PHD3–SUMO conjugation represses HIF1 transcriptional activity independently of PHD3 catalytic activity

Maintaining oxygen homeostasis is essential for most organisms as hypoxia, even when transient, could trigger irreversible damage. Adaptation to reduced oxygen availability is indeed a major physiological challenge during embryonic development and in adulthood, but it is also associated with many human diseases (Benizri et al., 2008). This adaptation is primarily led by activation of hypoxia inducible factor (HIF), which controls the expression of hundreds of direct target genes, secondary transcription factors and non-coding RNAs (Semenza, 2012). HIF is a heterodimer formed by the interaction of the constitutive HIFβ subunit with a tightly regulated HIFα subunit (Huang et al., 1996; Salceda and Caro, 1997; Wang et al., 1995). Oxygen-dependent hydroxylation by the prolyl hydroxylase domain (PHD)-containing proteins drives HIFα ubiquitylation and proteasomal degradation (Bruick and McKnight, 2001; Epstein et al., 2001; Ivan et al., 2001; Jaakkola et al., 2001). Hydroxylation is indeed limiting for HIFα degradation and hence HIF activation.

Caenorhabditis elegans and Drosophila melanogaster have a single PHD family member (EGL-9), whereas the genomes of higher metazoans encode at least three paralogous PHD proteins (PHD1, PHD2 and PHD3; also known as EGLN2, EGLN1 and EGLN3, respectively) (Bruick and McKnight, 2001; Epstein et al., 2001). All three PHD proteins catalyze site-specific hydroxylation of HIF1α polypeptides at P564 in vitro. However, PHD2 appears to be primarily responsible for basal HIFα levels, although PHD1 and PHD3 contribute to HIFα regulation in certain settings (Berra et al., 2003; Ginouvès et al., 2008; Takeda et al., 2006). Thereby, we found previously that PHD3, through isoform-specific regulatory mechanisms, participates in a negative-feedback loop in response to prolonged hypoxia (Ginouvès et al., 2008). Moreover, recent reports highlight specific HIF-independent targets for PHDs. In this regard, PHD1 controls cyclin D1 levels and thus mammary gland proliferation (Zhang et al., 2009). PHD2 has been found to modulate protein translation through regulation of eEF2 phosphorylation (Romero-Ruiz et al., 2012). PHD3 mediates oxygen-dependent stability of activating transcription factor 4 (ATF4), PKM2 and β₂-adrenergic receptor, and prevents degradation of myogenin (Fu et al., 2007; Köditz et al., 2007; Luo et al., 2011; Xie et al., 2009). PHD3 also negatively regulates NF-κB and Pax2 (Fu and Taubman, 2010; Xue et al., 2010; Yan et al., 2011). Furthermore, PHD3 appears to be necessary and sufficient for apoptosis of neuronal cells after nerve growth factor withdrawal (Lee et al., 2005). The absence of PHD3 is indeed related to the failure of developmental apoptosis, which might play a role in pheochromocytoma pathogenesis. Clearly if the PHD proteins are to be targeted, it will be important to understand the basis of their specificities.

Post-translational modifications (PTMs) act to fine-tune the functions of encoded proteins, and they play a crucial role in conferring plasticity. Among several PTMs, SUMOylation, the conjugation of the small ubiquitin-related modifier (SUMO) to target proteins, has emerged as a fundamental strategy to modulate protein function (Geiss-Friedlander and Melchior, 2007). SUMOylation requires a sequential enzymatic cascade implicating SUMO-specific E1 and E2 enzymes and, for some substrates, a SUMO-E3 ligase. As is the case for many other PTMs, SUMOylation is a reversible process, with SUMO proteases acting to remove the modification (Hay, 2007). SUMOylation occurs often, but not exclusively, at the lysine residue within the consensus SUMO-acceptor site sequence ψKxE/D, in which ψ is an aliphatic branched amino acid (Rodriguez et al., 2001). Altered patterns of protein SUMOylation are associated with stress, suggesting that SUMO modification is essential in the cellular adaptation to environmental changes (Tempé et al., 2008). Accordingly, previous reports have provided insight into the direct implications of SUMO conjugation on the regulation of essential factors required for hypoxia adaptation, such as HIFα, HIFβ, VHL and p300 (Núñez-O'Mara and Berra, 2013).

