Identification of the Hypoxia-inducible Factor 2α Nuclear Interactome in Melanoma Cells Reveals Master Proteins Involved in Melanoma Development*

Hypoxia-inducible factors (HIFs) are heterodimeric transcription factors that play a key role in cellular adaptation to hypoxia. HIF proteins are composed of an α subunit regulated by oxygen pressure (essentially HIF1α or HIF2α) and a constitutively expressed β subunit. These proteins are often overexpressed in cancer cells, and HIF overexpression frequently correlates with poor prognosis, making HIF proteins promising therapeutic targets. HIF proteins are involved in melanoma initiation and progression; however, the specific function of HIF2 in melanoma has not yet been studied comprehensively. Identifying protein complexes is a valuable way to uncover protein function, and affinity purification coupled with mass spectrometry and label-free quantification is a reliable method for this approach. We therefore applied quantitative interaction proteomics to identify exhaustively the nuclear complexes containing HIF2α in a human melanoma cell line, 501mel. We report, for the first time, a high-throughput analysis of the interactome of an HIF subunit. Seventy proteins were identified that interact with HIF2α, including some well-known HIF partners and some new interactors. The new HIF2α partners microphthalmia-associated transcription factor, SOX10, and AP2α, which are master actors of melanoma development, were confirmed via co-immunoprecipitation experiments. Their ability to bind to HIF1α was also tested: microphthalmia-associated transcription factor and SOX10 were confirmed as HIF1α partners, but the transcription factor AP2α was not. AP2α expression correlates with low invasive capacities. Interestingly, we demonstrated that when HIF2α was overexpressed, only cells expressing large amounts of AP2α exhibited decreased invasive capacities in hypoxia relative to normoxia. The simultaneous presence of both transcription factors therefore reduces cells' invasive properties. Knowledge of the HIF2α interactome is thus a useful resource for investigating the general mechanisms of HIF function and regulation, and here we reveal unexpected, distinct roles for the HIF1 and HIF2 isoforms in melanoma progression.

Hypoxia-inducible transcription factors (HIFs) 1 are central mediators of cellular adaptation to hypoxia that control the expression of genes involved in anaerobic metabolism, intracellular pH, angiogenesis, and cell growth and survival (1). HIF proteins are heterodimers consisting of an oxygen-regulated ␣ subunit and a constitutively expressed ␤ subunit (also referred to as the arylhydrocarbon receptor nuclear translocator, or ARNT). Three isoforms have been described for each subunit; most HIF function, however, has been attributed to HIF1␣, HIF2␣, and HIF1␤ (2). HIF1␣ is detected in almost all tissues, whereas HIF2␣ (also known as EPAS1, for endothelial PAS protein 1) is restricted to certain tissues. HIF2 is expressed notably in hypoxic endothelium but also in distinct cell populations in the brain, liver, kidney, lung, and intestine (3). HIF␣ proteins are regulated post-translationally as a function of the partial pressure of oxygen. In normoxic conditions, the protein is rapidly hydroxylated by prolyl hydroxylases and then recognized by the von Hippel-Lindau protein, a component of an E3-ubiquitin ligase complex that targets the ␣ subunit for degradation by the proteasome. When the partial pressure of oxygen is low, prolyl hydroxylase activity is inhibited, allowing ␣ subunit accumulation (4).
In the past decade, HIF protein expression has been examined extensively in tumor cells. Many immunohistochemical studies have clearly demonstrated high expression of HIF1␣ and/or HIF2␣ in many types of human primary tumors (5,6).
Moreover, numerous recent studies have revealed a significant correlation between HIF␣ subunit expression in human tumors and a poor prognosis (1). Other studies, however, have demonstrated distinct effects of the two HIF␣ subunits in some tumor cell types. For instance, in renal cell carcinoma models, HIF1␣ expression has been shown to decrease tumor size in a nude mouse xenograft model, whereas HIF2␣ expression has the opposite effect (7). Therefore, despite similarities in their structure and regulation, the two HIF␣ subunits display comparable but not identical behaviors in physiological and physiopathological conditions (8,9).
Malignant melanoma is a cancer arising from pigmentproducing cells called melanocytes, present predominantly in the skin. It is a very aggressive form of cancer that accounts for almost 80% of the casualties associated with skin cancer, and its incidence is currently growing faster than that of any other cancer worldwide, especially among young adults. Melanoma is treatable by surgical resection when caught in the early stages, yet it still causes tens of thousands of deaths each year worldwide. Although genetic and epigenetic alterations contribute to tumorigenesis (10), various stimuli coming from the tumor microenvironment appear to be required for the development of a malignant phenotype (11). Among the different elements of the melanoma microenvironment, oxygen availability is of major importance. Both HIF1 and HIF2 are up-regulated in melanoma cells (12,13). Although the importance of HIF1 in melanocyte survival and proliferation in hypoxia (14) and in melanoma carcinogenesis (15,16), progression (17), and invasion (18,19) has been studied extensively, the function of HIF2␣ in melanoma still remains unclear.
Protein-protein interactions play a key role in modulating protein functions. Thus, the identification of a protein's interactome might be a relevant way to uncover the role of proteins within the cell. Mass-spectrometry-based proteomics is the method of choice for addressing this issue in a systematic and relatively unbiased manner. Several strategies may be applied, including affinity purification with one or several tags, with or without prior cross-linking, combined or not with quantitative spectrometry (20 -22).
Here, we report an extensive study of HIF2␣ protein partners in the 501mel melanoma cell line using affinity purification and quantitative mass spectrometry. With this approach, we have described, for the first time, the global nuclear interactome of one HIF subunit. Among the HIF2␣ partners that we found quantitatively enriched in hypoxic melanoma cells, certain proteins relevant to melanoma biology were confirmed as HIF2␣ partners, leading to a better knowledge of melanoma biology.

