Glycogen synthase kinase-3 beta (GSK3β)-mediated phosphorylation of ETS1 promotes progression of ovarian carcinoma

ETS1 − an evolutionarily conserved transcription factor involved in the regulation of a number of cellular processes − is overexpressed in several malignancies, including ovarian cancer. Most studies on ETS1 expression have been focused on the transcriptional and RNA levels, with post-translational control mechanisms remaining relatively unexplored in the pathogenesis of malignancies. Here, we show that ETS1 forms a complex with glycogen synthase kinase-3β (GSK3β). Specifically, GSK3β-mediated phosphorylation of ETS1 at threonine 265 and serine 269 promoted protein stability, induced the transcriptional activation of matrix metalloproteinase (MMP)-9, and increased cell migration. In vivo experiments revealed that a GSK3β inhibitor was able to suppress both endogenous ETS1 expression and induction of MMP-9 expression. Upon generation of a specific antibody against phosphorylated ETS1, we demonstrated that phospho-ETS1 immunohistochemical expression in ovarian cancer specimens was correlated with that of MMP-9. Notably, the cumulative overall survival of patients with low phospho-ETS1 histoscores was significantly longer than that of those showing higher scores. We conclude that the GSK3β/ETS1/MMP-9 axis may regulate the biological aggressiveness of ovarian cancer and can serve as a prognostic factor in patients with this malignancy.


INTRODUCTION
Ovarian cancer is a deadly gynecologic malignancy frequently diagnosed at advanced stages. Primary debulking surgery followed by platinum-based adjuvant chemotherapy represents the mainstay of therapy. Although an initial response to primary treatment can be observed even in advanced-phase disease, recurrences and resistance to subsequent chemotherapy remain common [1]. In this scenario, a better understanding of the molecular mechanisms involved in ovarian tumorigenesis may pave the way to novel targeted treatment strategies for recurrent/advanced disease. AGING ETS1 belongs to the evolutionarily conserved ETS family of transcription factors [2] involved in the regulation of a number of cellular processes [3]. ETS1 has been found to be overexpressed in several solid tumors, including breast [4], colorectal [5], endometrial [6], lung [7], and ovarian [8] cancers. Moreover, an increased ETS1 expression has been associated with high tumor grading, poor differentiation, and more pronounced invasiveness [3,9] -ultimately being considered a marker of poor prognosis [10,11].
Several post-translational modifications -including phosphorylation -can influence ETS1 biological activity. Although phosphorylation by SRC kinase and protein kinase Cα (PKCα) can increase ETS1 stability [12,13], the opposite effect is observed when phosphorylation is catalyzed by calcium-dependent protein kinase II (CaMKII) [14,15]. Similarly, the transcriptional activity of ETS1 is activated by ERK1/2 phosphorylation [16] and blocked by CaMKII [14,15]. When stabilized and activated by kinases, ETS1 promotes cancer cell invasiveness by inducing the expression of matrix metalloproteinases (MMPs) [8,17]. Accordingly, a previous study conducted in breast cancer has shown that overexpression of ETS1 in the presence of epidermal growth factor (EGF) enhances MMP-9 promoter activity [18]. A strict association between ETS1 and MMP expression levels has been consistently demonstrated in ovarian and lung malignancies [8,19].
MMPs are zinc metalloenzymes that digest extracellular matrix proteins -ultimately playing a key role in cancer growth, invasion, and metastasis [20]. Specifically, upregulation of MMP-9 has been associated with the progression of several malignancies [21] -with its expression levels being positively regulated by numerous transcription factors, including NF-κB [22], AP1 [23], and ETS1 [24]. ETS1 is a helix-loop-helix protein capable of binding to the GGAA/T core motif in the MMP-9 promoter region [25]. Interestingly, the coexpression of ETS1 and MMP-9 portends a poor prognosis in solid malignancies [17,26].
Glycogen synthase kinase-3β (GSK3β) is a serine/threonine protein kinase that regulates several cellular processes -including proliferation, differentiation, and migration [27]. GSK3β substrateswhich are characterized by the presence of a common specific phosphoryl protein motif-(S/T)XXX(S/T)play a fundamental regulatory role in cell migration and invasiveness [28]. As a consequence, there is a growing interest in GSK3β as a therapeutic target in cancer [29].
In this study, we provide evidence that GSK3β phosphorylates ETS1 at two specific sites (threonine 265 and serine 269) that are part of the GSK3β phosphorylation motif. Site-directed mutagenesis of these residues reduced protein stability, MMP-9 transcriptional activation, and cell migration. The effect of a GSK3β inhibitor on endogenous ETS1 expression and induction of MMP-9 was investigated using in vivo experiments. Finally, we generated a specific antibody against phosphorylated ETS1 and examined its expression levels in human ovarian cancer specimens. Taken together, our data may shed more light into the role of the GSK3β/ETS1/MMP-9 axis in ovarian tumorigenesispotentially paving the way for novel therapeutics.

