Asparaginyl Hydroxylation of the Notch Ankyrin Repeat Domain by Factor Inhibiting Hypoxia-inducible Factor*

The stability and activity of hypoxia-inducible factor (HIF) are regulated by the post-translational hydroxylation of specific prolyl and asparaginyl residues. We show that the HIF asparaginyl hydroxylase, factor inhibiting HIF (FIH), also catalyzes hydroxylation of highly conserved asparaginyl residues within ankyrin repeat (AR) domains (ARDs) of endogenous Notch receptors. AR hydroxylation decreases the extent of ARD binding to FIH while not affecting signaling through the canonical Notch pathway. ARD proteins were found to efficiently compete with HIF for FIH-dependent hydroxylation. Crystallographic analyses of the hydroxylated Notch ARD (2.35Å) and of Notch peptides bound to FIH (2.4–2.6Å) reveal the stereochemistry of hydroxylation on the AR and imply that significant conformational changes are required in the ARD fold in order to enable hydroxylation at the FIH active site. We propose that ARD proteins function as natural inhibitors of FIH and that the hydroxylation status of these proteins provides another oxygen-dependent interface that modulates HIF signaling.

Although the post-translational hydroxylation of extracellular proteins such as collagen is well characterized, examples of similarly hydroxylated intracellular proteins are very limited (1). Recently, prolyl and asparaginyl hydroxylation of HIF␣ 7 has been shown to act in an unprecedented mode of intracellular signaling that links oxygen availability to the HIF transcriptional response (2). Three oxygenases (prolyl hydroxylase domain 1-3) hydroxylate two conserved HIF␣ prolines, signaling for proteasomal degradation (3). FIH catalyzes hydroxylation of the HIF1␣ C-terminal transactivation domain (CAD) at Asn-803, resulting in blockage of the HIF-p300/CBP interaction and inhibition of HIF-mediated transcription (4,5).
Since the discovery of the HIF hydroxylases, a fundamental question has been whether these enzymes have other substrates. To address this question, we have carried out studies aimed at identifying proteins that interact with FIH. Several proteins co-immunoprecipitating with FIH, including Notch3, contained an ARD. Notch is a transmembrane receptor that regulates cell fate decisions, differentiation, and proliferation (6). Mammals express four Notch proteins (N1-N4) that, upon binding of ligands such as Delta-1, stimulate cytoplasmic release of the Notch intracellular domain (ICD) (7). The notch ICD is recruited to target genes (e.g. hes-1) by the transcription factor CSL (CBF-1/suppressor of hairless/Lag-1), where, through its ARD, it coordinates the assembly of a nuclear transcriptional activation complex involving Mastermind-like (MAML) proteins and Ski-interacting protein (SKIP) (7,8). It has been reported that Notch signaling is sensitive to oxygen tension (9), and multiple direct and indirect interactions between the Notch and HIF signaling pathways have recently been described (9 -12).
Here we report that the Notch ARD is hydroxylated by FIH at conserved Asn residues. Taken together with the recent identification of ARD-containing FIH substrates in the NF-B system (13), these results imply that post-translational hydroxylation of ARDs is common. ARDs are found in over 200 proteins encoded by the human genome (SMART data base) (14) and generally comprise a variable number of 33 residue repeats folded into paired antiparallel ␣-helices linked by "␤-hairpin" type loops. The repeats typically stack to form an L-shaped profile with the concave face acting as a protein-protein interaction surface (15). Numerous structural studies have characterized the ARD fold, but none have used Asn-hydroxylated material (15). Our findings therefore raised important questions about the effect of hydroxylation on the ankyrin fold, how ARDs and HIF bind to the FIH active site, and whether ARD hydroxylation interacts with HIF signaling.
Crystallographic and kinetic studies demonstrated that binding of FIH requires major conformational changes to the ARD and that ARD hydroxylation is competitive with that of HIF␣CAD. The combined functional and biochemical studies imply that the hydroxylation status of cellular ARDs governs the availability of FIH for the hydroxylation of HIF, thus providing a novel oxygen dependent means of modulating HIF signaling.