EXPERIMENTAL PROCEDURES
Cell Culture-Melanoma cell lines (provided by Dr. Lionel Larue, Institut Curie, Orsay, France) were grown in RPMI 1640 (Invitrogen) containing 10% fetal calf serum (FCS) (Invitrogen), 50 g/ml penicillin-streptomycin mix (Invitrogen), and 500 g/ml hygromycin in order to select stably transfected cell lines. All cell lines were used within two months after the resuscitation of frozen aliquots and were authenticated on the basis of viability, recovery, growth, and morphology.
HIF␣ expression was induced by incubating cells for 12 h in a hypoxic chamber (0.1% O 2 ) or by adding 125 M CoCl 2 (Sigma), a hypoxia-mimicking reagent, to the cell culture medium for 12 h. For longer hypoxic assays, mild hypoxia was used (1% O 2 in a hypoxic chamber or 25 M CoCl 2 ). Cell number, viability, and proliferation rate were evaluated either simply by counting or by quantifying metabolically active cells (MTT assay). Briefly, 10,000 cells were plated in 0.1 ml of medium in 96-well plates. At the indicated times, 100 l of 1 mg/ml MTT (Sigma) solution in PBS was added to each well for 2 h. After removal of the medium, 100 l of DMSO was added to induce cell lysis and dissolve formazan crystals. The absorbance at 570 nm was determined using a microplate spectrophotometer (Quant, Bio-Tek, Winooski, VT).
Plasmid Construction and Establishment of Stable Cell Lines-The full-length coding region of human HIF2␣ or HIF1␣ (gifts from Dr. Richard Bruick of UT Southwestern Medical Center, Dallas, TX, and from Dr. Amato Giaccia of Stanford University, Stanford, CA, respectively) was cloned in-frame with a linker encoding an HA tag in a pCMV-hygro 2xFLAG expression vector (supplemental Figs. S1 and S6). The resulting vectors were transfected into 501mel cells, and stable transfectants were obtained after four weeks of hygromycin selection. A control cell line (Mock) was produced after transfection of a 501mel cell line with the empty pCMV-hygro 2xFLAG vector. For each construct, single clones were obtained by limiting dilution and were genotyped via PCR.
Immunoaffinity Purification for Mass Spectrometry Analysis-Nuclear extracts from stably transfected 501mel cells were prepared as described elsewhere (23), with minor modifications: before hypertonic treatment, nuclei were incubated for 10 min in buffer B (20 mM Tris-HCl pH 7.5, 5 mM MgCl 2 , 0.1 mM EDTA, 0.5 M CaCl 2, complete protease inhibitor mixture (Roche Applied Science), and 0.5 mM Pefabloc in the presence of PSC protector solution (Roche Applied Science)). After centrifugation, nuclear proteins were extracted by incubating the pellet in the same buffer containing 400 mM NaCl and 10% glycerol for 30 min at 4°C. The protein solution was concentrated via ultrafiltration (Vivaspin10, Vivascience Sartorius, Stonehouse, UK) at 4°C. Nuclear extracts were loaded onto a 4-ml 10%-50% glycerol gradient in buffer B containing 400 mM NaCl and subsequently were centrifuged at 200,000g for 4 h at 4°C. The glycerol gradients were then divided into 16 fractions of 0.25 ml and tested for HIF2␣ via immunoblotting. Fractions 2-8, which contained HIF2␣ protein, were pooled. The salt concentration was adjusted to 200 mM NaCl, and HIF2␣ protein complexes were purified via immunoprecipitation using anti-FlagM2 antibody-conjugated agarose beads (Sigma-Aldrich, France). Immunopurified complexes were washed twice with IP0.2 buffer (20 mM Tris-HCl pH 7.5, 200 mM NaCl, 0.5 mM MgCl 2 , 0.5% Nonidet P-40, 10 mM ␤-mercaptoethanol, and protease inhibitor mixture) and twice in IP0.4 (IP0.2 containing 400 mM NaCl) and eluted with a solution of 500 g/ml 3xFlag peptide in IP0.4 buffer (Sigma-Aldrich, France). As a negative control, the same purification was undertaken starting from the Mock cell line. Samples from HIF2␣ and Mock cell lines were separated by means of SDS-PAGE and submitted for proteomic analysis.
One-dimensional SDS-PAGE Fractionation, In-gel Digestion, and Nano-liquid Chromatography-electrospray Ionization LTQ-Orbitrap MS/MS Analysis-Before SDS-PAGE separation, proteins were reduced with 25 mM DTT in 25 mM Tris-HCl pH 6.8, 5% glycerol, and 1% SDS for 30 min at 56°C. Then, proteins were alkylated by the addition of 90 mM iodoacetamide for 30 min at room temperature in the dark. After protein separation via Nu-PAGE on a precast gradient 4%-12% acrylamide gel followed by Coomassie Blue staining, gel lanes were cut into 10 slices and incubated 3-fold on a shaker for 15 min at 37°C in 100 mM ammonium bicarbonate in 50% acetonitrile (ACN). The gel slices were then vacuum dried and rehydrated with 40 l modified sequencing-grade trypsin solution (10 ng/l; Promega, Madison, WI) in 50 mM ammonium bicarbonate for in-gel tryptic digestion. After absorption of the trypsin solution, 100 l of 50 mM ammonium bicarbonate was added. Trypsin digestion was performed overnight at 37°C. The resulting peptides were extracted from the gel slices through three incubations (15 min at 37°C) with shaking: a first incubation in 50 mM ammonium bicarbonate, and two subsequent incubations in 10% formic acid and ACN (1:1). The three successive extractions were pooled with the initial digestion supernatant, vacuum dried in a SpeedVac, and re-suspended in 14 l of 2% ACN and 0.05% TFA.
