Coiled-coil Coactivators Play a Structural Role Mediating Interactions in Hypoxia-inducible Factor Heterodimerization*

Background: Coiled-coil coactivators can enhance HIF-dependent gene transcription via direct interaction with the HIF/ARNT heterodimer. Results: ARNT uses the β-sheet of the PAS-B domain to recruit coiled-coil coactivators. Conclusion: Coiled-coil coactivators bridge HIF and ARNT via the PAS-B domain β-sheet contacts to both proteins to form a ternary structure. Significance: This work reveals the mechanism for assembling a coiled-coil coactivator complex with the HIF-2 transcription factor heterodimer. The hypoxia-inducible factor complex (HIF-α·aryl hydrocarbon receptor nuclear translocator (ARNT)) requires association with several transcription coactivators for a successful cellular response to hypoxic stress. In addition to the conventional global transcription coactivator CREB-binding protein/p300 (CBP/p300) that binds to the HIF-α transactivation domain, a new group of transcription coactivators called the coiled-coil coactivators (CCCs) interact directly with the second PER-ARNT-SIM (PAS) domain of ARNT (ARNT PAS-B). These less studied transcription coactivators play essential roles in the HIF-dependent hypoxia response, and CCC misregulation is associated with several forms of cancer. To better understand CCC protein recruitment by the heterodimeric HIF transcription factor, we used x-ray crystallography, NMR spectroscopy, and biochemical methods to investigate the structure of the ARNT PAS-B domain in complex with the C-terminal fragment of a coiled-coil coactivator protein, transforming acidic coiled-coil coactivator 3 (TACC3). We found that the HIF-2α PAS-B domain also directly interacts with TACC3, motivating an NMR data-derived model suggesting a means by which TACC3 could form a ternary complex with HIF-2α PAS-B and ARNT PAS-B via β-sheet/coiled-coil interactions. These findings suggest that TACC3 could be recruited as a bridge to cooperatively mediate between the HIF-2α PAS-B·ARNT PAS-B complex, thereby participating more directly in HIF-dependent gene transcription than previously anticipated.


The hypoxia-inducible factor complex (HIF-␣⅐aryl hydrocarbon receptor nuclear translocator (ARNT)) requires association with several transcription coactivators for a successful cellular response to hypoxic stress. In addition to the conventional global transcription coactivator CREB-binding protein/p300 (CBP/p300) that binds to the HIF-␣ transactivation domain, a new group of transcription coactivators called the coiled-coil coactivators (CCCs) interact directly with the second PER-ARNT-SIM (PAS) domain of ARNT (ARNT PAS-B)
. These less studied transcription coactivators play essential roles in the HIF-dependent hypoxia response, and CCC misregulation is associated with several forms of cancer. To better understand CCC protein recruitment by the heterodimeric HIF transcription factor, we used x-ray crystallography, NMR spectroscopy, and biochemical methods to investigate the structure of the ARNT PAS-B domain in complex with the C-terminal fragment of a coiled-coil coactivator protein, transforming acidic coiledcoil coactivator 3 (TACC3). We found that the HIF-2␣ PAS-B domain also directly interacts with TACC3, motivating an NMR data-derived model suggesting a means by which TACC3 could form a ternary complex with HIF-2␣ PAS-B and ARNT PAS-B via ␤-sheet/coiled-coil interactions. These findings suggest that TACC3 could be recruited as a bridge to cooperatively mediate between the HIF-2␣ PAS-B⅐ARNT PAS-B complex, thereby participating more directly in HIF-dependent gene transcription than previously anticipated.
Hypoxia-inducible factor (HIF) 3 proteins are the central regulators of the mammalian hypoxia response (1), consisting of an O 2 -regulated ␣ subunit (HIF-1␣, -2␣, and -3␣) and the stably expressed ␤ subunit (ARNT or HIF-␤) (2). Under hypoxia, stabilized HIF-␣ and ARNT subunits dimerize through the N-terminal basic helix loop helix (bHLH) and two Per-ARNT-Sim (PAS) domains. This heterodimer binds to the hypoxia-response element promoter with its N-terminal bHLH domain and controls the transcription of hundreds of target genes such as pro-angiogenic factors and metabolic enzymes (2,3).
