Genomic profiling of primary and recurrent adult granulosa cell tumors of the ovary

Adult-type granulosa cell tumor (aGCT) is a rare malignant ovarian sex cord-stromal tumor, harboring recurrent FOXL2 c.C402G/p.C134W hotspot mutations in 97% of cases. These tumors are considered to have a favorable prognosis, however aGCTs have a tendency for local spread and late recurrences, which are associated with poor survival rates. We sought to determine the genetic alterations associated with aGCT disease progression. We subjected primary non-recurrent aGCTs (n=7), primary aGCTs that subsequently recurred (n=9) and their matched recurrences (n=9), and aGCT recurrences without matched primary tumors (n=10) to targeted massively parallel sequencing of ≥410 cancer-related genes. In addition, 3 primary non-recurrent aGCTs and 9 aGCT recurrences were subjected to FOXL2 and TERT promoter Sanger sequencing analysis. All aGCTs harbored the FOXL2 C134W hotspot mutation. TERT promoter mutations were found to be significantly more frequent in recurrent (18/28, 64%) than primary aGCTs (5/19, 26%, p=0.017). In addition, mutations affecting TP53, MED12 and TET2 were restricted to aGCT recurrences. Pathway annotation of altered genes demonstrated that aGCT recurrences displayed an enrichment for genetic alterations affecting cell cycle pathway-related genes. Analysis of paired primary and recurrent aGCTs revealed that TERT promoter mutations were either present in both primary tumors and matched recurrences or were restricted to the recurrence and absent in the respective primary aGCT. Clonal composition analysis of these paired samples further revealed that aGCTs display intra-tumor genetic heterogeneity and harbor multiple clones at diagnosis and relapse. We observed that in a subset of cases, recurrences acquired additional genetic alterations not present in primary aGCTs, including TERT, MED12 and TP53 mutations and CDKN2A/B homozygous deletions. Albeit harboring relatively simple genomes, our data provide evidence to suggest that aGCTs are genetically heterogeneous tumors and that TERT promoter mutations and/or genetic alterations affecting other cell cycle-related genes may be associated with disease progression and recurrences.

Somatic TERT promoter mutations (C228T and C250T), associated with telomerase activation, have been reported at high frequency in cancers (12% overall) [11], including gynecologic malignancies such as ovarian clear cell carcinomas (16%) [12,13]. In addition, recent studies have reported a significantly higher frequency of the TERT C228T promoter hotspot mutations in recurrent (41-67%) than in primary aGCTs (22-29%) [14,15]. Patients with primary aGCTs harboring TERT promoter hotspot mutations were also found to have a significantly worse overall survival than those with wild-type TERT [14]. Furthermore, KMT2D inactivating mutations have been reported to be associated with recurrences in aGCTs [16].
Although TERT promoter and KMT2D mutations appear to provide the basis for recurrences in a subset of aGCTs, the genetic basis of the clinical behavior in a substantial proportion of recurrent aGCTs has not been defined to date. Here, we sought to compare the repertoire of somatic genetic alterations of (1) primary aGCTs that did not recur within at least four years of follow-up, (2) primary aGCTs that recurred and (3) aGCT recurrences. Furthermore, given that samples from primary aGCTs and their respective relapses were available, we have also compared the TERT mutation status between paired primary and recurrent aGCTs, and investigated whether aGCTs would display intra-tumor genetic heterogeneity and if specific genetic alterations would be selected during progression from primary tumor to recurrence.