To gain further insight into PHD3 regulation, we evaluated whether SUMO conjugates to PHD3. In the present study, we show for the first time that PHD3 can be a SUMO target. Surprisingly, our results demonstrate that PHD3 SUMOylation represses HIF1 transactivation activity without affecting HIF1α stability. Moreover, we present evidence that hypoxia modulates PHD3 SUMOylation and that this modification correlates with the inhibition of HIF1. PHD3 SUMOylation appears to be a newly identified additional step in the control of HIF.

To study new mechanisms of PHD3 regulation, we evaluated whether PHD3 can be modified by SUMO. We incubated [³⁵S]-methionine-labeled HA–PHD3 that had been in vitro transcribed and translated (IVTT) with a SUMOylation mixture, including the E1 (SAE1 and SAE2) and E2 (UBC9) enzymes, in the absence or in the presence of SUMO or GST–SUMO. As shown in Fig. 1A, we detected the non-modified PHD3 as well as higher-molecular-mass bands in the presence of SUMO1, 2 and 3 or GST–SUMO1, 2 and 3, suggesting that PHD3 is a target for SUMO. The identity of the high-molecular-mass band was revealed by double immunoblotting with anti-HA and anti-SUMO (Fig. 1B), which confirmed the presence of PHD3–SUMO conjugates.

Among SUMO proteins, SUMO1 shares only 50% identity with SUMO2 and SUMO3, which are almost identical (Su and Li, 2002). Hereafter, we use the nomenclature SUMO2/3 to refer to both proteins, as the two paralogs cannot be functionally distinguished. In view of the in vitro results, we next transfected (His)₆-tagged PHD3 together with HA–SUMO1 or HA–SUMO2/3 into COS-7 cells. PHD3 was purified under denaturing conditions on Ni-NTA beads, and the purified material was analyzed by western blotting (Fig. 1C). In the presence of SUMO2/3, we were able to detect high-molecular-mass PHD3 bands with the anti-(His)₆ antibody. The anti-HA antibody also recognized these bands, confirming that they corresponded to PHD3–SUMO conjugates. Note that SUMO2/3 was expressed at higher levels compared with SUMO1 (inset Fig. 1C), which might explain the absence of PHD3–SUMO1 conjugates (Fig. 1C). To validate the identification of PHD3 as a new SUMO target protein, we performed the reciprocal purification (Fig. 1D). We could show that PHD3 does co-purify with (His)₆-tagged SUMO2/3 when both proteins were ectopically expressed. Taken together, these data confirm that PHD3 can be a SUMO target protein not only in vitro but also in cultured cell lines.

Although PHD3 lacks any consensus SUMO acceptor site, an analysis of the sequence showed 12 evolutionary conserved lysines that might be potential SUMO-modified residues. To facilitate the identification of the target residue(s), we generated two overlapping truncated forms of PHD3 (PHD3 Nter and PHD3 Cter) that were subjected to in vitro SUMOylation assays as described previously. Interestingly, PHD3 Nter was no longer SUMOylated, whereas PHD3 Cter was still modified like the wild-type protein (Fig. 2A), suggesting that PHD3 is SUMOylated at the C-terminal region of the protein. We replaced each of the putative SUMO target lysines located within PHD3 Cter (K119, K154, K159, K172, K222, K223, K224 and K231) with arginines (R). However, none of the individual mutations prevented PHD3 SUMOylation (supplementary material Fig. S1). We next adopted a three-dimensional homology approach using an online server (http://www.sbg.bio.ic.ac.uk/phyre/) to better characterize the SUMO-modified residues (Kelley and Sternberg, 2009). This approach allowed us to identify two clusters of lysines at the C-terminal region of PHD3. These clusters correspond to lysines 154, 159 and 172 and lysines 222, 223, 224 and 231, respectively. Mutation of the two different clusters independently allowed us to identify lysines 222, 223, 224 and 231 as the PHD3 SUMOylation motif, at least in vitro (Fig. 2B).

To confirm these results, we transfected HEK293T cells with (His)₆-tagged SUMO2/3 together with wild-type PHD3 or PHD3 K222/223/224/231R. Consistent with the in vitro assays, PHD3 SUMOylation was almost abolished in the K222/223/224/231R mutant compared with that of the wild-type protein (Fig. 2C). Thus, we confirmed the PHD3 SUMOylation motif and accordingly identified the SUMOylation dead mutant (K222/223/224/231R), which we will refer to hereafter as PHD3ΔSUMO.

