Heat Shock Protein 90 Associates with the Per-Arnt-Sim Domain of Heme-free Soluble Guanylate Cyclase

Background: hsp90β binds to heme-free sGCβ1 subunit to enable heme insertion during maturation, but the mechanism is unclear. Results: Their interaction involves the hsp90β M domain and the Per-Arnt-Sim (PAS) domain of apo-sGCβ1. Conclusion: hsp90β may modulate the apo-sGCβ1 through its PAS domain, suggesting a means to aid heme insertion. Significance: This work expands our knowledge of how hsp90 regulates PAS-containing client proteins like sGC. Heat shock protein 90 (hsp90) drives heme insertion into the β1 subunit of soluble guanylate cyclase (sGC) β1, which enables it to associate with a partner sGCα1 subunit and mature into a nitric oxide (NO)-responsive active form. We utilized fluorescence polarization measurements and hydrogen-deuterium exchange mass spectrometry to define molecular interactions between the specific human isoforms hsp90β and apo-sGCβ1. hsp90β and its isolated M domain, but not its isolated N and C domains, bind with low micromolar affinity to a heme-free, truncated version of sGCβ1 (sGCβ1(1–359)-H105F). Surprisingly, hsp90β and its M domain bound to the Per-Arnt-Sim (PAS) domain of apo-sGC-β1(1–359), which lies adjacent to its heme-binding (H-NOX) domain. The interaction specifically involved solvent-exposed regions in the hsp90β M domain that are largely distinct from sites utilized by other hsp90 clients. The interaction strongly protected two regions of the sGCβ1 PAS domain and caused local structural relaxation in other regions, including a PAS dimerization interface and a segment in the H-NOX domain. Our results suggest a means by which the hsp90β interaction could prevent apo-sGCβ1 from associating with its partner sGCα1 subunit while enabling structural changes to assist heme insertion into the H-NOX domain. This mechanism would parallel that in other clients like the aryl hydrocarbon receptor and HIF1α, which also interact with hsp90 through their PAS domains to control protein partner and small ligand binding interactions.

Heat shock protein 90 (hsp90) 4 is an ATP-dependent molecular chaperone that plays important roles in cell proliferation, signaling, and survival (1,2). hsp90 stabilizes or activates numerous protein clients by binding to their unfolded, partly folded, or native-like forms (3,4). hsp90 functions as a homodimer, with each monomer consisting of a C-terminal dimerization domain (C), a middle domain (M) that is often involved in client binding, and an N-terminal domain (N) that contains the majority of the ATP-binding site and that participates in some client-binding interactions (5,6). The hsp90 dimer functions in client maturation by undergoing large conformational rearrangements that are governed by ATP site occupancy and hydrolysis, co-chaperone proteins, and posttranslational modifications (7)(8)(9). The molecular details of hsp90-client protein interactions, and their associated structural and functional impacts, are under active investigation (6, 10 -14).
Proteins that bind iron protoporphyrin IX (hemeproteins) are ubiquitous in nature and perform important roles in metabolism, energy transduction, communication, and host defense (15,16). hsp90 interacts with a number of hemeproteins, including the nitric-oxide synthases (17)(18)(19), eIF2a kinase (20), HAP1 (21), NADPH oxidases (22), and the intracellular signaling enzyme soluble guanylate cyclase (sGC) (23). Association with hsp90 has been shown to increase sGC lifetime and enzymatic activity in cells and tissues (23,24), but the mechanisms involved are unclear. In its active form, sGC is a heterodimer composed of similar ␣ and ␤ subunits, but only the ␤ subunit contains a binding site for heme (25,26). In recent studies, we found that hsp90 specifically associates with the heme-free form of the sGC␤1 subunit (apo-sGC␤1) in cells (27,28), con-sistent with hsp90's recognized role of stabilizing immature clients. Moreover, we found that hsp90 promotes heme insertion into apo-sGC␤1 in an ATP-dependent process. hsp90 subsequently dissociates from heme-loaded sGC␤1, which concurrently associates with an sGC␣1 partner subunit to form the mature sGC heterodimer (27,28).