HIF target gene regulation depends on the participation of several transcription cofactors. The intrinsically unfolded C-terminal transactivation domain of HIF-␣ subunit plays a major role in this process by directly interacting with the global transcription coactivator CBP/p300 (4). Interestingly, recent studies revealed a group of transcription coactivators involved in cancer development and progression, namely the coiled-coil coactivators (CCCs), could be recruited in a transactivation domain-independent manner (5)(6)(7). Three CCC family members have been described to date as follows: coiled-coil coactivator (8), thyroid hormone receptor interacting protein 230 (TRIP230) (9), and transforming acidic coiled-coil 3 (TACC3) (10). Under normal situations, these coactivators play an essential role in the hypoxia response by directly interacting with the ARNT subunit in a promoter-specific way. However, misregulation by overexpression or activating fusions caused by chromosomal translocations (e.g. FGF receptor-TACC3) is sufficient for transformation and is associated with the development of glioblastoma, renal cell carcinoma, and other cancers (11). * This work was supported, in whole or in part, by National Institutes of Health To investigate the molecular details of the HIF⅐CCC complex, we first characterized the structure of minimal interacting fragments of ARNT⅐TACC3 using x-ray crystallography and solution NMR spectroscopy. We previously reported that ARNT PAS-B utilizes its F␣-helix as the main interface to directly bind to the C terminus of TACC3 in vitro (6). However, crystal structures solved in this study directly demonstrate that the ␤-sheet of ARNT PAS-B serves as the TACC3-binding site. Subsequent solution NMR measurements reveal that the HIF-2␣ PAS-B ␤-sheet also binds to the TACC3 C terminus. Based on these observations, we generated a ternary complex model of ARNT PAS-B⅐TACC3⅐HIF-2␣ PAS-B that is supported by several lines of experimental data. The cooperative formation of a ternary complex among TACC3, ARNT PAS-B, and HIF-2␣ PAS-B described here could shed light on the gen-eral mechanism of the CCC protein recruitment in HIF signaling and provide a structural framework to inform future anticancer therapies.
Pulldown Assay-For pulldown experiments, 5 M purified His-ARNT PAS-B and 8 -15 M TACC3 were incubated with 15 l of nickel-nitrilotriacetic acid beads overnight at 4°C. Beads were washed twice with buffer before elution. Eluted protein was resolved in SDS-PAGE and stained with Coomassie Blue stain.
X-ray Crystallography-Single crystals of ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A were grown by hanging drop vapor diffusion against 1.0 M succinic acid (pH 6.5), in 25 mM Tris (pH 7.5), 17 mM NaCl, 5 mM ␤-mercaptoethanol buffer. Drops containing 2 l of 2.5 mg/ml ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A (molar ratio ϭ 1:2) were mixed with 0.7 l 1.0 M succinic acid (pH 6.5) (reservoir solution) and 0.3 l 30% sorbitol (additive). Crystals were observed within 2 days and reached maximum size within 1 week at 20°C. The crystals exhibited C2 space group symmetry with cell dimensions of a ϭ 116.85 Å, b ϭ 59.82 Å, c ϭ 73.51 Å, and ␤ ϭ 97.85°, contained two molecules of ARNT PAS-B and one molecule of GCN4-TACC3-CT D622A/E629A per asymmetric unit, and diffracted to a minimum Bragg spacing (d min ) of 3.15 Å when exposed to synchrotron radiation. The where the outer sum (h) is over the unique reflections, and the inner sum (i) is over the set of independent observations of each where n h is the number of observations of reflections h. c Information was not reported in this version of HKL3000 when processing the data. d Data are as defined by the validation suite MolProbity (19). diffracted to a minimum Bragg spacing (d min ) of 3.20 Å when exposed to synchrotron radiation.
Diffraction data for both complexes were collected at the Advanced Photon Source beamline 19-ID and were indexed, integrated, and scaled using the HKL-3000 (15) program package. Data collection statistics are provided in Table 1. For the ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A data, extending the resolution limit beyond 3.15 Å did not substantially improve calculated electron density maps and provided poorer refinement results, likely due to anisotropy that in turn limited data completeness in the highest resolution shells.
Phases for ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A were obtained via molecular replacement in the program Phaser (16) using a search model derived from the previously determined ARNT PAS-B domain crystal structure (PDB code 4EQ1 (17)). Two copies of ARNT PAS-B were located in the asymmetric unit, and inspection of the electron density maps revealed density corresponding to the coiled-coil domain, which was initially modeled as a pair of idealized polyalanine helices. Identification of the inter-helical disulfide bond allowed accurate assignment of GCN4-TACC3 density for residues 595-626.
Refinement was performed to a resolution of 3.15 Å with noncrystallographic symmetry restraints using the program Phenix (18) with a random 10% of all data set aside for an R free calculation. The final refinement included two ARNT PAS-B monomers (excluding extended loop residues 447-454) and one TACC3 homodimer (modeling residues 595-625 of monomer C and residues 596 -626 of monomer D). The R work is 0.247 and the R free is 0.272. A Ramachandran plot generated  (6). c and d, locations of sites on ARNT PAS-B ␤-sheet (c) and F␣ helix (d) to test the structural models of ARNT/TACC3 interactions. e, Ni 2ϩ pulldown assay and band quantification (mean Ϯ one S.D., n ϭ 2) shows that His-ARNT PAS-B ␤-sheet mutations modulate the binding of WT GST-TACC3, indicating that the ␤-sheet of ARNT PAS-B is the direct TACC3 binding interface. Similar to TACC3 interaction, nickel pulldown assay and band quantification (mean Ϯ one S.D., n ϭ 2) shows His-ARNT PAS-B ␤-sheet mutations change its affinity for WT GST-TRIP230(1583-1688), indicating that the ␤-sheet of ARNT PAS-B is the major interface for TRIP230 binding as well. f, nickel pulldown assay and band quantification show that His-ARNT PAS-B F␣ mutation is unable to alter its affinity with WT GST-TACC3(581-631) and WT GST-TRIP230(1583-1688), indicating that the F␣ is not the major CCC binding interface.