Immunoprecipitation and Pull-downs-Cells were lysed in IPϩ buffer (13). PK-tagged proteins were precipitated with conjugated agarose beads (Sigma). FLAG-FIH-inducible U2OS cells used for Notch interaction assays have been described (16); Notch receptors were immunoprecipitated with either rabbit anti-N2 (Chemicon), Rabbit anti-N3 (Santa Cruz Biotechnology, Inc., Santa Cruz, CA), or goat anti-N1 antibody (Santa Cruz Biotechnology). Rabbit anti-N2 and -N3 immunocomplexes were then Western blotted with the corresponding Notch antibody, followed by a TrueBlot TM horseradish peroxidase-conjugated secondary antibody (eBioscience). Goat anti-N1 immunocomplexes were Western blotted with mouse anti-N1 (Sigma), followed by standard horseradish peroxidaseconjugated secondary antibody. Endogenous N1⅐FIH complexes were immunopurified using a goat anti-N1 antibody (Santa Cruz Biotechnology) and either probed as described above for N1 or with anti-FLAG-horseradish peroxidase antibody (Sigma). GST-pull-down assays were performed with 293T cell extracts overexpressing the interaction partner of interest. Precleared extracts were incubated with glutathione beads coated with 5 g of GST or GST-bait. For purification of endogenous human N1, 20 ϫ 15-cm 2 plates of 293T cells were lysed in IPϩ and incubated with mouse anti-N1 antibody (Sigma) coupled to protein G-agarose. Beads were then washed and eluted in glycine, pH 2.5. PK-⌬N1 ICD⅐His-FIH pull-downs were undertaken by purifying PK-⌬N1 ICD from 293T cells or cells overexpressing FIH. Anti-PK-conjugated agarose beads were washed before adding His-FIH. PK-⌬N1 ICD⅐His-FIH complexes were washed and eluted in SDS sample buffer.
Mass Spectrometry-Liquid chromatography (LC)-MS used an Agilent 1100 LC system (Jupiter C4 column) and a Waters QTof micromass spectrometer in positive ion ESI mode. For in vitro competition assays, Fourier transform MS was used; 2-l samples were injected onto a Nano PepMap C18 column (75 m x 15 cm; Dionex) using a Dionex UltiMate 3000 LC system linked to a Bruker Apollo NanoElectrospray ion source (capillary voltage: 1400 V, dry temperature ϭ 150°C) and a Bruker APEX-Qe quadrupole hexapole-Fourier transform mass spectrometer (7T). MS/MS analyses of endogenous N1 hydroxylation and nondenaturing electrospray ionization MS methodology is described in the supplemental materials.
Statistical Analysis-Where possible, data are presented as the mean Ϯ S.D. Differences between means were considered significant when p was Ͻ0.05 using Student's t test.

Notch Receptors
Are Potential FIH Substrates-We identified four ARD-containing proteins (N3, Rabankyrin, Tankyrase, RNase L) in a screen for FIH-interacting proteins by immunoprecipitation of FLAG-tagged FIH from extracts of U2OS cells overexpressing this protein, followed by MS analyses. 8 Given the known links between Notch signaling and hypoxia (9), we focused studies on the Notch-FIH interaction. Initially, we tested whether other Notch receptors interact with FIH by immunoprecipitation (Fig. 1A). N1, N2, and N3 all bound to FIH, and interactions were enhanced by a cell-permeable FIH inhibitor (dimethyloxalylglycine), raising the possibility that Notch ICDs are FIH substrates.
Comparison of human N1-N4 sequences with HIF1␣CAD revealed the presence of two sites in the ARDs that together with HIF1␣CAD form a (D/E) Ϫ2 X Ϫ1 N consensus motif (supplemental Fig. S1). Both of the putative hydroxylation sites were present in the "␤-hairpin" type loops connecting individual ARs. The N1 N-terminal site ("Site 1") is positioned in the second AR, whereas the C-terminal site ("Site 2") is in the fourth. In contrast to N1-N3, N4 lacks the (D/E)XN motif; the acidic residue at the Ϫ2-position is absent from N4 Site 1, whereas Site 2 lacks the Asn residue itself, suggesting that N4 is not an FIH substrate.
FIH Hydroxylates N1 Asn-1945 and -2012-To determine whether Notch ARDs are indeed FIH substrates, we tested GST fusions of mouse Notch ARDs in assays based on 2OG (co-substrate) decarboxylation (18). GST-N1 ARD and GST-N3 ARD were similarly active, whereas GST-N4 ARD was not (Fig. 1B). Consistent with this, the ARDs of N1 and N3 (but not N4) were sufficient to interact with FIH in cells (supplemental Fig. S2). Thus, N1 and N3 ARD are FIH substrates in vitro, and the presence of a (D/E)XN motif may indicate whether an AR is an FIH substrate.
