Structure of Factor-inhibiting Hypoxia-inducible Factor (HIF) Reveals Mechanism of Oxidative Modification of HIF-1α*

The activity of the transcription factor hypoxia-inducible factor (HIF) is regulated by oxygen-dependent hydroxylation. Under normoxic conditions, hydroxylation of proline residues triggers destruction of its α-subunit while hydroxylation of Asn803 in the C-terminal transactivation domain of HIF-1α (CAD) prevents its interaction with p300. Here we report crystal structures of the asparagine hydroxylase (factor-inhibiting HIF, FIH) complexed with Fe(II), 2-oxoglutarate cosubstrate, and CAD fragments, which reveal the structural basis of HIF modification. CAD binding to FIH occurs via an induced fit process at two distinct interaction sites. At the hydroxylation site CAD adopts a loop conformation, contrasting with a helical conformation for the same residues when bound to p300. Asn803 of CAD is buried and precisely orientated in the active site such that hydroxylation occurs at its β-carbon. Together with structures with the inhibitors Zn(II) and N-oxaloylglycine, analysis of the FIH-CAD complexes will assist design of hydroxylase inhibitors with proangiogenic properties. Conserved structural motifs within FIH imply it is one of an extended family of Fe(II) oxygenases involved in gene regulation.

In hypoxic cells, activation of the HIF 1 transcriptional cascade directs a series of adaptive responses that enhance oxygen delivery or limit oxygen demand (1). Activation of HIF in cancer and ischemic/hypoxic vascular diseases has indicated a central role in human pathology (1). The transcriptional complex is composed of an ␣␤ heterodimer, HIF-␤ being a constitutive nuclear protein that dimerises with oxygen regulated HIF-␣ subunits (2). The activity of the HIF system is regulated by a series of Fe (II) and 2OG-dependent dioxygenases that catalyze hydroxylation of specific HIF-␣ residues. In normoxia, 4-hydroxylation of human HIF-1␣ at Pro 402 or Pro 564 by a set of HIF prolyl hydroxylase isozymes (PHD1-3) (3, 4) mediates HIF-1␣ recognition by the von Hippel-Lindau ubiquitin ligase complex leading to its proteasomal destruction (5)(6)(7)(8). In a complementary mechanism FIH (9) catalyzes hydroxylation of HIF-1␣ Asn 803 (10,11), which blocks interaction with the transcriptional coactivator p300 (12,13). In hypoxia, lack of hydroxylase activity enables HIF-␣ to escape destruction and become transcriptionally active. Inhibition of HIF hydroxylases by Fe (II) chelators and 2OG analogues activates the HIF transcriptional cascade even in normoxia (3,5,14). The HIF hydroxylases therefore provide a focus for understanding cellular responses to hypoxia and a target for therapeutic manipulation. Here we report crystal structures for the HIF asparagine hydroxylase (FIH) alone and complexed with CAD polypeptides, cosubstrates, and inhibitors.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and Crystallization-FIH, CAD 775-826 , and CAD 786 -826 were prepared as described (10). Selenomethionine (SeMet) substituted FIH was produced using a metabolic inhibition protocol and LeMaster media supplemented with 50 mg/liter L-selenomethionine. SeMet incorporation was Ͼ95% by electrospray ionization-mass spectrometry. Aerobic crystallization of SeMet FIH (at 11 mg/ml) was accomplished by hanging-drop vapor diffusion at 17°C. The mother liquor consisted of 1.2 M ammonium sulfate, 4% PEG 400 and 0.1 M Hepes pH 7.5. Crystallization of FIH-Fe-CAD fragment complexes was accomplished under an anaerobic atmosphere of argon in a Belle Technology glove box (0.3-0.4 ppm O 2 ) using the same mother liquor and a solution containing FIH (at 11 mg/ml), Fe 2ϩ (1 mM), 2OG/NOG (2 mM), and CAD fragment (1 mM). Crystallization of FIH-Zn-CAD fragment was accomplished aerobically under similar conditions. Peptides were either synthesized by solid phase peptide synthesis or purchased from Biopeptide Co. (San Diego, CA).