with MolProbity (19) indicated that 95.9% of all protein residues are in the most favored regions, and none occupies disallowed regions.
Phases for ARNT PAS-B E362R⅐TACC3-CT⅐HIF-2␣ PAS-B R247E crystals were also obtained via molecular replacement with Phaser. PAS domain search models were constructed from the previously determined ARNT PAS-B E362R⅐HIF-2␣ PAS-B R247E heterodimer coordinates (PDB code 3F1O (20)) by removal of N-and C-terminal residues 356 -361 and 448 -453 from ARNT PAS-B E362R, and residues 236 -242 and 345-349 from HIF-2␣ PAS-B R247E. The search model for the TACC3 domain was derived from residues 595 to 609 of the ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A structure, with all residues converted to alanine. Two copies of ARNT PAS-B E362R and two copies of the TACC3-CT homodimer were located in the unit cell. Inspection of these initial electron density maps revealed density corresponding to additional helical density at both the N and C termini of the TACC3-CT. Phases were further improved by density modification with 2-fold noncrystallographic symmetry averaging in the program Parrot (21) resulting in a figure-of-merit of 0.745. A model containing 92.7% of all residues in the two TACC3 domains was automatically generated in the program Buccaneer (22). The inter-helical disulfide bond used to confirm the register of the helices  was observed in one TACC3 homodimer (chains E and F) but was reduced in the second homodimer (chains B and C). Inspection of electron density maps after rigid body refinement of the ARNT PAS-B E362R⅐TACC3 heterodimer using Phenix revealed unmodeled density suggestive of another PAS domain. Identification of this as the HIF-2␣ PAS-B R247E domain (and not the ARNT PAS-B E364R) was confirmed via the statistics from molecular replacement in Phaser and inspection of kicked omit maps following refinement in Phenix. Additional residues for ARNT PAS-B E362R⅐TACC3⅐HIF-2␣ PAS-B R247E were manually modeled in the programs O (23) and Coot (24). Refinement was performed to a resolution of 3.20 Å using the program Phenix with a random 10% of all data set aside for an R free calculation. Noncrystallographic symmetry averaging and reference model restraints were used in initial rounds of refinement but were removed once the R free dropped below 0.30, and the geometry was stabilized. Because of the low resolution of the data, grouped isotropic as well as TLS atomic displacement parameters were refined. Mean atomic displacement parameters were 43.1-56.4 Å 2 for the domains of the ARNT PAS-B E362R⅐TACC3 heterodimer and 83.3 Å 2 for the HIF-2␣ PAS-B R247E domain. The current model contains two ARNT PAS-B E362R monomers (excluding extended loop residues 449 -451 for monomer A and residue 450 of monomer D); two TACC3 homodimers, including residues 582-631 of monomers B, E, and F, and residues 582-629 of monomer C; one monomer of HIF-2␣ PAS-B R247E, including residues 239 -344; and two sulfate anions. The final R work and R free values are 0.231 and 0.270. A Ramachandran plot generated with Molprobity indicated that 97.5% of all protein residues are in the most favored regions, and 0.2% (one residue) in the disallowed regions. A complete summary of phasing and model refinement statistics for both complexes are provided in Table 1.
Microscale Thermophoresis (MST)-MST experiments were conducted using a Monolith NT.115 (NanoTemper). ARNT PAS-B E362R was labeled with the blue fluorescent dye NT-495-NHS using a vendor-supplied kit (NanoTemper) and buffer exchanged to the assay buffer (25 mM Tris (pH 7.5), 17 mM NaCl, 0.05% Tween 20, and 5 mM ␤-mercaptoethanol). Band intensity c. The ARNT/TACC3 affinity improvement in both mutant complexes seems likely to be caused by alleviation of the electrostatic repulsion between these negatively charged groups in WT proteins. c, schematic summary of the relationship between interfacial charge-charge interactions and ARNT/TACC3 affinity (TACC3, orange; ARNT PAS-B, blue).