Although it was originally identified as a repressor of the HIF complex, additional roles have been reported for PHD3. We wondered whether SUMOylation might affect specific PHD3 functions. We initially focused on two signaling pathways, the HIF and NF-κB pathways, as they implicate PHD3 hydroxylase activity-dependent and -independent mechanisms, respectively. To address this issue, we co-transfected HEK293T cells with the different PHD3 mutants and luciferase reporter plasmids under the control of HIF- or NF-κB-responsive elements (HRE-Luc or NF-κB-Luc, respectively). In addition to PHD3ΔSUMO, we generated chimeric PHD3 mutants (PHD3–SUMO1 and PHD3–SUMO2/3) by fusing SUMO1 or SUMO2/3 to the C-terminal end of PHD3 (Fig. 3; supplementary material Fig. S2). This strategy has been successfully used to mimic constitutive SUMOylation and study the impact of SUMOylation on target proteins (Ross et al., 2002). Note that the two glycine residues, which are crucial for generating the covalent link between SUMO and the target proteins were removed in the chimera to exclude the possibility of the fusion protein acting as a SUMO-like protein.

Luciferase activity following the expression of wild-type PHD3 and H196A (the hydroxylase dead mutant) was consistent with the predicted impact of these proteins on hypoxia-induced HIF activation (Fig. 3A). PHD3–SUMO2/3 was a significantly stronger HIF repressor compared with wild-type PHD3, whereas PHD3ΔSUMO showed a clear tendency to be a less active repressor (Fig. 3A). This tendency is overcome by fusing SUMO2/3 to PHD3ΔSUMO, supporting the role of SUMO conjugation in PHD3-mediated HIF repression (Fig. 3A). Moreover, the effect of the SUMO2/3 chimera was specific, as PHD3–SUMO1 behaved similarly to the wild-type protein (supplementary material Fig. S2). In contrast to the effects on the HIF pathway, all of the PHD3 mutants inhibited TNFα-triggered activation of NF-κB in a similar manner. Therefore, our data imply that PHD3 SUMOylation optimizes its HIF repressor activity without affecting the NF-κB pathway (Fig. 3B).

To test whether endogenous SUMOylation at physiological levels contributes to HIF repression, HEK293T cells were transfected with short hairpin (sh)RNAs targeting HIF1α (used as an internal control), UBC9 or SUMO3 and the luciferase reporter plasmids. According to our data, UBC9 or SUMO3 silencing, and therefore the inhibition of the SUMOylation machinery, increased hypoxia-triggered HIF activation (supplementary material Fig. S2).

Although the functional consequences of SUMOylation depend on the modified protein, SUMO conjugation is very often associated with temporal and spatial regulation of its targets. Therefore, to evaluate the impact of SUMO conjugation on PHD3 behavior, we analyzed the stability of PHD3 and PHD3–SUMO conjugates. We co-transfected PHD3 and (His)₆-tagged SUMO2/3 plasmids. After transfection, we treated cells with the protein synthesis inhibitor cycloheximide and harvested cells at the indicated times (Fig. 4A). SUMOylated proteins were recovered on Ni-NTA beads, and PHD3–SUMO conjugates were identified by immunoblotting (Fig. 4A, left panel). In parallel, the expression of the non-modified PHD3 was resolved using total cellular extracts (Fig. 4A, right panel). No major differences in the stability of PHD3–SUMO conjugates compared with that of the non-modified PHD3 were observed. Furthermore, we evaluated the half-life of the different PHD3 mutants. As shown in Fig. 4B, the three constructs showed similar behavior, suggesting that SUMOylation does not affect PHD3 stability. To consolidate these data, we also evaluated the impact of Siah2, the ubiquitin E3 ligase that is reported to trigger PHD3 proteasomal degradation (Nakayama et al., 2004). We transfected cells with the different PHD3 mutants with and without Myc–Siah2 and analyzed their expression levels (Fig. 4C). Siah2 overexpression induced degradation of wild-type PHD3, PHD3–SUMO2/3 and PHD3ΔSUMO constructs to a similar extent. These results suggest again that SUMOylation is unlikely to affect PHD3 stability. The data shown are supported by the fact that PHD3ΔSUMO is similarly ubiquitylated to wild-type PHD3 (data not shown). Therefore, unlike other targets such as IkBα, SUMO would not compete with ubiquitin to stabilize PHD3 (Desterro et al., 1998).