Only heme-containing sGC␤1 can form heterodimers that bind and respond to nitric oxide (NO), the native sGC-activating molecule in signaling cascades (29). Thus, elucidation of the molecular details of the hsp90-sGC␤1 interaction is needed to understand hsp90's role in sGC maturation. To address this, we utilized fluorescence polarization measurements and hydrogen-deuterium exchange mass spectrometry (HxDMS) to characterize the interaction between hsp90␤1 and apo-sGC␤1, representative isoforms of the two proteins. We determined that hsp90␤ interacts not with the heme-binding H-NOX domain of heme-free sGC␤1 but rather with the adjacent Per-Arnt-Sim (PAS) domain, also termed the H-NOXA (H-NOX-associated) domain. We found that the M domain of hsp90␤ primarily mediates this interaction; moreover, we identified specific regions on the hsp90␤ M domain and sGC␤1 PAS domain that participate in or respond to the interaction. Our findings (i) provide unifying concepts regarding hsp90 interactions with PAS domain client proteins and (ii) suggest a possible mechanism whereby hsp90␤ anchors apo-sGC␤1 through its PAS domain for heme insertion during client maturation.

Experimental Procedures
hsp90␤ Plasmid Constructs, Protein Expression, and Purification-Dr. Tomasz Religa (Case Western Reserve University) kindly provided pet15MHL plasmids containing the coding sequences of full-length human hsp90␤ or the hsp90␤ N domain (residues 1-227) and a pet28MHL plasmid encoding the hsp90␤ M domain (residues 284 -543). These constructs add a TEV-cleavable hexa-histidine tag at the N or C termini of the respective constructs. We cloned the human hsp90␤ C domain (residues 621-724) into pGEX-4T1 plasmid to generate a construct with an N-terminal thrombin-cleavable GST tag. All constructs were sequenced for verification.
Plasmids coding for full-length hsp90␤, the N domain, or the M domain were transformed into Escherichia coli BL21codonϩ. Overnight cultures grown at 37°C in Luria Broth (LB) were used to inoculate 2 liters of Terrific broth. Expression cultures were grown at 37°C until they reached an A 600 of 0.8 -1.0, induced with 1 mM isopropyl ␤-D-1-thiogalactopyranoside, and shaken for 20 h at 25°C. Cells were harvested by centrifugation, and all subsequent steps were carried out at 4°C. Cell pellets were suspended in 50 ml of phosphate-buffered saline (PBS, 137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 7.4) supplemented with phenylmethanesulfonyl fluoride (PMSF), protease inhibitor mixture (aprotinin, leupeptin, pepstatin, and 4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride), and lysozyme and disrupted by sonication on ice. Supernatants were clarified by centrifugation and applied onto a gravity column of 20 ml of Ni-NTA-agarose resin (Qiagen) pre-equilibrated with PBS. The resin was washed with 100 ml of PBS followed by 100 ml of PBS ϩ 80 mM imidazole. Bound protein was eluted with 75 ml of PBS ϩ 200 mM imidazole and concentrated by centrifugal ultrafiltration. Concentrated protein was dialyzed into PBS ϩ 1 mM DTT, flash-frozen, and stored at Ϫ70°C.
The hsp90␤ C domain was expressed in E. coli BL21. Cells were grown, induced, and processed as described above. Clarified cell lysate was bound to 10 -40 ml of 50% glutathione-Sepharose-4B in PBS. The resin was washed five times in 30 ml of PBS per wash, and bound protein was eluted with 20 ml of fresh 10 mM glutathione, 10 mM Tris, pH 8.0. This solution was concentrated by ultrafiltration and dialyzed twice in 1 liter of PBS ϩ 1 mM DTT. GST-free hsp90␤ C domain was prepared by incubating the sample with 10 units of thrombin/mg of protein for several hours at room temperature, followed by tumbling for 30 min with 10 -40 ml of 50% GST-Sepharose-4B resin, centrifugation (250 rpm for 5 min), and collection of the supernatant. The resin was washed three times with 20 ml of PBS to recover the remaining C domain, and this solution was combined with the supernatant for concentration, flash-freezing, and storage at Ϫ70°C.