Titrations were conducted with a constant 500 nM fluorophorelabeled ARNT PAS-B (ARNT PAS-B E362R-fluor) against up to 500 M titrant in standard coated capillaries. Each data point was measured in triplicate. Single-site fitting was done by NanoTemper data analysis software. For bimodal MST data, a fitting model for the formation of ABB complexes was constructed in a Python script. The script uses a numerical method to determine the concentrations of all possible (macroscopic) species as follows: free A, free B, AB, and ABB. The overall thermophoretic signal (f) was considered to be a linear combination of the signals of the detectable species, f(B), f(AB), and f(ABB) (21). The two B-binding sites were constrained to have identical microscopic binding affinities, meaning that the macroscopic binding constant of the formation of the ABB complex was 4-fold higher than that for the formation of the AB complex. NMR Spectroscopy-NMR experiments were carried out at 25°C on Varian 600 and 800 MHz spectrometers. Chemical shift assignments of ARNT PAS-B and HIF-2␣ PAS-B were used as established previously (25,26). All NMR data were processed with NMRpipe/NMRDraw (27)  Paramagnetic Relaxation Enhancement-PRE was measured by comparing the backbone amide proton transverse ( 1 H N T 2 ) relaxation rates under paramagnetic and diamagnetic conditions. These relaxation rates were determined using a modified 15 N/ 1 H TROSY-HSQC experiment with delay times of 6.5, 8, 10, 13, 16, and 26 ms. Rates were calculated using NMRViewJ (Version 9.0.0, One Moon Scientific). Paramagnetic labeling with (1-oxyl-2,2,5,5-tetramethyl-⌬3-pyrroline-3-methyl) methanethiosulfonate (MTSL) or (1-oxyl-2,2,5,5-tetramethylpyrroline-3-yl) carbamidoethyl methanethiosulfonate (CMTSL) (Toronto Research Chemicals) at a specific cysteine site of target protein was conducted by incubating protein and spin label at a 1:2 molar ratio under a nonreducing condition and rotating overnight at room temperature. Labeled protein was purified by gel filtration through a Superdex 75 column. A diamagnetic control was generated by adding 10-fold sodium dithionite to the labeled protein and purging with N 2 (g) for 30 min to quench the spin label.

JOURNAL OF BIOLOGICAL CHEMISTRY 7713
C2 symmetry restraints were turned on to lock TACC3 as a parallel dimer in the docking process.

RESULTS
Overall Structure of ARNT⅐TACC3-The ARNT PAS-B domain and the last 20 amino acids of the TACC3 parallel dimer (residues 610 -631) of the TACC domain have been identified as the minimum binding fragments that are necessary and sufficient to form the ARNT⅐TACC3 complex (Fig. 1a) (6). However, high resolution structural studies of the wild-type ARNT PAS-B⅐TACC3 complex have been complicated by the moderate (low micromolar) affinity observed for this protein/ protein interaction.
To optimize the ARNT⅐TACC3 complex for crystallography, we conducted an alanine scan of the TACC3 C terminus (TACC3-CT), identifying multiple mutations that appeared to strengthen the ARNT PAS-B⅐TACC3-CT complex as accessed by coprecipitation (Fig. 1, a and b). Among these mutations, two changes to surface-exposed charged residues (D622A and E629A) exhibited the strongest effects, whereas a TACC3 D622A/E629A double mutant provided an even stronger result (Fig. 1b). An N-terminal GCN4 fusion with TACC3-CT(610 -631) D622A/E629A was used to ensure that the TACC3 double mutant was present as a constitutive homodimer (Fig. 1, c and  d) (14,29). Well dispersed peaks representing 43 out of 44 residues in the 15  and their corresponding peak broadening effects mapped on 15 N-labeled ARNT PAS-B (blue, less broadening; red, more broadening). TACC3 Q609C-MTSL induced the most substantial peak broadening effects, whereas TACC3 M605C-and S602C-MTSL induced less broadening consistent with further distance from ARNT PAS-B. Lower panel, peak broadening ratio distribution induced by the corresponding spin labels. Histograms of these ratios show mean peak intensity ratio histograms shift to a lower number (to the left, more broadening) when moving the spin label toward the C terminus of TACC3, demonstrating that the average distance between spin labels and ARNT PAS-B sites (r) follows the progression r S602C Ͼ r M605C Ͼ r Q609C .
