FIH is an oxygen sensor in ovarian cancer for G9a/GLP-driven epigenetic regulation of metastasis-related genes
Abstract
The prolyl hydroxlyases PHD1-3 and the asparaginly hydroxlyase FIH are oxygen sensors for HIF-driven transcription of hypoxia-induced genes, but whether these sensors affect oxygen- dependent epigenetic regulation more broadly is not known. Here we show that FIH exerts an additional role as an oxygen sensor in epigenetic control by the histone lysine methyltransferases G9a and GLP. FIH hydroxylated and inhibited G9a and GLP under normoxia. When the FIH reaction was limited under hypoxia, G9a and GLP were activated and repressed metastasis suppressor genes, thereby triggering cancer cell migration and peritoneal dissemination of ovarian cancer xenografts. In clinical specimens of ovarian cancer, expression of FIH and G9a were reciprocally associated with patient outcomes. We also identified mutations of FIH target motifs in G9a and GLP, which exhibited excessive H3K9 methylation and facilitated cell invasion. This study provides insight into a new function of FIH as an upstream regulator of oxygen-dependent chromatin remodeling. It also implies that the FIH-G9a/GLP pathway could be a potential target for inhibiting hypoxia- induced cancer metastasis.
INTRODUCTION
Hypoxia is a common feature of the microenvironment in pathologic and physiologic conditions. Based on ambient oxygen levels, eukaryotes optimally set the gene expression profile to maintain oxygen homeostasis and to readjust cell behavior and metabolism (1,2). Such adaptive processes are tightly controlled by oxygen sensors and their signaling pathways. In mammalian cells, dioxygenases, such as prolyl hydroxylase domain-containing proteins (PHD1-3) and factor inhibiting HIF (FIH), act as oxygen sensors for hypoxia-inducible factor -subunits (HIF-s) that transactivate many rescue genes under hypoxia. Under normoxia, PHDs hydroxylate HIF-s at proline residues, followed by recruitment of the von Hippel Lindau protein (pVHL), polyubiquitination, and proteasomal degradation of HIF-s (3,4). In addition, FIH hydroxylates HIF-s at asparagine residues, thereby repressing their activities by blocking the recruitment of p300/CBP coactivators (5,6). Under hypoxia, however, HIF-s are stabilized and activated due to limitation of both hydroxylase reactions, leading to robust expression of hypoxia-regulated genes. In the last two decades, HIF- hydroxylases and the HIF-mediated transactivation of rescue genes have been investigated intensively. In contrast, oxygen-sensitive epigenetic regulation at the chromatin level has not been a focus of interest despite the abundant supporting evidence. Indeed, multiple studies have demonstrated that hypoxia changes the post-translational modifications (PTMs) of histones H3 and H4 at a global level or at specific promoters (7). In particular, di- or tri- methylation of H3 lysine residues (K4, K9, K27, K36, or K79) were often reported to be induced during hypoxia (8-10). Given that the methylation at each H3K residue distinctly functions as an epigenetic mark (11), hypoxic alteration of H3K methylation may determine up- or down-regulation of specific gene sets. What is the oxygen sensor for oxygen-dependent histone modifications? We addressed this unanswered question in the present study.
Of the histone modifications, the status of H3K9 methylation is most dynamically changed under hypoxia. Surprisingly, we here found that the H3K9 methylation is FIH-dependent. FIH hydroxylatesG9a (EHMT2) at Asn779 and GLP (EHMT1) at Asn867 under normoxia, and the hydroxylated G9a and GLP lose their abilities to methylate H3K9me2/me3. Under hypoxia, G9a and GLP escape from Asn-hydroxylation and generate the repressive marks H3K9me2/me3, especially at the promoters of cell adhesion-related genes. Furthermore, we demonstrated that the FIH-G9a/GLP axis controls invasion and metastasis of ovarian cancer (OvCa) cells and it is also associated with the clinical outcomes of patients with ovarian cancer.Human embryonic kidney (HEK293 and HEK293T), human ovarian cancer (SKOV-3, ES-2 and PA- 1) cell-lines and various human cancer (A549, Hep3B, MCF7, DU145, HCT116 and 786-O) cell lines were obtained from American Type Culture Collection (ATCC; Manassas, VA). HEK293, HEK293T, Hep3B and HCT116 were cultured in DMEM, PA-1 in MEM, SKOV-3, ES-2, A549, MCF7, DU145 and 786-O in RPMI1640, supplemented with 10% heat-inactivated fetal bovine serum (FBS). SKOV- 3 and ES-2 stable cell lines expressing luciferase and G9a or N779Q mutant were selected with G418 and five clones were pooled to rule out the artifacts by plasmid insertion into genomes. Cell lines used in this study were authenticated by STR DNA profiling analysis, which were performed in Korean Cell Line Bank (Seoul, Korea). Mycoplasma contamination was routinely checked when cell growth or shape was changed. After thawing, cells were usually cultured for no more than 3 months. Cells were grown in a humidified atmosphere containing 5% CO2 at 37oC.
The ambient oxygen level was 20% for normoxia, 1% for moderate hypoxia, or near 0.1% for severe hypoxia.Culture media, FBS, dimethyloxalylglycine (DMOG), UNC0638, and antibodies against Myc-Tag,Flag-tag, and HA-tag were purchased from Sigma-Aldrich (St. Louis, MO). Antibodies against G9a(for ChIP), Histone-3, H3K9me1/2/3 and JNJ-42041935 were purchased from Merck Millipore (Billerica, MA); antibodies against FIH, G9a, β-tubulin, Suv39H1, EZH2, and HRP-conjugated secondary antibodies from Santa Cruz Biotechnology (Santa Cruz, CA); anti-GLP antibodies from Perseus Proteomics (Tokyo, Japan); anti-PHD1/2/3, anti-HIF-2 antibodies from Novus Biologicals (Littleton, CO). Anti-HIF-1 antibody was generated in rats against a human HIF-1 peptide (aa. 418-698), as previously described (12).The cDNA of G9a was cloned by RT-PCR and inserted into the HA-tagged pcDNA. The C-terminal HA-tagged GLP plasmid was purchased from GeneCopoeia Inc (Rockville, MD). The plasmids for FLAG/streptavidin-binding protein (F/S)-tagged G9a/GLP fragments were constructed from the G9a/GLP plasmid using PCR with pfu DNA polymerase and blunt-end ligation. Site-specific mutations of G9a and GLP were performed using PCR-based mutagenesis (Stratagene; Cedar Creek, TX). HA- and FLAG-FIH plasmids and its enzymatically inactive mutants were constructed, as previously described (13). For transient transfection of plasmids or siRNAs, cells at ~60% confluency were transfected using calcium phosphate in HEK293 cells or Lipofectamine 3000 reagent (plasmid) or Lipofectamine RNAiMAX (siRNA) reagent in SKOV-3, ES-2 and PA-1 cells. The nucleotide sequences of siRNAs and shRNAs are summarized in Supplementary Table S1.Cells were lysed in a histone extraction buffer containing 0.5% triton X-100, 2mM PMSF and 0.02% sodium azide, and centrifuged at 1000xg at 4°C for 10 min. The nuclear pellet was dissolved in 1ml of 0.2N HCl and rotated at 4°C overnight for acid extraction.