The sGC␤1 , sGC␤1(1-359), and sGC␤1(1-359)-H105F proteins were purified by anion exchange chromatography on a 25-ml Q-Sepharose column pre-equilibrated with buffer A. The column was washed with 3 column volumes of buffer A, and the sGC␤1 protein was eluted with a linear gradient of 50 mM to 1 M NaCl in buffer A. Fractions containing sGC␤1 protein were pooled, concentrated to 2 ml, and dialyzed into buffer A. The Q-Sepharose purification was then repeated using buffers adjusted to pH 8.5, and fractions containing the sGC␤1 proteins were pooled, concentrated, flash-frozen, and stored at Ϫ70°C.
The sGC␤1-PAS construct was purified on a gravity column containing 20 ml of Ni-NTA-agarose resin (Qiagen) pre-equilibrated with buffer A. The resin was washed with 100 ml of Hsp90 Interaction with Heme-free Soluble Guanylate Cyclase buffer A and 100 ml of buffer A ϩ 40 mM imidazole. Bound protein was eluted with 50 ml of buffer A ϩ 160 mM imidazole and concentrated to 2 ml by centrifugal ultrafiltration. Concentrated protein was dialyzed into buffer A ϩ 1 mM DTT, and the His 6 tag was removed by incubation with 1000 units of TEV protease (Invitrogen) overnight at 4°C. To remove uncut protein, the TEV digest was redialyzed twice into buffer A (without DTT) and applied to a gravity column containing 5 ml of Ni-NTA-agarose resin pre-equilibrated with buffer A. The flowthrough was retained, and 20 ml of buffer A ϩ 10 mM imidazole was used to elute residual sGC␤1-PAS from the beads. Residual sGC␤1-PAS was combined with the original flow-through, concentrated to 2 ml, dialyzed against buffer A ϩ 1 mM DTT, and flash-frozen and stored at Ϫ70°C. Protein molecular weight estimations were performed using a gel filtration column and protein standards.
Fluorescence Polarization Measurements-For labeling, selected proteins were dialyzed into 0.1 M sodium bicarbonate, pH 9.3. FITC, dissolved in the bicarbonate buffer, was added to the proteins at a ratio of 20 g of FITC to 1 mg of protein. After incubation for 1 h at room temperature in the dark, the reaction mixture was passed through a PD-10 column to capture excess FITC and to exchange the protein sample into PBS buffer. FITC labeling efficiency was calculated based on 495 nm based on a FITC extinction coefficient of 63,000 M Ϫ1 cm Ϫ1 at 495 nm.
FITC-labeled proteins were diluted in PBS in a fluorescence cuvette to a final concentration of 0.5 or 1 M, and unlabeled prospective binding partners were titrated into the cuvette. The excitation wavelength was 495 nm, and the emission was monitored at 520 nm. The fluorescence intensity was recorded multiple times after each addition, with polarization filters set in the parallel and perpendicular positions, and then used to calculate the average polarized parallel intensity (I 1 ) and average polarized perpendicular intensity (I 2 ). The fluorescence polarization at each titration point was calculated by the formula: (I 1 Ϫ I 2 )/(I 1 ϩ I 2 ). In addition, selected titrations were performed in triplicate using 0.2 M FITC-labeled sGC␤1(1-359)-H105F or sGC␤1-PAS in 100-l volumes in a 96-well plate, using a Flexstation 3 multimode microplate reader (Molecular Devices). The resulting binding curves were fit in Origin Lab 8.0 (Origin Lab Corp.) to calculate binding affinities. In some cases, the Flexstation titrations also included 1 mM ATP, ADP, or AMP-PNP. All titration experiments were repeated independently 2 or 3 times.