Since N1 is the best characterized family member, it was chosen for further analyses. To address whether N1 Site 1 and/or Site 2 are required for FIH activity, we compared N1 ARD N1945A, N1 ARD N2012G, and N1 ARD N1945A/ N2012G with wild-type N1 ARD and HIF1␣CAD. 2OG decarboxylation with N1 ARD was much greater than with HIF1␣CAD, consistent with hydroxylation at two sites ( Fig.  1C). Mutation of either site individually reduced decarboxylation to levels comparable with that observed with HIF1␣CAD; mutation of both Asn residues reduced it to control levels. MS analysis of the N1 ARD/FIH reaction products confirmed that Asn-1945 and -2012 were Ͼ95% hydroxylated (supplemental Fig. S3). Time course analyses suggest that Asn-1945 is more rapidly and more completely hydroxylated by FIH under nonsaturating conditions however. 9 Next we asked whether Notch ARDs are hydroxylated by FIH in vivo. To generate sufficient material with both substrate sites amenable for MS, we expressed a C-terminally truncated N1 8 M. Cockman, unpublished data. 9 C. Coles, unpublished data. ICD K2002E mutant (⌬N1 ICD) in 293T cells. PK-tagged ⌬N1 ICD was immunopurified; MS identified 35% hydroxylation at Asn-1945 and 3% at Asn-2012 (Fig. 1D). These results support the kinetic data showing Asn-1945 as the preferred site of hydroxylation. 9 FIH siRNA prior to overexpression of ⌬N1 ICD demonstrated that FIH was required for hydroxylation at both A ϩ16-Da shift is observed in the y ion series from y8, which corresponds to fragments containing Asn-1956. Several tryptic peptides from the N1 cytoplasmic domain were detected by MS analysis of immunopurified N1. 10 No peptides from the extracellular domain were detected. The most N-terminal fragment detected (residues 1666 -1673 of human N1) implies that the N1 purified was the membrane-bound form prior to ligand-induced cleavage. 10 sites. Consistent with the role for a dioxygenase, incubation of cells in 1% O 2 reduced Asn hydroxylation. Conversely, overexpression of FIH (but not the hydroxylation-inactive D201A mutant) increased hydroxylation at both sites to almost 100% (Fig. 1D). Incubation of PK-⌬N1 ICD-transfected 293T cells with dimethyloxalylglycine also significantly reduced Asn hydroxylation (from 38.9 to 16.8% N1945 hydroxylation), whereas postlysis incubation with the FIH inhibitor N-oxalylglycine had no significant affect (supplemental Fig. S4). Finally, purification of overexpressed N3 ICD from 293T cells also demonstrated ϳ50% hydroxylation at Site 1. 10 Endogenous N1 Receptor Is Hydroxylated-We next sought to assess the hydroxylation status of the endogenous human N1 Notch receptor. The interaction of the receptor with endogenous FIH was confirmed by co-immunoprecipitation (supplemental Fig. S5); we then determined the hydroxylation status of endogenous N1 directly, by immunopurification of N1 from 293T cells, followed by MS analyses. A single unhydroxylated peptide with Asn-2023 (equivalent to Asn-2012 of mouse N1) was detected by MS/MS. 11 Two peptides containing Asn-1956 (equivalent to Asn-1945 of mouse N1) were detected, one unhydroxylated ( Fig. 2A) and the other hydroxylated at Asn-1956 (Fig. 2B). LC-MS data indicated that 96% of the Asn-1956-containing peptide was hydroxylated in 293T cells (Fig. 3A). Similar analysis identified 30% Asn-1956 hydroxylation in HeLa-S3 cells. 11 Incubating 293T cells in 0.5% oxygen for 48 h prior to endogenous N1 purification and LC-MS analyses reduced Asn-1956 hydroxylation from 96 to 40% (Fig.  3A). Likewise, siRNA knockdown of FIH reduced Asn-1956 hydroxylation from 84% (control siRNA) to 25% (Fig. 3B).
These results confirm that the hydroxylation status of endogenous ARDs is not only dependent on FIH but also cellular oxygen availability. We next sought to understand the role of Asn hydroxylation and in particular whether it regulates Notch signaling activity.