The extract was centrifuged at 3000xg for 10 min, and the supernatant was collected. The extract was incubated in a 20% trichloroacetic acid (TCA) on ice for 30 min, and precipitated histones were collected by centrifugation at 10000xg for 105min. Histone pellet was washed twice with ice-cold acetone to remove acid, and air-dried. Histone extracts were dissolved in a 50 mM Tris/HCl (pH 7.4) solution and protein amount was quantified using BCA.H3 and H4 Histone modifications were measured using Epiquik Histone Modification Multiplex Assay Kit (Epigentek; NY, USA). One l of histone extracts (50 ng/l) was mixed with 49 l of the antibody buffer, and the mixture was applied to each well in the microplate at 37°C for 1 hr. After three times washing with a buffer, 50l of the detection antibody solution (1:1000 dilution) was applied to the microplate for 1 hr. After the antibody solution was washed out, 100 l of the developer solution was added to each well and incubated in the dark for 1 to 10 minutes. The specific absorbance at a 450 nm wavelength was calculated by subtracting the 655 nm absorbance from the 450 nm absorbance.Cell lysates were separated on SDS-polyacrylamide (8-12%) gels, and transferred to an Immobilon-P membrane (Millipore; Bedford, MA). Membranes were blocked with a Tris/saline solution containing 5% skim milk and 0.1% Tween-20 for 1 hr and incubated overnight at 4°C with a primary antibody diluted 1:1,000-10,000 in the blocking solution. Membranes were incubated with a horseradish peroxidase–conjugated secondary antibody (1:5,000) for 1 hr at room temperature, and stained using the ECL-plus kit (Amersham Biosciences; Piscataway, NJ). To analyze protein interactions, cell lysates were incubated with anti-FIH, anti-G9a, or anti-GLP antibody overnight at 4°C, and theimmune-complexes were pulled down with protein A/G beads (Santa Cruz Biotechnology). Otherwise, cell lysates were incubated with EZview Red anti-HA or anti-Flag affinity gel (Sigma-Aldrich; St. Louis, MO) for 4 hr at 4°C. The bound proteins were eluted in a denaturing SDS sample buffer or HA/FLAG-tag peptides, and loaded on SDS-PAGE.
All Western blotting experiments were performed three or more times.Cells were centrifuged at 1,000xg for 5 min, and resuspended with a lysis buffer containing 10 mM Tris/HCl (pH 7.4), 10 mM KCl, 1 mM EDTA, 1.5 mM MgCl2, 0.2% NP-40, 0.5 mM dithiotheritol, 1 mM sodium orthovanadate, and 400 M PMSF. The cell lysates were separated into pellet (for nuclear fraction) and supernatant (for cytosolic faction) using centrifugation at 1,000xg for 5 min. One packed volume of a nuclear extraction buffer (20 mM Tris/HCl, pH 7.4, 420 mM NaCl, 1 mM EDTA, 1.5 mM MgCl2, 20% glycerol, 0.5 mM dithiotheritol, 1mM sodium orthovanadate, and 400M PMSF) was added to the pellet, and vortexed intermittently at low speed on ice for 30 min. The nuclear extracts were cleaned at 20,000xg for 10 min and stored at -70°C.Cells were fixed with 3.7% formaldehyde for 10 min and permeabilized with 0.1% Triton X-100 for 30 min. Cells were incubated in PBS containing 0.05% Tween-20 and 3% bovine serum albumin for 1 hr, and further incubated overnight at 4°C with a primary antibody (1:100 to 1,000 dilution). Cells were incubated with Alexa Fluor 488 or Alexa Fluor 594-conjugated secondary antibodies (Molecular Probes; Eugene, OR) for 1 hr. To stain actin filaments (F-actin), cells were incubated with Alexa- Fluor 633 phalloidin (Invitrogen; Carlsbad, CA) for 30 min, stained with DAPI (Sigma-Aldrich; St. Louis, MO) for 30 min, and mounted in Faramount aqueous mounting medium (Dako; Glostrup, Denmark). Immunostained cells were observed under Carl Zeiss LSM510 META confocalmicroscope.F/S-G9a fragments, HA-G9a/GLP-WT and NQ mutant and HA-FIH were separately expressed in HEK293T cells, and purified by anti-FLAG and anti-HA affinity beads, respectively. Recombinant GST-FIH and GST-G9a/GLP peptides were expressed in E. coli, purified by GSH-affinity beads, and eluted by GSH.
The mixtures of peptides were incubated at 4°C for 1 hr, further incubated with the affinity beads at 4°C for 4 hr, centrifuged, and immunoblotted. To analyze ARDs binding to histone H3K9me1/me2, an in vitro hydroxylation assay was performed with recombinant GST-FIH, GST- G9a/GLP_ARD-WT and NQ mutant peptides as described above. After the reaction, 120 g of Biotin-conjugated H3K9me1 or me2 peptide (aa. 1-100; Epigentek) was added to the assay mixture and incubated with at 4°C for 1 hr. The peptides were pulled down using streptavidin affinity beads and immunoblotted. Recombinant GST-G9a/GLP_ARD and GST-FIH peptides were incubated with 1 mM - ketoglutarate, 1 mM ascorbate, and 100 M FeSO4 in a hydroxylation buffer (40 mM Tris-HCl pH7.5, 10 mM KCl, 3 mM MgCl2) at 37°C for 1 hr. G9a/GLP_ARD peptides were electrophoresed, digested in gel, and analyzed by LC-MS/MS.After in vitro hydroxylation, proteins in the reaction solution were electrophoresed on SDS-PAGE. The gel slice were digested with trypsin and endoproteinase AspN, and subjected to a nano-flow ultra- performance liquid chromatography/ESI/MS/MS with a mass spectrometer (Q-tof Ultra global), comprising a three-pumping Waters nano-LC system, a stream selection module, and MassLynx 4.0controller (Waters, Milford, MA). Five l of digested peptides was dissolved in buffer C(water/acrylonitrile/formic acid; 95:5:0.2, V:V:V), injected on a column, and eluted by a linear gradient of 5~80% buffer B (water/acrylonitrile/formic acid, 5:95:0.2) over 120 min. MS/MS spectra were processed and analyzed using ProteinLynx Global Server (PLGS).The dioxygenase activity of FIH was measured using the CO2 capture assay, as previously described(14).