Peptide mixtures were analyzed via nano-LC-MS/MS using an Ultimate3000 system (Dionex, Amsterdam, The Netherlands) coupled to an LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) operating in positive mode. Five microliters of each sample were loaded onto a C 18 precolumn (300 m inner diameter ϫ 5 mm; Dionex) at 20 l/min in 2% ACN and 0.05% TFA. After 5 min of desalting, the precolumn was switched online with the analytical C 18 column (75 m inner diameter ϫ 15 cm; PepMap C 18 , Dionex) equilibrated in 100% solvent A (5% ACN, 0.2% formic acid) and 0% solvent B (80% ACN, 0.2% formic acid). Peptides were eluted by using a 0%-50% gradient of solvent B for 60 min at a flow rate of 300 nl/min. The LTQ-Orbitrap XL was operated in data-dependent acquisition mode with Xcalibur software (version 2.0 SR2, Thermo Fisher Scientific). Survey scan MS spectra were acquired in the Orbitrap in the 300 -2000 m/z range with the resolution set to a value of 60,000 at m/z 400. Up to five of the most intense multiply charged ions (2 ϩ and 3 ϩ ) per survey scan were selected for collisioninduced dissociation fragmentation, and the resulting fragments were analyzed in the linear trap (LTQ). Dynamic exclusion was used within 60 s to prevent repetitive selection of the same peptide.
Database Search and Data Analysis-Mascot Daemon software (version 2.3.2, Matrix Science, London, UK) was used to perform database searches in batch mode with all of the raw files acquired for each sample. To extract peak lists automatically from Xcalibur raw files, the Extract msn.exe macro provided with Xcalibur was used through the Mascot Daemon interface. The following parameters were set for the creation of the peak lists: parent ions in the mass range of 300 -4500, no grouping of MS/MS scans, and no threshold. A peak list was created for each analyzed fraction (i.e. gel slice), and Mascot searches were performed for each fraction. Data were searched against all entries in the International Protein Index human protein database (v3.86, released September 7, 2011; 91,522 sequences). Cysteine carbamidomethylation was set as a fixed modification; methionine oxidation and protein N-terminal acetylation were set as variable modifications for all Mascot searches. Proteins were digested by trypsin, and two missed trypsin cleavage sites were allowed. The mass tolerances in MS and MS/MS were set at 10 ppm and 0.6 Da, respectively, and the instrument setting was specified as "ESI-Trap." To calculate the false discovery rate (FDR), the search was performed using the "decoy" option in Mascot. Mascot results were parsed and quantified with the in-house-developed software Mascot File Parsing and Quantification (MFPaQ), version 4.0.0 (24). Peptide identifications extracted from Mascot result files were validated at a final peptide FDR of 1 (at least one peptide with a score of 40, or at least two peptides with a score of 20). The FDR was calculated at the protein level (FDR ϭ number of validated decoy hits/number of validated target hits ϫ 100). If the same set of peptides belongs to several proteins, the first protein listed is the one that contains the most peptides. Unambiguous protein identification is provided by a proteotypic peptide (i.e. a peptide with a unique amino acid sequence among all proteins). Known contaminants excluded were keratins and trypsin.
Label-free Quantification-The quantification of proteins was performed using the label-free module implemented in the MFPaQ v4.0.0 software (25). Briefly, for each sample, the software used the validated identification results and extract ion chromatograms of the identified peptide ions in the corresponding raw nano-LC/MS files, based on their experimentally measured retention time and monoisotopic m/z values. In order to perform protein relative quantification in different samples, a protein abundance index (PAI) was calculated and defined as the average of the extract ion chromatogram area values from the three most intense peptides identified for the protein in Mock (501mel cell line stably transfected with an empty vector) and H2 (501mel cell line stably transfected with an HIF2␣ vector) samples. The integration of quantitative data across the fractions was performed by summing the PAI values for fractions adjacent to the fraction with the best PAI (the same three consecutive fractions for all the samples to be compared). Four biological replicates were analyzed in order to minimize potential experimental deviations. Proteins were subsequently screened for the following: (1) known nuclear localization, (2) proteins detected in at least two independent experiments with FDRs lower than 1%, and (3) proteins enriched more than 4-fold in H2 relative to Mock assays (see supplemental Fig. S3 and Table I). The retained proteins were further analyzed for gene ontology and protein interaction network based on literature investigations and Ingenuity Pathway Analysis.
Alternatively, 5 g of antibody (supplemental Table S3) was incubated with the protein extract for 3 h at 4°C with rotation prior to the addition of 40 l of protein G Plus agarose beads (Santa Cruz Biotechnology, Santa Cruz, CA). The mixture was incubated for 1 h at 4°C with rotation to capture immune complexes, and then the beads were pelleted via brief centrifugation and washed three times with WIP buffer. Bound proteins were eluted by means of boiling with 50 l of SDS-PAGE loading buffer. Protein content was analyzed via Western blot (antibodies are listed in supplemental Table S3).
Fluorescence Microscopy of the Actin Cytoskeleton-Cells were grown on coverslips for at least 16 h and were subsequently washed, fixed for 15 min in 3.7% paraformaldehyde, permeabilized, blocked, and incubated for 2 h with rhodamine-conjugated phalloidin (Molecular Bioprobes, Eugene, Oregon, USA) in phosphate-buffered saline (PBS) containing 1% BSA (PBS-BSA). Subsequently, the coverslips were washed with PBS-BSA, air dried, and embedded in Dako Cytomation Fluorescent Mounting Medium. Fluorescence was detected using a Leica microscope (Leitz DMRB; 100ϫ objective). Images were acquired as sets of color images and prepared using Photoshop software.