Peptide Mapping by HPLC-Tandem Mass Spectrometry-hsp90␤ M domain (8 g) or sGC␤1(1-359)-H105F (10 g) in 100 l of 40 mM EPPS, pH 7.6, 150 mM NaCl was mixed with 10 l of 1 mM HCl to lower the pH to 2.4. Samples were digested by incubation with 10 g of porcine pepsin (1 mg/ml), dissolved in 0.05% TFA, for 5 min on ice. The final hsp90␤ M and sGC␤1(1-359)-H105F concentrations were 2 M. The digested sample was injected into a micropeptide trap (Michrom Bioresources) connected to a C18 HPLC column (5 cm ϫ 1 mm, Alltech) coupled to a ThermoElectron LTQ ion-trap mass spectrometer (ThermoElectron). Peptic fragments were eluted using a gradient of acetonitrile (Burdick and Jackson) at a flow rate of 50 l/min for a tandem mass spectrometry experiment to sequence each of the peptic fragments. Peptic fragments were identified using the search algorithm SEQUEST (ThermoElectron) and by manual inspection.
Hydrogen/Deuterium Exchange Reactions of Individual Proteins and Proteins in Complex-Purified hsp90␤ M domain and sGC␤1(1-359)-H105F were utilized singly or were incubated together (20 M each) for 45 min at room temperature in 40 mM EPPS buffer, pH 7.6, 150 mM NaCl to form a complex. Each protein sample was rapidly diluted 10-fold with 40 mM EPPS buffer, pD 7.6, 150 mM NaCl dissolved in D 2 O (Cambridge Isotope Laboratories) at 25°C to initiate deuterium exchange. Reactions were quenched at different time points by addition of 1 mM HCl to lower the pH to 2.4, followed by quick-freezing in a dry ice-ethanol bath. Samples were stored at Ϫ80°C before further analysis.
Isotope Analysis by HPLC-Electrospray Ionization Mass Spectrometry-Frozen hydrogen/deuterium exchange samples were quickly thawed and digested with 10 g of pepsin (1 mg/ml, dissolved in 0.05% TFA, and H 2 O) for 5 min on ice immediately followed by injection into a micropeptide trap connected to a C18 HPLC column (5 cm ϫ1 mm, Alltech) coupled to a ThermoElectron LTQ ion-trap mass spectrometer. Peptic peptides were eluted in 15 min using a gradient of 5-45% acetonitrile at a flow rate of 50 l/min. The micropeptide trap and C18 HPLC column were immersed in a water/ethanol bath at 0.2°C during the entire run to minimize back exchange. Peaks were identified manually, and centroid masses were calculated using the software MagTran (Amgen, Thousand Oaks, CA). A fully deuterated reference sample for each protein was used to correct for deuterium loss during sample handling. Deuterium incorporation for each peptide fragment was calculated and is presented as normalized percentage of deuterium uptake (% deuteration or % deuterium uptake) using Equation 1, where m t is the centroid mass of the deuterated peptic fragment at incubation time t, and m 0% and m 100% are the centroid masses of the unlabeled and fully deuterated peptic fragments, respectively. All deuterium uptake experiments were performed in triplicate for individual proteins and their complexes.
Structural Modeling-Homology models of sGC␤1 domains and hsp90␤ (full-length or domains) were generated using I-Tasser (31), either without explicitly specified templates or using homologous structures/models from the protein model portal (32) as inputs. The best I-Tasser outputs were manually adjusted in Coot (33) and refined using ModRefiner (34). We generated all molecular graphics using PyMOL 1.7 (Schrödinger, LLC).