Recombinant GST-FIH, GST-G9a/GLP_ARD-WT and NQ mutant peptides were purified fromE. coli. G9a or GLP peptides and FIH peptides were incubated mixed with 4 mM ascorbate and 1.5 mM FeSO4 in 35 l of reaction buffer (20mM Tris-HCl/pH7.5, 0.5 mM DTT, 0.1% BSA, and 150 mM NaCl). The enzyme reaction was started with 5 l of [1-14C]-labeled 2-oxoglutarate with a specific activity of 58.7 mCi/mmol (Perkin-Elmer, Wellesley, MA). A filter paper soaked in 30 mM Ca(OH)2 solution was immediately placed in the airtight screw-cap microcentrifuge tube. Each reaction was timed individually at 37°C for 20 min and terminated by removal of filters. Air dried filter paper was soaked in a scintillation cocktail and [14C]-labeled CO2 was counted with a liquid scintillation counter.The histone methyltransferase activity was measured according to a protocol provided by Fingerman et al. (15). Biotin-conjugated histone-3 peptide (aa. 1-21; Merck Millipore) was mixed with affinity gel-purified G9a/GLP and/or recombinant GST-FIH. The mixtures (58 l) were incubated with 1 mM α-ketoglutarate, 1 mM ascorbate, and 100 M FeSO4 in the hydroxylation buffer at 30°C for 30 min, and further reacted with 2 l of [3H]-labeled-S-adenosyl-l-[methyl-3H]methionine (SAM) with a specific activity of 13.3 Ci/mmol (Perkin-Elmer) at 30°C for 1 hr. Ten l of the mixture was spotted onto a P81 phophocellulose membrane, washed with 0.2 M ammonium bicarbonate, and dried. Themembrane was soaked in a scintillation cocktail and [3H]-labeled H3 was counted with a liquid scintillation counter.Histone H3 full length peptide (Sigma-Aldrich; St. Louis, MO) was mixed with affinity gel-purified G9a/GLP and/or GST-FIH. The peptide mixtures were subjected to in vitro hydroxylation assay, as described before, and further reacted with 1M of S-(5’-adenosyl)-L-methionine (SAM) (Sigma- Aldrich; St. Louis, MO) at 37°C for 2 hr. After then, reaction mixtures were loaded for immunoblotting.Atomistic molecular dynamics simulationAll-atom atomistic MD simulations were performed to test binding stability of H3K9me2 bound to GLP_ARD (aa. 775 – 998).
The initial bound state of unmodified ARD was obtained from the protein data bank (PDB ID code 3B95) and hydroxylated ARD was prepared by hydroxylation of N867 residue at -carbon position, where H3K9me2 is encapsulated within a partial hydrophobic cage formed by three tryptophans (W874, W879, W912) and a glutamate residue (E882) of ARD. Each ARD system was modeled to be solvated by about 30,600 water molecules with 100 mM NaCl in the simulation box with a periodic boundary condition. The CHARMM27 force field was used for natural amino acid residues while the force fields for the non-standard residues N867-OH and H3K9me2 were generated using ForceField Toolkit module implemented in VMD (16). With this system constitution, NPT-ensemble MD simulations were carried out at 310 K and 1 bar using GROMACS5.1.2 program (17). The modified Berendsen thermostat and Parrinello-Rahman barostat were used for maintaining temperature (310 K) and the pressure (1 bar). A full system periodic electrostatics was employed by the particle-mesh Ewald method with a 1 Å grid spacing. The cutoff and switching distances for van der Waals force were set to be 12 Å and 10 Å, respectively. The bonds involving hydrogen were constrained to be rigid by using the LINCS algorithm (18). The MD system was10equilibrated for 30 ns with a 2 fs time step and the structure analysis was done for simulating a further 20 ns run with recording every 2 ps.
The chromatin protein extraction was performed according to a protocol provided by Yang et al. (19). Cells were lysed in buffer A (10 mM HEPES pH 7.5, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 0.15% Nonidet P-40, 1% protease inhibitor cocktail, and 1mM DTT) for 10 min. After centrifugation at 12,000 × g for 30 sec, the pellets were washed twice in PBS and then lysed in buffer B (3 mM EDTA, 0.2 mM EGTA, 1% protease inhibitor cocktail, and 1 mM DTT) for 30 min. After centrifugation at 12,000 × g for 3 min, the supernatant was assumed to contain soluble nucleoproteins, and the pellets were assumed to contain the chromatin fraction.Total RNAs were extracted using TRIZOL (Invitrogen; Carlsbad, CA), and reverse-transcribed in a reaction mixture containing M-MLV Reverse Transcriptase (Promega, Madison, WI), RNase inhibitor, dNTP, and random primers at 46°C for 30 min. Quantitative real-time PCR on 96-well optical plates was performed in the qPCR Mastermix (Enzynomics, Daejeon, Korea), and fluorescence emitting from dye-DNA complex was monitored in CFX Connect Real-Time Cycler (BIO-RAD, Hercules, CA). The mRNA values of targeted genes were calculated relative to GAPDH expression for each sample. All reactions were performed as triplicates. The nucleotide sequences of PCR primers are summarized in Supplementary Table S2.Cells were fixed with 2 mM disuccinimidyl glutarate (Santa Cruz) for 45 min, crosslinked with 1% formaldehyde at 37°C for 10 min, treated with 150 mM glycine. After cells were lysed with 0.5% NP-40, the nuclear fraction was extracted with 1% SDS and sonicated to cut genomic DNAs into 300-11500bp fragments. Soluble chromatin complexes were precipitated with 1-5 g of anti-H3K9me2, anti- G9a or anti-GLP antibody overnight at 4°C. Immune complexes were precipitated with pre-blocked protein A/G beads at 4°C for 4 hr. The beads were sequentially washed with low salt buffer, high salt buffer, LiCl wash buffer, and TE buffer. The bound chromatin complexes were eluted in a ChIP direct elution buffer at 65°C for 30 min and cross-link was reversed overnight at 65°C. DNAs were extracted by phenol-chloroform and precipitated with ethanol. The precipitated DNAs were resolved in distilled water and analyzed by real-time PCR.