Cell Migration and Invasion Assays-Equal numbers of cells (1 ϫ 10 5 ) in serum-free RPMI were added to the upper compartments of the chamber (ThinCerts ® , 12 wells, 8-m pores (Greiner Bio-One, Frickenhausen, Germany)) and kept in an incubator at 37°C for 20 h. As a chemoattractant, the lower compartment contained RPMI supplemented with 10% FCS. At the end of the incubation time, cells from the upper surface of the filter were wiped off with a cotton swab. The lower surface of the filter was stained with Toluidine Blue O (1% w/v, Sigma) in 1% sodium tetraborate. Cells that had migrated to the bottom of the chamber were quantified via optical density measurement (570 nm) after solubilization in lysis buffer (6.25 mM Tris-HCl pH 6.8, 10% glycerol, 2% SDS, 5% ␤-mercaptoethanol). At least three independent sets of experiments were carried out, each in duplicate. For Matrigel assays, the filter separating the two compartments was pre-coated with 50 l of 0.125 g/l Matrigel (BD Biosciences).
Statistical Analyses-The values reported in the tables and figures are means Ϯ S.E. of the mean. The percentages of melanoma cells in the migration and invasion assays were compared between Mock-, HIF1␣-, and HIF2␣-transfected cells using Student's t test.

RESULTS
Identification of HIF2␣-interacting Proteins-501mel cells express the HIF2␣ subunit when cultured under hypoxic conditions or in the presence of the hypoxia-mimicking agent cobalt chloride (supplemental Fig. S1D). However, to facilitate HIF2␣-interacting protein purification, we chose to stably transfect this human melanoma cell line with a plasmid construction that allowed the expression of a Flag/HA fusion protein. This epitope-tagging strategy allows a generic purification procedure that results in cleaner samples and a higher yield than traditional immunoprecipitation experiments (22). Transient interactions between transcription factors and most of their partners make the identification of this type of interactome harder (22). To address this problem, we chose to perform a one-step Flag-affinity-based protein purification protocol rather than tandem affinity purification, and we combined this with quantitative MS analysis. To discriminate between genuine partners and contaminants, label-free quantitative MS analysis was performed on immunoprecipitates of Flag-HIF2␣ prepared from stably transfected cells and from a control cell line that does not express the tagged bait. The label-free MS method allows the identification of contaminants, which are present in similar amounts in both control and Flag-HIF2␣-expressing cells, and their differentiation from true interactors, which are significantly enriched in Flag-HIF2␣ expressing cells.
The expression, correct functionality, and nuclear location of exogenous tagged HIF2␣ protein were verified via Western blot analysis, reporter gene assay, and immunofluorescence microscopy (supplemental Fig. S1). One clone (named H2) exhibiting a high total HIF2␣ expression (about 10-fold higher than in the Mock cell line, which was stably transfected with the empty vector) was selected. Exogenous tagged HIF2␣ protein overexpression partially bypasses the proteasomal degradation capacity of the cell; therefore, the recombinant protein was also detected in normoxic conditions. As expected, higher amounts of the recombinant HIF2␣ protein were produced under hypoxic conditions (supplemental Fig.  S1D). Accordingly, as normoxic and hypoxic HIF proteins present different functional characteristics, we performed coimmunoprecipitation experiments under hypoxia.
The strategy followed in order to identify nuclear proteins associated with HIF2␣ is presented in Fig. 1A. Nuclear extracts were prepared from H2 cells cultured in hypoxia-mimicking conditions, and soluble nuclear protein complexes were separated using a glycerol gradient to reduce sample complexity. The HIF2␣ content of the different fractions of the glycerol gradient was analyzed via Western blotting using an anti-HIF2␣ antibody (Fig. 1B). HIF2␣-positive fractions were pooled (fractions 2 to 8). Immunoprecipitation of HIF2␣ with its protein partners was then performed using an anti-Flag monoclonal antibody, and the proteins were subsequently eluted by Flag peptide competition. Nuclear extracts from the Mock cell line were treated similarly and used as a negative control. The Flag-tagged protein was not expressed in this cell line. HIF2␣ was identified in the H2-immunopurification (IP) but not in the Mock-IP control lane (Fig. 1C). Aliquots of the IP fractions were loaded for SDS-PAGE and silver stained. A protein band that might correspond to HIF2␣ and specifically enriched protein bands corresponding to HIF2 partners were observed in the H2-IP lane (Fig. 1D). The low background in the Mock-IP samples and the presence of specific bands in the H2-IP samples indicate the specificity of the IP protocol. The remaining IP fractions were subjected to analytical SDS-PAGE. The gel was cut into slices and subsequently treated with trypsin. Peptides were extracted and analyzed via nano-LC-MS/MS on an LTQ-Orbitrap-XL. The identification of peptides was performed using the Mascot Daemon software for searching databases, combined with a label-free protein quantification analysis with in-house-developed MFPaQ software (24). Prior to quantification, a list of proteins exhibiting an FDR lower than 1% was automatically validated. Protein quantification was then performed using the listed proteins. Label-free quantification is based on a comparison of the PAI of each protein in two different conditions. The PAI (average of area under the curve) is calculated from the three most intense peptides of each protein. To minimize biological and analytical variations, four independent experiments were performed, leading to the identification of, on average, 624 proteins (643, 597, 663, and 592 proteins, respectively, for the first, second, third, and fourth experiments). The reproducibility among these four experiments was evaluated using a four-way Venn diagram that compared the proteins identified in each biological replicate (supplemental Fig. S2A). Of these 624 proteins, 57% were common among the four experiments. Multiple-step filtering was performed to retain highconfidence interacting partners. First, proteins lacking nuclear gene term ontology were excluded. Then, proteins had to be detected in at least two independent experiments with FDRs lower than 1%. Additionally, proteins had to display a greater than 4-fold enrichment between H2 and Mock assays (supplemental Fig. S3 and Table I). This quality-control filtering validated 70 proteins as specific direct or indirect HIF2␣interacting proteins (HIF2␣ ϭ EPAS1; Table I). Of the selected proteins, 71% were encountered in all experiments (supplemental Fig. S2B), with more than 87% being identified in at least three of them.