FIH Does Not Regulate the Notch/ CSL Pathway-Gustaffson et al. reported that FIH overexpression suppressed N1 ICD activity in mouse embryonic teratocarcinoma P19 cells (11). In agreement, we found that FIH overexpression in P19 cells significantly suppressed the ability of N1 ICD to activate a Notch 12XCSL luciferase (CSL-luc) reporter (Fig. 4A, left). However, mutant FIH D201A also inhibited N1 ICD activity as effectively as wild-type FIH, implying that the effect is independent of hydroxylation. Moreover, FIH siRNA did not significantly enhance N1 ICD activity (compared with control siRNA), suggesting that, at least under these conditions, endogenous FIH is not exerting tonic control of N1 ICD activity (Fig.  4A, right). Similar reporter assays in HeLa and 293T cells did not show inhibition of N1 ICD activity by FIH, probably due to lower levels of overexpression (supplemental Fig. S6). Again, FIH siRNA did not affect N1 ICD activity in these cells. 10 To test whether effects of FIH might be restricted to chromatinized Notch targets, we went on to assess endogenous Notch activity using a co-culture assay. HeLa cells overexpressing (FIH transfection) or underexpressing (siRNA) FIH were seeded onto mouse L-cells or L-cells overexpressing the Notch ligand Delta-1 (20). After 24 h, to allow Notch cleavage and signal transduction, cell extracts were immunoblotted for 10 M. Coleman, unpublished data. 11 M. Coleman, M. Edelmann, and B. Kessler, unpublished data. Hes-1 induction. FIH intervention did not significantly affect Delta-mediated Notch activation and Hes-1 induction (Fig.  4B); FIH also did not affect the induction of endogenous Hes-1 by overexpression of N1 ICD in HeLa cells. 10 To test the effect of N1 ARD hydroxylation on the interaction with proteins that bind at or near the ARD (7, 21), GST-ARD or GST-ARD (OH) (100% Asn-1945 and 20% Asn-2012 hydroxylation) was incubated with 293T extracts overexpressing proteins of interest. Fig. 4C shows that GST-ARD and GST-ARD (OH) captured similar levels of MYC-Deltex, HA-SKIP, or MYC-CSL. We used a different approach to study the N1 ARD-MAML1 interaction, since overexpressed MAML1 did not bind GST-ARD. GST or GST-⌬MAML1 (amino acids 7-254 of human MAML1 containing the Notch binding domain) was incubated with extract from PK-⌬N1 ICD-transfected cells overexpressing (FIH transfection) or underexpressing (siRNA) FIH (see Fig. 1D). PK-⌬N1 ICD bound to GST-⌬MAML1, but the interaction was independent of Asn-1945/2012 hydroxylation (Fig. 4D). Therefore, FIH-mediated hydroxylation does not regulate the interaction of N1 with known ARD-interacting  Fig. 1D). Complexes were probed with anti-PK to detect ⌬N1 ICD.
proteins, consistent with the lack of effect of FIH-dependent hydroxylation on the Notch/CSL pathway.
Structure of Asn-hydroxylated N1 ARD-FIH-mediated hydroxylation of the Notch ARD is a post-translational modification likely to be common to many other ARDs. Since structural analyses of all ARDs to date have used unhydroxylated material (15), we investigated the structural consequences of FIH-mediated ARD hydroxylation by crystallography. These studies also aimed to define the regio-and stereoselectivity of ARD hydroxylation.
Hydroxylated N1 ARD (N1 ARD (OH)) was produced by co-lysis of bacteria separately producing Histagged FIH (His-FIH) and GST-N1 ARD with co-factors and co-substrates. Under these conditions, the N1ARD was hydroxylated to Ͼ95% at Asn-1945 and Ͻ10% at Asn-2012. Following removal of the GST tag, N1 ARD (OH) was crystallized, and the structure was solved to 2.35 Å resolution by molecular replacement using human N1-(1873-2115) (Protein Data Bank code 1YYH) as a search model (22) (supplemental Table S1). The N1 ARD (OH) structure contains six observed ARs (ARs 2-7) that fit the defined consensus. Like other ARD folds, N1ARD (OH) is curved with a slight twist along its longest axis with neighboring repeats held together by hydrophobic forces that extend throughout the ARD core (Fig. 5A).
