Adult‐type granulosa cell tumor of the ovary: a FOXL2‐centric disease

Abstract Adult‐type granulosa cell tumors (aGCTs) account for 90% of malignant ovarian sex cord‐stromal tumors and 2–5% of all ovarian cancers. These tumors are usually diagnosed at an early stage and are treated with surgery. However, one‐third of patients relapse between 4 and 8 years after initial diagnosis, and there are currently no effective treatments other than surgery for these relapsed patients. As the majority of aGCTs (>95%) harbor a somatic mutation in FOXL2 (c.C402G; p.C134W), the aim of this study was to identify genetic mutations besides FOXL2 C402G in aGCTs that could explain the clinical diversity of this disease. Whole‐genome sequencing of 10 aGCTs and their matched normal blood was performed to identify somatic mutations. From this analysis, a custom amplicon‐based panel was designed to sequence 39 genes of interest in a validation cohort of 83 aGCTs collected internationally. KMT2D inactivating mutations were present in 10 of 93 aGCTs (10.8%), and the frequency of these mutations was similar between primary and recurrent aGCTs. Inactivating mutations, including a splice site mutation in candidate tumor suppressor WNK2 and nonsense mutations in PIK3R1 and NLRC5, were identified at a low frequency in our cohort. Missense mutations were identified in cell cycle‐related genes TP53, CDKN2D, and CDK1. From these data, we conclude that aGCTs are comparatively a homogeneous group of tumors that arise from a limited set of genetic events and are characterized by the FOXL2 C402G mutation. Secondary mutations occur in a subset of patients but do not explain the diverse clinical behavior of this disease. As the FOXL2 C402G mutation remains the main driver of this disease, progress in the development of therapeutics for aGCT would likely come from understanding the functional consequences of the FOXL2 C402G mutation.


Introduction
Adult-type granulosa cell tumors (aGCTs) are the most common type of malignant ovarian sex cordstromal tumor (SCST) [1]. They account for 2-5% of all ovarian malignant tumors and 90% of malignant SCSTs [2]. aGCTs occur in perimenopausal women with a median age of diagnosis of between 50 and 54 years [3,4].These tumors arise from the granulosa cells of the ovarian follicles and have distinct clinical features, such as their ability to secrete estrogen and inhibins [5]. Due to their indolent growth and unique hormonal activity, these tumors are usually diagnosed at an early stage and are treated with surgery [6][7][8]. A somatic missense mutation (c.C402G; p.C134W) in the transcription factor FOXL2 was identified in over 95% of aGCTs and can be used as a clinical diagnostic assay for differential diagnosis [9][10][11][12]. In our previous study by McConechy et al, we showed that, in a molecularly defined cohort of aGCTs (FOXL2 C402G mutation positive), the overall survival did not differ from age-matched, population-based controls [13]. However, one-third of aGCT patients relapse, and there are no effective treatments for relapsed patients with inoperable tumors [14,15].
Due to the propensity of aGCT to recur years after initial diagnosis (typically 4-8 years, with the median being 7.2 years in molecularly defined cohorts), the challenge is identifying patients who are at high risk for relapsed disease and require prolonged surveillance [13,14,16]. As the FOXL2 C402G mutation is present in essentially all aGCTs, there has been much research on secondary mutations to explain the diverse clinical behavior of these tumors. Our research group was the first to describe TERT C228T promoter mutations in 51 of 229 (22%) primary and 24 of 58 (41%) recurrent aGCTs [17]. Likewise, Alexiadis et al performed targeted TERT promoter sequencing and found that a higher frequency of recurrent tumors harbor the TERT C228T promoter mutation [18]. Hillman et al were the first to identify KMT2D truncating mutations in aGCT and suggested that KMT2D inactivation may increase the risk of tumor recurrence [19]. Da Cruz Paula et al used targeted sequencing of known cancer genes and identified genetic alterations in cell cycle-related genes, including TP53, TERT promoter, MED12, and CDKN2A/B, suggesting that these mutations may play a role in recurrence [20,21]. Furthermore, Roze et al suggest that patients with TP53 mutations represent a high-grade subgroup of aGCT [21].
In this study, we used whole-genome sequencing (WGS) on 10 aGCTs and their matched peripheral blood to identify novel somatic mutations in aGCT.
Validation was performed using targeted sequencing on an independent cohort of 83 primary and recurrent aGCTs collected from four international institutions. We identified KMT2D inactivating mutations in 10 of 93 aGCTs (10.8%). A splice site mutation was identified in the candidate tumor suppressor WNK2, and nonsense mutations were identified in NLRC5 and PIK3R1. Missense mutations were identified in cell cycle-related genes TP53, CDKN2D, and CDK1.