The identified HIF2␣ protein partners were grouped according to their biological function based on literature investigations and function analysis using Ingenuity Pathway Analysis from Ingenuity. Proteins belonged to diverse biological categories and were known to participate in various cellular functions and activities, including transcription regulation (n ϭ 51), nuclear cytoskeleton organization, mitosis and nucleocytoplasmic transport (n ϭ 4), ubiquitin-dependent protein catabolism (n ϭ 6), and protein modification (n ϭ 3) (Fig. 2A). Thirteen proteins found in the present study had been previously reported to interact directly or indirectly with HIF2␣, HIF1␣, or HIF1␤ (ARNT1) (supplemental Table S1). Moreover, peptides belonging to several previously characterized partners of HIF subunits were also detected in our conditions but did not fulfill all the screening criteria. For example, peptides belonging to HDAC1-3, Hsp90 (HSP90A), and Cullin B (CUL2), well-known interacting partners of HIF␣ subunits, were detected in only one of four experiments or were insufficiently enriched relative to the control (Ͻ4-fold enriched) (supplemental Table S1). These proteins therefore did not meet our stringent criteria and could not be firmly assigned as HIF2␣ partners. A functional network analysis was performed to test for possible complex formation between several HIF2␣ partners (Fig. 2C). Among the 70 nuclear protein partners of HIF2␣ identified in the present study, 60 were already known to interact directly with at least one other protein on the list, and 182 direct interactions were revealed in this protein interaction network. Among these interactions, some well-known complexes such as mediator or SMARC complexes were detected (Fig. 2C). Indeed, 20 proteins belonging to the multiprotein mediator complex (27) were identified as HIF2␣ partners (Table I; Figs. 2B and 2C). Almost all of the subunits thought to be part of the mediator complex were identified, which is remarkable, as these proteins are usually found in small amounts in cells (28). MED21 was the only missing subunit; this might be explained by its relatively small size (144 residues). Also, the two subunits recovered with the lowest scores, MED18 and MED20, are not part of the mediator core complex and therefore are more labile (29) (supplemental Fig. S4A). Similarly, subunits involved in SWI/SNF (SMARC) complex formation (30), and thus in chromatin remodeling, have also been shown to be HIF2␣ protein partners. One of these SMARC complexes has been entirely identified in our study: BRM/BRG1 complex (BRM ϭ SMARCA2, BRG1 ϭ SMARCA4; Figs. 2B and 2C, supplemental Fig. S4B). CBP (CREBBP) and P300 (EP300) were also identified as HIF2␣ interacting partners. These proteins, which exhibit histone acetyltransferase activity, are major partners of FIG. 1. Purification of HIF2␣-interacting proteins. A, schematic presentation of the strategy used to identify the HIF2␣ nuclear interactome in 501mel cells. B, nuclear extracts from 501mel cells expressing Flag/HA-tagged HIF2␣ (H2) or from Mock cells, cultured with CoCl 2 , were fractionated on a glycerol gradient. Fractions were analyzed via Western blotting with an anti-HIF2␣ antibody. I/20, input/20. C, immunoblot analysis of nuclear extract (NE), unbound (U), wash (W), and Flag immunoaffinity-purified fractions (IP) of Mock and H2 cell lines using an antibody against HIF2␣. D, purified HIF2␣ complexes (IP) were separated on a 4%-12% gradient protein gel and revealed by means of silver staining. The band that might correspond to HIF2␣ protein is indicated by an arrowhead. Nuclear extracts (NE) were used as loading controls.

TABLE I HIF2␣ protein partners identified via mass spectrometry
The proteins listed have been identified and quantified in at least two of four independent experiments with (i) an FDR Ͻ 1 and (ii) a ratio of PAI assay/PAI control Ͼ 4. Accession IDs, molecular weights (MW), protein names, the number of Specific peptides identified (the number of quantified peptides is indicated in parentheses), sequence coverage, Mascot score, and PAI ratios are indicated in the table. Score column: two score values are provided, separated by a semicolon. The left-hand value corresponds to the protein score obtained in the H2 assay. The right-hand value corresponds to the protein score obtained in the Mock assay. A dash (-) indicates that protein peptides have not been sequenced or detected. Ratio column: ratios are the PAI of the protein in the assay divided by the PAI of the protein in the control. The PAI is the average of the extract ion chromatogram area values for the three most intense peptides identified for a protein. Spe: proteins were not detected in the Mock protein extract and were thus Specific for H2 cells. Proteins have been categorized according to their biological function. The bait (HIF2␣ ϭ EPAS1) is in bold red. Proteins whose interaction with HIF2␣ has been further validated in the current study are in bold black (TFAP2A ϭ AP2␣). Annotated MS/MS Spectra of single peptide-based protein identifications are provided in supplemental Fig. S9. both HIF␣ subunits and interact with their C-terminal transactivation domains (31). CBP and P300 displayed more than 60% sequence identity. In the four independent proteomic experiments carried out in the current study, only 8 and 11 peptides among the 15 and 20 identified peptides for CBP and P300, respectively, strictly discriminated both proteins (Table I and supplemental Fig. S5). CBP/P300, mediator, and the SMARC complex are ubiquitous, and some of these proteins have already been identified as HIF1 and/or HIF2 partners (supplemental Table S1). Indeed, these interactions might not be specific to melanoma cells. Nevertheless, the presence of these partners not only confirms the quality and biological relevance of the identified interactome, but also suggests that they might play a role in melanoma progression, as discussed below. Interestingly, among all the identified HIF2␣ partners, several proteins were already known to play key roles in melanoma development, such as microphthalmiaassociated transcription factor (MITF), SOX10, and AP2␣ (TFAP2A) (32)(33)(34). Because of the novelty of these results, we decided to focus our study on these interactions.