Crystallographic Data Collection and Structure Solution-Crystals were cryocooled by plunging into liquid nitrogen and x-ray data were collected at 100 K using a nitrogen stream. Cryoprotection was accomplished by sequential transfer into a solution containing 1.2 M ammonium sulfate, 3% PEG 400, 0.1 M Hepes, pH 7.5, and 10% followed by 24% glycerol. A three-wavelength multiple anomalous dispersion data set was collected to 2.9-Å resolution on beamline 14.2 of the Synchrotron Radiation Source, Daresbury, UK. Data from crystals of FIH-CAD complexes were collected on beamlines 14.2, 9.6, or 9.5 using ADSC Quantum 4 (14.2 and 9.6) or MarCCD detectors (9.5). All data were processed with MOSFLM and the CCP4 suite (15). The crystals belonged to space group P4 1 2 1 2. The crystallographic asymmetric unit contains one FIH molecule. Six selenium positions were located and phases calculated using SOLVE (16). Density modification, which increased the figure of merit from 0.56 to 0.66, was performed using RESOLVE (17).
Refinement-An initial model was built using O (18) and refined against the SeMet data (remote wavelength) using CNS (19 ¶ Supported by a German Deutsches Akademischer Austauschdienst fellowship. 1 The abbreviations used are: HIF, hypoxia-inducible factor; CAD, C-terminal transactivation domain; FIH, factor inhibiting HIF; 2OG, 2-oxoglutarate; NOG, N-oxaloylglycine; SeMet, selenomethionine; PEG, polyethylene glycol; DSBH, double-stranded ␤-helix; CBP, CREB-binding protein (where CREB is cAMP-response element-binding protein). of simulated annealing followed by grouped B-factor refinement brought the R free to 36.2%. Following further rebuilding and refinement, which brought the R free to 32.3%, the model was transferred to the 2.15-Å data set. Rebuilding and refinement using REFMAC5 (20), including addition of iron, substrate and solvent molecules, and refinement of TLS parameters brought the conventional R-factor to 17.8% and the R free to 21.3%. The following residues are missing in the current model: 1-15 and 304 -306 of FIH and 786 -794, 807-811, and 824 -826 of the CAD fragment. According to PROCHECK (21) there are no Ramachandran disallowed residues, and 90.7% of residues have most favorable backbone conformations. For the CAD peptide, 77.8% of residues are in the most favorable region with the remaining 22.2% in additionally allowed regions.
Other structures were solved by molecular replacement using the coordinates from the 2.15-Å data and refinement using REFMAC5. In all structures electron density for the iron and 2OG/NOG was visible throughout refinement. Significant positive difference electron density was observed between the iron and the CAD Asn 803 ␤-carbon. Since B-factor differences between FIH and CAD imply that the CAD is not at 100% occupancy, this may represent an alternative binding mode for the 20G 1-carboxylate in the absence of substrate, although it could also be due to a ligating water molecule, again in the absence of substrate.

RESULTS AND DISCUSSION
To obtain an FIH-CAD complex without oxidation of the CAD or the Fe (II) , anaerobic conditions were employed to crystallize FIH in the presence of Fe (II) , 2OG, and various CAD polypeptides from 7-52 residues. Crystals were also obtained anaerobically for FIH complexed with Fe (II) and the FIH inhibitor NOG and aerobically for FIH complexed with Zn (II) and NOG. The structures were solved by molecular replacement using a model obtained by multiple anomalous dispersion on selenomethionine-substituted apo-FIH.
Crystalline FIH-CAD complexes were obtained with CAD 786 -826 and CAD 775-826 . Crystallization attempts with CAD 787-806 , HIF-2␣ CAD 850 -862 (equivalent to HIF-1␣ CAD 802-814 ), and CAD 800 -806 did not result in FIH-CAD complexes. Thus, a CAD peptide of greater than 20 residues was required (see below). Structures 1-3 (Table I) are of the FIH-CAD complexes, while structure 4 is of FIH in the absence of CAD. Unless otherwise indicated, the following discussion refers to structure 1.