We next analyzed the effect of SUMO on PHD3 intracellular localization. We transfected the different HA-tagged constructs into HeLa cells and performed immunofluorescence assays (supplementary material Fig. S3). In agreement with previous reports (Metzen et al., 2003), wild-type PHD3 showed nuclear and cytoplasmic localization. PHD3 SUMO mutants exhibited similar localization compared with that of the wild-type protein, indicating that SUMO does not affect PHD3 localization.

Among the complex mechanisms underlying HIF activation, PHD-mediated regulation of HIFα stability through hydroxylation plays an essential role. Thus, one scenario could be that SUMOylation optimizes PHD3 hydroxylase activity. We tested this possibility using a commercially available antibody that specifically recognizes hydroxylated P564 (P564OH), as we have confirmed (supplementary material Fig. S4A,B). In contrast to our initial hypothesis, wild-type PHD3, PHD3–SUMO2/3 and PHD3ΔSUMO proteins were equally competent to hydroxylate HIF1α P564, either in the presence or in the absence of the proteasome inhibitor, MG132 (Fig. 5A,B). Hence, we found no evidence that SUMOylation directly affects PHD3 hydroxylation activity, at least on HIF1α P564, and therefore, HIF1α protein levels.

An alternative scenario could be that SUMOylation affects HIF1 transactivation activity per se. To investigate this issue, we analyzed the impact of PHD3 mutants on the transcriptional activity triggered by a constitutively expressed HIF1α (HIF1-DM). As shown in Fig. 5C, the expression of wild-type PHD3 significantly repressed HIF-dependent activation of the HRE-Luc reporter gene. Surprisingly, PHD3ΔSUMO was a weaker inhibitor, whereas PHD3–SUMO2/3 and PHD3ΔSUMO–SUMO2/3 behaved as stronger inhibitors compared with the activity of the wild-type protein (Fig. 5C). Note that mutation of the two proline residues released HIF1α from PHD3-dependent degradation, but it did not prevent inhibition of HIF-dependent transactivation. By contrast, silencing of endogenous UBC9 or SUMO3 expression by shRNAs significantly reduced PHD3-mediated HIF repression (supplementary material Fig. S4C). To determine whether PHD3 hydroxylase activity was implicated in the repression of HIF1 transactivation activity, we transfected cells as described previously but including the PHD3 H196A mutant (supplementary material Fig. S4D). This experiment clearly showed that PHD3 catalytic activity is not required to repress HIF transactivation.

Finally, we analyzed the impact of PHD3 mutants on the expression levels of two endogenous HIF1 target genes – CA9 and GLUT1 (also known as SLC2A1). Supporting our data on reporter gene expression, wild-type PHD3 repressed CA9 and GLUT1 mRNA levels. Furthermore, PHD3ΔSUMO and PHD3–SUMO2/3 were a weaker and a stronger repressor, respectively, compared with the activity of the wild-type protein (Fig. 5D). Taken together, these data demonstrate that PHD3 represses HIF1 transcriptional activity in a hydroxylase-independent manner and also that PHD3 SUMOylation optimizes such repression.

We have previously shown that PHD3 participates in a feedback loop to suppress HIF1 activation in response to prolonged hypoxia (Ginouvès et al., 2008). The role of PHD3 SUMOylation in the repression of HIF-dependent transactivation prompted us to investigate whether hypoxia might regulate the conjugation of SUMO to PHD3. In line with our previous work, we conducted a timecourse study to evaluate the levels of PHD3–SUMO conjugates at different time-points of hypoxia (Fig. 6A). Interestingly, we observed that PHD3 SUMOylation was reduced after 4 h of hypoxia, whereas the levels of SUMOylated PHD3 recovered at prolonged hypoxia. These results demonstrate that hypoxia modulates PHD3 SUMOylation and show that the levels of PHD3–SUMO inversely correlate with an active HIF response.

The hypoxic response is primarily triggered by the inactivation of enzymes that belong to the family of 2-oxoglutarate and iron-dependent dioxygenases. Consequently, we wondered whether the activity of these enzymes might also affect PHD3 SUMOylation. Thus, we expressed PHD3 together with (His)₆-tagged SUMO2/3 and treated cells with DMOG, a pan-hydroxylase inhibitor. We found no evidence that hydroxylase activity affects PHD3 SUMOylation (Fig. 6B). Therefore, hypoxia-mediated modulation of PHD3 SUMO conjugation is independent of dioxygenase activity.