Validation of HIF2␣ Interaction with MITF, SOX10, and AP2␣-Binding of MITF, SOX10, and AP2␣ to HIF2␣ was confirmed by means of either co-immunoprecipitation using both tagged and endogenous proteins or reciprocal co-immunoprecipitation using endogenous proteins, followed by Western blot analysis. Cell lysates from Mock and H2 cells cultured under hypoxia-mimicking conditions were prepared and immunoprecipitated with an anti-FLAG antibody targeting exogenous HIF2␣ protein. Immunoprecipitated complexes were then separated via SDS-PAGE and analyzed via Western blot. Antibodies raised against HIF1␤ (as a positive control), actin (as a negative control), MITF, SOX10, and AP2␣ were used. As expected, all proteins, except actin, were detected in the immunoprecipitate of H2-nuclear extract (Fig. 3A). No signal was observed in the Mock-nuclear extract immunoprecipitate (Fig. 3A). Note that no significant difference in the levels of expression of these proteins was observed in any of the 501mel clones used in the present study, except for HIF2␣ (see "Input" lanes in Fig. 3A). The interaction between HIF2␣ and its partners was further confirmed in hypoxic 501mel cells using an anti-HIF2␣ antibody targeting endogenous HIF2␣ (Fig. 3B). Again, all proteins were detected in the immunoprecipitate of 501mel protein extracts, clearly demonstrating that endogenous HIF2␣ was also present in protein complexes containing HIF1␤, MITF, SOX10, or AP2␣. Reciprocal coimmunoprecipitations were carried out in untransfected 501mel cells to confirm the interactions between HIF2␣ and its newly identified partners. Lysates from hypoxic 501mel cells were immunoprecipitated with anti-SOX10, anti-MITF, or anti-AP2␣ antibodies. Immunoprecipitated complexes were separated via SDS-PAGE and analyzed via Western blot (Fig.  3C). HIF2␣ was detected in the three different immunoprecipitates, confirming the interaction between HIF2␣ and these three proteins.
Interaction of Newly Identified HIF2␣ Partners with HIF1␣-To see whether HIF1␣ could also interact with HIF2␣ partners in melanoma cells, similar co-immunoprecipitation experiments were performed with the HIF1␣ protein as bait. The strategy to assess the presence of proteins in HIF1␣containing protein complexes was similar to that used to validate interactions between HIF2␣ and its partners. First, stable 501mel cell lines overexpressing tagged HIF1␣ were established. The expression, functionality, and subcellular localization of the tagged protein were verified (supplemental Figs. S6A, S6B). A clone, named H1, was selected that expressed about 2-fold more HIF1␣ than the Mock cell line (supplemental Fig. S6C). Immunoprecipitation of tagged HIF1␣ protein was performed on lysates of both Mock and H1 clones cultured under hypoxia-mimicking conditions (Fig. 4A). As expected, HIF1␤ co-immunoprecipitated with HIF1␣, whereas actin did not. SOX10 and MITF both co-immunopre-cipitated, although less than when HIF2␣ was used as bait. This might be explained by a lower expression of the HIF1␣ tagged protein in the H1 clone than of HIF2␣ in the H2 clone. Interestingly, we did not detect AP2␣ as an HIF1␣ partner. Co-immunoprecipitations and reciprocal co-immunoprecipitations were also performed with protein extracts from untransfected 501mel cells cultured under hypoxia (Figs. 4B, 4C), from which we conclude that in 501mel cells, SOX10 and MITF interact with both HIF␣ subunits, whereas AP2␣ is specifically associated with HIF2␣ protein complexes.

Interactions between HIF␣ Subunits and AP2␣, MITF, and SOX10 Proteins Are Observed in Other Melanoma Cell
Lines-To verify that the interactions between HIF1␣ or HIF2␣ and their partners also occur in other melanoma cell lines, co-immunoprecipitations were performed in a panel of melanoma cell lines expressing both HIF1 and HIF2 proteins. The G1, SK-MEL-28, and Lu1205 cell lines were cultured under hypoxia-mimicking conditions. Cell lysates were prepared and co-immunoprecipitation was carried out using either HIF1␣or HIF2␣-specific antibodies. As observed in the 501mel cell line, MITF and SOX10 co-immunoprecipitated with both subunits, whereas AP2␣ interacted only with HIF2␣ (Fig. 5). Note that neither MITF nor AP2␣ was detected in the Lu1205 cell line.
To determine whether the presence of HIF proteins modifies AP2␣ functions, we first analyzed the effect of HIF2␣ or HIF1␣ overexpression on the proliferation, invasion, and migration of 501mel cells. The growth rates of untransfected 501mel, CoCl 2 were immunoprecipitated with anti-Flag beads. Detection of tagged HIF1␣ or co-immunoprecipitated partners was performed via immunoblotting with anti-HIF1␣, anti-HIF1␤, anti-SOX10, anti-MITF, anti-AP2␣, or anti-actin antibodies. I, input (cell lysate); IP, immunoprecipitated fraction. B, endogenous HIF1␣ protein was immunoprecipitated from 501mel cells cultured under hypoxia using anti-HIF1␣ antibodies (IP) or a control isotype-matched antibody (IgG) followed by immunoblotting with the indicated antibodies. C, lysate from untransfected 501mel cells cultured in hypoxic conditions was immunoprecipitated with anti-SOX10, anti-MITF, anti-AP2␣ antibodies (IP), or a control isotype-matched antibody (IgG). Immunoprecipitates were analyzed via Western blotting using the indicated antibodies. I, input (cell lysate). In all panels, a representative blot is shown from at least three independent experiments.
Mock-, HIF1␣-, and HIF2␣-transfected cells were compared under hypoxia-mimicking conditions. Numbers of viable cells were evaluated daily using the MTT assay. No statistical significance was observed between the growth rate curves of the different cell lines (Fig. 6A); therefore, in our conditions, HIF␣ subunit overexpression did not affect cell proliferation. Migrating cells are characterized by being more spread on the substratum and by remodeling of the actin cytoskeleton with the formation of stress fibers (35). Rhodamin-phalloidin staining was used to assess actin cytoskeleton organization in the cell lines overexpressing HIF␣ subunits (Fig. 6B). Interestingly, whereas the H2 cells (overexpressing HIF2␣) displayed a more dendritic phenotype (a hallmark of differentiated melanocytes) and only few stress fibers, the H1 cells (overexpressing HIF1␣) were more spread and had an extensive network of stress fibers.