DNAs were amplified over 40 PCR cycles (95°C/53°C/70°C, 20 sec at each phase) by real-time PCR. ChIP-qPCR results were represented as percentage of IP/input signal (% input). All reactions were performed as triplicates.For the ovarian cancer patients gene set enrichment analysis (GSEA), the mRNA data sets from ovarian cancer patients were imported from The Cancer Genome Atlas (TCGA). A formatted GCT file was used as input for the GSEA algorithm v2.0 (available from: http://www.broadinstitute.org/gsea). For grouping TCGA dataset, the values of the EHMT2 (G9a) and HIF1AN (FIH) were used as criteria standard for low expression and high expression groups.To assess the invasive potential, the Transwell chambers with a matrigel-coated membrane with 8 m pore size was used. Cells were seeded onto upper chambers in serum-free medium at a density of 1X104 (SKOV-3) or 2X104 (PA-1) cells/well and 1ml of FBS-containing medium was placed in the lower chamber. After 18 hr-incubation, cells were fixed with 4% paraformaldehyde and stained with Hematoxylin and Eosin (Sigma-Aldrich; St. Louis, MO). Cells on upper side of the filters were removed with cotton-tipped swabs and invaded cells on the down side were viewed and photographedunder BX53-P polarizing microscope (Olympus; Tokyo, Japan). The invaded cells were countedusing the ImageJ software (NIH, Bethesda, MD).All animal studies were carried out with an approved protocol proposal from the Seoul National University Institutional Animal Care and Use Committee (approval #, SNU-160303-1). SKOV-3 cell lines harboring the CMV-luciferase/IRES/GFP plasmid or the CMV-luciferase/IRES/G9a (or N779Q) plasmid were selected with G418, and five clones per cell line were pooled to rule out the artifacts due to genomic insertion. These cells were infected by Lenti-viral shRNAs 1 hr prior to peritoneal implantation. Female (8 weeks old, Balb/cSlc-nu/nu) mice are used for establishing ovarian cancer xenografts.
Cancer cells were injected into the lower right of quadrant of the mouse abdomen, as previously performed (20). For imaging tumors in live animals, VivoGlo luciferin (Promega) was dissolved in sterilized PBS (finally 40 mg/ml). Mice were anaesthetized with isoflurane and injected intraperitoneally with 100 l of the luciferin solution. After 5 min, images were acquired with the Xenogen IVIS® Lumina series II and analyzed using the LivingImage 2.11 software package (Xenogen Corp, Alameda, CA). The peritoneal dissemination and metastasis were monitored once a week.Human ovarian cancer tissue-arrays were obtained from SuperBioChips Lab (Seoul, South Korea). Patients’ information and cancer stages are summarized in Supplementary Table S3. The tissue slides were autoclaved to retrieve antigens, treated with 3% H2O2, and incubated with a primary antibody overnight at 4°C. The sections were biotinylated with a secondary antibody at room temperature for 1 hr. The immune complexes were visualized using the Vectastatin ABC kit (Vector Laboratories, Burlingame, CA), and tissue slides were counterstained with hematoxylin for 10 min. The slides were viewed and photographed at 4 high-power fields in each slide. The expression level was evaluatedbased on intensity and stained cell number and scored from 0 to 5.All data were analyzed using the Microsoft Excel 2013 or the GraphPad Prism 5.0 software, and results were expressed as means and standard deviations (SDs). The unpaired, two-sided Student’s t- test was used to compare results of protein expression, ROI flux, or cell number. Correlation values(R) are calculated by Pearson correlation coefficient test. Cancer specific survival in tissue microarray was assessed using the Kaplan-Meier method and compared with the log-rank test. All statistical significances were considered when P values were less than 0.05.
RESULTS
The modifications of histones H3 and H4 were analyzed in HEK293 cells exposed to normoxia or hypoxia. Histone H3 is more dynamically modified in altered oxygen levels (Fig. 1A; Supplementary Fig. S1A), compared with histone H4 (Supplementary Fig. S1B). In particular, the H3K9me2/me3 levels increased most distinctly in hypoxia (> 3-fold versus the normoxic levels), whereas the H3K9me1 level dropped. This conversion of H3K9me1 to H3K9me2/me3 occurred as early as 8 hours and increased up to 72 hours in HEK293 cells (Fig. 1B). The hypoxia-induced methylation of H3K9 was also observed in various cancer cell lines (Fig. 1C; Supplementary Fig. S2A). Since PHDs and FIH are the only known sensors for oxygen-dependent gene regulation, we examined whether the sensors participate in the oxygen-dependent methylation of H3K9. Surprisingly, H3K9 methylation was stimulated even under normoxia by FIH knock-down, which was not further augmented under hypoxia (Fig. 1D). The H3K9 di-/tri-methylation by FIH knock-down occurred commonly in various cell lines (Fig. 1E; Supplementary Fig. S2B). Since FIH inhibits HIF-1/2 functionally (5,6), we checked the possibility that FIH indirectly regulates H3K9 methylation through the inhibition. As shown in Fig. S2C, FIH knockdown enhanced the H3K9me2/3 levels even under HIF-1/2 knockdown, supporting the direct role of FIH in H3K9 methylation. Because FIH was characterized to have a residual activity to HIF-1 hydroxylation at 1% O2 due to a high affinity to oxygen (21,22), we tested whether H3K9 methylation is further increased at a lower O2 level. As expected, the H3K9me2/3 levels were higher at ~0.1% O2 than at 1% O2 (Supplementary Fig. S2D), suggesting that FIH controls the H3K9 methylation depending on the O2 tension. Given that the H3K9 methyltransferases G9a and GLP have been reported to be activated under hypoxia (23), we tested the possibility that G9a and GLP activities are controlled by FIH.