PHDs drive oxygen-dependent HIFα hydroxylation, which triggers protein ubiquitylation and further degradation by the proteasome. Therefore, oxygen is a master regulator of PHDs, although additional players have been shown to modulate these enzymes. The present study focused on SUMOylation as a potential PHD3 regulator. So far, hypoxia-dependent ubiquitin modification, catalyzed by the ubiquitin E3 ligase Siah2, is the only PTM described for PHD3 (Nakayama et al., 2004). Such ubiquitylation leads to PHD3 targeting for proteasomal degradation, and therefore modulates the response to hypoxia. In this report, we have demonstrated for the first time that PHD3 can be a SUMO target in vitro and in cultured cells. However, as reported for most of the SUMO targets, only a small fraction of PHD3 appears to be modified (Da Silva-Ferrada et al., 2012).

Although the in vitro SUMOylation assay showed one major PHD3–SUMO band, our mutational analysis indicates that mutation of four lysines (corresponding to the minimal PHD3ΔSUMO mutant) was required to abolish PHD3 SUMOylation. Based on the close proximity of the four lysines within the cluster, it is tempting to speculate that although they are all likely to be modified, SUMOylation only occurs on one of the residues each time. PHD3 SUMOylation takes place at a non-consensus site in the C-terminal end of the protein. However, the SUMO acceptor site in PHD3 is contained in the context of an alpha helix, as has been reported for substrates that do not conform to the consensus acceptor motif (Knipscheer et al., 2008; Pichler et al., 2005).

The unique PHD3–SUMO band detected in vitro also contrasts with the multiple SUMO-conjugates detected in cultured cells, which might result from poly-SUMOylation by SUMO2/3. Indeed, SUMO2 and SUMO3, but not SUMO1, contain an internal canonical SUMOylation site (K¹¹) that allows the formation of SUMO chains (Tatham et al., 2001). Besides the migration pattern of PHD3–SUMO conjugates, our data showing that SUMO2/3-conjugates are more readily detected in cultured cells as well as the functional assays support a central role for SUMO2/3 in the modification of PHD3. Otherwise, the shift of some SUMO-conjugates does not properly fit with the addition of 1, 2, 3 or more SUMO molecules. Thus, we cannot exclude the possibility that this shift might correspond to additional PTMs. Moreover, the interplay between SUMOylation and other PTMs has been extensively reported.

Our data provide new evidence for a dual role of PHD3 in HIF regulation as reported previously for EGL-9 in Caenorhabditis elegans (Shao et al., 2009) and PHD2 in mammals (Ozer et al., 2005; To and Huang, 2005). However, these results disagree with a more recent report claiming that PHD3 is a HIF1 coactivator through hydroxylation-mediated recruitment of PKM2 (Luo et al., 2011). In addition to the canonical role as the oxygen sensor that regulates HIFα stability, Powell-Coffman's group showed evidence that EGL-9 (the unique nematode ortholog of the PHD proteins) inhibits HIF1 transcriptional activity through a mechanism that does not implicate the hydroxylase activity. Supporting this prior study, our findings show the first evidence for a role of a mammalian PHD (in this case PHD3) as a HIF repressor independent of the hydroxylase activity. Moreover, we show here that PHD3 SUMOylation by SUMO2/3 mediates the non-canonical role of PHD3 as a HIF1 transcriptional repressor. Direct SUMOylation of transcription factors, which is mostly associated with transcriptional repression, arises as one of the most frequent functions attributed to SUMO in eukaryotic cells (Ouyang et al., 2009). By contrast, our data support an indirect role of PHD3 SUMOylation in transcriptional repression, as recently shown for IκBα (Mulero et al., 2013).

Contrary to the global SUMOylation increase induced by hypoxia, de-SUMOylated HIF1α, HIF2α, p300 and CBP correspond to the stable and/or active entities responsible for HIF activation (Cai et al., 2010; Cheng et al., 2006; Huang et al., 2009; van Hagen et al., 2010). Our results support this model, because PHD3 de-SUMOylation mirrors maximum HIF activation (Fig. 7). Whether hypoxia affects the machinery for PHD3 SUMOylation and/or de-SUMOylation or whether a second PTM is involved in this regulation remains to be explored.