Subjects and samples
Following approval by the Institutional Review Board (IRB) of the authors' institutions, we retrieved representative hematoxylin and eosin and unstained tissue sections from formalin-fixed paraffin embedded aGCTs from Memorial Sloan Kettering Cancer Center (MSKCC, NY, USA), Fudan University Cancer Center (Shanghai, China), Hospital Universitario de Bellvitge (Barcelona, Spain), Hospital Universitario Arnau de Vilanova (Lleida, Spain), and Cleveland Clinic (OH, USA). Patient consents were obtained according to the protocols approved by the local IRBs of the authors' institutions. Samples were anonymized prior to analysis. Samples from 40 cases were reviewed by eight pathologists (RB, SES, MV, CGP, XM-G, BPR, JSR-F, and DFD) following the criteria put forward by the World Health Organization [17]. Only cases where a consensus diagnosis of aGCT was achieved were included in this study (n = 38), and two cases were excluded. Patients were defined as having primary non-recurrent aGCTs if no recurrence was detected within at least 48 months of follow-up, based on the median and range of time-to-recurrence of aGCTs previously reported [4][5][6]. Our final series included 47 samples from 38 patients: (1) primary non-recurrent aGCTs (n = 10), (2) primary aGCTs that subsequently recurred (n = 9) and their matched recurrences (n = 9 from nine patients), and aGCT recurrences without matched primary tumors (n = 19; Table 1 and Supplementary  Table S1). Surgical staging was performed according to the 2014 International Federation of Gynecology and Obstetrics system [18].
Targeted capture massively parallel sequencing using the Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) [11,19] was performed on tumor-normal pairs from 26 patients, including seven primary non-recurrent aGCTs, nine primary recurrent aGCTs and their matched recurrences, and ten aGCT recurrences. The remaining three primary nonrecurrent aGCTs and nine aGCT recurrences from 12 additional patients did not yield sufficient DNA for MSK-IMPACT sequencing, and, therefore, were subjected to Sanger sequencing analyses to assess the presence of FOXL2 and TERT promoter hotspot mutations (see below; Supplementary Table S1).

Microdissection and DNA extraction
Representative sections of tumor tissue samples were microdissected under a stereomicroscope (Olympus SZ61) to ensure a tumor cell content >80%, as previously described [20,21]. DNA from tumor and matched normal tissues was extracted using the DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturers' instructions. DNA of sufficient quantity/ quality was obtained for Sanger sequencing for all 38 cases (47 samples) and for targeted massively parallel sequencing for 26 cases (35 samples; see below).
Assessment of FOXL2 and TERT promoter hotspot mutations by Sanger sequencing PCR amplification of FOXL2 and TERT promoter hotspot loci was performed using the AmpliTaq Gold 360 Master Mix kit (Life Technologies, ThermoFisher Scientific) using previously described primers [22,23]. PCR fragments were cleaned using ExoSAP It (ThermoFisher Scientific) and Sanger sequenced as previously described [22].

Pathway analyses
A MSigDB and DAVID pathway analysis was performed based on genes affected by nonsynonymous pathogenic somatic mutations, amplifications, or homozygous deletions in primary non-recurrent aGCTs (n = 7), primary aGCTs (n = 9) and matched recurrences (n = 9) and aGCT recurrences (n = 10) [38]. Pathways found to be significantly enriched (p < 0.01) were selected as previously reported [39]. In addition, a mutual exclusivity analysis was performed using combinations of mutually exclusive alterations (CoMET) with the use of a pair-wise Fisher's exact test to detect the presence of significant pairs of genes [40].

Mutation-based tree construction
The mutation-based trees of the primary and matched aGCT recurrences were constructed using Treeomics [41] based on all synonymous and nonsynonymous mutations identified, as previously described [24]. For these analyses, a given mutation was considered "shared" if it was present in both the primary and matched aGCT recurrences. We defined mutations present only in the primary tumor or only in the recurrence as "private to the primary aGCT" and "private to the aGCT recurrence", respectively.

Statistical analysis
The frequencies of somatic mutations affecting cancer genes in primary and recurrent aGCTs were compared using two-tailed Fisher's exact test. Mutual exclusivity was tested using two-tailed Fisher's exact test. The fraction of genome affected by CNAs in primary and recurrent aGCTs was aGCT adult-type granulosa cell tumor, N/A not applicable. a Primary non-recurrent aGCTs versus primary recurrent aGCTs, using Student's t test and Fisher's exact test. evaluated using a Mann-Whitney U test. All p values were two-tailed, and 95% confidence intervals were adopted for all analyses.