A distinguishing feature of the electron density map for the N1 ARD (OH) structure relative to other ARD structures was clear evidence for hydroxylation at the pro-S position of the ␤-carbon of Asn-1945 (no such electron density was observed at Asn-2012) (Fig. 5B). This observation supports assignments, based on NMR and synthetic standards, that FIH catalyzes the hydroxylation of HIF1␣CAD Asn-803 in an analogous manner (23). It is also the first direct crystallographic evidence for the stereochemistry of post-translational Asn hydroxylation in an intracellular mammalian protein. Interestingly, The conformation of the hydroxylated N1 ARD is very similar to that observed in previously published structures of unhydroxylated N1 ARDs (0.41 Å root mean square deviation based on C-␣ atoms in the N1 ARD (OH) and human N1-(1873-2115) structures (22)), indicating that hydroxylation does not significantly affect the conformation of the crystalline ARD fold (Fig. 5C). Furthermore, the positions of Asn-2012 and hydroxy-Asn-1945 are distal to any of the protein-protein-DNA interactions in the DNA-bound Notch transcriptional complex (24,25). These observations are consistent with results showing that hydroxylation does not significantly regulate the Notch/CSL pathway or the interaction of N1 with known ARD binding partners.
Crystal Structures of N1 Site 1 and Site 2 Peptides Bound to FIH-The hydroxylation sites in IB and Notch ARDs are located within the "␤-hairpin" type loops that connect ARs ( Fig. 5A and supplemental Fig.  S1). In contrast, HIF1␣ Asn-803 is located in the CAD, which does not have discernible secondary structure when isolated in solution (26). The HIF1␣CAD peptide binds to the surface of FIH in an extended groove that runs between the active site and the dimer interface, with the target Asn at the apex of a tight turn (19). Thus, in order for the ARD substrates to bind FIH productively, they must either undergo a considerable and unprecedented structural rearrangement to enable them to adopt a largely extended conformation, or they must bind to FIH in a different manner to HIF1␣CAD.
To investigate how AR substrates bind to FIH, we initially endeavored to solve structures of FIH in com-plex with recombinant full-length N1 ARD. Since this approach has not yet yielded FIH⅐N1 crystals, we obtained structures of FIH in complex with Fe(II), the cosubstrate 2OG, and the two N1 substrate peptides, N1-(1930 -1949) (site 1) to 2.4 Å (Fig.  6A) or N1(1997-2016) (site 2) to 2.6 Å (supplemental Fig. S7), under anaerobic conditions. FIH crystallized as a homodimer with each monomer adopting the double-stranded ␤-helix fold characteristic of the 2OG oxygenases and binding a single Fe(II), 2OG, and N1 peptide (Fig. 6A and supplemental Fig.  S7A; see supplemental Fig. S7 for a detailed analysis of the FIH⅐N1 peptide structures) (supplemental Table S1). Similar to HIF1␣CAD, the target Asn together with adjacent residues forms a tight turn in the FIH active site. In each of the FIH⅐N1 structures, the target asparagines are buried such that their C-3 prohydrogen projects toward the Fe(II) center ( Fig. 6C and supplemental Fig. S7D), consistent with the observed stereo-and regioselectivity of ARD hydroxylation by FIH (Fig. 5B).
The structural results reveal that binding of an AR to FIH occurs with induced fit involving conformational changes to both FIH and the ARD. Upon binding of N1 peptides, the FIH Trp-296 side chain moves relative to its position in the FIH⅐Fe⅐2OG structure; this movement accommodates the hydrophobic side chain of the residue N-terminal to the target N1 Asn. The same Trp-296 movement was observed for N1-(1997-2012) and N1-(1930 -1949) (despite the smaller side chain of Ala-1944 versus Val-2011), suggesting that some variability at this position is tolerated.
The backbone atoms of N1- (1930 -1949), N1-(1997-2016), and HIF1␣-(795-805) adopted very similar conformations when bound to FIH (Fig. 6D), and although there are differences in detail, the octahedral Fe(II) coordination chemistry (involving His-199, Asp-201, and His-279) and substrate binding mode of the N1 peptides was similar to that observed in the FIH⅐HIF1␣CAD⅐2OG complex ( Fig. 6C and supplemental Figs. S7D and S8) (19). As for HIF1␣CAD, the N1 peptides appear to bind primarily by backbone and hydrophobic interactions. The lack of specific side chain interactions other than the hydroxylated Asn appears to enable FIH to accommodate many different AR sequences at its active site. A leucinyl residue located 8 residues N-terminal to the Asn substrate (Leu-1937 in N1 site 1 and Leu-2004 in N1 site 2) makes a distinct interaction with a hydrophobic pocket on the surface of FIH that is likely to be important for binding and catalysis (Fig. 6E). Interestingly, this residue is conserved in all identified FIH substrates, including HIF1␣ (supplemental Fig. S1) (13). Therefore, the preferred FIH substrate motif (D/E) Ϫ2 X Ϫ1 N may be more precisely described as L Ϫ8 (D/E) Ϫ2 X Ϫ1 N.