First, we validated the contribution of G9a or/and GLP to H3K9 methylation using G9a/GLP-targeting siRNAs. H3K9 methylation under hypoxia was almost completely blocked by co-silencing G9a and GLP or by a G9a/GLP inhibitor (Fig. 1F; Supplementary Fig. S2E). We tested the possibility that FIH controls the G9a/GLP-driven methylation of H3K9. FIH negatively regulated the di-/tri-methylation of H3K9 under normoxia, but not at a 1% or lower oxygen level (Fig. 1G). To examine whether hydroxylase activity is required for FIH inhibition of H3K9 methylation, we used two FIH mutants (W296H and W296A) lacking catalytic activity (24). Although the FIH mutants bound to G9a and GLP, they failed to inhibit H3K9 methylation (Fig. 1H), which indicates that the hydroxylation process is essential for FIH action against H3K9 methylation. Because FIH did not affect G9a and GLP expression, we next investigated how FIH functionally regulates G9a and GLP. FIH was originally found to hydroxylate HIF-1 at Asn803 (5). Later it was also reported to target non-HIF proteins containing the ankyrin-repeat domain (ARD), such as Notch1-3, IκBα, P105, ASPP2, TRPV3, and OTUB1 (25-30). As putative ARDs were identified in G9a and GLP on the CD- search BLAST_NCBI program, we tested the possibility that FIH targets G9a and GLP. Endogenous (Fig. 2A; Supplementary Fig. S3A) and ectopically expressed (Fig. 2B) G9a and GLP proteins were co-immunoprecipitated with FIH. These interactions were stronger under hypoxia than under normoxia (Supplementary Fig. S3B and S3C), which suggests that FIH does not dissociate from G9a and GLP unhydroxylated under hypoxia (27). In contrast, FIH did not interact with the histone methyltransferases lacking ARD, such as EZH2 and SUV39H1 (Supplementary Fig. S3D and S3E). Since SUV39H1 is also known to generate H3K9me3 (31), we examined whether SUV39H1 contributes to the hypoxia-induced trimethylation of H3K9. In SUV39H1-knocked down cells, the H3K9me3 level was profoundly enhanced during hypoxia though it was slightly reduced under normoxia (Supplementary Fig. S3F). This further supports our notion that G9a and GLP largely contribute to hypoxia-induced H3K9 methylation.
In immunofluorescence analyses, FIH was localized mainly in the cytoplasm, whereas G9a and GLP were in the nucleus (Supplementary Fig. S3G). The subcellular localization of these proteins was confirmed by immunoblotting of the cellular fractions; FIH was found to associate with G9a and GLP in the cytoplasm (Fig. 2C). As was reported for HIF-1 (32), FIH is likely to target G9a and GLP during their transient residence in the cytoplasm. As expected, FIH interacts with the ARDs of G9a and GLP (Fig. 2D and Supplementary Fig. S3H). These results prompted us to evaluate the FIH- mediated hydroxylation of asparagine residues in ARDs. The ARDs in G9a and GLP function to recognize H3K9me1/me2 (33). Therefore, we examined whether FIH regulates this function of ARD in an O2-dependent manner by hydroxylating the conserved asparagine residue. Upon in vitro co- incubation of recombinant G9a/GLP-ARD and FIH peptides in a hydroxylation buffer, LC-MS/MS analyses revealed that the mass of a peptide segment in each ARD was shifted by +16 Da (Fig. 2E; Supplementary Fig. S4A-C). Accordingly, N779 of G9a and N867 of GLP are likely to be hydroxylated by FIH. To further validate the ability of FIH to hydroxylate asparagine residues, G9a/GLP-ARD peptides and their mutants (G9a-N779Q and GLP-N867Q; “NQ mutants”) were subjected to [14C] CO2 capture assay. Wild-type ARDs markedly promoted decarboxylation of 2- oxoglutarate, but the NQ mutants did not (Fig. 2F; Supplementary Fig. S4D). Nonetheless, the NQ mutations did not interfere with FIH binding to G9a or GLP (Supplementary Fig. S4E and S4F), which suggests that the asparagine residues are the hydroxylation sites, but not the docking sites, for FIH. Collectively, these findings strongly indicate that the asparagine residues are hydroxylated by FIH. Fig. 2G shows that the FIH-hydroxylated motifs are well conserved among FIH substrates.
We next investigated the functional consequences of FIH-mediated hydroxylation in G9a and GLP. In an in vitro assay of methyltransferase, FIH inactivated wild-type G9a and GLP, but not the NQ mutants (Fig. 3A; Supplementary Fig. S5A). Initially, the SET domains of G9a/GLP recognize H3K9 and convert it to H3K9me1. In the following reaction, the ARDs recognize H3K9me1 and then the SETs additionally methylate the residue to generate H3K9me2 and H3K9me3 (34). To further understand which step of the H3K9 methylation is regulated by FIH, we measured H3K9me1/2/3 levels in an in vitro methylation assay. FIH inhibited the catalytic activity of G9a/GLP to generate H3K9me2/3, but did not affect those of NQ mutants (Supplementary Fig. S5B). The H3K9me1 increment by FIH may result from the blocked conversion from H3K9me1 to H3K9me2. In HEK293 cells, FIH overexpression attenuated the G9a/GLP-driven di-/tri-methylation of H3K9, but failed to inhibit the methyltransferase activities of NQ mutants (Fig. 3B). However, FIH did not affect the mono-methylation of H3K9 because ARD functions in the recognition of methylated H3K9 rather than unmethylated H3K9 (33). These findings suggest that FIH inhibits di-/tri-methylation of H3K9 through the Asn-hydroxylation of ARDs in G9a and GLP. Given the aforementioned function of ARD, we examined the possibility that the Asn- hydroxylation of ARD interferes with G9a/GLP binding to methylated H3K9. In an in vitro binding analysis, ARD peptides were co-precipitated with biotinylated H3K9me1 and H3K9me2, but the interactions were substantially reduced after FIH hydroxylated ARDs (Fig. 3C; Supplementary Fig. S5C). In contrast, NQ-mutated ARD peptides bound consistently to H3K9me1/me2, regardless of the FIH reaction. For a better understanding of the interaction between ARD and methylated H3K9, we performed an atomistic molecular dynamics (MD) simulation. The stability of the ARD-H3K9me2 interaction was analyzed in two model systems for control and N867-hydroxylated GLP_ARD (aa. 775 – 998). The H3K9me2 residue is encapsulated within a hydrophobic pocket formed by three tryptophan residues (W874, W879, W912) and one glutamate (E882) residue in ARD. The position of H3K9 within the pocket was monitored by analyzing trajectories obtained from isothermal-isobaric MD runs at 310 K and 1 bar, which are shown in Supplementary Movies S1 and S2.