Analysis of crystallographic symmetry revealed a dimeric form of FIH, consistent with native gel-electrophoresis analysis FIG. 1. The FIH-CAD complex (a-c, structure 1; d, structure 2). a, FIH monomer. The CAD peptide is shown as a ball-and-stick representation in red and the DSBH motif in green. b, FIH dimer. The two molecules of FIH are in dark and light blue, the DSBH motif is in green, and the CAD peptide is in red. c, the 2OG binding site with bound NOG is shown in yellow. The Fe (II) is colored pink, and the 2mF o Ϫ DF c electron density map is contoured at 1.5 . d, orientation of CAD Asn 803 at the FIH active site. The 2OG and CAD peptide are shown in yellow. (data not shown). The dimer interface involves the two Cterminal helices of each molecule in an interlocking arrangement predominantly involving hydrophobic interactions (Fig.  1, a and b). This unusual interface buries a surface area of 3210 Å 2 , which is large by comparison with other dimeric proteins of this size (25). The core of FIH comprises a double-stranded ␤-helix (DSBH or jellyroll) motif formed from eight ␤-strands, ␤8 -␤11 and ␤14 -␤17 (Fig. 1a) that is characteristic of the 2OG oxygenase superfamily. Residues 220 -259 form an insert between strands 4 (␤11) and 5 (␤14) of the DSBH. One face of the DSBH is flanked by an additional four ␤-strands from the N-terminal region to form an eight-membered antiparallel ␤-sheet (Fig.  1a). Interestingly, the N-terminal strand ␤1 bisects the face of the DSBH opposite to the active site (Fig. 1a). The ␤1 strand has a 360°twist located at a PXXP sequence, in between its interactions with ␤14 and ␤2. A similarly positioned ␤-strand is found in most 2OG dependent oxygenases, although not always from the same region of the protein. The sheet-helix-sheet motif formed by ␤1, ␣1, and ␤2 is conserved in all enzymes of this class except proline 3-hydroxylase. The DSBH topology of FIH defines it as a member of a superfamily of DSBH proteins, including non iron-containing enzymes such as the Cu (II) utilizing quercetin 2,3-dioxygenase (26) and Mn (II) utilizing type II phosphomannose isomerase.
The structures unexpectedly reveal the existence of two distinct FIH-CAD interaction sites, one involving the hydroxylation site itself (CAD 795-806 , site 1) and a second lying to the C-terminal side of this site (CAD 813-822 , site 2) (Fig. 2, a and b). The binding sites involve contact surface areas of 1640 Å 2 and 1080 Å 2 , respectively, and CAD residues in these regions are highly conserved in all known HIF-1␣ and HIF-2␣ sequences. Kinetic analyses were employed to investigate the relative importance of sites 1 and 2. CAD fragments shorter than 20 residues are not efficient in vitro substrates (Table II). Those containing site 1 only are hydroxylated by FIH 2 but less efficiently than those containing both sites, consistent with the crystallographic data. Electron density for site 1 is of good quality, with only the side chain of Tyr 798 poorly defined, while that for site 2 is at a lower level and quality, probably reflecting weaker binding at this site (Fig. 2c). CAD 804 -806 , and presumably also CAD 807-811 for which density was not observed, do not form direct interactions with FIH.
At site 1, HIF-CAD 795-803 residues are bound in a groove (Fig. 2b) and adopt a largely extended conformation linked to FIH by ten hydrogen bonds. Asn 803 of CAD is completely buried at the active site and lies directly adjacent to the Fe (II) . CAD Asn 803 and Ala 804 form a tight turn, stabilized by a hydrogen bond between the backbone carbonyl of Val 802 and NH of Ala 804 , which projects the side chain of Asn803 toward the Fe (II) . This side chain is precisely orientated by three hydrogen bonds to enable hydroxylation at the pro-S position of the ␤-carbon (Fig. 1d), consistent with NMR assignment of hydroxylation at this site. 2 The primary amide of CAD Asn 803 is sandwiched between FIH residue Tyr 102 and the Fe (II) . It forms hydrogen bonds with the side chains of FIH residues Gln 239 and Arg 238 (Fig. 1d), residues located on the insert to the DSBH motif, rationalizing the unusual selectivity of FIH for asparagine over aspartate (10). Interestingly, the substrate and Fe (II) binding sites are directly linked, since the backbone nitrogen of CAD Asn 803 also forms a hydrogen bond (ϳ3 Å) with the carboxylate oxygen of Asp 201 that is not complexed to the iron. Six additional hydrogen bonds stabilize the binding of FIH to 2 L. A. McNeill, unpublished results. CAD 795-801 . In contrast with site 1, site 2 is bound on the FIH surface and involves only two hydrogen bonds (Fig. 2a). CAD 816 -823 of site 2 form an ␣-helix, in exact agreement with the structure of this region in complex with CBP/p300 (12,13). As in that complex, the highly conserved Leu 818 , Leu 819 , and Leu 822 sit in a hydrophobic pocket on the surface of FIH (Fig.  2a); it is not possible for CAD to bind simultaneously to CBP/ p300 and FIH.
The extended loop conformation adopted by the CAD residues at site 1 contrasts with the ␣-helical conformation adopted by the same residues when complexed with the 1st transcriptional adaptor zinc-binding domain (TAZ1) of CBP/p300 (12,13). The disordered structure observed for the CAD, and other HIF-␣ residues (7), when free in solution may thus reflect a requirement to adopt more than one conformation for complex formation with different proteins. The changes in the conformation of CAD on binding are complemented by changes in FIH revealing an induced fit binding process. Most strikingly Trp 296 of FIH undergoes a 50°rotation about C ␤ -C ␥ to accommodate CAD Val 802 , while both Tyr 102 and Tyr 103 become more ordered. This ordering of FIH on substrate binding may be reflected in the significant improvement in resolution for the structures obtained with CAD fragments bound over those without.