In silico modeling analyses indicated that in order for the substrate Asn to bind the FIH active site and for the AR substrate to bind to FIH in the extended conformation described above, there must be a conformational change in the ARD. 12 To detail these structural alterations, we directly compared the conformation of the N1 substrate peptide when bound to FIH with that of the same residues in the ARD fold (Fig. 6F). Aside from the overall extension of the ␤-hairpin loop, there are other major changes in conformation. Notably, there is significant rotation in the peptide bond between the Ϫ1 residue and the target Asn (supplemental Table S2) relative to the conformation in the N1 ARD structure (Fig. 6F and supplemental Fig.  S7E). In addition, Leu-1937 and -2004 are located near the center of the H2 helix of the preceding AR, which must unfold in order for the leucinyl side chain to become buried in the hydrophobic pocket of FIH ( Fig. 6F and supplemental Fig. S7E).
Asn Hydroxylation Regulates the FIH-N1 ARD Interaction-Reduced binding of hydroxylated product versus substrate is precedented for 2OG oxygenases, such as in the case of the hydroxylation reaction catalyzed by clavaminic acid synthase (27). In the case of Notch, Asn hydroxylation creates a hydrogen bond to the acidic residue at the Ϫ2-position of the substrate motif (Fig. 5B) that may also stabilize the ␤-hairpin loop of the AR and therefore hinder the conformational changes required for productive binding of the AR substrate to FIH. To test these possibilities, we investigated the binding of unhydroxylated or hydroxylated N1 ARD (100% Asn-1945 and 20% Asn-2012) to FIH using electrospray ionization MS under nondenaturing conditions (Fig. 7A and supplemental Fig. S9). In the absence of added Fe(II) and 2OG, K d values of 4.0 and 48.9 M, respectively (excluding data for FIH 2 (N1 ARD) 2 ; supplemental Fig. S9), demonstrated tighter binding of the unhydroxylated N1 ARD. The K d value of FIH and His-HIF1␣CAD (105.9 M) showed that N1 ARD binds significantly tighter. The analysis also demonstrated that two N1 ARD molecules can bind to dimeric FIH simultaneously under these conditions (supplemental Fig. S9).
To confirm the effects of hydroxylation on FIH binding, we undertook pull-down assays using PK-⌬N1 ICD that was either completely (⌬N1 ICD-OH; FIH overexpression) or incompletely (⌬N1 ICD-H/OH) hydroxylated. The two forms of N1 ICD were incubated with His-FIH, and complex formation was assessed (Fig. 7B); partially hydroxylated ⌬N1 ICD captured ϳ40% of His-FIH, whereas Ͻ2% of His-FIH was captured by ⌬ICD-OH. Together, these results agree with observations  made with other 2OG oxygenases and may be consistent with a role for the hydrogen bond between the hydroxylated Asn and the Ϫ2 acidic residue in hindering the conformational change required for FIH binding.
Notch ARDs Compete with HIF␣CAD-The structural results reveal that ARs bind to the substrate binding groove of FIH essentially as for HIF␣CAD. Given the widespread extent of AR sequences closely related to N1 Site 1 and 2 (15), the structures indicate that many AR sequences may be accommodated within the FIH active site. This, along with the observation that N1 ARD binds FIH tighter than HIF␣CAD, raised the possibility that FIH-mediated HIF␣CAD hydroxylation may be competitively inhibited by ARD proteins. To test this, we incubated N1 ARD, HIF1␣CAD, and FIH in vitro (Fig. 7C, left). HIF1␣CAD hydroxylation was dramatically reduced in the presence of an equimolar amount of N1 ARD. Conversely, FIH-mediated N1 Asn-1945 hydroxylation was rapid but relatively unaffected by the addition of HIF1␣CAD (Fig. 7C,  right). N1 ARD Asn-2012 hydroxylation was minimal over this time course and was also unaffected by HIF1␣CAD. To determine whether similar competition was observed in vivo, we overexpressed a GAL4 DNA binding domain (DBD) fusion of HIF1␣CAD and PK-⌬N1 ICD either in isolation or in combination, prior to purification and LC-MS analyses of Asn hydroxylation sites. Fig. 7D shows that, in an identical manner to the in vitro data, N1 is a potent inhibitor of HIF1␣CAD hydroxylation, whereas N1 Asn hydroxylation is not significantly affected by HIF1␣CAD.