ETS1 interacts with GSK3β
The potential interaction between ETS1 and GSK3β was investigated by means of immunoprecipitation. After pulling-down of endogenous ETS1 with a specific antibody, endogenous GSK3β was identified. Notably, we also observed an endogenous ETS1-GSK3β complex in ovarian cancer cell line ( Figure 1A). Co-localization of the two proteins within cells was confirmed by confocal microscopy ( Figure 1B). Serial truncated GSK3β and ETS1 constructs were used to map the interaction domains of ETS1 and GSK3β. The kinase domain of GSK3β (56-353/GSK3β) was found to be critical for the ETS1 interaction ( Figure 1C). Interestingly, deletion of the central ETS1 region (1-197/ETS1 and 363−486/ETS1) reduced its interaction with GSK3β ( Figure 1D). Altogether, these data indicate that 1) the kinase domain of GSK3β is responsible of its interaction with ETS1, 2) the central region of ETS1 (197−363) is necessary for its interaction with GSK3β.
To identify the amino acid residues phosphorylated by GSK3β, the wild-type ETS1 products of the in vitro kinase assay were digested with trypsin and subjected to phosphosphopeptide enrichment for LC-MS/MS. The AGING  mono-and double-phosphorylation events of the ETS1 tryptic peptide (LTQSWSSQSSFNSLQR) were determined by LC-MS/MS. The extracted ion chromatography spectra are shown in Figure 2C. There were no differences in the monophsophorylation site forms of ETS1 (upper panel, Figure 2C). However, there were two distinct double-phosphorylated forms of the ETS1 tryptic peptide -which were denoted by black (peak1) and white (peak2) arrows in Figure 2C (lower panel). When the relative intensities of the two doublephosphorylated peptide forms were analyzed with and without GSK3β treatment, the signal of peak2 doublephosphorylated peptide was found to be markedly increased upon GSK3β exposure. Two different doublephosphorylated isoforms can be targeted by GSK3β ( Figure 2B). Therefore, two synthetic doublephosphorylated peptides (pT2pS6, also known as T265/S269 and pS6pS10, also known as S272/S276) were generated to identify the exact phospho-amino acid residues in the GSK3β-enhanced ETS1 phosphopeptide (peak2). As shown in Supplementary Figure 1A, 1B, the fragmentation profile (MS/MS spectra) of the peak2 precursor ion (m/z 1008.41, z = 2, upper panel) showed non-phosphorylation signals in the y7 and y8 fragments (m/z 851.44 and 938.47, respectively). Notably, the corresponding masses were identical (y7, m/z 851.43; y8, 938.47) to those observed in the pT2pS6 spectrum profile (middle panel) but were different from those of the pS6pS10 peptide fragmentation spectra (lower panel). These results clearly indicate that phosphorylation on the first threonine and the second serine residues of the ETS1 peptide occurs only upon GSK3β treatment. To further confirm the presence and the role of this doublephosphorylated form, the pT2pS6 synthetic peptide was used as an immunogen to generate a polyclonal antibody capable of detecting this specific phospho-ETS1 isoform in cell lines and tissue specimens.