In HIF1␣ and HIF2␣ stably transfected cell lines, both HIF␣ isoforms are expressed. To decipher the role of each HIF␣ isoform in the invasive capacity of the 501mel cell line, siRNAmediated knock-down experiments were carried out. siRNA targeting HIF1␣ or HIF2␣ or an irrelevant siRNA was transfected into 501mel cells. Cells were placed under hypoxiamimicking conditions 36 h after transfection, and invasion assays were performed. The knock-down efficiency was as-sayed using RT-qPCR and Western blotting (supplemental Figs. S7A, S7B). For both HIF1␣ and HIF2␣, more than 95% of the protein was depleted. As expected, whereas HIF1␣ knock-down moderately reduced the invasive capacity of cells relative to the control (1.26-fold, p value ϭ 0.006), HIF2␣ knock-down strongly enhanced invasion (2.14-fold, p value ϭ 0.002) (Fig. 6D).
In conclusion, HIF1 enhanced the invasive capacity of 501mel cells, whereas HIF2 severely decreased it, possibly through its interaction with AP2␣.
Melanoma Cells Co-expressing AP2␣ and HIF2␣ Exhibit Decreased Invasive Capacities-Six different human mela- Invasion assays were performed using Matrigel invasion chambers (right panel). D, wild-type 501mel melanoma cells were transfected with the indicated siRNA (siC for non-targeting siRNA and siH1 or siH2 for, respectively, HIF1␣ or HIF2␣ targeting siRNA). Thirty-six hours after transfection, cells were depleted in FCS and treated with CoCl 2 for 12 additional hours. Cell invasion assays were performed as above. The results were expressed as the ratio of the number of invasive cells of the indicated cell lines to the number of invasive cells of the control siRNA-transfected cell line. Error bars represent S.E. (n ϭ 4 for migration assay, n ϭ 3 for invasion assay). *p value Ͻ 0.01; **p value Ͻ 0.001. noma cell lines were used in an attempt to correlate the expression of both AP2␣ and HIF2␣ proteins with the cells' invasive capacities. Cells were cultured under normoxia or hypoxia-mimicking conditions. Western blot analysis of AP2␣ expression discriminated two groups of melanoma cell lines: G1, 501mel, Daju, and SK-MEL-28 clearly expressed AP2␣, and Lyse and Lu1205 had no or little detectable AP2␣ protein.
Hypoxia did not change AP2␣ expression levels in any cell line considered (Fig. 7A). All melanoma cell lines tested expressed HIF1␣ under hypoxia, whereas only a subset expressed HIF2␣. G1 and 501mel exhibited large amounts of HIF2␣ under hypoxic conditions, whereas both SK-MEL-28 and Lu1205 cell lines expressed small amounts, and Lyse and Daju cell lines had no detectable HIF2␣ (Fig. 7A). In order to evaluate the possible role of HIF␣ subunit expression on the invasive properties of melanoma cell lines in relation to AP2␣ status, Matrigel invasion assays were performed with cells cultured for 24 h either under normoxia (no expression of HIF␣ subunits) or under hypoxia. For each cell line, the percentage of invasive cells under hypoxia was divided by the percentage of invading cells measured under normoxia. Melanoma cells expressing high levels of HIF2␣ under hypoxia (G1 and 501mel) exhibited a decrease in their invasive capacities under low oxygen pressure relative to normoxic conditions. Conversely, melanoma cells expressing only HIF1␣ under hypoxia (Lyse and Daju) showed a greater capacity to invade Matrigel under low oxygen pressure. Cells expressing low amounts of HIF2␣ under hypoxia (SK-MEL-28 and Lu1205) showed an intermediate modification of their invasive abilities (Fig. 7B). Contrary to an earlier report (34), we found no correlation between AP2␣ expression and invasive capacity in either normoxic or hypoxic conditions (supplemental Fig. S8). We therefore postulated that simultaneous expressions of both AP2␣ and HIF2␣ were required in order to alter cell invasion potential. To test this hypothesis, cell lines naturally expressing high or low amounts of AP2␣ (Daju or Lu1205 cell lines, respectively) were transiently transfected with an HIF2␣ expression vector or an empty vector (Mock). The transfected cells strongly expressed HIF2␣ even under normoxia, as confirmed by Western blot analysis. In these conditions, the amount of HIF2␣ detected in the transfected cells was comparable to the physiological level found in the G1 cell line cultured under hypoxia (Fig. 7C; see inset). Matrigel invasion assays were performed under normoxia 36 h after transfection. HIF2␣ overexpression in the AP2␣-negative cell line Lu1205 did not change its invasive ability. In contrast, the HIF2␣-transfected Daju cell line exhibited a significantly altered invasion capacity (0.51-fold, p value ϭ 0.001), suggesting that the simultaneous high expression of AP2␣ and HIF2␣ weakens the invasive capacities of melanoma cells (Fig. 7C).
As summarized in the model proposed in Fig. 7D, our compelling results demonstrate that a high HIF2␣ level under hypoxia leads to a decrease in the invasive capacities only of cells expressing high levels of AP2␣. In other cases, hypoxia increases the invasive properties of melanoma cell lines as already described (19). DISCUSSION The oxygen distribution in skin is highly heterogeneous. Oxygen tension at the dermal-epithelial junction where melanocytes reside is about 5%, whereas it is about 0.5% around hair follicles (36). Oxygen availability has important physiological consequences, including the mediation of cellular transformation, especially during the melanocyte-melanoma transition (14). Cellular adaptation to hypoxia is facilitated by the expression of HIF␣ subunits, mainly HIF1␣ and HIF2␣. HIF proteins are transcription factors that, under hypoxia, enable the activation of target genes involved in metabolism regulation, stress adaptation, growth, migration and invasion, drug resistance, and apoptotic cell death (37).