To determine whether competition promotes HIF1␣CAD activity in vivo, we used a UAS luciferase reporter system that is activated by the unhydroxylated GAL4DBD-HIF␣CAD⅐p300 complex. HeLa cells were transfected with GAL4DBD-HIF␣CAD and UAS-luciferase in the presence/absence of FIH siRNA or Notch ARD overexpression (Fig.  7E). As anticipated, FIH siRNA dramatically induced HIF␣CAD transcriptional activity and luciferase expression, confirming this system as an assay of FIH activity. Similar to FIH siRNA, N1 ARD overexpression induced significant levels of HIF␣CAD transcriptional activity. This effect was dependent on Asn-1945 and Asn-2012, since N1 ARD N1945A/N2012G was much less effective. Similarly, HIF1␣CAD activity was significantly induced by wild type N1 ICD overexpression, but not N1 ICD N1945A/N2012G. Overexpression of N3 ARD also induced luciferase, but N4 ARD, which is not an FIH substrate (Fig. 1B), did not (Fig. 7E). The ARD-mediated effects described here were specific, since they were not seen in GAL4DBD-VP16 or CSL-luc transcriptional reporter assays (supplemental Fig. S10).

DISCUSSION
Despite the widespread occurrence of post-translational hydroxylation in extracellular proteins, the hydroxylation of alkyl amino acid side chains of intracellular proteins has been thought to be rare (1). Together with our recent work on IB and p105 (13), we have identified the interaction of FIH with a total of eight ARD-containing proteins, including Notch. We show here that selected ARs in the ARDs of Notch receptors are targets for FIH-mediated Asn hydroxylation; two Asn residues in an L Ϫ8 (D/E) Ϫ2 X Ϫ1 N motif are hydroxylated to different extents both in vitro and in the endogenous protein. Considering the prevalence of this FIH substrate motif in these and the ϳ200 other human ARD proteins, the results imply that FIHmediated Asn hydroxylation is a common post-translational modification.
Structural analyses of hydroxylated N1 ARD revealed that although a hydrogen bond was formed between the hydroxyl group of Asn-1945 and the carboxylate group of Asp-1943, there was no major change in the crystalline conformation of the ARD compared with reported N1 structures (Fig. 5). Consistent with this, Asn hydroxylation did not affect the interaction of the N1 ARD with components of the Notch transcriptional complex (Fig. 4, C and D). Although it is possible that hydroxylation regulates another Notch activity, such as that in the Deltex-dependent pathway (28), we did not observe any affect of hydroxylation on the interaction of N1 ARD with Deltex (Fig. 4C).
The crystallographic analyses of the hydroxylated Notch ARD and of Notch peptides bound to FIH revealed that the canonical ARD fold must undergo major conformational changes in order to bind the FIH active site. Although work with short Notch substrate peptides may not exactly reflect the ARD-FIH interaction in solution, it is unclear at present how the ARDs of Notch and IB␣ unfold in order for the target Asn to bind productively to the FIH active site (13). In vitro this process is apparently mediated by FIH itself and may be related to the relatively tight binding of FIH (via interactions at the active site and beyond) to these ARD-containing proteins, compared with HIF1␣CAD (Fig. 7A). It will now be of interest to determine if the ability of FIH to partially unfold and hydroxylate ARDs extends to all members of the ARD family or is limited to a subset that are unusually flexible. Recent biophysical studies have revealed that some ARDs are more flexible than crystallographic studies imply, with the extent of conformational flexibility varying for different ARs within the overall ARD (29). Folding of the Drosophila N1 ARD proceeds according to a discrete pathway via an intermediate ensemble with significant structure in ARs 3-5; ARs 2, 6, and 7 only become structured in the step leading from the intermediate ensemble to the native state (30). A molten globule structure is proposed for folded IB␣; AR 3 is the most compact, followed by ARs 2 and 4, with ARs 1 and 6 being solvent-exposed (29). The preference for hydroxylation at AR 2 Ͼ AR 4 in N1 (Fig. 1D), and for AR 6 Ͼ AR 5 in IB␣ (13) may therefore reflect the ease of unfolding of ARs at the ARD termini.