GSK3β phosphorylates ETS1 to maintain protein stability
To investigate the impact of GSK3β-mediated ETS1 phosphorylation, we initially determined the protein turnover rate in 293 cells overexpressing wild-type ETS1 and ETS1-3A in presence of cycloheximide (in order to inhibit protein synthesis). Degradation of ETS1-3A occurred faster than that of wild-type ETS1 ( Figure  3A). Furthermore, GSK3β inhibitor (LY2090314) did not alter the stability of ETS1 ( Figure  3B). Because conjugation of ubiquitin with proteins plays a key role in protein breakdown, ETS1 and ETS1-3A were pulled-down and probed with an anti-ubiquitin antibody. Ubiquitin levels were then assayed in the presence of the proteasome inhibitor-MG132. The results revealed that levels of ubiquitin-labeled ETS1-3A were higher than those of wild-type ETS1 ( Figure  3C). Moreover, the GSK3β inhibitor LY2090314which has already undergone clinical testing [30]induced ETS1 degradation in ovarian and endometrial cancer cells ( Figure 3D), an effect blocked by MG132 in the presence of GSK3β siRNA ( Figure 3E). Taken together, these data suggest that ETS1 phosphorylation by GSK3β regulates its stability.

GSK3β phosphorylates ETS1 to regulate MMP-9 expression
Because MMP-9 is one of the ETS1 downstream target genes [18], we investigated the extent to which ETS1 and ETS1-3A promoted MMP-9 transcriptional activity. First, we analyzed the MMP-9 promoter to identify ETS1 responsive regions. Overexpression of ETS1 was able to induce both MMP-9 protein and activity in ARK2 cells (Supplementary Figure 2A). After transfection with ETS1 and different truncated MMP-9 promoter constructs, ETS1 was found to induce a 10fold increase in promoter activity when the -580 promoter construct was used (Supplementary Figure  2B). All of the subsequent experiments were therefore performed with this construct. To confirm the role of GSK3β in promoting MMP-9 expression, we examined its induction in cancer cells co-transfected with the ETS1 expression vector and GSK3β siRNA. Silencing of endogenous GSK3β was found to block MMP-9 promoter activity ( Figure 4A), as well as its mRNA ( Figure 4B) and protein ( Figure 4C) expression. The induction levels of MMP9 in overexpressed ETS1-3A cell lines showed no difference in the absence or presence of GSK3β siRNA ( Figure 4C). Because we have previously shown that LY2090314 can inhibit murine ovarian cancer-MOSEC cells growth [31], we also examined the endogenous ETS1 and MMP-9 protein levels in the same experimental tumors of LY2090314-treated mice. LY2090314 was found to inhibit both ETS1 and MMP-9 protein expression ( Figure 4D). Furthermore, wild-type ETS1, but not ETS1-3A, was able to induce MMP-9 promoter activity in ARK2, MDAH2774, and SKOV3 cells ( Figure 5A). p300 is a histone acetyltransferase which is able to bind ETS1 to serve as a transcription coactivator [32,33]. To shed more light on the potential role of p300 in ETS1mediated MMP-9 regulation, the interaction of p300 with ETS1 and ETS1-3A was examined. ETS1-3A was found to interact with p300 at a lesser extent than ETS1 ( Figure 5B). In addition, ETS1-3A had a lower capacity to bind the MMP-9 promoter (MMP9 promoter -166 region in Supplementary Figure 2B) compared with ETS1. Results of chromatin immunoprecipitation experiments revealed that a lower amount of p300 was AGING  recruited by ETS-3A on the MMP-9 promoter region ( Figure 5C). MMP-9 induction was less prominent in ETS1-3A-overexpressing cancer as compared with that observed when ETS1 was overexpressed ( Figure 5D).
Taken together, these results indicate that ETS1 phosphorylation catalyzed by GSK3β can promote its interaction with p300 and induce the expression of MMP-9 as a downstream target gene.

GSK3β mediates ETS1 phosphorylation and regulates cell migration
Stable expression of Flag-ETS1 was found to induce MMP-9 protein levels, but not in Flag-ETS1-3A in 293 cells. (Figure 5E). Cell migration ability was tested using the transwell migration assay and the wound healing assay. Flag-ETS1-expressing cells were characterized by a higher cell migration ability compared with Flag-ETS1-expressing 3A cells ( Figure  5F). Results of the wound closure assay revealed that the cell-free gap area for Flag-ETS1 was lower than that observed for ETS1-3A ( Figure 5G). These findings indicate that the GSK3β/ETS1/MMP-9 axis may regulate tumor cell migration.