Protein function is frequently regulated by assembly into complexes. This aspect is particularly important for transcription factors, which interact with co-activators or repressors and with the transcription machinery to regulate gene expression. Indeed, transcription factor functions are profoundly remodeled by protein-protein interactions. Although several partners of HIF1 or HIF2 have already been described (supplemental Table S1), no global interactome analysis of HIF proteins is currently available. Furthermore, none of the wellcharacterized protein partners of these subunits have been identified in melanoma cells. In the present study, we performed a full screening of the HIF2␣ nuclear interactome in 501mel cells. The majority of the HIF2␣ protein partners identified are involved in the initiation of gene transcription. Mediator and SMARC complexes were found associated with the HIF2␣ subunit. Both protein complexes have already been shown to regulate melanoma development. The deletion of MED1, a component of the mediator complex, is known to increase melanoma aggressiveness (38). In the same way, when underexpressed in a cell line exhibiting a BRAF mutation (V600E) as in 501mel cells, SMARCB1, a component of the SMARC complex, prevents senescence and promotes melanomagenesis (39,40).
The current study also shows for the first time that HIF2␣ interacts with the products of two well-characterized melanoma master genes, MITF and SOX10. These direct or indirect interactions were also observed with the HIF1␣ subunit. SOX10 plays a key role in the development of the melanocyte lineage, mediated in part by transcriptional control of MITF expression (41). SOX10 is expressed in most, if not all, primary and metastatic melanoma cells. Moreover, this transcription factor regulates the expression of nestin, a marker of melanocytic stem cells that correlates with poor prognosis (33,42). MITF also has a key role regarding melanocytes and melanoma in that it regulates cell cycle progression, survival, and differentiation (19,43,44). Under hypoxia, MITF stimulates HIF1␣ expression in melanocytes, resulting in enhanced cell survival and improved resistance to UVB mutations leading to the malignant transformation of melanocytes to melanoma (45). In melanoma, HIF1␣ also indirectly represses MITF expression (19,46). In the 501mel cell line used in the current study, MITF down-regulation was not observed under hypoxia, even though this cell line carries a BRAF V600E mutation known to stimulate HIF1␣ expression (47).
As reported above, the capacity of 501mel cells to invade an extracellular matrix is modified when the expression of the HIF␣ subunits is modulated. MITF is also known to regulate the aggressiveness of melanoma cell lines. Cells expressing small amounts of MITF are more invasive, and cells expressing large amounts of MITF are more differentiated (48,49). However, several data argue against a key role of the regulation of MITF expression and function by HIF isoforms in explaining the observed effect on cell invasion. We found that neither HIF overexpression nor HIF down-regulation using specific siRNA lead to a modification of MITF expression (supplemental Fig. S7). Concerning its function, we found that the expression of a panel of MITF target genes was not globally modified by HIF1␣ or HIF2␣ overexpression or downregulation (data not shown). Therefore, the biological significance of the interaction between HIF1␣ or HIF2␣ and MITF, and also SOX10, deserves further investigation. In contrast to MITF and SOX10, the transcription factor AP2␣ was identified in the current study as a partner of HIF2␣, but not of HIF1␣, in the 501mel melanoma cell line (Figs. 3 and 4). We further demonstrated that the co-expression of AP2␣ and HIF2␣ has functional consequences for the invasive properties of tumor cells (Figs. 6 and 7). In many tumor types, including melanoma, AP2␣ acts as a tumor suppressor gene by activating p21Waf1/Cip1 expression and thereby inducing cell cycle arrest (50). Quantitative analysis of a large retrospective cohort of patients suffering from melanoma reveals that a high cytoplasmic-to-nuclear ratio of AP2␣ correlates with a poor prognosis. In addition, a loss of AP2␣ expression has been associated with malignant transformation and melanoma progression (34,51). Also, a recent in vitro study demonstrated that AP2␣ transcription factor overexpression could reduce cell invasion (52). Consistent with this observation, the very aggressive melanoma A375SM exhibits greatly reduced invasive capacities, tumor growth rates, and ability to metastasize when transfected with a plasmid allowing AP2␣ overexpression (53). Conversely, the overexpression of a dominant-negative mutant of AP2␣ in a nonmetastatic melanoma cell line, SB2, increases the aggressiveness of the cell line (53). In 501mel cells, AP2␣ interacts with HIF2␣ but not with HIF1␣ (Figs. 3 and 4). HIF2␣ overexpression in 501mel cells notably reduced cell invasion, whereas HIF1␣ overexpression significantly enhanced this property (Fig. 6). Similar defects in invasion potential have been observed in other human melanoma cell lines expressing HIF2␣ under hypoxia. Strikingly, HIF2␣ overexpression in a cell line expressing very small amounts of AP2␣ had no effect on cell invasion, whereas HIF2␣ overexpression in AP2␣-expressing cells was associated with a decrease in invasion potential (Fig. 7). Thus, we propose that the interaction between these two transcription factors in hypoxic conditions might modify the target genes they interact with and, consequently, reduce cell invasion properties.
In summary, our data strongly suggest that the simultaneous presence of both AP2␣ and HIF2␣ in melanoma cells might be a useful prognostic factor for patients. It will be very interesting to test this hypothesis in immunohistochemistry studies on large collections of melanoma tumors. Drugs targeting HIF proteins are currently under clinical assessment in patients. Our work underlines that, although important, the function of the HIF␣ subunits might be modulated by partners that confer opposing properties. Therefore, the functions of HIF1␣ and HIF2␣ must be investigated specifically in each human tumor type.