Here, PHD3 SUMOylation is highlighted as an additional step among the extensive and dynamic mechanisms that are likely to be required to subtly adjust HIF function and to ensure survival regardless of oxygen availability. This complexity reflects not only the crucial role of HIF1 in maintaining oxygen homeostasis but also the need to safeguard against its inappropriate activation. Previous reports have linked the PHD3–HIF and SUMOylation pathways with pathologies such as cancer and ischemia (Cheng et al., 2006; Chen et al., 2011; Flotho and Melchior, 2013; Fox et al., 2011; Loinard et al., 2009; Mo et al., 2005; Silveirinha et al., 2013; Su et al., 2010; Wang and Banerjee, 2004; Xue et al., 2010). In light of these findings, there is a great interest in the development of drugs targeting these pathways. However, targeting global SUMOylation and/or the HIF response might have widespread toxicity and side effects. We anticipate that deciphering the specific SUMO E3 ligase and/or SUMO protease involved in PHD3 SUMOylation would facilitate the identification of more specific and secure targets and, therefore, provide alternative therapies.

The HA–PHD3 Nter construct was generated by digestion of the pcDNA3-HA-PHD3 plasmid with BamHI and EcoRI, filling to give blunt ends, and re-ligation. HA–PHD3 Cter was generated by PCR amplification using HA–PHD3 as the template and the oligos 5′-GCAAATACAGATCTATGGAGAGGTCTAAG-3′ and 5′-CTGATGGCGGCCGCTCAGTCTTCAGTGAG-3′. The PCR product was digested with BglII and NotI and cloned into the pcDNA3-HA vector that had been digested previously with BamHI and NotI. The pcDNA4-PHD3-(His)₆ construct was generated by digestion of HA–PHD3 with HindIII and EcoRI and cloning into the pcDNA4-(His)₆ plasmid. The pcDNA4-PHD3 SUMO1-(His)₆ chimera was generated by PCR amplification of SUMO1-(His)₆ with the oligos 5′-CATGGTGCTGCAGTGTCTGAC-3′ and 5′-GTGGATCCTAACTCGAGGTTTG-3′. The PCR product was digested with PstI and XhoI and cloned into the pcDNA4-PHD3-(His)₆ plasmid that had been digested previously with the same enzymes. pcDNA4-PHD3 SUMO2/3-(His)₆ was generated using the same strategy and the oligos 5′-CATGGTGCTGCAGTGTCCGAG-3′ and 5′-GAATTCCTAACTCGAGGTCTG-3′. The PHD3 mutants were obtained using the QuikChange® II XL Site-Directed Mutagenesis Kit (Stratagene). Primer sequences used for mutagenesis are listed in supplementary material Table S1. pCMV-Myc-HIF wild type and DM (P402/564A), pcDNA3-HA-PHD3, CMV-βgalactosidase, pRE-Δtk-Luc-HRE, pcDNA3-SUMO1-(His)₆, pcDNA3-SUMO2/3-(His)₆, pcDNA3-HA-SUMO1 and pcDNA3-HA-SUMO2/3 were as described previously (Berra et al., 2003; Desterro et al., 1998; Ginouvès et al., 2008; Tatham et al., 2001; Vertegaal et al., 2004). All of the constructs were sequence verified. Validated pools of shRNAs are from Open Biosystems.

According to the manufacturer's protocol, the indicated PHD3 constructs were synthesized using the TNT T7 Quick Coupled Transcription/Translation Kit (Promega) supplemented with L-[³⁵S]-methionine (PerkinElmer). A volume of 2 µl of the [³⁵S]-labeled IVTT PHDs were used for SUMOylation assays (50 mM Tris-HCl pH 7.5, 5 mM MgCl₂) in the presence of SAE1 and SAE2 (0.109 µM, Enzo/BioMol), ATP (2 mM), creatine phosphate (10 mM), creatine kinase (3.5 U/ml), pyrophosphatase (1 U/ml) and recombinant UBC9 (3.6 µM), SUMO (80 µM) or GST–SUMO (50 µM). After 2 h at 30°C, the reactions were stopped and resolved by SDS-PAGE. Migrated gels were stained with Coomassie Brilliant Blue, dried and exposed. All of the SUMOylation assays were performed at least three times.