GNC4 fusion protein is well folded in solution (Fig. 1c). Nicely superimposed peaks in the 15 N/ 1 H TROSY spectra and the constant molar mass determined by size-exclusion multiangle laser light scattering further confirmed that the GCN4 fusion protein is a homogeneous homodimer at various concentrations in solution (Fig. 1, c and d). The GCN4-TACC3-CT D622A/ E629A protein can also compete away the WT GST-TACC3-(561-631) from a pre-formed complex with ARNT PAS-B in a pulldown assay, demonstrating that this stabilized dimeric high affinity TACC3 construct shares the same binding interface on ARNT PAS-B as WT TACC3 (Fig. 1e). The high affinity ARNT PAS-B⅐GCN4-TACC3-CT D622A/ E629A complex readily crystallized, ultimately providing a 3.15 Å dataset that was phased by molecular replacement using a crystal structure of the isolated ARNT PAS-B domain as a search model (Table 1 and Fig. 2a, PDB code 4LPZ) (17). As anticipated, GCN4-TACC3-CT D622A/E629A crystallized as a parallel coiled-coil dimer. Although the TACC3 portion of the fusion protein (Fig. 2a, orange, residue 610 -631) contacts the ␤-sheet of ARNT PAS-B, burying a surface area of 610 Å 2 , the GCN4 tag (magenta) does not directly contact ARNT. Interestingly, the ARNT⅐TACC3 binding interface shown in this structure is different from a previous NMR-guided computational model where TACC3 makes primary contacts with ARNT PAS-B on the F␣-helix (Fig. 2b) (5). This observation makes it critical to further validate the current crystal structure.
Validating the Binding Interface in ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A Crystal Structure by Mutagenesis-To confirm the ARNT⅐TACC3 binding interface revealed in our crystal structure of the complex, we extensively mutated the ARNT⅐TACC3 binding interface on the ARNT PAS-B side. As expected, single and triple mutants on the ␤-sheet of the PAS domain attenuated TACC3 binding (Fig. 2, c and e), although mutations on the ARNT F␣-helix, a surface that is not involved in the crystallographically defined binding interface, did not show comparable reduction in TACC3 binding (Fig. 2, d  and f), confirming the ARNT PAS-B ␤-sheet as the major TACC3 binding interface.
We additionally tested the ARNT PAS-B complex with another CCC protein, TRIP230, testing whether ARNT PAS-B recruits a different CCC protein via the same ␤-sheet interface, as suggested by prior competition assays (6). As was the case for TACC3, most ␤-sheet mutants weakened the ARNT⅐TRIP230 interaction, whereas the F␣-helix mutants showed minimum effects, demonstrating that the ARNT PAS-B ␤-sheet is essential for CCC protein recruitment (Fig. 2, e and f).
Electrostatic Interactions Are Critical for ARNT⅐TACC3 Binding-A close look at the binding interface of ARNT PAS-B and GCN4-TACC3-CT D622A/E629A revealed many interprotein contacts involving charged residues. Mutations at two such residues on ARNT PAS-B (H378D and R379E) markedly reduced complexation with TACC3 in pulldown assays (Fig. 2,  c and e), likely by removing an interface-spanning salt bridge by exchanging a stabilizing ϩ/Ϫ charged residue pair for a repelling Ϫ/Ϫ charged pair (Fig. 3a). However, ARNT PAS-B E362R and E455R showed enhanced complex formation with TACC3 (Fig. 2, c and e), likely due to the formation of a new salt bridge with TACC3 Asp-622 and Asp-623, respectively, by flipping a Ϫ/Ϫ charged residue pair to a ϩ/Ϫ charged pair (Fig. 3a). 15 N/ 1 H HSQC spectra of all four mutants overlapped nicely with the WT ARNT PAS-B spectrum (Fig. 3b); the most substantial differences were observed with E455R, which exhibited peak loss in ϳ20% of residues suggestive of exchange broadening from increased dynamics, but otherwise it retained many peaks with characteristic chemical shifts identical to wild type. Taken together, these NMR data demonstrate that the binding perturbation observed in pulldown assays was not caused by protein unfolding or other artifacts (Fig. 3b).  Likewise, the high affinity mutants used in crystallography appear to work by alleviating electrostatic repulsion present at the ARNT⅐TACC3 interface by removing one negatively charged residue in the Ϫ/Ϫ charged residue pair consisting of TACC3 Asp-622 and ARNT PAS-B Glu-362. To test this hypothesis, a mutagenesis study with different combinations of charged residues at the ARNT Glu-362 and TACC3 Asp-622 positions was conducted to examine the binding affinity alteration. As expected, the highest affinity enhancement was observed in the ARNT E362R⅐TACC3 WT, ARNT WT⅐TACC3 D622R, and ARNT WT⅐D622K complex, where a pair of ϩ/Ϫ residues face each other at the binding interface. Moderate affinity enhancement was shown in the ARNT WT⅐TACC3 D622A complex, where one of the negatively charged residues was replaced with a neutral residue. Minimal effects were observed