Identification of recurrent FHL2-GLI2 oncogenic fusion in sclerosing stromal tumors of the ovary

Sclerosing stromal tumor (SST) of the ovary is a rare type of sex cord-stromal tumor (SCST), whose genetic underpinning is currently unknown. Here, using whole-exome, targeted capture and RNA-sequencing, we report recurrent FHL2-GLI2 fusion genes in 65% (17/26) of SSTs and other GLI2 rearrangements in additional 15% (4/26) SSTs, none of which are detected in other types of SCSTs (n = 48) or common cancer types (n = 9,950). The FHL2-GLI2 fusions result in transcriptomic activation of the Sonic Hedgehog (SHH) pathway in SSTs. Expression of the FHL2-GLI2 fusion in vitro leads to the acquisition of phenotypic characteristics of SSTs, increased proliferation, migration and colony formation, and SHH pathway activation. Targeted inhibition of the SHH pathway results in reversal of these oncogenic properties, indicating its role in the pathogenesis of SSTs. Our results demonstrate that the FHL2-GLI2 fusion is likely the oncogenic driver of SSTs, defining a genotypic–phenotypic correlation in ovarian neoplasms.

S ex cord-stromal tumors are a heterogeneous group of benign and malignant neoplasms derived from the primitive sex cords or stromal components of the ovary 1 . Sclerosing stromal tumors (SSTs) represent an uncommon subtype of sex cord-stromal tumor (<5%) [1][2][3] , occurring in young adults in the second and third decades of life 4 . Patients typically present with abdominal or pelvic pain and menstrual irregularities in the setting of a unilateral pelvic mass. SSTs are usually hormonally inactive, however androgenic changes in the setting of pregnancy or virilization have been reported 4 . Although regarded as a benign neoplasm, its clinical presentation, imaging findings and elevated CA-125 levels may mimic those of malignant ovarian neoplasms, resulting in young women undergoing radical surgical resection 2,5,6 . In addition, the panoply of histologic features SSTs display may pose diagnostic challenges and result in misdiagnoses 2,7 . Immunohistochemistry may be helpful as these tumors are positive for sex cord markers including calretinin and FOXL2 [8][9][10] ; however, understanding the genetic underpinning of SSTs may aide in the diagnosis and management of these rare ovarian neoplasms.
The genomic landscape of common types of ovarian cancer has been thoroughly investigated 11 , and the genetic underpinnings of some rare types of ovarian cancer have been identified. For instance, SMARCA4 loss-of-function mutations have been documented in small cell carcinomas of the ovary hypercalcemic type 12 and FOXL2 p.C134W hotspot mutations have been described in >97% of adult-type granulosa cell tumors 1,9 , the most common sex cord-stromal tumor. These seminal studies indicate the vast potential for the discovery of unique genomic drivers in rare types of ovarian tumors 13 . In addition, DICER1 mutations have been detected in a subset of Sertoli-Leydig cell tumors and other non-epithelial ovarian cancers 14,15 . The genetic landscape of other sex cord-stromal tumors, including SSTs, however, is currently unknown.
We posited that if SSTs are driven by a pathognomonic genetic alteration, this information could be used for the development of ancillary markers to mitigate the diagnostic challenges posed by these rare tumors. In this study, we sought to define the repertoire of genetic alterations in SSTs, using a combination of wholeexome sequencing, targeted massively parallel sequencing and RNA-sequencing. Our analyses reveal the presence of a highly recurrent FHL2-GLI2 fusion transcript or GLI2 rearrangements in SSTs. Functional analyses in vitro establish that expression of the FHL2-GLI2 fusion increases signaling via the Sonic Hedgehog (SHH) pathway and results in the acquisition of oncogenic properties, which can be reversed through its chemical inhibition, thereby establishing a genotypic-phenotypic correlation and the importance of the SHH pathway in the biology of these tumors.

Results
Clinical and histologic features of SSTs. SSTs were retrieved from the authors' institutions, following approval by the institutional review boards (IRBs)/local ethics committees, and patient consents were obtained where appropriate. Following central pathology review, 26 tumors were classified as SSTs and included in this study (Supplementary Table 1, Supplementary Fig. 1). Patient median age at diagnosis was 29 (range  years, and all patients underwent surgical resection without any further adjuvant treatment (Supplementary Table 1). Histologically, SSTs were characterized by alternating areas of hypercellularity and hypocellularity imparting a vague lobulated architecture. An often prominent component of staghorn vessels, as well as varying numbers of spindle and luteinized stromal cells with overall bland cytologic features and overall low mitotic and proliferation rates were noted (Fig. 1a, Supplementary Table 1, Supplementary Fig. 2).
SSTs display few mutations and copy number alterations. To determine whether SSTs are underpinned by a pathognomonic somatic mutation, DNA samples from eight SSTs and matched normal tissues were subjected to whole-exome sequencing (WES; n = 3, median depth of coverage of tumor 108×, range 50×−226×, and normal samples 22×, range 20×−144×) or to targeted massively parallel sequencing due to the limited yields of DNA available (n = 5, median depth of coverage of tumor samples 698×, range 423×−855×) using the Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT 16 ) assay ( Supplementary Fig. 1, Supplementary Table 1). We reasoned that if SSTs were to harbor a somatic mutation at the same frequency of FOXL2 mutations in adult-type granulosa cell tumors 9 , eight samples would confer >90% of statistical power to detect a pathognomonic mutation.