Given snapshot images of MD-simulated structures for ARD and H3K9me2 after a 30 ns run, the unhydroxylated ARD stably holds H3K9me2 in the hydrophobic pocket, but the N867-hydroxylated ARD does not because the pocket is open (Fig. 3D, top). The stability of the ARD-H3K9me2 interaction was verified by analyzing the pair correlation function g(r) between H3K9me2 and the pocket residues. H3K9me2 is surrounded by the four residues with radii of 3~6 Å in control ARD, but W879 in N867- OH ARD moves away (peak r = 11.5 Å) from other residues because the face of an indole group of W879 is pivoted toward N867-OH through interaction with P876 (Fig. 3D, bottom). These findings strongly support our notion that FIH-hydroxylated G9a and GLP cannot generate the gene-repressive marks H3K9me2/me3. Next, we examined whether FIH is responsible for the oxygen-dependent expression of G9a/GLP target genes. MAGEA2/8 (melanoma-associated antigen 2 and 8) genes have been identified as targets of G9a via genomic analysis in ES cells (35). DHFR (dihydrofolate reductase) and MLH1 (MutL homolog 1) genes were also reported to be repressed by the hypoxia-induced methylation of H3K9 (36,37). Under normoxia, the transcripts of these G9a/GLP target genes were downregulated by FIH knock-down, which was fully recovered by co-silencing G9a and GLP or by treatment with a G9a/GLP inhibitor UNC0638 (Fig. 3E). However, under hypoxia, the target gene expressions were consistently reduced regardless of FIH expression. Also, the mRNA levels of G9a/GLP target genes were decreased by the overexpression of wild-type G9a or GLP and further reduced by hydroxylase inhibitor, dimethyloxaloylglycine (DMOG). However, the gene repression by NQ-mutated G9a or GLP was unaffected by DMOG (Supplementary Fig. S5D and S5E). The level of H3K9me2 on these gene promoters was increased during hypoxia and consistently elevated by FIH knock-down (Fig. 3F). The G9a or GLP recruitment to target loci was augmented by DMOG, but not by a PHD-specific inhibitor JNJ-42041935 (Fig. 3G; Supplementary S5F). The NQ mutants of G9a or GLP enrichment at target loci was not affected by DMOG (Supplementary Fig. S5G and S5H). In addition, the global chromatin abundance of G9a/GLP was increased by FIH knockdown under normoxia, but was constitutively high under hypoxia (Supplementary Fig. S5I). These findings suggest that FIH participates in O2-dependent regulation of G9a/GLP target genes by hydroxylating the specific asparagine residues of G9a and GLP, which is illustrated in Fig. 3H.
To explore the role of the FIH-G9a/GLP axis in human cancer on a large scale, we examined mRNA expression of FIH and G9a/GLP using The Cancer Genome Atlas (TCGA) database. Of 30 major cancers, ovarian cancer (OvCa) expresses G9a and GLP at a higher level but does FIH at a lower level (Fig. 4A; Supplementary Fig. S6A). Based on the reciprocal expression of G9a and FIH, we selected OvCa as a subject to study the disease significance of the FIH-G9a axis. To get a clue to the axis’s role, we performed a gene set enrichment assay (GSEA) on TCGA database (GSE82191) containing mRNA expression profiles in 478 ovarian cancer tissues. High and low expression groups were categorized based on the mean ±1 standard deviation (SD) values (Supplementary Fig. S6B). In GSEA analysis of high versus low expression, 142 gene sets negatively correlate with low G9a (Supplementary Table S4), whereas 32 gene sets positively correlate with high FIH (Supplementary Table S5). All acquired gene sets were ontologically categorized based on description of each gene sets from GSEA MsigDB. Consequently, the anti-metastatic properties gene set group is the largest portion in both of G9a- and FIH-related gene sets (Fig. 4B). These gene sets show a negative correlation with G9a but a positive correlation with FIH (Supplementary Fig. S6C). In particular, the gene set ‘Aigner_ZEB1_Targets’ is commonly included in G9a- and FIH-related gene sets. This gene set contains many genes essential for epithelial differentiation and cell adhesion, and CDH1, CDH11, TSPAN15, and PKP3 genes are ranked in the lists of core enrichment genes of this gene set (Supplementary Fig. S6D). As down-regulation of cell adhesion molecules promotes cancer metastasis (38), we investigated the role of the FIH/G9a axis in terms of ovarian cancer metastasis.
EMT is a phenotypic conversion in which epithelial cells acquire fibroblast-like properties, including reduced intercellular adhesion and increased motility. In PA-1 and SKOV-3 ovarian cancer cells, FIH knock-down and/or hypoxia stimulated the formation of lamellipodia, which was reversed by co-silencing G9a and GLP (Fig. 4C). In addition, FIH knock-down increased the expression of mesenchymal markers (vimentin, -SMA) but reduced that of epithelial markers (E-cadherin, - catenin), with little change in the expression of N-cadherin or ZEB-1. However, these effects of FIH knock-down were reversed by G9a/GLP knock-down and were not substantially observed under hypoxia (Supplementary Fig. S6E). We next explored whether cell invasion is regulated via the FIH- G9a/GLP axis in OvCa cells. FIH knock-down or hypoxia promoted OvCa cell invasion across a Matrigel-coated membrane, which was inhibited by co-silencing G9a and GLP (Fig. 4D; Supplementary Fig. S7A and S7B). Overexpression of wild-type G9a or GLP enhanced cell invasion, which was augmented by FIH inhibition using siRNA or DMOG. Interestingly, NQ-mutated G9a or GLP stimulated cell invasion to a greater extent; this effect was not enhanced in OvCa cells by FIH inhibition (Fig. 4E; Supplementary Fig. S7C and S7D) or FIH knock-down (Supplementary Fig. S7E and S7F). Hence, our results suggest that FIH controls G9a/GLP-mediated EMT and invasiveness of OvCa cells in an O2-dependent manner.