Immunohistochemical expression of phospho-ETS1
Thr265, Ser269, and MMP-9 in human ovarian cancer Specific polyclonal antibodies against phospho-ETS1 Thr265, Ser269 were generated to investigate its immunohistochemical expression in ovarian cancer tissues. Using cell lysates obtained from the breast cancer cell line MDA-MB-231 as a positive control, the antibody was found to specifically detect phospho-ETS1 Thr265, Ser269 both in the absence and presence of the control peptide (i.e., without phosphorylated residues). However, the signal was no longer evident when the antibody was pre-incubated with the phosphorylated peptide (Supplementary Figure 3A), and downregulated in the purified ETS1 protein from GSK3β inhibitor treated cells (Supplementary Figure  3B). Similar findings were obtained when ovarian cancer tissue was used (Supplementary Figure 3C). These results confirmed the specificity of the antiphospho-ETS1 Thr265, Ser269 antibody. We subsequently analyzed the expression of phospho-ETS1 in nine different ovarian cancer cell lines. Although all of them were found to express total ETS1, the phospho-ETS1 signal was evident in four cell lines only (TOV112D, OV90, MDAH2774, and SKOV3 cells) in which MMP-9 was expressed as well (Supplementary Figure 4A). Knockdown of endogenous GSK3β in ovarian cancer cells inhibited both phospho-ETS1 and MMP-9 expression (Supplementary Figure 4B).  Figure 4C).
We also analyzed phospho-ETS1 and MMP-9 in ovarian cancer tissues. The results revealed that MMP-9 expression was higher in tissue areas characterized by high phospho-ETS1 expression ( Figure 6A, upper panel) as compared with areas showing lower phospho-ETS1 expression ( Figure 6A, lower panel). Moreover, advanced serous and endometrioid ovarian malignancies were characterized by a high phospho-ETS1 expression (reflecting the results obtained in SKOV3 and MDAH2774 cells, respectively ( Figure  6B). Kaplan-Meier analysis revealed that the cumulative overall survival ( Figure 6C) and recurrencefree survival ( Figure 6D) of patients with a phospho-ETS1 histoscore ≤ 20 were significantly higher than those cases with scores >20 in endometrioid type ovarian cancer -suggesting that it may serve as a prognostic biomarker in this malignancy.