SSTs displayed a low mutation burden, with a median of 16 (range 4-36) somatic mutations identified by WES, of which 9 (range 3-10) were non-synonymous. MSK-IMPACT sequencing detected a median of 2 (range 1-3) somatic mutations, with only 1 (range 0-3) being non-synonymous (Supplementary Table 2). None of the mutations identified were recurrent in the SSTs analyzed. In fact, the majority of mutations were non-pathogenic missense mutations not affecting hotspot residues, and three of the five SSTs subjected to MSK-IMPACT sequencing did not harbor any mutations affecting the 410 cancer-related genes tested ( Supplementary Fig. 3). One potentially pathogenic nonsynonymous somatic mutation was identified, a subclonal ATR (p.R1183*) nonsense mutation in SST2 ( Supplementary Fig. 3, Supplementary Table 2). Genome-wide copy number analysis revealed a paucity of copy number alterations, and the eight SSTs analyzed displayed diploid/near-diploid genomes. No homozygous deletions or amplifications were detected in any of the SSTs studied ( Supplementary Fig. 3).
Taken together, SSTs are characterized by a low mutation burden, low levels of genomic instability, and the lack of recurrent somatic mutations or copy number alterations.
SSTs of the ovary harbor a recurrent FHL2-GLI2 fusion gene. RNA-sequencing of eight SSTs revealed a recurrent in-frame FHL2-GLI2 fusion gene with a high driver probability as defined by OncoFuse 17 in 50% (4/8) of the cases analyzed (Supplementary  Table 3). This fusion gene resulted in a chimeric transcript composed of exons 1-5 of FHL2 (chr2:105984027) and exons 8-13 of GLI2 (chr2:121732551), likely through a paracentric inversion of chromosome 2q (Fig. 1b). FHL2 (Four And A Half LIM Domains 2) is a member of the four and a half LIM domainonly family, which is constitutively expressed in ovarian tissues 18,19 , whereas the oncogene GLI2 (GLI Family Zinc Finger 2) encodes a zinc finger transcription factor, which comprises a DNA binding domain, a GLI2 repression domain at the N-terminus and an activation domain at its C-terminus 20 (Fig. 1b). In this fusion, two of the four LIM protein domains of FHL2 were retained, while the repression domain of GLI2 was lost (Fig. 1b). The fusion of FHL2 to the 3′-end of GLI2 resulted in increased expression of exons 9-13 of GLI2, including the DNA binding and activation domains (Fig. 1c). Based on the domains involved in the chimeric transcript and the patterns of exonic transcription in SSTs harboring the FHL2-GLI2 chimeric gene (Fig. 1c), the most parsimonious explanation is that the 5′ aspects of FHL2 are juxtaposed with the DNA binding and activation domains of GLI2, coupled with a loss of the GLI2 repression domain 18,19,21 . As important mediators of the SHH pathway 22 , the GLI family proteins, GLI1-3, play an integral role in embryogenesis and are regarded as potent oncogenes 23,24 . Specifically, GLI2 functions as a transcription factor binding DNA through zinc finger motifs 22,24 , and GLI1 has been found to constitute an oncogenic partner in fusion genes described in gastroblastoma 25 , soft tissue sarcomas 26 , and fibromyxomas 27 .
One of the initial eight SSTs of the discovery cohort, SST27, harbored a FHL2-GLI2 fusion involving exons 7-13 of GLI2, rather than exons 8-13 of GLI2 found in the remaining 16 SSTs; this chimeric FHL2-GLI2 transcript in SST27 also leads to a juxtaposition of the 5′ elements of FHL2 with GLI2 lacking the repression domain whilst leaving the DNA binding and activation domains intact ( Supplementary Fig. 4a, Supplementary Table 3). In addition, SST12, which lacked the FHL2-GLI2 fusion as assessed by RNA-sequencing and RT-PCR, was found to harbor a DYNLL1-GLI2 fusion gene with a high driver probability as defined by OncoFuse 17 (Supplementary Table 3  O v a ri a n s e ro u s (n GLI2 rearrangement FHL2-GLI2 fusion chimeric transcript was composed of exons 1-2 of DYNLL1 and exons 8-13 of GLI2 ( Supplementary Fig. 4b).
FHL2-GLI2 fusion gene as a defining feature of ovarian SSTs.
To determine whether the FHL2-GLI2 fusion gene would constitute a pathognomonic genetic alteration for SSTs, we investigated the presence of the FHL2-GLI2 and the DYNLL1-GLI2 fusions in other sex cord-stromal tumors by RT-PCR and FISH, and across 33 common cancer types from The Cancer Genome Atlas (TCGA) retrieved from the Tumor Fusion Gene Data Portal 28 . Whilst the FHL2-GLI2 and DYNLL1-GLI2 rearrangements were present in 17/26 (65%) and 1/26 (4%) of SSTs, respectively, none of the other ovarian sex cord-stromal tumors tested by RT-PCR, namely adult-type granulosa cell tumors (n = 28), fibromas, fibrothecomas and thecomas (n = 14) and Sertoli-Leydig cell tumors (SLCTs, n = 6), and none of the 9950 tumors of 33 cancer types from TCGA 28 harbored the FHL2-GLI2 or DYNLL1-GLI2 fusion genes (Fig. 1h, i, Supplementary Fig. 4, Supplementary Table 4). In addition, no GLI2 break-apart signals/rearrangements were found using FISH in other sex cordstromal tumors (n = 9; Supplementary Fig. 5, Supplementary Table 4). Taken together, these findings provide evidence to suggest that FHL2-GLI2 and DYNLL1-GLI2 fusions might be pathognomonic for SSTs of the ovary.