in both the Ϫ/Ϫ pairs present in the ARNT WT⅐TACC3 WT complex, and for ϩ/ϩ pair complexes such as ARNT E362R⅐TACC3 D622R and ARNT E362R⅐ACC3 D622K (Fig. 4, a and c). We observed such a salt bridge in another ARNT⅐TACC3 complex crystal structure solved to 3.2 Å resolution (ARNT PAS-B E362R⅐TACC3(585-631) WT) (Figs. 4b and 5a and Table 1, PDB code 4PKY). Superimposition of this structure with the previous ARNT PAS-B⅐GCN4-TACC3-CT D622A/E629A crystal structure showed excellent similarity (r.m.s.d. ϭ 0.9 Å over 153 aligned C␣ carbons; Fig. 5b), suggesting that the different mutations involved in stabilizing the two complexes had minimal effects on the observed structures. The stabilizing effects of ARNT Glu-362 and TACC3 Asp-622 mutations further establish their importance as critical contact spots at the binding interface, with changes to their charges substantially affecting binding. These results also confirmed that the ARNT PAS-B ␤-sheet, where residue 362 is located, is directly associated in TACC3 complexation.  (20)) (peak broadening is colored from less (blue) to more (red) with the largest effects apparent on the ␤-sheet). Residue Arg-247 was substantially affected by the TACC3 binding (black arrow). d, 15 N-labeled HIF-2␣ PAS-B⅐HisG␤1 TACC3(585-631) interaction shows comparable peak broadening effects as the 15 N-labeled HIF-2␣ PAS-B/ARNT PAS-B interaction, demonstrating that the TACC3-induced peak broadening is not due to nonspecific aggregation. e, HADDOCK (28) model of the binary HIF-2␣ PAS-B⅐TACC3 complex was generated by utilizing the top 20 most severely broadened HIF-2␣ PAS-B residues in 15 N-labeled HIF-2␣ PAS-B⅐TACC3 titration experiment and all solvent-exposed TACC3 residues in TACC3-CT as active residues (see Table 2). The most abundant cluster from the prediction is shown here, where TACC3 binds to HIF-2␣ PAS-B at the ␤-sheet. f, TACC3-Q609C-MTSL induces significant PRE effects at positions adjacent to the spin label as predicted in the HIF-2␣ PAS-B⅐TACC3 HADDOCK model, demonstrating that the model accurately represents the solution complex.
ARNT⅐TACC3 Interaction in Solution Is Similar to the Crystal Structure-Although our two independent crystal structures and accompanying mutagenesis data supported the use of an ARNT ␤-sheet binding mode, we sought to further validate this with additional structural information from the ARNT PAS-B⅐TACC3 complex in solution. To do so, we titrated TACC3(585-631) into 15 N-labeled ARNT PAS-B, with complex formation monitored by 15 N/ 1 H HSQC spectra. We observed intermediate chemical exchange behavior from the ARNT signals as TACC3 was titrated, suggesting a low to midmicromolar range affinity. The most severely broadened residues were localized to the ␤-sheet of ARNT PAS-B (Fig. 6, a and  b) (17,26), consistent with this site binding TACC3. Analyzing these data with the protein rigid body docking program HADDOCK (28) found TACC3 interacting with the ARNT PAS-B ␤-sheet in a fashion similar to that observed in the crystal structure (Fig. 6c). Superposition of this solution NMRguided model with the ARNT PAS-B E362R⅐TACC3(585-631) crystal structure revealed good agreement between solid state and solution measurements (r.m.s.d. ϭ 2.1 Å over 163 aligned C␣ carbons).
To further bolster support of the present HADDOCK-calculated model, we obtained long range distance restraints between sites on the TACC3 dimer and those in bound ARNT PAS-B domains using PRE (30). In this experiment, TACC3(585-631) residue Gln-609 was mutated to a cysteine (Q609C), facilitating the cross-linking of the nitroxyl spin label MTSL to this position (Fig. 6d). T 2 relaxation rates were measured from 15 N-labeled ARNT PAS-B residues by TROSY-HSQC in the presence of TACC3(585-631) Q609C-MTSL under both oxidative (paramagnetic status) and reducing conditions (diamagnetic status, negative control). The observed net change in relaxation rates measured under oxidative and reducing conditions were converted to PRE values (s Ϫ1 ). Residues close to the MTSL unpaired electron (less than 25 Å) should have larger differences in relaxation rates between these two conditions and correspondingly larger PRE values (highlighted in red spheres, Fig. 6d), whereas residues that are far away from the spin label should have smaller PRE values. For TACC3 Q609C-MTSL, 15 N-labeled ARNT PAS-B amide groups with the largest PRE values mapped to ARNT PAS-B sites proximal to TACC3 Gln-609 in the crystal structure (Fig.  6d). MTSL labeling at other TACC3 sites more distal to ARNT PAS-B in the crystal structure (S602C and M605C) attenuated the observed PRE effects (Fig. 6e). These solution NMR experiments collected from wild-type proteins provide key independent confirmations of the complex arrangement observed in the ARNT PAS-B⅐TACC3 crystal structures.