Our results suggest that in addition to long range determinants of selectivity, FIH-catalyzed ARD Asn hydroxylation is also governed by the sequence immediately surrounding the target Asn. First, there is striking conservation of these residues between FIH substrates (supplemental Fig. S1) (13). Second, N4 site 1 is not hydroxylated (Fig. 1B), probably due to site 1 lacking the acidic residue at the Ϫ2-position (supplemental Fig. S1) and the presence of a proline at the Ϫ1-position that would be predicted to disrupt the tight turn required for productive binding at the FIH active site (Fig. 6C) (19). Interestingly, it is also probable that the hydrogen bond formed between the hydroxyl group of the target Asn and the acidic residue at the Ϫ2-position of the substrate motif might stabilize the ␤-hairpin type loop and limit the conformational change required for FIH binding, thus contributing to the more than 10-fold reduction in affinity for FIH we observed after N1 Asn-1945 hydroxylation (Fig. 7, A and B). The acidic residue at the Ϫ2-position is present in at least one AR of many ARDs. Therefore, this internal hydrogen bond may be a general consequence of FIH-mediated ARD hydroxylation.
Binding studies indicated that the affinity of FIH for the unhydroxylated N1 ARD was more than 20-fold greater than that for the unhydroxylated HIF␣CAD. In keeping with this, we have been able to co-immunoprecipitate a variety of ARD-containing proteins from cell extracts under conditions which did not capture HIF␣, 8 suggesting that this differential affinity is operative in vivo and extends to other ARD family members. We also observed a striking preference for ARD hydroxylation over HIF␣CAD hydroxylation in competition assays, with the presence of NI ARD sequences greatly reducing HIF␣CAD hydroxylation but not vice versa. These results were consistently observed both in vitro using purified proteins (Fig. 7C) and in vivo in transfected cells (Fig. 7D). Moreover, co-transfection of N1 and N3 ARDs, but not mutant N1 or the non-FIH substrate N4, reversed FIH-mediated suppression of HIF␣CAD activity (Fig. 7E). Taken together with the knowledge that ARD proteins are ubiquitous and in some cases abundant, these findings strongly suggest that ARDs may limit the availability of FIH for suppression of HIF␣CAD activity. These findings are consistent with, and potentially provide an explanation for, otherwise puzzling observations of strong HIF transcriptional activity under circumstances when effective FIH suppression might be anticipated. Thus, stabilized HIF␣ proteins in von Hippel-Lindau-deficient cells (31) and HIF␣ proteins that escape destruction following overexpression (32) generate strong transcriptional responses despite the presence of FIH, whereas in normoxic cells, modest increases in FIH expression suppress persistent HIF target gene expression (16), again implying that normal levels of FIH are rather surprisingly incompletely active on HIF.
Given the (i) large effects of hydroxylation on the affinity of the N1 ARD for FIH, (ii) tighter binding of unhydroxylated ARDs to FIH than HIF␣, and (iii) observation of substantially reduced hydroxylation of endogenous human N1 after exposure of cells to 48-h hypoxia, it is possible that in vivo competition between the ubiquitous ARD proteins and HIF␣ for FIH is also modulated by the hydroxylation status of cellular ARDs. Thus, due to its relatively high affinity for unhydroxylated ARDs, FIH may be more effectively sequestered in ARD complexes under conditions in which hydroxylation is compromised, such as chronic hypoxia, thereby enhancing the effect of hypoxia on HIF signaling beyond what might be predicted from kinetic studies of FIH-dependent HIF␣CAD hydroxylation alone. Under such conditions, it is also possible that FIH affects the activity of ARD proteins independently of hydroxylation, as observed with N1 ICD following FIH overexpression in P19 cells (Fig. 4A). Interestingly, evidence for physical association of HIF and Notch (11) raises the further possibility of local competition between different FIH binding sites in such complexes.
Previously, the only characterized interfaces between oxygen and HIF regulation were the hydroxylation of HIF␣ Pro and Asn residues that either enable (HIF␣-pVHL) or disable (HIF␣-p300) protein-protein interactions important in transcriptional regulation (33). ARD hydroxylation provides a new and more complex interface between HIF and oxygen, which has the potential to amplify the oxygen-dependent regulation of HIF transcriptional activity and contribute to physiological tuning of hypoxia signaling. The proposal that FIH has a range of substrates also has implications for the development of selective therapies that target HIF activation via hydroxylase inhibition.