Consistent with the RNA-sequencing and RT-PCR findings, RNA in situ hybridization (RNA-ISH) demonstrated mRNA expression of the specific forms of the chimeric FHL2-GLI2 fusion gene in the SSTs harboring each of the two FHL2-GLI2 fusion genes (Fig. 1f, Supplementary Fig. 6, Supplementary Table 4). In addition, immunohistochemistry with anti-GLI2 antibodies revealed moderate to high levels of GLI2 protein expression in SSTs with and without the FHL2-GLI2 fusion (Supplementary Figs. 2 and 4), respectively, supporting the notion that SSTs may be a disease driven by activation of GLI2.
FHL2-GLI2 expression results in oncogenic properties. Based on the known biological roles of FHL2 and GLI2, we hypothesized that cells expressing the FHL2-GLI2 fusion would exhibit oncogenic behavior through overexpression of the DNA binding and activation domains of GLI2 coupled with the loss of its repression domain. The FHL2-GLI2 fusion gene encompassing exons 8-13 of GLI2, recurrently detected in SSTs, as well as truncated (t) GLI2 lacking the repression domain were evaluated in multiple cell models, including immortalized mesenchymal stem cells (MSCs), HEK-293 cells, and in cells derived from cancers with known deregulation of the SHH pathway 22,29,30 , namely medulloblastoma (DAOY) and basal cell carcinoma (BCC) cells. MSC, HEK-293, DAOY, and BCC cells stably expressing tGLI2 or the FHL2-GLI2 fusion displayed a significant increase in proliferation and colony formation as compared to the same cells expressing empty vector, wild-type FHL2, or wild-type GLI2 (Fig. 2a, b). In addition, a scratch assay analysis revealed increased migration of MSCs, HEK-293, DAOY, and BCC cells upon stable expression of tGLI2 and FHL2-GLI2 as compared to cells expressing empty vector, wild-type FHL2 or wild-type GLI2 (Fig. 2c). Akin to the partners involved in oncogenic fusion genes, an increase in proliferation, colony formation and migration in cells transduced with wild-type GLI2 was observed, however, cells harboring tGLI2 or the FHL2-GLI2 fusion consistently significantly exhibited a more overt phenotype (Fig. 2a-c).
FHL2-GLI2 expression results in SST phenotypic features. We next sought to define whether the expression of the FHL2-GLI2 fusion would result in the acquisition of phenotypic features of SSTs in the cell models tested. Calretinin, a calcium-binding protein encoded by the CALB2 gene, is primarily expressed in specific subtypes of neuronal tissue, but has also been found to be present in non-neoplastic human ovaries and sex cord-stromal tumors 31 . Specifically, calretinin is expressed in theca interna, hilus and stromal cells 31 , and has proven a useful marker in the diagnosis of sex cord-stromal tumors 32 . Based on this understanding, we posited that cells expressing the fusion FHL2-GLI2 would adopt an SST-like phenotype and express calretinin. We observed significantly higher levels of CALB2 transcripts in MSC and HEK-293 cells stably expressing tGLI2 or the FHL2-GLI2 fusion as compared to empty vector controls or cells  expressing wild-type FHL2 or wild-type GLI2 (Fig. 3a). These findings were confirmed at the protein level; western blot and immunofluorescence analyses demonstrated increased levels of calretinin in MSC and HEK-293 cells expressing tGLI2 or the fusion gene FHL2-GLI2 than in controls (Fig. 3b, c). In addition, we tested FOXL2, a highly specific marker of ovarian sex cordstromal tumors 10 , and also found FOXL2 gene expression levels to be higher in cells transduced with tGLI2 and the FHL2-GLI2 fusion gene than in cells expressing empty vector controls, wildtype FHL2 or GLI2 alone (Fig. 3d), providing further evidence that the presence of the FHL2-GLI2 fusion gene results in the acquisition of some of the cardinal phenotypic characteristics of SSTs.
FHL2-GLI2 expression results in SHH pathway activation. GLI2 has been reported to play pivotal roles in the regulation of the SHH signaling pathway [22][23][24] . To define whether the FHL2-GLI2 fusion gene would result in activation of this pathway in SSTs, we performed a differential gene expression analysis between the eight human SSTs subjected to RNA-sequencing, common-type high-grade serous ovarian carcinomas derived from TCGA 11 (n = 16) and other sex cord-stromal tumors (n = 11). This analysis revealed an enrichment of SHH pathway genes, including GLI1, GLI2, HHIP, PTCH1, and PTCH2, in SSTs as compared to high-grade serous ovarian cancers and other sex cord-stromal tumors (Fig. 4a). To determine specifically whether SHH target genes, including PTCH1, PTCH2, HIF1A, HHIP, HHAT, GLI1/2/3, SMO, and SUFU, are co-expressed with the FHL2-GLI2 fusion gene in SSTs, we subjected SSTs (n = 11) and other ovarian sex cord-stromal tumors (n = 9; thecomas, n = 3; fibromas, n = 3; GCTs n = 2; and SLCT, n = 1) to NanoString gene expression analysis. The SHH GLI2 target genes tested were found to be co-expressed with the FHL2-GLI2 fusion gene. In addition, we observed that the SHH pathway genes were expressed also in SSTs with other GLI2 rearrangements, as well as in SSTs lacking GLI2 rearrangements based on the methods employed. Regardless of the presence and type of GLI2 rearrangements, SSTs were found to display significantly higher SHH pathway enrichment scores than other sex cord-stromal tumors (p < 0.001, Wilcoxon rank test; Fig. 4b, Supplementary Fig. 7). Given the evidence that the SHH pathway activation would potentially mediate the oncogenic properties of the FHL2-GLI2 fusion gene in human SSTs, we investigated the impact of the FHL2-GLI2 fusion expression on this pathway in our in vitro models. Consistent with the observations derived from the analysis of human SSTs, stable expression of tGLI2 and the FHL2-GLI2 fusion in MSC, HEK-293, DAOY, and BCC cells resulted in a significant increase in the expression levels of the SHH receptor PTCH1 and/or GLI1/CCND1 as compared to empty vector, wild-type FHL2 and wild-type GLI2 (Fig. 4c). For BCCs, given the very low endogenous levels of GLI1 gene/protein expression, CCND1 rather than GLI1 was assessed. Again, increased PTCH1 and GLI1/CCND1 gene expression was also observed upon wild-type GLI2 expression, however at levels significantly lower than in tGLI2 and FHL2-GLI2 cells, both of which harbor GLI2 lacking the repression domain (Fig. 4c). This finding was further supported by quantitative western blot analyses, which demonstrated increased levels of PTCH1 and GLI1 or CCND1 proteins upon tGLI2 and FHL2-GLI2 expression in all four cell models, although the increases in PTCH1 were not significant in the DAOY and BCC cells (Fig. 4d).