HIF-2␣ PAS-B Also Directly Interacts with the TACC3 C Terminus-TACC3-dependent transactivation of HIF genes requires the formation of a HIF-2␣, ARNT, and TACC3 ternary complex (6). We monitored formation of this complex using MST, a technique that detects macromolecular associations by changes in thermophoretic mobility, which in turn depends on complex molecular weight, hydration, and other parameters (31). MST data collected while titrating TACC3 into a sample of ARNT PAS-B E362R revealed three distinct states with increasing TACC3 concentrations (Fig. 7a) as follows: a free ARNT PAS-B domain; a 2:1 ARNT⅐TACC3 dimer complex; and a final complex with a 1:1 ARNT⅐TACC3 dimer (state iii). These data established the potential for a TACC3 dimer to simultaneously engage two PAS-B domains, as required for formation of the ternary complex.
Using conditions we identified to make the 1:1 ARNT PAS-B E362R⅐TACC3 dimer complex, we titrated HIF-2␣ PAS-B into these samples and controls of ARNT PAS-B E362R alone (Fig.  7b). As expected, HIF-2␣ PAS-B formed ternary complexes with the ARNT PAS-B E362R⅐TACC3 complex. Notably, this interaction was an order of magnitude tighter than the HIF-2␣⅐ARNT PAS-B E362R interaction on its own (Fig. 7b). Although the MST data establishes the existence of this threeway complex, they do not provide complementary views of the structural arrangements within the complex. A key piece of inferential information about this arrangement was provided by examining the effects of adding an allosteric HIF-2␣ inhibitor, compound 2 (32), to the ternary complex. This chemical binds within the HIF-2␣ PAS-B domain, triggering structural changes on its ␤-sheet and is known to down-regulate HIF-2driven transcription in cells. Here, we found that compound 2 weakened the binding of HIF-2␣ PAS-B to the ARNT⅐TACC3 binary complex by approximately an order of magnitude (Fig.  7c). This result discounts any potential nonspecific interaction between HIF-2␣ PAS-B interacting with the ARNT⅐TACC3 complex and further implicates the use of the HIF-2␣ PAS-B ␤-sheet as the binding surface for this interaction.
Based on these NMR mapping data, we generated a HADDOCK model of the TACC3 C-terminal dimer bound to HIF-2␣ PAS-B on the ␤-sheet (Fig. 8e and Table 2). We further validated this HIF-2␣ PAS-B⅐TACC3 HADDOCK model by PRE experiments where TACC3(585-631) Q609C-MTSL induced PRE effects on several 15 N-labeled HIF-2␣ PAS-B ␤-sheet residues that are adjacent to the spin label (red stick) (Fig. 8f).
Model of the ARNT⅐TACC3⅐HIF Ternary Complex-The preceding biophysical characterization and computational model of the HIF-2␣ PAS-B⅐TACC3 complex provides the structural basis for an initial view of ARNT⅐TACC3⅐HIF ternary complex assembly. By comparing the TACC3-binding modes with both HIF-2␣ and ARNT PAS-B, we hypothesized that TACC3 could use both sides of the coiled-coil dimer to simultaneously recruit HIF-2␣ and ARNT via their PAS-B ␤-sheets to assemble the ternary complex observed by MST. As a starting model for the ARNT⅐TACC3⅐HIF ternary complex, we superimposed the TACC3 dimers from the experimental ARNT PAS-B⅐TACC3 crystal structure (PDB code 4LPZ) and HIF-2␣ PAS-B⅐TACC3 HADDOCK model (Fig. 9a). Assembled together into this model, the ARNT and HIF-2␣ PAS-B ␤-sheets are able to simultaneously contact opposite surfaces of the TACC3 dimer.
To validate this model, we obtained distance restraints between ARNT PAS-B S451C-CMTSL and 15 N-labeled HIF-2␣ PAS-B residues by PRE under conditions with or without TACC3 (Fig.  9b). In the absence of TACC3, only a few HIF-2␣ PAS-B residues (located at the beginning of the A␤ strand) were affected  by the ARNT-bound spin label (Fig. 9c). These sites agree with the anti-parallel assembly in our previously determined HIF-2␣⅐ARNT PAS-B crystal structure (Fig. 9d) (20), suggesting that some of this complex formed under these solution conditions. These PRE effects on 15 N-labeled HIF-2␣ PAS-B were substantially enhanced in the presence of TACC3, especially for residues in the H␤ and I␤ strands, indicating both a closer proximity and different arrangement to the ARNT PAS-B S451C-CMTSL spin label in the presence of TACC3 (Fig. 9e). This result is well accommodated with a ternary structure model where most of the highly affected residues (spheres) are clustered in the vicinity of ARNT PAS-B S451C-CMTSL spin label (red sticks) (Fig. 9f). Taken together, this ternary complex model successfully integrates data from multiple biophysical and biochemical approaches, including the effects of an allosteric HIF-2␣ PAS-B inhibitor (32), giving us confidence in its representation of the solution complex of these three proteins.