Clinico-pathologic features of primary and recurrent aGCTs
Our series encompassed aGCTs from 38 patients, including 10 patients with non-recurrent aGCTs (primary nonrecurrent aGCTs) and 28 patients with recurrent disease. Of the 28 patients with recurrent disease, we analyzed samples from the primary tumor (primary recurrent aGCT) and matched recurrences from nine patients, and from the recurrent tumors only from 19 patients (aGCT recurrences; Supplementary Table S1). The median age of patients at aGCT diagnosis was 52.5 years (range 34-87 years) in primary non-recurrent aGCTs (n = 10), 62 years (range 41-83 years) for patients with primary aGCT that developed recurrences (n = 9) and 56 years (range 34-89 years) for patients with aGCT recurrences without available matched primary tumors (n = 19). No significant differences in age at diagnosis were observed between primary non-recurrent aGCTs and primary aGCTs that recurred (p = 0.74, Student's t test, Table 1, Supplementary Table S1). All patients (10/10, 100%) with primary non-recurrent aGCTs had early stage disease (IA and IC) at the time of diagnosis, whereas 1 (11%) and two patients (22%) with primary aGCTs that recurred had stage II and stage III disease at the time of diagnosis, respectively. No significant differences in stage were observed between primary non-recurrent aGCTs and primary aGCTs that recurred (p = 0.087, Fisher's exact test), or other clinic-pathologic factors assessed (Table 1,  Supplementary Table S1).
Analysis of the somatic mutation data obtained from MSK-IMPACT further revealed that aGCTs overall displayed a relatively low mutation burden, with a median of 3 (range 1-7) somatic mutations in the genes analyzed, of which two (range 1-6) were nonsynonymous (Supplementary Table S3). Despite the numerically higher number of somatic mutations identified in aGCT recurrences (median 3, range 2-7), no statistically significant differences in the mutational burden were observed when compared with primary non-recurrent aGCTs (median 2, range 1-5; p = 0.47, Fisher's exact test) or primary recurrent aGCTs (median 2, 1-5; p = 0.86, Fisher's exact test). Recurrent mutations affecting known cancer-related genes such as GNAQ and KMT2C were identified, however none of these was statistically different between the groups (Fig. 1b). Also, inactivating KMT2D mutations, which have been reported to be associated with recurrence in aGCTs [16], were only found in a single sample in our study and affected a primary non-recurrent aGCT (aGCT35-P; Fig. 1b, Supplementary Table S3). In contrast, we identified TP53 pathogenic mutations only in aGCT recurrences (16%). Of note, the TP53-mutant aGCT recurrences did not harbor TERT promoter mutations (Fig. 1b); however formal mutual exclusivity analysis using CoMET showed that TERT promoter and TP53 mutations were not significantly mutually exclusive in aGCT recurrences (p = 0.062, Fisher's exact test, Fig. 2a), likely due to the low number of samples/TP53 mutations. We further found a subclonal pathogenic mutation affecting TET2 (5%) that was restricted to an aGCT recurrence lacking a TERT promoter mutation (aGCT82-R; Fig. 1b, Supplementary Fig. S1), and STAG2 and IDH1 pathogenic mutations in primary recurrent aGCTs (11%) and their matched aGCT recurrences (5%) but not in primary non-recurrent aGCTs (0%) (Fig. 1b). None of these differences reached statistical significance, however, likely due to the small sample size.

Recurrent aGCTs harbor somatic genetic alterations affecting the cell cycle pathway
Given the distinct genetic alterations observed in primary non-recurrent aGCTs and aGCT recurrences, we sought to determine the signaling pathways that are enriched in aGCT recurrences. For this, we performed a pathway analysis using the genes that were either affected by nonsynonymous pathogenic somatic mutations, amplifications, or homozygous deletions. This analysis revealed that there was an enrichment in cell cycle pathway genes that were altered in aGCT recurrences but not in primary aGCTs (Supplementary Table S4). As mentioned above, pathogenic TP53 alterations were solely found in aGCT recurrences (Fig. 2c, top), whereas CDKN2A/B homozygous deletions were identified in both aGCT recurrences and primary recurrent aGCTs but not in primary non-recurrent aGCTs (Fig. 2c, bottom left). In contrast, CDKN1B homozygous deletions were identified at different frequencies in primary non-recurrent aGCTs (14%), primary aGCTs that recurred (11%) and in aGCT recurrences (5%; Fig. 2c, bottom right). These findings suggest that in addition to the described TERT promoter mutations [14,15] alterations in cell cycle-related genes and apoptosis might also play a role in the progression of aGCTs.