These findings in human SSTs and in vitro models provide evidence to suggest that the oncogenic behavior of SSTs can at least in part be explained by activation of the SHH pathway, and the potential of GLI2 rearrangements to serve as a marker for these uncommon ovarian neoplasms.
FHL2-GLI2 activates the SHH pathway via GLI2 target promoters. Given the role of the SHH pathway in the oncogenesis of SSTs, we sought to define the mechanistic basis of the activation of the SHH pathway in these tumors. GLI2, a key member of the SHH pathway, has a nuclear localization where it acts as a transcription factor 33 . Immunofluorescence utilizing an anti-FLAG antibody revealed that akin to wild-type GLI2, which displayed a nuclear localization in HEK-293 cells expressing the wild-type GLI2 vector, tGLI2 and the FHL2-GLI2 fusion proteins were expressed in the nuclear compartment (Fig. 5a). These findings demonstrate that deletion of the N-terminal domain of GLI2 does not abrogate its nuclear localization.
To test whether tGLI2 and the FHL2-GLI2 fusion proteins would activate GLI-responsive (RE) elements akin to wild-type GLI2, we performed a luciferase reporter assay with a specific GLI-RE containing GLI DNA binding sites in HEK-293, DAOY and BCC cells stably expressing empty vector, wild-type FHL2, wild-type GLI2, tGLI2, and FHL2-GLI2 (Fig. 5b, Supplementary Fig. 8). Increased activation of GLI-RE was seen in cells expressing wildtype GLI2, tGLI2, and the FHL2-GLI2 fusion but GLI-RE activation was most enhanced in tGLI2 or FHL2-GLI2 fusion gene expressing cells (Fig. 5b). Furthermore, we observed that mutation of the R338A-K339A amino acids in the Zinc Finger 5 of GLI2, which are highly conserved between all GLI family members and reported to be required for DNA binding 33 , resulted in inhibition of the capacity of the FHL2-GLI2 fusion protein to activate the SHH pathway (Fig. 5b). These data provide additional evidence that DNA binding of the FHL2-GLI2 fusion protein is necessary for the downstream activation of the SHH pathway.
We next investigated the ability of the FHL2-GLI2 fusion protein, which lacks the N-terminal binding domain, to bind to the Suppressor of fused homolog (SUFU) protein. SUFU is a known negative regulator of the SHH signaling pathway, which inhibits GLI2 through direct binding to specific regions located in the N-terminal and C-terminal regions 34 . Immunoprecipitation of SUFU revealed that both the FHL2-GLI2 fusion and tGLI2 proteins do not bind or only weakly bind to SUFU as compared to the wild-type GLI2 protein in HEK-293 cells (Fig. 5c). Luciferase experiments confirmed that SUFU overexpression is able to suppress SHH activity in HEK-293 cells expressing tGLI2 and the FHL2-GLI2 fusion protein (Fig. 5c). Inhibition of SHH activation in cells expressing tGLI2 or the FHL2-GLI2 fusion protein, which lack the N-terminal region of GLI2, likely stems from the binding of SUFU to the retained C-terminal domain of GLI2 34 . These findings were further supported by treatment of , and human basal cell carcinoma (BCC) cells stably expressing empty vector (control), wild-type FHL2 (FHL2), wild-type GLI2 (GLI2), truncated GLI2 (tGLI2), or the FHL2-GLI2 fusion. b Representative images of colony formation assay of MSCs, HEK-293, DAOY, and BCC cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2 (scale bars, 5 mm; top). Quantification of the number of colonies/ well compared to control (bottom). c Wound healing assay of MSCs, HEK-293, DAOY, and BCC cells stably expressing control, FHL2, GLI2, tGLI2 or FHL2-GLI2. The migratory effects/wound area was assessed at 0 and 24 h (Scale bar, 500 μm; top) and quantified (bottom). In a-c, data are representative of at least three independent experiments. Error bars, s.d. of mean; n.s., not significant; *P < 0.05, **P < 0.01, ***P < 0.001; two-tailed unpaired t-test.
HEK-293, DAOY, and BCC cells with GANT61, a small molecule inhibitor that specifically targets the GLI1/2 DNA binding site, which resulted in decreased proliferation in all cells, but predominantly in DAOY and BCC cells expressing tGLI2 or FHL2-GLI2 (Fig. 5d). HEK-293 cells were found to be markedly sensitive to GANT61 treatment and proliferation was decreased in all vectors and conditions tested.