The human sGC␤1 H-NOX model was generated based on the equivalent domain of a Nostoc sp. PC7120 H-NOX protein (35,36). The sGC␤1 PAS domain was modeled as a monomer and dimer using PAS domains from Manduca sexta sGC␣ and signal transduction histidine kinase of Nostoc punctiforme sp. PCC73102 as templates (37,38). We modeled a short segment of the coiled-coil domain of human sGC␤1 based on the nearly identical equivalent in rat sGC␤1 (39). The crystal structures of Hsp90 Interaction with Heme-free Soluble Guanylate Cyclase AUGUST 28, 2015 • VOLUME 290 • NUMBER 35 the human hsp90␤ N and M domains were solved previously ( (40) and Protein Data Bank code 3PRY). We generated a model of the hsp90␤ C domain based on the crystal structure of the M and C domains of human hsp90␣ (41). The I-tasser Z-scores of all domain models were greater than Ϫ0.25, indicating a high likelihood that the domains had correct folds and correct overall tertiary structure. We further estimated the accuracy of the individual domain models using ModEval (42). For each model, the predicted root mean square deviation from the true structure was less than 3 Å, whereas the estimated Native Overlap 3.5 score (the fraction of model C␣ atoms within 3.5 Å from their true positions) was above 85%. These scores suggest that the models have correct overall folds and realistically represent the individual domain structures.
A multidomain model of sGC␤1(1-359) was generated using a previously described model of truncated M. sexta sGC␣ that encompasses the H-NOX, PAS, and coiled-coil regions (43). The M. sexta model was based on restraints derived from small angle x-ray scattering and chemical cross-linking experiments, and our current model satisfies the equivalent distance restraints. We superimposed individual human sGC␤1 domain models onto their counterparts in the M. sexta model, added missing or linker residues using Coot (33), and refined the structure using FG-MD (44) and ModRefiner (34). We generated models of the full-length hsp90␤ dimer in a similar manner. Experimental structures or models of the hsp90␤ N, M, or C domains were superimposed onto their counterparts in a closed structure of the Saccharomyces cerevisiae hsp90 dimer (45) or onto an open dimer of the E. coli hsp90 orthologue HtpG (46).

Results
Properties of the sGC␤1 and hsp90␤ Proteins- Fig. 1 illustrates the protein constructs we used in the study. The sGC␤1(1-359) construct is similar to others that have been extensively studied (26) and is composed of the N-terminal regulatory domain (H-NOX or heme domain), the PAS domain, and an N-terminal portion of the coiled-coil domain. This construct lacks most of the coiled-coil domain and the entire catalytic domain. Unlike the full-length sGC␤1 protein, however, it can be expressed well in E. coli (26). The sGC␤1(1-197) and sGC␤1(202-342) (sGC␤1-PAS) constructs, respectively, correspond to the individual H-NOX and PAS domains, including inter-domain linker sequences. All proteins were expressed in E. coli. The sGC␤1(1-359) purified as a minor fraction of heme-enriched and a major fraction of heme-free (apo) protein; the two forms were separated using an anion exchange column. The sGC␤1(1-359)-H105F variant was at least 95% heme-free and was predominantly dimeric with some higher oligomers as purified (data not shown), confirming that the Phe substitution resulted in the intended heme-binding defect (25,26). The sGC␤1-PAS protein purified as a mixture of apparent dimeric and trimeric forms. The sGC␤1  protein has also been extensively studied (25). In our hands, this construct was predominantly heme-free after purification even without the H105F mutation (data not shown).
Analysis of Binding Interactions between hsp90␤ and Apo-sGC␤1-Previous studies that used affinity resins or antibody capture established that hsp90 associates with apo-sGC␤1 in cells, primarily through the hsp90 M domain (23,24). To understand the interaction in greater detail, we utilized fluorescence polarization measurements on recombinant purified forms sGC␤1 and hsp90␤. These measurements detect binding between a fluorophore-labeled protein and an unlabeled partner protein, and they have been used previously to characterize binding between hsp90 and clients or hsp90-cochaperones (10,50,51).