TERT promoter mutations in primary aGCTs and matched recurrences
To investigate the role of TERT promoter mutations in the progression of aGCTs, we next assessed the TERT promoter mutation status in primary aGCTs and their matched recurrences using MSK-IMPACT and Sanger sequencing. Of the nine pairs of primary and recurrent aGCTs included in this study, five harbored TERT mutations in at least one of the samples of a given patient based on MSK-IMPACT sequencing (Fig. 3). We found that in two cases (aGCT77 and aGCT80), the primary lesion lacked TERT promoter mutations, but that the matched aGCT recurrences harbored a clonal C250T mutation (aGCT77) or a subclonal C228T b Fraction of the genome altered in primary non-recurrent aGCTs, primary recurrent aGCTs, and aGCT recurrences. c Frequency of lossof-function somatic genetic alterations affecting genes in the canonical cell cycle pathway. Genes are depicted in blue rectangles, and the percentage of primary non-recurrent aGCTs (Prim), primary recurrent aGCTs (Prim Rec), and aGCT recurrences (Rec) altered is shown below each gene. mutation (aGCT80). This finding provides evidence to suggest that, in these cases, TERT mutations were either selected from a minor subclone not detected in the sequencing of the primary tumor or were acquired during disease progression. In contrast, the primary aGCT18 and aGCT43 both harbored clonal C228T TERT promoter hotspot mutations, which were preserved in the respective recurrences. Finally, a subclonal C250T was found in both the primary and matched recurrence of case aGCT76 (Fig. 3).

Clonal composition analysis of paired primary and aGCT recurrences
To interrogate the genetic alterations in addition to TERT promoter mutations that might be associated with clinical progression of aGCTs, we performed a clonal composition analysis of the nine primary aGCTs and their matched recurrences. Our analyses revealed that both the primary aGCTs and their matched recurrences shared clonal mutations in FOXL2 (p.C134W) in all nine cases analyzed (Fig. 4, Supplementary Fig. S2). Furthermore, primary tumors and their matched recurrences also shared somatic mutations affecting KMT2C (p.A1685S), MYOD1 (p. S260F), KMT2D (p.C5481F), PIK3R1 (p.W624R), and the TERT promoter (Fig. 4, Supplementary Fig. S2). We observed, however, that in a subset of cases, the recurrences acquired additional somatic mutations or CNAs not present in the primary aGCT: we identified clonal MED12 (aGCT77, p.Q2076dup), clonal SH2D1A (aGCT78, p. P97S), and subclonal TET2 (aGCT82, p.C1281Vfs*82) mutations restricted to the recurrences. Furthermore, two aGCT recurrences acquired alterations in cell cycle-related genes such as TP53 mutations (aGCT78, p.F338Lfs*7; aGCT82, p.Y236H) or CDKN2A/B homozygous deletions (aGCT79), which were not detected in their respective primary tumors (Fig. 4).