next checked the expression of G9a/GLP-controlled cell adhesion genes, such as CDH1, MASPIN and DSC3 (39,40). Repression of CDH1 (E-cadherin gene) is a hallmark of EMT. MASPIN (mammary serine protease inhibitor) suppresses the invasion and metastasis of cancer cells, and DSC3 (desmocollin 3) is a component of intercellular desmosome cell-cell junctions. As expected, the mRNA levels of CDH1, MASPIN and DSC3 were decreased under hypoxia. FIH knock-down reduced their levels under normoxia, which did not occur under hypoxia (Fig. 4F; Supplementary Fig. S8A and S8B). G9a/GLP knock-down and UNC0638 confirmed that these genes are negatively regulated by G9a and GLP. Ectopic expression of G9a or GLP resulted in repression of these genes, which was augmented by DMOG (Fig. 4G; Supplementary Fig. S8C and S8D). Compared with wild-type G9a and GLP, NQ-mutants exerted a greater effect on repression of these genes, but this was not enhanced by FIH inhibition. Based on ChIP analyses, the recruitment of G9a or GLP to the CDH1 promoter was enhanced by FIH inhibition (Fig. 4H). Concomitantly, the H3K9me2 level at the CDH1 promoter was increased under hypoxia or by FIH knock-down (Fig. 4I). Since FIH is already inactivated under hypoxia, FIH knock-down and hypoxia may not show an additive effect. The recruitment of wild-type G9a or GLP to the CDH1 promoter was also facilitated by FIH inhibition. Notably, NQ-mutants showed a greater potential for binding the CDH1 promoter than wild-type G9a/GLP, but did not respond to FIH inhibition (Fig. 4J). These results suggest that binding of G9a and GLP to their target genes is restricted under normoxia due to FIH-mediated Asn-hydroxylation, but this restriction is absent under hypoxia.
We examined the growth and metastasis of OvCa xenografts by injecting cancer cells into mouse peritoneum, as previously described (41). We established SKOV-3 and ES-2 cell lines stably co- expressing luciferase and G9a, and luciferase activity and G9a expression were verified (Supplementary Fig. S9A and S9B). Based on the knock-down efficiencies of five shRNAs targeting different sites of FIH mRNA (Supplementary Fig. S9C), the 4th shFIH lentivirus was transduced into the stable cell lines. On the next day, the cells were intraperitoneally injected into female athymic nude mice, and the mice were sacrificed 8 weeks post-injection. SKOV-3-bearing mice in the G9a- WT/shFIH, G9a-NQ/shEGFP, and G9a-NQ/shFIH groups showed significant weight loss, whereas ES-2-bearing mice in these groups showed weight gain due to massive ascites (Supplementary Fig. S9D). The volume of aspirated ascites was larger in the ES-2 group than in the SKOV-3 group (Supplementary Fig. S9E). Because ES-2 tumors spread out more quickly than SKOV-3 tumors, the ES-2 xenograft study was terminated on the 28th day after cell injection. Bioluminescence imaging analyses revealed that peritoneal dissemination was more severe in G9a-WT/shEGFP tumors than in shEGFP control tumors. In both tumors, G9a-WT/shFIH tumors exhibited greater expansion in the peritoneum than G9a-WT/shEGFP tumors. G9a-NQ tumors also spread out rapidly in the peritoneum irrespective of FIH expression (Fig. 5A and 5B). Bioluminescence images of all mice are shown in G9a-NQ/shFIH group died more early than those in the shEGFP or G9a-WT/shEGFP group (Fig. 5C). None of mice died in the SKOV-3 xenograft groups until experiments were terminated. After the abdominal skin was removed, ascites could be observed in the G9a-WT/shFIH, G9a-NQ/shEGFP, and G9a-NQ/shFIH groups (Fig. 5D). In these groups, large solid tumors (> 1 cm3) were adherent to the omentum, intestines and pelvic fat, and small tumors (< 0.25 cm3) were found in the posterior peritoneal walls (Fig. 5D). To examine cancer metastasis, ovaries, kidneys, spleens, lungs, liver, and hearts were isolated (Fig. 5E). OvCa cells metastasized more frequently in the G9a-WT/shEGFP group than in the shEGFP control group, and this was promoted by FIH knock-down and G9a-NQ overexpression (Fig. 5F). Therefore, FIH likely prevents peritoneal dissemination and metastasis of OvCa cells through Asn-hydroxylation of G9a under normoxia. Therefore, when FIH is inactivated under hypoxia, activation of G9a and GLP may promote cancer metastasis. The FIH, G9a, GLP, and E-cadherin levels in human ovarian cancer specimens, of which clinical information is summarized in Supplementary Table 3, were analyzed immunohistochemically (Fig. 6A). G9a expression was correlated positively, and FIH and E-cadherin expression negatively, with increasing malignancy grade (Fig. 6B). FIH and G9a expression was reciprocally associated with E- cadherin expression (Fig. 6C). Next, we investigated whether the FIH-G9a axis is related to the clinical outcomes of ovarian cancer. As expected, FIH downregulation and G9a upregulation were associated with a poor survival rate (Fig. 6D). GLP expression was also correlated with poor survival, albeit not significantly so. This clinical evidence supports the animal results in terms of demonstrating the effect of FIH against G9a-induced peritoneal metastasis of OvCa. To further evaluate the relationship between FIH-mediated Asn-hydroxylation of G9a/GLP and promotion of cancer development, we searched for G9a/GLP mutations in the Catalogue of Somatic Mutations in Cancer (COSMIC) database (42). However, no mutation was found at N779 of G9a or at N867 of GLP. Instead, several amino acid residues near the hydroxylated site were mutated in lung and liver cancers (Supplementary Fig. S12A). We tested whether the mutations affect the FIH- mediated Asn-hydroxylation of G9a/GLP. GLP_Q869H was hydroxylated by FIH, but G9a_D782G, GLP_D865V, and GLP_V866I were not (Fig. 6E; Supplementary Fig. S12B). Furthermore, based on the levels of H3K9me2/me3, FIH regulation of mutated G9a and GLP was examined in transfected cells. As was shown with N779Q-G9a, H3K9 was constitutively methylated by D782G-G9a, D865V- GLP and V866I-GLP H3K9, but not by Q869H-GLP, irrespective of FIH expression (Fig. 6F). Toward this end, we determined whether such mutations affected the malignant behavior of cancer cells. In Matrigel invasion analyses, the invasive potential of SKOV-3 cells was augmented under normoxia by ectopic expression of D782G-G9a, D865V-GLP or V866I-GLP (Fig. 6G; Supplementary Fig. S12C). These results suggest that some cancer cells exhibit increased G9a or GLP expression due to point mutations in the FIH target motif, which increases their capacity for invasion. DISCUSSION Cells express diverse rescue genes for adaptation to hypoxic stress, in which HIF-1/2 play crucial roles. In contrast, overall levels of the repressive marks H3K9me2/me3 are also increased during hypoxia, which implies that numerous