Finally, we sought to define whether the FHL2-GLI2 fusion protein binds to GLI target genes. We performed chromatin immunoprecipitation (ChIP) with FLAG antibody, which demonstrated the recruitment of the FHL2-GLI2 fusion protein on two different regions of both the GLI1 and PTCH1 promoters (promoters 1 and 2) in MSC and HEK-293 cells expressing this fusion gene (Fig. 5e) 35,36 . These results support the notion that   Fig. 6a-c) and cyclopamine (Supplementary Fig. 9a-c), both highly specific inhibitors of SMO 37 , we observed a significant effect on proliferation and colony formation in cells expressing tGLI2 and the FHL2-GLI2 fusion protein (Fig. 6a, b, Supplementary Fig. 9a, b). We also observed a significant decrease in migration of MSC and/or HEK-293, DAOY, and BCC cells expressing tGLI2 and the FHL2-GLI2 fusion protein as compared to controls when treated with Vismodegib ( Fig. 6c) or cyclopamine ( Supplementary Fig. 9c). Taken together, these results implicate the SHH pathway in the phenotype observed in cells stably transduced with the fusion FHL2-GLI2 and provide evidence for the underlying biological basis and pathogenesis of SSTs of the ovary.

Discussion
Here, we demonstrate that SSTs of the ovary harbor recurrent GLI2 rearrangements, with FHL2 being the most frequent partner gene. We further show through an analysis of other ovarian sex cord-stromal tumors and 33 different cancer types that this FHL2-GLI2 fusion is likely pathognomonic for SSTs, providing a genotypic-phenotypic correlation in ovarian neoplasms.
GLI-transcription factors have been shown to be mediators of SHH signaling, and constitutive activation of the SHH/GLI signaling cascade has been implicated in numerous human malignancies, including skin, lung and prostate 22,38 . N-terminally truncated Gli2, unlike the full-length protein, has been reported to activate Shh target genes in transgenic mouse embryos 39 , suggesting that the modulation of the N-terminal repression domain is a mechanism of Shh signaling activation. The SHH pathway has been shown to play a prominent role in the communication between granulosa cells and developing theca cells 40,41 , and FHL2 is ubiquitously expressed in benign ovarian tissue and ovarian granulosa cells 19 . The FHL2-GLI2 fusion gene identified here likely has an oncogenic mechanism consistent with that of several other fusion genes, where the expression of the 3′ partner gene is driven by the regulatory elements of the ubiquitously expressed 5′ partner gene 42 . In addition, the exons of GLI2 included in the FHL2-GLI2 fusion gene lack the GLI2 repression domain but maintain its DNA binding domain. Hence, we propose a model where the FHL2-GLI2 fusion gene results in the overexpression of a truncated GLI2 (lacking its repression domain), given that its expression is under the FHL2 promoter, a protein ubiquitously expressed in ovarian stromal cells (i.e., the likeliest cell of origin of SSTs). Consistent with this model, the FHL2-GLI2 fusion and tGLI2 lacking the repression domain display nuclear localization (Fig. 5a) and result in activation of the SHH signaling pathway (Fig. 5b). In addition, the FHL2-GLI2 fusion protein was found to bind to the promoters of the GLI2 target genes PTCH1 and GLI1 (Fig. 5e), and was observed to require an intact GLI2 DNA binding site for the activation of the SHH pathway (Fig. 5b). Furthermore, the oncogenic effects of FHL2-GLI2 expression observed in cells could be reversed through pharmacological inhibition of the SHH pathway using three chemical inhibitors, further demonstrating the role of the GLI2-mediated SHH signaling pathway activation and dependency in cells expressing FHL2-GLI2.
We have demonstrated here that SSTs of the ovary are driven by GLI2 activation through the expression of GLI2-rearrangements. Whilst other rare types of ovarian neoplasms have been found to be underpinned by highly recurrent mutations, such as FOXL2 hotspot mutations in adult-type granulosa cell tumors 9 or SMARCA4 loss-of-function mutations in small cell carcinomas of the ovary hypercalcemic type 12 , SSTs were found to be characterized by recurrent and likely pathognomonic FHL2-GLI2 fusion genes. Other rare types of tumors with a relatively favorable clinical behavior have also been shown by our group and others to be driven by activating mutations and/or fusion genes. For example, adenoid cystic carcinomas of the breast are primarily driven by the MYB-NFIB fusion gene in~80% of cases 43,44 ; in breast adenoid cystic carcinomas lacking these fusions, MYBL1 rearrangements and MYB gene amplification have been reported as alternative mechanisms resulting in MYB overexpression and activation of MYB targets 45 . Likewise, >70% of polymorphous adenocarcinomas of the salivary glands are characterized by pathognomonic PRKD1 hotspot mutations; 46 those with wild-type PRKD1 have been shown to harbor PRKD1/ 2/3 rearrangements 47 . Consistent with these observations, in this study, we identified a DYNLL1-GLI2 fusion gene in an SST lacking the FHL2-GLI2 fusion gene. Further studies are warranted to define the fusion partners of the SSTs with GLI2 rearrangements but lacking FHL2-GLI2 and DYNLL1-GLI2 fusions, and to determine the mechanisms leading to GLI2 and SHH signaling pathway activation in SSTs lacking fusion genes involving GLI2.