Discussion
Despite their relatively indolent behavior, management of a subset of aGCTs remains challenging due to their unpredictable behavior and late relapses. Identification of markers predictive of disease recurrences/metastases has been the subject of considerable interest. Here we not only confirm the presence of TERT promoter hotspot mutations in aGCTs but also find that somatic genetic alterations affecting cell cycle progression and apoptosis-related genes may play a role in the progression from primary aGCTs to recurrences. Recent studies have reported a highly recurrent somatic mutation (C228T) in the promoter region of TERT in aGCTs [14,15]. In our study, in addition to the C228T mutation (80%), we also identified recurrent C250T TERT hotspot mutations in 20% of the aGCTs harboring TERT promoter mutations. Whilst TERT promoter mutations have been reported to be an early genetic event in several cancer types [42][43][44], in aGCTs one study has suggested that these mutations are a late event [14]. The overall frequency of TERT promoter hotspot mutations in our series was significantly higher in aGCT recurrences than in primary aGCTs. TERT promoter mutations have been reported to lead to increased TERT expression and telomerase activation, to overcome the proliferative barrier imposed by telomere shortening, and to promote both immortalization and tumorigenesis [45]. Of note, the analysis of paired primary and recurrent aGCTs revealed that TERT promoter mutations were likely acquired during disease progression in two cases (aGCT77 and aGCT80), whereas in three cases (aGCT18, aGCT43 and aGCT76) these mutations were present in both primary tumors and matched recurrences. These findings suggest that depending on the context and activation of other signaling pathways, TERT promoter mutations may either be an early event and play a role in the development of more aggressive primary disease, that has the potential to spread and recur; or, in other contexts, TERT promoter mutations may be a late event and acquired during progression, disease spread or recurrence [14,15]. Further studies are required to understand the interplay between TERT promoter mutations and other genetic or epigenetic alterations, and TERT's activation and role in aGCT maintenance and progression. It is unlikely, however, that TERT promoter mutation status alone would be the sole genetic predictor of recurrence. In our series, only one third of primary aGCTs with recurrences harbored a TERT hotspot mutation, as did two primary aGCTs (20%) without subsequent recurrences after more than 8 years of follow-up.
In our series, TP53 mutations were identified in 16% of aGCT recurrences but were not present in primary aGCTs. Importantly, we observed that these TP53 mutations were acquired during disease progression in cases that did not harbor TERT promoter mutations. Alexiadis et al. also reported a TP53 mutation in a recurrent aGCT that also lacked TERT promoter mutations but they had a lower frequency of TP53 mutations overall in their series (4%, 1/22) [15]. Our findings suggest that TP53 may play a role in the progression of aGCTs, in particular in those lacking TERT promoter mutations.
Apart from TP53, we observed an enrichment of alterations in cell cycle genes in aGCT recurrences, including CDKN2A/B homozygous deletions. Given the small numbers assessed, further studies are warranted to define the frequency and role of cell cycle-related genes in the progression of aGCTs. Whilst homozygous deletions or lossof-function mutations of CDKN2A/B are frequent events in various human primary solid and hematopoietic neoplasias [46,47], these have not been previously reported in aGCTs. Gene expression analyses comparing wild-type FOXL2 and mutant FOXL2 C402G (p.C134W) transfected COV434 aGCT cells in vitro revealed the presence of differentially expressed genes associated with cell death and cell proliferation in mutant FOXL2 cells, such as CDKN1A, CDKN2A, and CDK6 [48]. Inactivating mutations in the chromatin remodeling gene KMT2D have been recently reported to be strongly associated with aGCT recurrences [16]. The frequency of KMT2D loss-of-function mutations was very low in the aGCT analyzed here (one truncating mutation identified in one primary non-recurrent aGCT35). Furthermore, even when taking all nonsynonymous KMT2D mutations into account, no difference in frequency between primary aGCTs with/without recurrences and aGCT recurrences could be identified.
Clonal composition analysis revealed that primary aGCTs and their matched recurrences, despite their generally simple genomes with few mutations and CNAs, display intra-lesion heterogeneity harboring clonal and subclonal mutations. In the progression from primary to metastatic disease, the acquisition of additional mutations, LOH of the wild-type allele of a given mutated gene as well as clonal shifts of genes affected by somatic mutations have been described [49][50][51]. Through the analysis of paired primary and matched recurrent aGCTs, we found that the acquisition of mutations, including those affecting TP53, TET2, MED12, and SH2D1A, was most commonly associated with disease progression.
Our study has important limitations. The sample size of the study is small, given the rarity of aGCTs, and larger validation studies are warranted. In addition, given the multi-institutional nature of our study, survival analyses could not be performed. Furthermore, our findings may not be applicable to the general population, given that the majority of the patients included in our cohort were treated at tertiary referral centers that tend to see higher risk populations. The recurrence rates observed here were similar to those reported in the literature, however [5,7]. Given the limited amounts of DNA available from these lesions, we restricted our sequencing analysis to 410-468 cancer-related genes. We cannot rule out, however, that there are other genes which may play a role in the progression of aGCTs. Nonetheless, our findings provide support to the notion that TERT promoter hotspot mutations are the most recurrent genetic events affecting aGCTs and might be associated with disease progression in a subset of cases. We further identified genetic alterations affecting cell cycle-related genes, which may be associated with aGCT progression. Finally, our data suggest that whilst aGCTs harbor simple genomes, intra-tumor heterogeneity is present in this rare subtype of pure sex cord tumor.