Materials and methods
Clinical specimens for exploratory WGS Ten representative fresh-frozen aGCTs were selected from OVCARE's Gynecological Tissue Bank in Vancouver, Canada, for WGS. Patient tumor and peripheral blood samples were collected at diagnosis during standard-of-care debulking surgery. For DNA isolation from banked samples, frozen tissue samples were cryosectioned, and sections adjacent to the scrolls submitted for sequencing were stained with hematoxylin and eosin to assess tumor cell content. These slides were reviewed by two of the following gynecologic pathologists: HMH, ANK, HLC, and BG, and the final cases with >80% tumor cellularity were selected by BG. All cases were positive for the FOXL2 (c.C402G; p.C134W) mutation. In brief, WGS was performed using Illumina HiSeq 2500 v4 chemistry (Illumina Inc., Hayward, CA, USA) with the polymerase chain reaction (PCR)-free protocol to eliminate PCR-induced bias and to improve coverage across the genome. The average read depth was 46.6X per sample (range: 36-60X). WGS analysis comparing tumor tissue to matched peripheral blood was performed to identify somatic alterations in each case, including singlenucleotide variants (SNVs), small insertions and deletions (indels), copy number alterations (CNAs), and structural variants (SVs). This cohort was previously published by our research team, and detailed methods are available [22].

Case inclusion
All cases were tested for FOXL2 C402G mutation using an SNP genotyping assay and were confirmed to contain the FOXL2 C402G mutation. As the validation cohort consisted of FFPE material with no matched normal tissue available, the cases included in the final cohort reported were selected if they satisfied the following two conditions: (1) two libraries (A and B) per case were constructed successfully and sequenced, and (2) FOXL2 C402G mutation was detected using the targeted sequencing variant analysis pipeline to call mutations in the gene panel. Condition 1: Two libraries (A and B) per case are important to eliminate any sequencing artifacts that may have resulted from the formalin fixation process. Libraries A and B were constructed with different probes to generate amplicons covering the 39 genes selected in our custom gene panel. Both libraries A and B were sequenced separately and processed independently through the variant analysis pipeline. After the variants were called for both libraries A and B, an intersection function was performed that generated the list of variants that were present in both libraries A and B. The assumption that formalin fixation introduces SNVs at random enabled the removal of these randomly generated SNVs that were only present in one of the two libraries. The mutations reported in the Results section were present in both libraries A and B for each case. Condition 2: The detection of the FOXL2 C402G mutation using the targeted sequencing variant analysis pipeline was employed to ensure that the targeted sequencing pipeline was highly sensitive. The aim of the study was to report on a molecularly defined aGCT cohort, where all cases harbor the FOXL2 C402G mutation. Therefore, cases were tested beforehand using an SNP genotyping assay to confirm the presence of the FOXL2 C402G mutation, and only cases where the FOXL2 C402G mutation was called using the targeted sequencing variant analysis pipeline were included in the validation cohort. This ensured that the pipeline was highly sensitive in detecting variants in the remaining genes in the panel. Due to the stringency required to remove FFPE artifacts and the fact that some of the cases are over 50 years old with no matched normal tissue available, it was expected that there would be a high number of cases excluded from the validation cohort. A less rigorous approach would lead to the inclusion of an overwhelming number of false-positive mutations. A flowchart outlining the number of cases excluded at each step is illustrated in Figure 1. was confirmed prior to targeted sequencing. The lowest variant allele frequency (VAF) in which FOXL2 C402G was detected was 8.5% using the variant analysis pipeline on the targeted sequencing data. This established the limit of the detection for identifying the FOXL2 C402G mutation and was thus chosen as the VAF for variant reporting of the remaining genes in the panel. Furthermore, the FOXL2 C402G mutation is expected to be clonal with secondary subclonal mutations. For missense mutations, three tools (PolyPhen-2, SIFT, and MutationAssessor) were selected to predict the pathogenicity of each variant using Ensembl's Variant Effect Predictor. Variants that were predicted by two or more tools to have no effect on protein function were removed. Therefore, only missense mutations that were predicted as pathogenic by two or more tools were reported. A flowchart describing the steps of variant filtering is illustrated in supplementary material, Figure S1.