DISCUSSION
The results of our study indicate that GSK3β phosphorylates ETS1 at T265/S269 and maintains its protein stability. Upon phosphorylation by GSK3β, ETS1 binds the histone acetyltransferase p300 and activates MMP-9 transcription by acting on its promoter ( Figure 7). Notably, the use of LY2090314 -a specific GSK3β inhibitor -was found to inhibit both ETS1 and MMP-9 expression in vivo. Altogether, our data suggest additional roles of GSK3β/ETS1/MMP-9 pathway in ovarian malignancies.
GSK3β is multifunctional kinase that has been implicated in various aspects of tumorigenesis. As a tumor suppressor, GSK3β can phosphorylate and inactivate several oncogenes, including β-catenin, c-Myc, and cyclin D1 -ultimately inhibiting tumor cell proliferation [28]. Previous studies have also shown that GSK3β inhibition can interfere with certain oncogenic pathways -including notch and hedgehog (Hh) [34,35]. There is also evidence to suggest that the lysine-specific histone demethylase 1A (KDM1A/LSD1) oncoprotein is stabilized by GSK3βmediated phosphorylation [31,36]. In addition, clinical studies have shown that GSK3β overexpression portends a poor prognosis in patients with ovarian and lung malignancies [37,38]. These findings demonstrate that GSK3β may exert AGING in ovarian cancer tissues. Tissue with high and low expressed phospho-ETS1 are shown in the upper and lower panels, respectively. (B) Advanced-stage malignancies (stages 3−4) showed a higher immunohistochemical expression of phospho-ETS1 as compared with earlystages tumors (stages 1−2) both in serous (black square) and endometrioid (black triangle) ovarian carcinomas. Asterisks in B denote statistical significance (*p < 0.05). Kaplan-Meier plots of overall (C) and recurrence-free survival (D) in endometrioid type ovarian cancer patients (n = 29) were constructed according to phospho-ETS1 histoscores. The optimal cutoff for phospho-ETS1 histoscore in the prediction of overall survival and recurrence-free survival was 20. GSK3β is also involved in epithelial-mesenchymal transition (EMT) through the transcriptional factor Snail to repress E-cadherin [39]. In addition, it activates MMP transcription [40] via phosphorylation-mediated Snail degradation -ultimately preventing EMT [41]. Although this observation may suggest an inverse relationship between GSK3β and MMP protein expression, our immunohistochemical data indicate that both GSK3β and MMP-9 are overexpressed in advanced-stage ovarian cancer [42]. It is conceivable that -in addition to the GSK3β/Snail pathway -the GSK3β/ETS1/MMP-9 axis may be involved in ovarian cancer spread.
In this study, we have also shown that GSK3β-mediated phosphorylation of promoted ETS1 protein stability as well as its transcriptional activity and protein-binding capacity. Compared with wild-type ETS1, ETS1-3A was unable to induce MMP-9 promoter activityultimately preventing MMP-9 protein expression ( Figure 5A, 5C). Similar findings were obtained when cancer cells were co-transfected with wild-type ETS1 and the shGSK3β vector ( Figure 5). Overexpression of ETS1 was capable of inducing MMP-9 expression, but this effect was no longer evident when GSK3β was knocked-down. These data indicate that GSK3β regulates ETS1 transcriptional activity and is capable of blocking the expression of ETS1 downstream genes. We also demonstrated that ETS1-3A displayed a reduced capacity to bind both the co-transcriptional activator p300 and the MMP-9 gene promoter ( Figure  5B, 5C).
The ETS1 protein is characterized by the presence of different domains − including a pointed domain, a transactivation domain, and an auto-inhibitory module (which is distinct from the ETS DNA-binding domain) [3]. Importantly, the capacity of ETS1 to bind DNA is regulated by the auto-inhibitory module. Four α-helices of this module are capable of interacting with the ETS DNA-binding domain [3]. Notably, the phosphorylation status of the serine-rich region (SRR, serine residues 251, 257, 282, 285) located in the proximity of the autoinhibitory module inhibits DNA binding. Phosphorylation of the SRR catalyzed by CaMKII can decrease both DNA binding capacity and ETS1 transcriptional activity [15]. Our results indicate that GSK3β is able to phosphorylate the threonine 265 and serine 269 residues located in the SRR region. Sitedirected mutagenesis of these amino acids or GSK3β silencing were found to abrogate ETS1 transcriptional ability, its interaction with p300, as well as its DNA binding capacity (Figure 4, Figure 5A-5C). We therefore speculate that GSK3β and CaMKII can both control the ability of ETS1 to bind DNA, albeit in an opposite manner (i.e., GSK3β promotes and CAMKII abrogates ETS1 binding to DNA). To our knowledge, this is the first comprehensive description of how two different kinases can regulate ETS1 biological activity.
We conclude that the GSK3β/ETS1/MMP-9 axis regulates the biological aggressiveness of ovarian cancer. Specifically, we demonstrated that 1) GSK3βmediated phosphorylation of ETS1 regulates its protein stability and transcriptional activity and 2) the expression levels of phospho-ETS1 are associated with ovarian cancer stage and can be used as a biomarker to predict overall survival in this malignancy.

Transfection of small interfering (si) RNA
SKOV3, MDAH2774 and 293 cells (5 × 10 5 cells in a 6-well plate) were transfected with double-stranded RNA (50 nM) using the Lipofectamine RNAimax reagent (Invitrogen). GSK3β siRNA was from Santa Cruz (Santa Cruz Biotechnology). After 72 h of transfection, western blot was used to confirm silencing of the target genes.

