G.M. Arnau, T.P. Holloway, T. Semple, S.M. Grimmond, M. Köbel, H. Rizos, A. deFazio

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Etemadmoghadam, W.J. Azar, T. Moujaber, D.W. Garsed, Australian Ovarian Cancer Study Group, R. Sharma, J. Li, E.L. Christie, A.-M. Patch, J. George, G. Au-Yeung, J.V. Pearson, N. Waddell, S.M. Grimmond, M. Köbel, H. Rizos, I.B. Lomakin, D.D.L. Bowtell, A. deFazio

Writing, review, and/or revision of the manuscript: D. Etemadmoghadam, W.J. Azar, Y. Lei, T. Moujaber, D.W. Garsed, C.J. Kennedy, S. Fereday, Y.-E. Chiew, J. Hendley, The Australian Ovarian Cancer Study Group, P.R. Harnett, E.L. Christie, A.-M. Patch, G. Au-Yeung, J.V. Pearson, N. Waddell, M. Köbel, H. Rizos, I.B. Lomakin, D.D.L. Bowtell, A. deFazio

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Lei, T. Moujaber, C.J. Kennedy, S. Fereday, C. Mitchell, J. Hendley, Australian Ovarian Cancer Study Group, J. Li, G.M. Arnau, T. Semple, D.D.L. Bowtell

Study supervision: Australian Ovarian Cancer Study Group, H. Rizos, D.D.L. Bowtell, A. deFazio

The AOCS gratefully acknowledges the cooperation of the participating institutions in Australia and also acknowledges the contribution of the study nurses, research assistants, and all clinical and scientific collaborators. AOCS Study Group members are listed in Supplementary Information. The investigators would like to thank all of the women, and their families, who participated in these research programs. The authors acknowledge assistance from Prue Cowin and Franco Caramia.

This work was supported by Cancer Council New South Wales (RG-15-23), The National Health and Medical Research Council of Australia (NHMRCID631701), Worldwide Cancer Research (09-0676), and Cancer Australia (1004673). The Australian Ovarian Cancer Study was supported by the U.S. Army Medical Research and Materiel Command under DAMD17-01-1-0729, The Cancer Council Victoria, Queensland Cancer Fund, The Cancer Council New South Wales, The Cancer Council South Australia, The Cancer Foundation of Western Australia, The Cancer Council Tasmania, and the National Health and Medical Research Council of Australia (NHMRC; ID400413, ID400281). The AOCS gratefully acknowledges additional support from S. Boldeman, the Agar family, Ovarian Cancer Australia and Ovarian Cancer Action (UK). The Gynaecological Oncology at Westmead, a member of the Australasian Biospecimen Network-Oncology group, was supported by grants from the NHMRC (ID310670 and ID628903) and the Cancer Institute New South Wales (12/RIG/1-17 and 15/RIG/1-16). A. de Fazio is supported by the University of Sydney Cancer Research Fund and Cancer Institute NSW, through the Sydney West Translational Cancer Research Centre. T. Moujaber is a recipient of a Westmead Medical Research Foundation PhD Scholarship.

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.