genes other than HIF target genes are epigenetically repressed. This work was initiated by the following question: which factor is responsible for hypoxia-induced epigenetic repression? We here revealed for the first time that G9a and GLP are substrates for FIH. Under normoxia, FIH hydroxylates G9a/GLP at specific asparagine residues within ARDs in an oxygen-dependent manner. Then, the hydroxylated G9a/GLP cannot further methylate H3K9me2/me3 because the hydrophobic pocket for holding H3K9 is deformed. Under hypoxia, however, G9a and GLP escape from Asn-hydroxylation, leading to epigenetic repression by generating H3K9me2/me3. The hypoxic induction of G9a protein is controversial. A few research groups argued that G9a is upregulated at the post-translational level under hypoxia (23,43), but other groups showed that G9a expression is constant regardless of the O2 level (44). In fact, we could not observe any change in G9a and GLP expression under moderately and severely hypoxic conditions in all experiments. Moreover, a recent report suggested that PHD1 hydroxylates G9a under normoxia and this promotes the polyubiquitination and proteasomal degradation of G9a (45). In our experimental conditions, however, PHD1-3 knockdown failed to increase G9a and GLP protein levels under normoxia, while it clearly stabilizes HIF-1 or HIF-2 (Fig.1D). The oxygen-dependent degradation of G9a remains to be clarified. G9a and GLP are generally known to catalyze mono- and di-methylation, rather than tri- methylation, of H3K9. However, the H3K9 tri-methylation was substantially blocked by co- knockdown of G9a and GLP or by a G9a/GLP inhibitor (Fig. 1F; Supplementary Fig. S2E), which suggests that G9a and GLP largely contribute to the H3K9 tri-methylation. We considered two possible involvements of G9a and GLP in H3K9 tri-methylation; the enzymes directly methylate H3K9me2 or they indirectly enhance the H3K9me3 level by increasing the H3K9me2 level. To further investigate such roles of G9a and GLP, we performed the in vitro assay for H3K9 methylation, and found that H3K9me3 as well as H3K9me1/2 can be generated by purified G9a or GLP peptide (Supplementary Fig. S5B). This strongly indicates that G9a and GLP can catalyze the tri-methylation of H3K9. Indeed, the H3K9me3 generation by G9a and GLP has been also demonstrated by other research (46-49). In addition to the FIH-G9a/GLP axis, the Jumonji C domain-containing histone demethylases (JHDMs) are thought to take part in oxygen-dependent chromatin remodeling. JHDMs, which belong to the superfamily of 2-oxoglutarate and Fe2+-dependent dioxygenases, hydroxylate the methyl group in a lysine residue of histones, and in turn the oxidized methyl is released as formaldehyde (50). As this hydroxylation occurs O2-dependently, hypoxia-induced hypermethylation of histones was suggested to arise from the inactivation of JHDMs (10). If so, H3K9 methylation under hypoxia may be determined bidirectionally by G9a/GLP activation and JHDMs inhibition. However, how the two events work cooperatively in oxygen-dependent epigenetic regulation is an open question. FIH and HIF-1 are localized in the cytoplasm and the nucleus, respectively. Nonetheless, FIH can hydroxylate HIF-1 in the cytoplasm before synthesized HIF-1 enters into the nucleus (51). Likewise, the Asn-hydroxylation of G9a/GLP seems to occur while they are transiently in the cytoplasm prior to nuclear translocation. Although the nuclear-cytoplasmic shuttling of G9a/GLP has not been demonstrated, it was reported that substantial amounts of G9a and its splicing variant are present in the cytoplasm (52). In addition, a proteomic analysis revealed that many non-histone proteins in the cytoplasm are methylated by G9a/GLP (53), which indirectly supports the cytoplasmic location of G9a/GLP. Indeed, we also detected a small but significant portion of G9a/GLP in the cytoplasmic fraction and identified their interactions with FIH (Fig. 2C). Therefore, it is suggested that G9a/GLP are targeted by FIH in the cytoplasm and the hydroxylated G9a/GLP translocate to the nucleus. The FIH regulation of G9a/GLP may alter the expression of diverse genes. In this study, we focused on their roles in hypoxia-induced cancer metastasis. Indeed, G9a/GLP exert pro-metastatic effects in several types of cancers by inducing the EMT. The findings reported herein suggest that the FIH-G9a/GLP axis determines EMT induction and cell invasion by epigenetically repressing the metastasis suppressor genes CDH1, DSC3, and MASPIN. Furthermore, a xenograft assay showed that G9a/GLP activation by FIH inhibition promoted ovarian cancer metastasis, and a large-scale genome analysis suggested that FIH and G9a reciprocally regulated anti-metastatic gene sets. Therefore, FIH and G9a/GLP may be potential targets for preventing cancer metastasis. Although G9a and GLP are similar in structural and functional aspects, G9a appears to be more closely associated with carcinogenesis and cancer progression than GLP. Indeed, overexpression and gene amplification of G9a were observed in a number of human cancers, but GLP was observed in only a few cases (54). In addition, G9a, not GLP, was often evaluated as a marker of aggressive cancers (55-57). For these reasons, we performed an in vivo metastasis experiment using ovarian cancer cell lines overexpressing G9a or its NQ-mutant, and successfully showed the involvement of the FIH/G9a axis in cancer metastasis (Fig. 5). However, we do not mean that the FIH/GLP axis is not critical for ovarian cancer metastasis because GLP also repressed adhesion-related genes and promoted cell invasion in this study (Fig. 4 and Supplementary Fig. S6); whether or not G9a and GLP play distinct roles in metastasis remains to be investigated. Interestingly, a recent publication showed that FIH hydroxylates the Asn22 residue of the This hydroxylation restricts the OTUB1 interaction with cell metabolism-related proteins (30), which implicates an oxygen-dependent regulation of cell metabolism. Furthermore, OTUB1 also promotes the progression of ovarian cancer by deubiquitinating FOXM1 (58). Taken together, FIH may be responsible for the hypoxia-induced ovarian cancer progression by hydroxylating and controlling multiple tumor-promoting proteins. In conclusion, this study provides insight into a novel function of FIH as a key upstream regulator of oxygen-dependent epigenetic regulation. Interestingly, the hypoxia-induced inactivation of FIH plays two distinct roles in gene regulation—induction of hypoxia-adaptive genes by activating HIF- 1/2 and repression of metastasis suppressor MS8709 genes by activating G9a/GLP. Therefore, FIH appears to act as a control that directs gene expression according to the oxygen level.