This study has several limitations. Due to the retrospective multi-institutional nature of our study, follow-up information was unavailable, and the impact of the FHL2-GLI2 fusion on patient outcome could not be assessed. We did not identify a GLI2 rearrangement or alternative driver genetic alteration in 19% (5/26) of the SSTs studied. In addition, in three SSTs, FISH Fig. 3 FHL2-GLI2 expression results in the expression of sex cord markers in vitro. a Quantitative assessment of CALB2 transcripts in immortalized mesenchymal stem cells (MSCs) and HEK-293 cells stably expressing empty vector (control), wild-type FHL2 (FHL2), wild-type GLI2 (GLI2), truncated GLI2 (tGLI2), or the FHL2-GLI2 fusion. Expression levels were normalized to GAPDH expression, and comparisons of mRNA expression levels were performed relative to control. b Representative western blot analysis of calretinin protein levels in MSCs and HEK-293 cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2. Tubulin was used as protein loading control. Quantification (bottom) of protein levels as compared to control. c Representative confocal micrographs of immunofluorescence analysis of calretinin (green) and 4-6-diamidino-2-phenylindole (DAPI, blue) in MSCs and HEK-293 cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2 (scale bars, 50 μm). Quantification (bottom) of calretinin intensity/cell relative to control. d Quantitative assessment of FOXL2 transcripts in MSCs and HEK-293 cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2. Expression levels were normalized to GAPDH expression, and comparisons of mRNA expression levels were performed relative to control. In a-d, data are representative of at least three independent experiments. Error bars, s.d. of mean; n.s., not significant; *P < 0.05, **P < 0.01, ***P < 0.001; two-tailed unpaired t-test.    n.s.  Fig. 4 Expression of the FHL2-GLI2 fusion results in the activation of the Sonic Hedgehog pathway. a Differential gene expression analysis of human sclerosing stromal tumors (SSTs) subjected to RNA-sequencing (n = 8, this study) and high-grade serous ovarian carcinomas (n = 16; The Cancer Genome Atlas) and other sex cord-stromal tumors (SCSTs, n = 11; this study). Gene expression fold-change is color-coded according to the legend. Only genes significantly differentially expressed (P < 0.05; two-tailed unpaired t-test) are shown. CPM, count per million. b Expression levels of Sonic Hedgehog (SHH) pathway genes in human SSTs (n = 11) and other sex-cord stromal tumors (n = 9) as defined using NanoString. Expression levels and SHH enrichment scores are color-coded according to the legends. ***P < 0.001, Wilcoxon rank test. Hierarchical clustering was performed using complete linkage and Euclidian distance. c Quantitative assessment of the Sonic Hedgehog pathway PTCH1 and GLI1 transcripts in immortalized mesenchymal stem cells (MSCs), HEK-293 and medulloblastoma (DAOY) cells and of PTCH1 and CCND1 transcripts in human basal cell carcinoma (BCC) cells stably expressing empty vector (control), wild-type FHL2 (FHL2), wild-type GLI2 (GLI2), truncated GLI2 (tGLI2), or FHL2-GLI2. Expression levels were normalized to GAPDH expression, and comparisons of mRNA expression levels were performed relative to control. d Representative western blot analysis of PTCH1 and GLI1 protein expression in MSC, HEK-293 and DAOY cells and of PTCH1 and CCND1 in BCC cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2. Tubulin was used as protein loading control. Quantification (below) of protein levels as compared to control. In c-d, data are representative of at least three independent experiments. Error bars, s.d. of mean; n.s., not significant; *P < 0.05, **P < 0.01, ***P < 0.001; two-tailed unpaired t-test.
revealed the presence of GLI2 rearrangements, however the partner genes could not be identified. Of note, GLI2 protein expression and SHH pathway activation were also observed in SSTs lacking any of the GLI2 rearrangements identified here (Fig. 4, Supplementary Figs. 2 and 6), providing evidence to suggest that mechanisms other than the FHL2-GLI2 and DYNLL1-GLI2 fusion genes may lead to the constitutive activation of GLI2 in these lesions. Further studies, ideally utilizing frozen samples of SSTs with GLI2 rearrangements other than those described in this study or lacking GLI2 rearrangements altogether are warranted to define their driver alterations.   Despite these limitations, we report on the pathognomonic fusion gene FHL2-GLI2 in SSTs of the ovary, which results in the activation of the SHH pathway. Our findings provide evidence to suggest that GLI2 rearrangements are likely drivers of SSTs and may serve as potential ancillary markers for the diagnosis of these tumors, and that GLI2 should be included in panels surveying fusion genes in human cancers. Finally, our results on the SHH pathway activation in SSTs provide a potential therapeutic avenue for patients with SSTs, with FDA-approved SHH pathway inhibitors being available.

Methods
Subjects and samples. Following approval by the Institutional Review Boards (IRBs)/local ethics committees of the authors' institutions, unstained tissue sections from formalin-fixed paraffin-embedded (FFPE) SSTs were retrieved from Fudan University Cancer Center's consultation cases (Shanghai, China) and Massachusetts General Hospital (MGH, MA, USA). Patient consents were obtained following the respective IRB protocols approved by the authors' institutions, and samples were anonymized before analysis. All tumors were centrally reviewed by four pathologists (R.B., J.S.R.-F., R.A.S. and E.O.), and 26 tumors were classified as SSTs and included in this study.