WGS on exploratory cohort
Ten aGCTs (eight primary and two recurrent) and their matched normal DNA samples were subjected to WGS as previously described [22]. The median age of diagnosis was 51 years for aGCT, and the clinical characteristics are described in Table 1. All aGCTs were heterozygous for the FOXL2 C402G mutation.
Somatic variations were detected in the tumor genome of each patient, including SNVs, small indels, CNAs, and SVs. The overall mutational burden for these 10 aGCTs was low as previously described in the Wang et al's study [22]. SNVs and small indels are described in Table 1. The majority of mutations reported was identified in only 1 of 10 aGCTs (10%) except for the TERT C228T promoter mutation that was present in 5 of 10 aGCTs (50%). Missense mutations in BMP7 and MPHOSPH8 were identified in 2 of 10 aGCTs (20%). Copy number analysis revealed loss of chromosome 22 in 6 of 10 aGCTs (60%) and gain of chromosome 14 in 2 of 10 aGCTs (20%). Concurrent loss of chromosome 22 and gain of chromosome 14 was observed in 1 of 10 aGCTs (10%). No gain of chromosome 12 was found. These results confirm previous reports of CNAs in aGCTs, including loss of chromosome 22 in 30-56% and gain of chromosome 14 in 20-27% [18,21,[23][24][25]. Other structural rearrangements were sporadic and infrequent (see supplementary material, Figure S2).

Targeted sequencing on validation cohort
Our validation cohort encompassed 83 aGCTs, including 61 primary and 22 recurrent tumors, collected from four international institutions. For the 58 patients with age data, the median age at diagnosis was 52 years (range: 26-94 years). The clinical features for each cohort are described in Table 2. All cases included in the validation cohort have confirmed FOXL2 C402G Cases where both libraries were successful were sequenced, and variants were called (n = 129). Cases where the previously confirmed FOXL2 C402G mutation was not detected using the targeted sequencing variant analysis pipeline were excluded (n = 46). The final cohort consisted of cases with two libraries and where FOXL2 C402G mutation was detected using the targeted sequencing variant analysis pipeline (n = 83).

246
JA Pilsworth et al  247 Adult-type granulosa cell tumor: mutation profiles mutation determined by an SNP genotyping assay, and this mutation was detected in the targeted sequencing for all cases reported (Figure 1). Two cases harbor FOXL2 C402G hemizygous mutations, which appear as a homozygous state. The remaining cases were FOXL2 C402G heterozygous. For the 73 aGCTs with available TERT C228T promoter mutation data, the frequency was 18 of 73 aGCTs (24.7%), which is similar to our original TERT C228T promoter mutation publication that contains additional cases along with 73 cases of the current validation cohort (75/287; 26.1%). The following missense mutations were mutated in two aGCTs: KMT2D (c.C3893T; p.S1298F) and NLRC5 (c.G823A; p.E275K). All genes mutated in our panel are shown in Figure 2 and in supplementary material, Table S2. KMT2D inactivating mutations (nonsense, indel, or splice site mutations) were present in 9 of 83 aGCTs in our validation cohort, and an indel (frameshift mutation) was reported in 1 case from the exploratory cohort (10/93; 10.8%). Missense mutations in KMT2D were observed in 9 of 93 aGCTs (9.7%); however, these mutations have not previously been reported, and therefore, their relevance to aGCT development is unknown. The missense mutations annotated in dark blue in Figure 2 are putative driver events and include mutations in FOXL2, TP53, and KDR. The Mutual exclusivity of TP53 and TERT C228T promoter mutations was observed for the three cases with available TERT C228T promoter mutation data.