RNA extraction and real-time Q-PCR
Total RNA was isolated with the TRIzol reagent (Invitrogen). First-strand cDNA for RT-QPCR was synthesized with an oligo-T primer using the Superscript III first strand synthesis kit (Invitrogen). The SYBR green gene expression assay was used to investigate expression levels of MMP-9, ETS1; and GSK3β. The housekeeping gene GAPDH was used as an internal control. The primers used in the SYBR green gene expression assay are shown in Supplementary  Table 1.

Western blot analysis
The protocol used for western blot has been previously described in detail [47]. In brief, Cell lysates were prepared with RIPA buffer (150 mM NaCl, 20 mM Tris-Cl pH7.5, 1% Triton X-100, 1% NP40, 0.1% SDS, 0.5% deoxycholate) containing freshly added proteinase and phosphatase inhibitors (Bionovas, Toronto, Canada). Protein concentrations were measured with the Bradford method. Each sample (100 µg) was subjected to electrophoresis in 10% SDS-polyacrylamide gels and subsequently transferred onto nitrocellulose membranes. All experiments were conducted using commercially available antibodies and corresponding horseradish peroxidase-conjugated antibodies (Santa Cruz Biotechnology). The specific rabbit polyclonal antibody against phospho-ETS1-Thr265, Ser269 was produced by ABclonal (Woburn, MA, USA) using the following immunogen: CDRL(Tp)QSW(Sp)SQSS (phospho-ETS1-Thr265, Ser269). After immunization for five times in two months, the antibodies were purified by antigen affinity chromatography. Chemiluminescence reagents were obtained from Millipore. The signal intensity of the autoradiograms was quantified using the ImageJ software (https://imagej.nih.gov/ij/) after normalization for the corresponding actin intensity. The antibodies used for western blot are shown in Supplementary Table 2.

Protein purification and in vitro kinase assay
Halo-tag ETS1 was purified with Halo Tag Protein purification system (Promega). In briefly, Halo-ETS1 transfected 293 cell pellets were lysed in Halo Tag purification buffer (1x PBS, 1mM DTT, 0.005% IGEPAL-CA630 (Sigma) and protease inhibitor). After freeze-thawing three times, the supernatants were added into HaloLink resin (Promega) according manufacturer instructions. Halo-ETS1 was released from the resin by using HaloTEV enzyme to cleavage TEV cutting site between Halo tag and ETS1 protein. The purified ETS1 protein was incubated with GSK3β (New England BioLabs) in GSK3 reaction buffer (20mM Tris-HCl, 10mM MgCl2, 2mM DTT and 200µM ATP) at 3° C for 30 mins. Finally, the phosphorylation levels of purified ETS1 were performed by western blot with antiphosphoserine/threonine antibody (Cell signaling). The reagents and antibodies used for in vitro kinase assay are shown in Supplementary Table 2.

Chromatin immunoprecipitation (ChIP) assay
The chromatin immunoprecipitation assay was performed as described previously [31]. In brief, Cells were treated with 1% formaldehyde at room temperature for 10 min to cross-link proteins to DNA. The reaction was stopped through the addition of glycine (0.125 M). Cells were scraped from culture dishes into PBS containing proteinase inhibitors. Pellets were resuspended in lysis buffer (5 mM PIPES/KOH pH 8.0, 85 mM KCL, 0.5% NP-40) and left on ice for 10 min. Nuclei collected by centrifugation were lysed with a nuclear lysis buffer (50 mM Tris pH 8.1, 10 mM EDTA, 1% SDS) and incubated on ice for 10 min. Lysates were sonicated to obtain chromatin having a mean length of approximately 600 bp. After centrifugation, 1 mg of protein supernatant was diluted 5-fold in ChIP dilution buffer (0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris pH 8.1, 167 mM NaCl). Immunocomplexes were precipitated overnight at 4° C with 1 µg of specific antibodies, followed by four sequential washes with low salt, high salt, LiCl, and Tris-EDTA buffer (pH 7.8), respectively. The washed immunoprecipitate was then eluted in elution buffer (1% SDS, 0.1 M NaHCO3) at room temperature and subjected to two 15-min runs of vortex and centrifugation. Formaldehyde cross-links were reversed by incubation in a water bath at 65° C for 5 h. After treatment with proteinase K and subsequent phenol/chloroform extraction, DNA fragments were recovered by ethanol precipitation and amplified by QPCR on MMP-9 promoter -166 region [31]. The primers and antibodies used for ChIP assay are shown in Supplementary Tables 1, 2.