Whole-exome and targeted massively parallel sequencing. Microdissected tumor and normal DNA from 3 SSTs was subjected to whole-exome sequencing (WES), and microdissected tumor DNA from 5 SSTs to MSK-IMPACT sequencing ( Supplementary Fig. 1, Supplementary Table 1), a massively parallel sequencing assay targeting all exons and selected introns of 410 cancer-related genes 16 . Sequencing was performed at Memorial Sloan Kettering Cancer Center's (MSKCC's) Integrated Genomics Operation (IGO) using validated protocols 50,51 . Sequencing data were analyzed and mutations identified using validated bioinformatics methods 50,51 . Mutations affecting hotspot codons were annotated according to Chang et al. 52 . Copy number alterations (CNAs) and loss of heterozygosity (LOH) were defined using FACETS 53 as described 50,51 . ABSOLUTE (v1.0.6) 54 was employed to determine the cancer cell fraction (CCF) of each mutation 50,51 .
RNA-sequencing and fusion gene identification. Eight SSTs were subjected to paired-end RNA-sequencing (Supplementary Fig. 1; Supplementary Table 1) using validated protocols at MSKCC's IGO 51,55 . Read pairs supporting fusion transcripts were identified using INTEGRATE 56 , deFuse 57 and FusionCatcher 58 , as described 51,55 . To filter out common alignment artifacts and normal transcriptional variation, we removed fusion gene and read-through candidates that were also found in a set of 297 normal samples from The Cancer Genome Atlas (TCGA) 59 . The remaining in-frame candidate fusion genes were annotated to predict their oncogenic/driver potential using OncoFuse (v1.0.9b2) 17 . The presence of identified fusion gene candidates was assessed in other cancer types using the Tumor Fusion Gene Data Portal 28 , which comprises 20,731 fusion genes across 33 cancer types (n = 9950; October 2019).
Differential gene expression analysis. For differential gene expression analysis, RNA-sequencing data of 8 SSTs (discovery cohort), of 11 other sex cord-stromal tumors (this study; adult-type granulosa cell tumors, n = 6; juvenile-type granulosa cell tumors, n = 2; and Sertoli-Leydig cell tumors, n = 3) and of high-grade serous ovarian carcinomas derived from TCGA 11 (n = 16; GDC Legacy Archive, https:// portal.gdc.cancer.gov/legacy-archive/) were employed. Transcriptomic analysis was performed in the R environment using edgeR (v3.24.3) 60 . Genes with less than 0.2 count per million (CPM) were labeled as poorly expressed across all libraries and removed from further analyses. Raw counts were normalized using an upper quartile of M-values approach 61 , and dispersion estimates over all genes were carried out with the quantile-adjusted conditional maximum likelihood method 62 . Resulting p-values were corrected for multiple testing using the Benjamini-Hochberg false discovery rate 63 . Furthermore, Reads Per Kilobase Million (RPKM) values were estimated from raw counts normalizing for sequencing depth and adjusting for length of genes.
FHL2-GLI2 fluorescence in situ hybridization. FHL2-GLI2 fusion gene assessment by FISH was performed using validated protocols at MSKCC's Molecular Cytogenetics Core 64 . Probes for detection of the FHL2-GLI2 fusion gene consisted of a three-color probe mix, composed of bacterial artificial chromosomes (BACs) mapping to 5′ GLI2 (orange), 3′ GLI2 (red), and 5′ FHL2 (green). To screen for the presence of GLI2 rearrangements in SSTs lacking FHL2-GLI2/DYNLL1-GLI2 fusions by RNA-sequencing and/or RT-PCR, probes for the detection of the GLI2 gene consisted of a dual-color probe mix, composed of the same BACs as above mapping to 5′ GLI2 (yellow) and 3′ GLI2 (green). At least 10 images per tumor region were captured and at least 50 non-overlapping interphase nuclei with welldelineated contours were analyzed for the presence of the FHL2-GLI2 fusion or GLI2 rearrangements. SSTs were considered positive for the FHL2-GLI2 fusion gene or for GLI2 rearrangements if >15% of the cells displayed at least one 5'FHL2-3'GLI2 fusion or 5'GLI2 or 3'GLI2 signal.
NanoString gene expression analysis. A NanoString nCounter gene expression assay using a custom CodeSet panel of SSH pathway genes was performed according to the manufacturer's instructions (NanoString Technologies). NanoString nCounter data were analyzed using the R Bioconductor package NanoStringDiff 65 , an approach which facilitates analysis of nCounter data using a negative binomial generalized linear model (GLM) with differential expression analysis performed using model-based linear contrasts, and associated inference based on likelihood ratio tests. The negative binomial GLM incorporates normalization factors for positive controls, background noise and the 5 housekeeping genes. Single sample Gene Set Enrichment (ssGSEA) analysis 66 was used to define the Sonic Hedgehog (SHH) enrichment score for each sample. Hierarchical cluster analysis of gene expression data was performed using complete linkage and Euclidian distance.
Reverse transcription PCR (RT-PCR) for fusion validation. Total RNA was reverse-transcribed using SuperScript VILO Master Mix (Life Technologies; Thermo Fisher Scientific), according to the manufacturer's instructions. PCR amplification of 10 ng of cDNA was performed using specific primer sets designed based for each fusion gene and each breakpoint (Supplementary Table 3) 51 . PCR fragments were purified (ExoSAP-IT, Affymetrix) and sequenced on an ABI 3730 capillary sequencer using the ABI BigDye Terminator chemistry (v3.1, Life Technologies) according to the manufacturer's protocol. Sequences of the forward   70 . Quantification and analysis were performed using ImageJ. Experiments were performed in triplicate.
Proliferation assay. Cells were seeded in 96-well plates (1000 cells/well; n = 6 for each cell line and condition). Proliferation rate was assessed using the Cell Titer-Blue Cell Viability Assay (Promega) 46,50 . Absorbance was detected with 560 nm excitation and 590 nm emission using a Victor X4 Multimode Plate Reader (PerkinElmer). Experiments were performed in triplicate.