We measured the residual fluorescence polarization of FITC-labeled sGC␤1(1-359)-H105F titrated with hsp90␤ or with individual hsp90␤ N, M, or C domains. We observed a concentration-dependent, saturable increase in the residual polarized fluorescence emission in titrations with full-length hsp90␤ or its M domain. In contrast, we observed only weaker or no binding between sGC␤1(1-359)-H105F and the hsp90␤ N domain, and no apparent binding to the hsp90␤ C domain ( Fig. 2A). We concluded that sGC␤1(1-359)-H105F interacted directly with hsp90␤, primarily through the hsp90␤ M domain. We obtained similar results using a purified heme-depleted form of wild-type sGC␤(1-359) instead of the H105F mutant (data not shown). In a complementary experiment, we titrated unlabeled sGC␤1(1-359)-H105F against FITC-labeled hsp90␤ M domain (Fig. 2B). The results confirmed that the binding interaction did not depend on which protein was labeled with FITC.
We utilized FITC-labeled apo-sGC␤1  or sGC␤1 PAS proteins to further define which sGC␤1 domains interact with hsp90␤. Titration of FITC-labeled PAS domain with either hsp90␤ or the M domain led to saturable increases in the residual polarized emission (Fig. 3A). In contrast, we observed no increase when labeled apo-sGC␤1  was titrated with full-length hsp90␤ or with the hsp90␤ M domain (Fig. 3B). We carried out similar titrations to determine dissociation constants for the relevant interactions. We measured the binding of either sGC␤1(1-359)-H105F or sGC␤1 PAS to hsp90␤ or its M domain (Fig. 4, A-D). The calculated dissociation constants ranged between 1 and 1. HxDMS Studies-We utilized HxDMS to identify regions within the hsp90␤ M domain and in apo-sGC␤1(1-359)-H105F that mediate or respond to their interaction. This approach was used previously to identify regions in hsp90 that bind cochaperones or small molecule inhibitors (10, 11, 52, 53).
Impact of Complex Formation on hsp90␤ M Domain-Binding to sGC␤1(1-359)-H105F altered the exchange behaviors of several reporter peptides within the hsp90␤ M domain (Figs. 5,  A and B, and 7). Two peptides near the N terminus showed the largest percentage change in rapidly exchanging protons due to complex formation (Fig. 5B, 10 s histogram). Several reporter peptides located in the middle and C-terminal regions of the M domain displayed differences in the extent of exchange that accrued over time (Fig. 5B, 3000 s histogram), which reflected reduced or increasing exchange progression as a result of complex formation (Fig. 7). Four M domain peptides exhibited 7-18% decreases in exchange, while two peptides exhibited 6 -8% increases (Fig. 5B, 3000 s histogram). These results identify specific sequences within the hsp90␤ M domain that may interact directly with sGC␤1(1-359)-H105F and/or respond structurally upon complexation. Fig. 5C maps these sequences onto a crystal structure of the hsp90 M domain (55,56). The sequences that exhibited reduced exchange upon complexation at 3000 s form two discrete noncontiguous surfaces, one each on the N-and C-terminal halves of the hsp90␤ M domain. Surface-exposed helices on each of these surfaces (␣7 and ␣11, respectively) manifest the greatest decreases in exchange. Interestingly, the two surfaces are flanked and separated by elements that are less affected or even undergo an increase in exchange, indicating a local relaxation of structure (for example, at the edge of ␤12 residues 362-367 and the first half of helix ␣8 residues 420 -436).    AUGUST 28, 2015 • VOLUME 290 • NUMBER 35

Impact of Complex Formation on sGC␤1(1-359)-H105F-
Complex formation with the hsp90␤ M domain increased the extent of rapid proton exchange for one mapping peptide located at the N terminus of the H-NOX domain, but it did not affect exchange in any other segment of the H-NOX (Figs. 6, A  and B, and 8). In conjunction with our fluorescence polarization data, these HxDMS data suggest that the H-NOX domain does not interact directly with the hsp90␤ M domain, but it still dynamically responds to complex formation.