DNA transfection and luciferase reporter assay
The protocols for DNA transfection and luciferase reporter assays have been previously reported [48,49]. Briefly, ovarian cancer cells (SKOV3 and MDAH2774) were trypsinized and re-suspended in serum-free RPMI medium (concentration: 10 million cells/mL). Cell suspensions (200 µL) were mixed with 5 µg DNA, transferred to a 2 mm-gap electroporation cuvette, and pulsed at 120 V for 70 msec with a BTX ECM2001 electroporator (Genetronics Inc., San Diego, CA, USA). Cells were subsequently re-seeded into a 6-well plate and maintained in DMEM/F12 with 10% fetal bovine serum for 24 h. ARK2 cells were transfected with Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. Finally, luciferase activity was measured using the Dual luciferase reporter assay system (Promega).

Transwell migration assay
Cell migration ability was assessed using a chemotaxis chamber (8µm pore, Corning, NY, USA). In brief, cells (1×10 5 in 200 μL of serum free DMEM/F12) were loaded onto the upper part of the chamber, followed by the addition of serum-free DMEM/F12 (800 μL) containing 5 μg/mL fibronectin to the 24-well plate [50]. After insertion of the plate to the migration chamber, cells were passed through the polycarbonate membrane (pore size: 8 μm) located between the two chambers. Upon 24 h incubation at 37° C, the membrane was fixed in methanol for 10 min and stained in 20:2 water/Giemsa solution (Sigma-Aldrich) for 1 h. Cells that migrated onto the membrane were counted under a microscope.

Wound closure assay
Wound closure assay was performed using a two-well migration insert with a defined cell-free gap (ibidi, Munich, Germany). Cells (density: 1 × 10 5 ) were seeded overnight into the migration insert, which was subsequently removed to create the cell-free gap. Upon addition of serum-free medium, images of the cell-free gap were acquired at baseline and after 24 h of incubation. Cell-free gap areas were quantified with the ImageJ software (https://imagej.nih.gov/ij/) after normalization for the corresponding gap area at baseline.

Immunohistochemistry
This study was approved by the institutional review board of Chang Gung Memorial Hospital, Linkou (IRB No. 201601560B0). Informed consent was waived for the retrospective nature of this study. Formalin-fixed, paraffin-embedded ovarian cancer specimens (46 serous adenocarcinomas and 29 endometrioid adenocarcinomas) were sectioned at 4 μm thickness and deparaffinized with xylene. Sections were dehydrated through a series of graded ethanol baths and stained with specific antibodies using an automated Leica Bond-MAX stainer (Leica Biosystem, Wetzlar, Germany). Expression levels of phospho-ETS1 and MMP9 were investigated by immunohistochemistry using anti-phospho-ETS1 Thr265/Ser269 antibody (custom-made; ABclonal Technology, Woburn, MA, USA) and MMP9 antibody (Abcam). Histoscores (0-300) were calculated as staining intensities (0-3) multiplied by the percentage of positively stained tumor cells (0-100%). The antibodies used for immunohistochemistry assay are shown in Supplementary Table 2.

Protein digestion
Enriched protein extracts were reduced with dithiothreitol, alkylated with iodoacetamide, and digested with sequencing-grade trypsin (Promega). The tryptic digests were acidified with trifluoroacetic acid and desalted using Sep-Pak C18 cartridges (Waters Associates, Milford, MA, USA).