Discussion
Here, we report the genomic analysis of 93 aGCTs, a malignant SCST of the ovary associated with late recurrence. The current study is a targeted genomic analysis of a molecularly defined aGCT cohort, where all tumors harbor the FOXL2 C402G mutation. Our research team initially described TERT C228T promoter mutations in 51 of 229 primary (22%) and 24 of 58 recurrent (41%) aGCTs, suggesting that these mutations occur later in tumorigenesis. The majority of the previously described cohort is included in this study, and TERT C228T promoter mutations are annotated in Figure 2 and Table 2. Multiple research groups have also observed TERT promoter mutations with a higher frequency of mutations in recurrent aGCTs (6/9, 67%; 1/28, 64%) compared to primary aGCTs (5/17, 29%; 5/19, 26%) [18,20]. In our study and that of Alexiadis et al, seven and five patients whose tumors were initially wild-type for the TERT promoter acquired the C228T mutation in their recurrent tumors, respectively [17,18]. These data support our statement that TERT promoter mutations are acquired during tumor progression after the initial FOXL2 C402G driver mutation. Although targeting telomerase as an anticancer treatment may represent an attractive strategy for recurrent aGCT, the development of efficacious cancer-specific telomerase inhibitors remains an ongoing challenge [26]. Using a combination of WGS and targeted sequencing, the results of the current study demonstrate that KMT2D undergoes frequent inactivating mutations in aGCTs. We report a similar frequency of KMT2D inactivating mutations (10/93, 10.8%) as the Hillman et al's study, which reported that 11 of 79 (13.9%) aGCTs harbored KMT2D inactivating mutations. Hillman et al described inactivating mutations in KMT2D as a recurrent event (10/43, 23%) in these tumors compared to primary tumors (1/35, 3%) [19]. However, in our dataset, we observed a similar frequency of inactivating mutations in KMT2D in primary (8/69, 11.5%) and recurrent (2/24, 8.3%) aGCTs. Aligned with our current results, Da Cruz Paula et al identified only one inactivating KMT2D mutation in a primary nonrecurrent aGCT in a cohort consisting of 10 primary nonrecurrent aGCTs (defined by no recurrent tumor within 4 years of initial diagnosis), 28 recurrent aGCTs, and 9 matched primary aGCTs with known recurrences [20]. The occurrence of KMT2D inactivating mutations in 10.8% of our aGCT cohort not only suggests that they are a pathogenic driver mutation in a subset of aGCTs but also that these mutations do not correlate with recurrence. We also observed KMT2D missense mutations in a subset of aGCTs (9/93; 9.7%), but these variants are of unknown significance. A limitation of this study was that the validation cohort did not have matched normal tissue, and thus, some reported missense mutations may be rare SNPs. Further research into the role of KMT2D in aGCT development is warranted, specifically exploring the epigenetic landscape of these tumors to identify aberrant histone marks and methylation patterns.
Additional inactivating mutations, including a splice site mutation in the candidate tumor suppressor WNK2 and nonsense mutations in the newly discovered protein NLRC5, are reported. WNK2 was recently discovered to be epigenetically silenced through promoter hypermethylation in gliomas and meningioma, suggesting its role as a growth suppressor [27][28][29]. WNK2 was also found to be downregulated via epigenetic silencing in early pancreatic ductal adenocarcinoma and may support cell proliferation through the Mitogen-activated protein kinase (MAPK) signaling pathway [30]. WNK2 somatic mutations and copy number loss were reported in hepatocellular carcinoma, resulting in lower WNK2 protein levels, which were associated with early tumor recurrence linked to enhanced ERK1/2 signaling [31]. Moreover, WNK kinases have been linked to Wnt/β-catenin signaling, a known pathway involved in granulosa cell development, and loss-of-function WNK genes resemble canonical Wnt pathway mutants [32]. NLRC5 has recently been linked to the regulation of cancer immune evasion, but its role in tumor development is controversial [33]. Some research studies report that NLRC5 elicits antitumor immunity, while other studies found that it promoted tumorigenesis and progression [33]. In fact, NLRC5 has been shown to regulate the Wnt/β-catenin signaling pathway to promote cell proliferation and migration in both clear cell renal carcinoma and hepatocellular carcinoma [34,35].