Analysis of the active site reveals that the Fe (II) is bound in an almost octahedral manner by the side chains of His 199 , Asp 201 , and His 279 , and the 2-oxo and 1-carboxylate groups of 2OG (Fig. 1c). While bidentate chelation of 2OG with Fe (II) in this structure was anticipated, the binding interactions for the 2OG 5-carboxylate, which forms hydrogen bonds with the side chains of Lys 214 , Thr 196 , and Tyr 145 , are unprecedented. In many 2OG oxygenases Arg and Ser/Thr residues fulfil an anal-ogous role. FIH is further unusual in that Lys 214 is on the fourth DSBH ␤-strand, whereas previously assigned basic 2OG-5-carboxylate binding residues are at the beginning of the eighth DSBH strand.
In the FIH-CAD complexes (structures 1-3 in Table I), there is a vacant position opposite His 279 (Fig. 1, c and d) revealing that the enzyme is primed for dioxygen binding (27), although there could be rearrangement of the ligands during the catalytic cycle (28). As with other 2OG oxygenases it is assumed that following binding of dioxygen, decarboxylation of 2OG yields an iron-oxo species [Fe (IV) ϭ O 7 Fe (III) -O ⅐ ] that effects oxidation at the ␤-carbon of CAD Asn 803 (Fig. 3a). Interestingly, accommodation of a dioxygen ligand (or an alternate 2OG conformation) opposite His 279 may require disruption of the unusual hydrogen bond between Asp 201 and CAD Asn 803 (the Fe (II) and Asn 803 ␤-carbon are only ϳ4.9 Å apart) that is observed in the anaerobic enzyme substrate complex (Fig. 1d). This hydrogen bond may have energetic consequences for the binding of dioxygen that are relevant to the oxygen sensing function of the enzyme.
To understand the mechanisms of action of hydroxylase inhibitors, structures were also obtained for FIH complexed with NOG and with Zn (II) . FIH was demonstrated to bind Zn (II) in an identical manner to Fe (II) (structure 3), consistent with the metal-mediated mimic of hypoxia being due to displacement of Fe (II) from the active site of HIF hydroxylases. FIH-CAD structures with NOG reveal that like 2OG it is ligated to Fe (II) in a bidentate manner (Fig. 1c) and imply that it is an inhibitor due to decreased susceptibility to attack by an iron bound (su)peroxide. Kinetic analyses of a series of inhibitors based upon N-oxaloyl amino acids demonstrated that the R-enantiomer (IC 50 Ͻ0.4 mM) of N-oxaloylalanine was significantly more  potent than the S-enantiomer (IC 50 2.5 mM). Analysis of the 2OG binding pocket in FIH suggests that the binding of the S-enantiomer is disfavored by interactions with Thr 196 and Ile 281 in the 2OG binding pocket. Since a reversed selectivity (i.e. the S-enantiomer was more potent) was observed both for procollagen prolyl hydroxylase and the PHD isozymes) (5), it should be possible to develop selective inhibitors for individual types of HIF hydroxylase based on such structural constraints.
The unusual and precise structural determinants of both CAD and 2OG binding to FIH may aid inhibitor design via linkage of the 2OG and CAD binding sites and development of heterocyclic compounds that mimic the tight turn adopted by the CAD 802-804 when complexed to FIH. Recognition of post-translational hydroxylation as a major mode of regulation of the HIF pathway raises an important question as to its general role in biological signaling. Sequence analyses based on the FIH structure indicate that the 2OG oxygenase superfamily extends further than has previously been foreseen (Fig. 3b). Of particular interest are similarities with the JmjC homology region of the jumonji transcription factors (10,29). These proteins are predicted to have a DSBH core and have been implicated in diverse biological processes such as cell growth and heart development. Conserved HX(D/E) residues had been identified in some JmjC domains but not assigned as an iron binding motif (29). In the light of the FIH structure it is clear that many JmjC proteins have conserved residues including both this motif and the newly defined 2OG 5-carboxylate binding site involving FIH residues Lys 214 and Thr 196 on the fourth strand of the DSBH (Fig. 3b). The structure therefore implies that FIH is a one of a large family of iron-and 2OG-dependent oxygenases that are involved in the regulation of gene expression.