DISCUSSION
Our HIF⅐TACC3 structural study reveals a unique mode of transcription activator/coactivator recruitment in the HIF-mediated hypoxia response, adding to the previously described canonical recruitment of p300 coactivator with transcriptional activation domains in the HIF-␣ C terminus (4). Here, we solved two crystal structures of ARNT PAS-B⅐TACC3 complexes and characterized the wild-type ARNT⅐TACC3 interac-tion in solution. These results consistently show that the ARNT PAS-B ␤-sheet provides the direct interface for TACC3 binding. Interestingly, this ␤-sheet interaction surface has also been found for other intermolecular interactions, such as with the HIF-␣ PAS-B domains (20), and more broadly with PAS domains in other bHLH/PAS transcription factors (33) and other types of proteins (13). Although this result might initially suggest competition between TACC3 and HIF-␣ PAS-B for ARNT binding, our MST data clearly demonstrate a cooperative effect among the three proteins (Fig. 7b).
To provide a structural basis for this cooperativity, we generated a data-driven model where TACC3 mediates a ternary complex with both PAS-B domains. Our model suggests that the ARNT PAS-B ␤-sheet can be bound either directly by HIF-2␣ (20) or TACC3 itself; if the latter, a new HIF-2␣ PAS-B-binding site is available on the opposite side of the TACC3 dimer to form the ternary complex. As a result, recruiting TACC3 to the ␤-sheet of ARNT PAS-B will not impair the integrity of the larger HIF-␣⅐ARNT heterodimer nor its transcription factor activity (Fig. 10).
We suggest that the alternative ARNT PAS-B binding modes indicated by our work, direct binding to HIF-2␣ PAS-B or in a ternary complex with HIF-2␣ and TACC3, may be in equilibrium in the full-length heterodimeric transcription factor. As such, mutations or other factors that shift this equilibrium Such modifications do not happen on ARNT and TACC3 so the ARNT⅐TACC3 complex is readily assembled in the nucleus. Hypoxia stops the hydroxylation, allowing HIF-2␣ to accumulate in the nucleus and complex with ARNT⅐TACC3. Here, we propose that TACC3 simultaneously engages the PAS-B domains from both HIF-2␣ and ARNT PAS-B, directing recruitment of this CCC protein to hypoxia-responsive enhancer (HRE) sites, controlling target gene transcription (blue arrows).
would be expected to shift the ability of the HIF transcription factor to activate gene transcription, with accompanying potential for misregulation. Notably, two such alterations are correlated within patient-derived tumor samples of various urogenital cancers (34), including a somatic TACC3 alteration in the ARNT⅐HIF-␣ interacting region (D622N in mouse sequence; D829N in human) and up-regulation of TACC3 levels through gene amplification in 5-10% of these cancers. We postulate that these effects could stabilize the formation of the HIF-␣⅐TACC3⅐ARNT ternary complex, either by a stabilizing mutation that functions similarly to our TACC3 D622A mutant in our crystallography study or by mass action, with the net effect of inappropriately enhancing HIF-dependent gene up-regulation.
On a technical point, it is worth noting that the binding models we describe here for ARNT⅐TACC3 interactions differ somewhat from a previous NMR-guided model (Fig. 2b) (6). Our initial model, generated with a more limited set of experimental data and before the solutions of our crystal structures, identified the F␣-helix as the principal TACC3-binding site. With the new data presented here, we rigorously tested predictions of both the F␣-helix and ␤-sheet-directed binding models in light of the differences among them. Several lines of evidence, particularly the minimal effects of ARNT F␣-helix mutants on TACC3 binding and the largest magnitude NMR broadening effects clustering on the ARNT ␤-sheet, unambiguously establish that the current structure represents the solution structure of the ARNT PAS-B⅐TACC3 complex.
Closing with the broader context for this work, the ARNT⅐CCC interaction plays an important role in mammalian hypoxia response. Because ARNT is the common partner for all three HIF-␣ paralogs, blocking ARNT⅐CCC interactions could affect all HIF-␣-containing transcription factors. This suggests that directly targeting the ARNT⅐CCC complex may represent an interesting approach to regulate HIF activity in cancer therapy, complementing previously described methods that target specific HIF-␣ isoforms (32,35). In such work, our in vitro small molecule inhibitors that target a water-binding cavity inside ARNT PAS-B, induce conformational changes, and disrupt full-length ARNT/TACC3 interactions (17) represent a starting point for the development of such approaches. As such, further refinement of these inhibitors, particularly to optimize cellular and in vivo potency, could be a promising therapeutic route for HIF-dependent cancer treatment and a powerful tool for studying CCC function in hypoxia signaling pathways.