Scratch wound healing assay. Cells were seeded in 24-well plates at 90-95% confluence. On the following day, a scratch was made on the cell monolayer using a 1 ml pipette tip across the center of the well. Phase-contrast images were obtained at 0 and 24 h following scratch wounding, using an EVOS XL Core Microscope (ThermoFisher Scientific), the wound area was measured using ImageJ, and the data analyzed to determine percent of wound closure, as described 50 . Experiments for each condition were performed in triplicate in at least 3 independent experiments. Fig. 6 Inhibition of the Sonic Hedgehog pathway by Vismodegib reverses oncogenic properties. a Cell titer blue proliferation assay of immortalized mesenchymal stem cells (MSCs), HEK-293, medulloblastoma (DAOY) and human basal cell carcinoma (BCC) cells stably expressing empty vector (control), wild-type FHL2 (FHL2), wild-type GLI2 (GLI2), truncated GLI2 (tGLI2) or FHL2-GLI2 treated with 250 nM Vismodegib or vehicle control (DMSO). b Representative images of colony formation assay of MSC, HEK-293, DAOY, and BCC cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2 treated with Vismodegib 500 nM or vehicle control (DMSO). Scale bars, 5 mm. Quantification of the number of colonies/well compared to control (bottom). c Wound healing assay of MSC, HEK-293, DAOY, and BCC cells stably expressing control, FHL2, GLI2, tGLI2, or FHL2-GLI2 treated with 250 nM Vismodegib or vehicle control (DMSO). The migratory effect/wound area was assessed at 0 and 24 h and quantified compared to DMSO (bottom). Vismo, Vismodegib. Scale bars, 500 μm. In a-c, data are representative of at least three independent experiments. Error bars, s.d. of mean; n.s., not significant; *P < 0.05, **P < 0.01, ***P < 0.001; two-tailed unpaired t-test.
Colony formation assay. Cells were seeded in 6-well plates (500-100 cells/well). After 10 days, cells were fixed and stained using the sulforhodamine B (SRB) protocol, as described 46,50 . In brief, cells were fixed at 4°C for 1 h using Trichloroacetic acid (TCA) (Sigma), washed then stained on a rocking platform for 30 min with 0.04% SRB in 1% acetic acid. Plates were imaged using Zeiss Observer Z1, ×5/0.5NA objective and Hamamatsu Flash V3 cMOS camera. Motorized stage was calibrated to tile-image the entirety of each well. Stitched images were analyzed in ImageJ/FIJI using customized macro script. Gaussian and median filters were used to blur the image before colonies were detected by intensity threshold. Touching colonies were separated using Watershed algorithm. The number of colonies and the area of each were analyzed. Quantification of the number of colonies per well and colony size was performed using ImageJ. Experiments were performed in triplicate.
Immunofluorescence. MSCs HEK-293 cells were grown on glass coverslips and fixed with 4% formaldehyde (ThermoFisher Scientific), blocked in 10% normal Goat Serum (Vector Laboratories, S-1000) then incubated with primary antibody against Calretinin (Santa Cruz, sc-565956, 1:100) followed by Alexa Fluorconjugated secondary antibody Alexa 647 goat anti-mouse (Life Technologies, A21236, 1:500). For GLI2 nuclear localization studies, HEK-293 cells in slide chambers were fixed with ice cold solution 1:1 acetone:methanol for 15 min at −20°C. After washing 3 times with PBS the samples were blocked with blocking solution (PBS; 5% goat serum). The samples were incubated overnight at 4°C with anti-FLAG M2 antibody (Sigma, F-1804, 1:200 in PBS/1% BSA), followed by Alexa 647 tagged secondary goat anti-mouse antibody (Life Technologies, A21236, 1:500). Slides were mounted using Gold Prolong antifade reagent containing DAPI (ThermoFisher Scientific, P36930). Fluorescent images were acquired at MSKCC's Molecular Cytology Core using a TCS SP5 upright Confocal microscope (Leica Microsystems), with a ×63/1.40 oil objective, HyD hybrid detectors, and the Leica Application Suite Advanced Fluorescence (LASAF) acquisition software (Leica Microsystems). DAPI and Alexa647 were excited with 405 nm and 650 nm lasers, respectively, and imaged in the range of 410-460 nm and 660-700 nm, respectively. Linear LUT was used at full range. Images were acquired as LIF files (Leica image file format; Leica Microsystems) at a resolution of 1024 × 1024 pixels, a scaling of 0.12 micron/pixel, and an 8-bit depth and processed using Fiji and ImageJ. No post-acquisition processing was performed, besides minor adjustments of brightness and contrast, applied equally to all images. ImageJ software was used to quantify the number of pixels per cell analyzing at least five representative images (×40 field) for each condition. Calretinin images were analyzed using ImageJ/FIJI with a custom macro. The number of cells within each image were counted by segmenting the DAPI channels. After setting an appropriate threshold, total intensity of calretinin was measured for the image then normalized to the number of cells. Experiments for each condition were performed in triplicate in at least 3 independent experiments.
Statistical analysis. Statistical analysis was performed using Prism 7 (GraphPad). Student's two-tailed t-test was employed for the comparison of means in parametric data. The heteroscedasticity was assessed for each comparison, and homoscedastic or heteroscedastic t-tests were employed as appropriate, as described 50 . Wilcoxon rank test was used for the comparison of the SHH enrichment scores. A P value of <0.05 was considered significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.