Tandem mass spectrometry-based protein identification
Tandem mass spectrometry was performed on Q Exactive TM HF mass spectrometer (Thermo Fisher) coupled with an ACQUITY UPLC M-Class system (Waters Associates). Peptide mixtures were injected onto a trap column (Symmetry C18, 100 A, 5 µm, 180 µm × 20 mm, Waters Associates) connected with an analytic column (BEH C18 Column, 130 A, 1.7 µm, 75 µm × 250 mm, Waters Associates) and separated through a gradually increasing buffer B gradient (100% acetonitrile in 0.1% formic acid) over 75 min (flow rate: 260 nL/min). Peptide spectra were acquired in the positive ion mode (scan range: 375−1300 m/z). The ten most abundant precursor ions were dynamically selected for further fragmentation in the high-collision dissociation mode (with a normalized collision energy set at 28). The resolution of the full MS scan was set at 60000 with an m/z of 200 and an automatic gain control (AGC) of 3e6. The following parameters were used for the MS/MS scan: resolution, 15000; AGC target, 2e5; maximum injection time, 50 msec. The dynamic exclusion of the selected precursor ions was set between 20 and 30 sec.

MS-based proteomic data analysis by proteome discoverer software and mascot search engine
The MS raw files were uploaded into the Proteome Discoverer (version 2.1, Thermo Scientific) software.
The peak list for protein identification was generated using the Mascot search algorithm (version 2.5.1, Matrix Science, London, UK) against the Swiss-Prot human protein database. Carbamidomethylation of cysteine was set as a static or fixed modification. Oxidized methionine, phosphorylated serine or threonine, N-terminal glutamate conversion to pyroglutamate, and N-terminal acetylation were considered as dynamic modifications. Searches were done by allowing a single missed cleavage and with a mass accuracy set to 6 ppm and ±0.02 Da. In order to increase identification confidence, an overall false discovery rate below 1% was used. Moreover, only proteins with at least two unique peptides were accepted.

Data analysis
Overall and recurrence-free survival curves were plotted with the Kaplan-Meier method. The optimal cutoff values for phospho-ETS1 histoscores were identified with the Youden's index -with the goal of maximizing the sum of sensitivity and specificity to discriminate different events [51]. All calculations were performed with the SPSS 20.0 statistical package (SPSS Inc., Chicago, IL, USA). Two-tailed p values < 0.05 were considered statistically significant.

In vivo animal model
The experimental protocols and animal care were approved by the animal care and use committee (IACUC) in Chang-Gung memorial hospital (IACUC No.2016113001). All tumor tissues were from our previous study [31]. In briefly, after 1x10 6 murine ovarian surface epithelial cells (MOSEC cells) were injected to peritoneal cavity of C57BL/6 mice one week later, mice were intraperitoneally treated with vehicle control (20% Water:20% DMSO: 60% chromophore) or GSK3β inhibitor-LY2090314 (10 mg/kg) twice a week. One month later, mice were sacrificed and tumors in peritoneal cavity were collected for further detecting ETS1, MMP9 and actin with western blot.

Abbreviations
GSK3β: glycogen synthase kinase-3β; MMP-9: matrix metalloproteinase-9; PKCα: protein kinase Cα; CaMKII: calcium-dependent protein kinase II; EGF: epidermal growth factor; VEGF: vascular endothelial growth factor; ERK: extracellular signal-regulated kinases; NF-κB:  Table. The direction of the Y-ion position started from the end of the peptide C-terminus. Phosphorylated residues of the two synthetic peptides (pT2S6 and pS6S10) are shown in red and correspond to the Y11 and Y15 fragment ions for pT2S6 and to the Y7 and Y11 fragment ions for pS6S10. (B) The sequence and phosphorylation sites of the potential GSK3β-catalyzed double-phosphorylation event (peak2 in Figure 2C) was analyzed with LC-MS/MS (upper panel). Using two synthetic peptides (pT2S6 and pS6S10) as reference standards, the phosphorylated peptide in peak2 displayed the first phosphorylation event at Y11 (upper panel, red) but not at Y7. The pT2S6 peptide exhibited the first phosphorylation event at Y11 (middle panel, red). However, the pS6S10 peptide was characterized by a first phosphorylation event at Y7, followed by a second phosphorylation at Y11 (lower panel, red). The fragment ion spectra of the candidate for GSK3β-catalyzed phosphorylation were remarkably more similar to the pT2pS6 fragmentation profile -without phosphorylation ions being evident between the y5 and y8 fragments.