COS7, HeLa and HEK293T cells were cultured in Dulbecco's modified Eagle medium supplemented with 5% FBS. Cells were incubated at 37°C at 95% humidity and 5% CO₂. Hypoxic conditions were generated by incubation of cells in an anaerobic workstation (In vivo₂ 400, Ruskinn). For hypoxic kinetics, HEK293T cells (supplemented with 10% FBS, 20 mM HEPES and ampicillin-streptomycin) were exposed to 1% O₂ for 7, 5 or 3 days or 24, 16 or 4 h prior to lysis. During longer timecourses, cells were split and seeded inside the workstation to obtain 80% confluence at the time of lysis. Cell extraction was performed inside the anaerobic workstation to avoid reoxygenation. Cells were transfected at 60–70% confluence 24 h post-seeding using the indicated DNA with FuGENE6 (Roche), Lipofectamine 2000 (Invitrogen) or ArrestIN (Open Biosystems) as transfection reagents.

To purify on Ni-NTA beads, cells were lysed (6 M guanidium, 0.1 M Na₂HPO₄/NaH₂PO₄, 0.01 M Tris-HCl pH 8.0) and the extracts were incubated with the beads (Qiagen) in the presence of 10 mM β-mercaptoethanol for 2.5 h at room temperature. After two sequential washes using urea buffers (8 M urea, 0.1 M Na₂HPO₄/NaH₂PO₄, 0.01 M Tris-HCl, 10 mM β-mercaptoethanol) equilibrated first at pH 8.0 and then at pH 6.3, the purified proteins were eluted with 3× disruption buffer (150 mM Tris-HCl pH 6.8, 6% SDS, 30% glycerol, 10% β-mercaptoethanol, Bromophenol Blue) containing 200 mM imidazole. For total cell extracts, cells were lysed with Laemmli buffer (50 mM Tris-HCl pH 6.8, 1.25% SDS, 15% glycerol). Following quantification using the Lowry assay, the proteins were resolved by SDS-PAGE and electrophoretically transferred onto a PVDF membrane (Millipore). The following antibodies were used for immunoblotting: mouse anti-HA.11 (BAbCO/Covance), mouse anti-β-actin (Sigma), rabbit anti-SUMO2/3 (Eurogentec), mouse anti-(His)₆ (Novagen), mouse anti-Myc (Cell Signaling Technology), rabbit anti-HIF-P564OH (Cell Signaling Technology), rabbit anti-HIF1α (Richard et al., 1999). Immunoreactive bands were visualized with ECL.

Cells were lysed in 25 mM Tris phosphate pH 7.8, 8 mM MgCl₂, 0.5% Triton X-100, 7.5% glycerol and 1 mM DTT. Luciferase activity measurement was performed using the Steadylite plus™ High Sensitivity Luminescence Reporter Gene Assay System (PerkinElmer). β-galactosidase activity measurement was performed using the Galacto-Light Plus system (Applied Biosystems).

Total RNA was isolated using the RNeasy Mini Kit (Qiagen), reverse transcribed and amplified with the quantitative PCR MasterMix Plus for SYBR Green I (Roche). PCR was carried out in a C1000™ Thermal cycler (Bio-Rad). RPLP0 was used for normalization. The primer sequences are available upon request.

HeLa cells were fixed in 3% paraformaldehyde for 30 min at 37°C. After several washes with PBS, cells were permeabilized for 5 min with 0.2% Triton X-100 and blocked for 30 min in PBS solution containing 0.2% gelatin and 2% bovine serum albumin (BSA). Then cells were incubated with mouse anti-HA.11, donkey anti-mouse-IgG conjugated to Alexa Fluor 488 (Invitrogen) and DAPI. Cells were analyzed with a fluorescent microscope, the Axio Imager D1 (Zeiss), using structured illumination (Apotome accessory) with a 63× objective. Images were taken with an Axiocam HRm camera (Zeiss) operated by AxioVision Rel 4.8 software.

Data are expressed as the mean±s.e.m. Differences were examined by Student's t-test (two-tailed) between two groups or by one-way analysis of variance (ANOVA) within multiple groups. P<0.05 was considered statistically significant.

We thank Dr Roser Buscà (IRCAN, Nice, France) for critical reading of the manuscript and Dr Cormac T. Taylor (Conway Institute-University College Dublin, Dublin, Ireland), Dr María del Mar Vivanco (CIC bioGUNE, Derio, Spain) and Dr Sylvie Gisselbrecht (Cochin Hospital, Paris, France) for generously providing NF-κB-Luc, SV-40-β-galactosidase and pBK-CMV-Myc-Siah2 plasmids. We are indebted to Dr Amandine Ginouvès and Dr Melanie Berta for their support.