249
Adult-type granulosa cell tumor: mutation profiles Various signaling pathways have been explored in the development of aGCT, including transforming growth factor beta (TGF-β), phosphatidylinositol-3-kinase; serine/threonine kinase (PI3K/AKT), GATA4, and vascular endothelial growth factor (VEGF). These factors have been shown to play important roles in granulosa cell proliferation, apoptosis, or angiogenesis. In our aGCT cohort, we observed a small number of mutations in members of these pathways including GL1, BMP7, SMAD3 (TGF-β pathway), PIK3R1 (PI3K/AKT pathway), GATA4, and KDR/VEGFR2. However, these mutations occurred in a small percentage of aGCT patients and do not provide any prognostic or predictive value. In addition, mutations in cell cycle-related genes, including TP53, CDKN2D, and CDK1, were identified.
Da Cruz Paula et al used targeted sequencing of over 400 cancer-related genes on a cohort of 38 aGCT patients. They identified mutations present in recurrent aGCTs that were not present in primary aGCT, including TERT promoter mutations (C228T and C250T), MED12, and TP53, as well as CDKN2A/B homozygous deletions, suggesting that genetic alterations in cell cycle-related genes may be associated with recurrence [20]. In the current cohort, TP53 mutations were observed in three primary and one recurrent aGCTs, CDKN2D mutations were observed in one primary and one recurrent aGCT, and a CDK1 mutation was observed in one primary aGCT. The frequency of mutations in cell cycle-related genes is low in our cohort; thus, it is difficult to draw conclusions regarding their association with recurrence.
Most recently, Roze et al used WGS of 33 aGCT patients and identified TP53 mutations in three patients (3/33; 9.1%) with higher tumor mutational burden and mitotic activity. They propose that tumors with TP53 mutations define a high-grade subgroup of aGCT and suggest that a personalized medicine approach is required for treatment of aGCTs. Furthermore, they state that the absence of the FOXL2 C402G mutation does not exclude an aGCT diagnosis and that TP53, TERT C228T promoter, and DICER1 mutations may drive tumorigenesis [21]. The data presented here also show that aGCT can occasionally harbor other driver mutations, including KMT2D (10/93; 10.8%) and TP53 (4/93; 4.8%). However, only a small minority of patients would benefit from a genomic-driven approach to treatment. The FOXL2 C402G mutation is the dominant genetic event, and greater benefits would be derived from targeting the downstream mutations.
Inclusively, this study and other research studies have not identified any common actionable features in aGCT. Although it is still possible that features such as promoter methylation could be discovered in noncoding regions, it appears that aGCT is not a disease that could be stratified into different treatment groups by genomic features. Our data and those of others suggest that this is a highly specific clinical entity driven by the FOXL2 C402G mutation, and treatment strategies focusing on targeting FOXL2 or the downstream consequences are more likely to be successful in treatment compared to a precision medicine approach. Two recent publications have shown initial evidence that the FOXL2 C402G mutation alters DNA-binding specificity of the FOXL2 C134W protein (FOXL2 C134W ). Carles et al identified unique targets of FOXL2 C134W , including SLC35F2, a solute transporter that transports sepantronium bromide, which is a transcriptional suppressor of survivin. They found that the immortalized granulosa cell line SVOG3e transduced with inducible FOXL2 C134W showed an increased sensitivity to YM155, a small molecule inhibitor of survivin, compared to FOXL2 wild-type or empty vector. The authors suggest YM155 as a potential therapeutic strategy for aGCT [36]. Weis-Banke et al show that the mutant FOXL2 C134W acquires the ability to bind SMAD4, which forms a FOXL2 C134W /SMAD4/SMAD2/3 complex that is able to bind to a novel hybrid DNA motif and induces the transcription of genes involved in epithelial-tomesenchymal transition (EMT). They show that the ablation of SMAD4 or SMAD2/3 in the immortalized granulosa cell line HGrC1 reduces binding of FOXL2 C134W to these hybrid sites and decreases expression of these EMT-related genes, suggesting the possibility of targeting this FOXL2 C134W -SMAD4 interaction for therapeutic purposes [37]. These studies may lead to the first generally applicable, biologically informed therapeutic strategies for aGCT of the ovary.