In contrast, complex formation with the hsp90␤ M domain resulted in more extensive differences in exchange within the PAS domain (Figs. 6, A and B, and 8). The percentage of rapidly exchanging protons increased in two PAS peptides from 5 to 45% (residues 251-65) and from 40 to 57% (residues 237-50) of total exchangeable protons (Fig. 6B, 10 s histogram, and 8). The exchange accrued over time also increased in these two pep-tides and in a third PAS peptide (residues 304 -20). Together, these peptides correspond to portions of the central ␤-sheet (␤C, ␤D, ␤F, and ␤G) of the PAS domain and two helical segments (␣B and ␣C), representing a substantial increase in dynamics over approximately one-third of the domain (Fig. 6B,  3000 s histogram). At the same time, complex formation strongly reduced rapid exchange within two peptides (residues 266 -276 and 332-351) that otherwise rapidly exchanged 35 and 58% of their total exchangeable protons (Figs. 6B, 10 s histogram, and 8). These sequences respectively correspond to helix ␣E, which flanks one side of the PAS domain, as well as the C-terminal linker sequence (residues 332-351) that connects the PAS domain to the coiled-coil domain of sGC␤1. Complex formation also reduced slower exchange kinetics in these two segments by 60 and 45% (Fig. 6B, 3000 s histogram). These findings are consistent with our fluorescence polarization

Hsp90 Interaction with Heme-free Soluble Guanylate Cyclase
experiments, which showed that the PAS domain interacts directly with the hsp90␤ M domain. The above-mentioned sequences correspond to regions within the PAS domain that respond to the complex formation and possibly interact directly with hsp90␤. Fig. 6C illustrates our exchange results in the context of a structural model for sGC␤1(1-359)-H105F that is based on existing structural and biophysical studies (37,38,43,57).
Modeling Complexes of sGC␤1  with hsp90␤-The segments of the hsp90␤ M domain that were protected by sGC␤1(1-359) define essentially two surface-exposed interfaces. Full-length hsp90␤ dimers undergo conformational Hsp90 Interaction with Heme-free Soluble Guanylate Cyclase AUGUST 28, 2015 • VOLUME 290 • NUMBER 35 changes between open and closed forms in nucleotide-and cochaperone-driven reaction cycles (5,6). We built homology models of full-length hsp90␤ dimers based on available crystal structures and determined that the protected surfaces are solvent-exposed in both closed and open conformations (Fig. 9, A  and B). These data support the notion that these surfaces correspond to direct binding sites for the sGC␤1 PAS domain. Correspondingly, two distinct segments on sGC␤1 also exhibited decreased exchange upon complex formation. One of these, the linker segment that connects the PAS and coiled-coil domains of sGC␤1 is structurally flexible and could adopt different orientations relative to the PAS domain. We therefore carried out computational docking of sGC␤1(1-359) onto open and closed hsp90␤ dimers using HADDOCK (47)(48)(49), specifying the two most strongly protected segments of each binding partner as potential interaction sites. As illustrated in Fig. 9, C-F, the lowest-energy ensembles of models readily accommodate such interactions. All low energy models involved packing of the PAS domain onto hsp90␤ helix ␣7 and nearby surfaces. The flexible linker segment of sGC␤1 extends toward and packed against surfaces formed by helices ␣10 and ␣11 of the M domain. Although no explicit interprotein restraints were specified during the docking, no low-energy models from the docking exhibited the opposite arrangement.

Discussion
hsp90 binds to apo-sGC␤1 and drives its heme insertion in an ATP-dependent process (27). We utilized purified hsp90␤, apo-sGC␤1(1-359), a heme-deficient sGC␤1(1-359) mutant, and individual domains to investigate how hsp90␤ and sGC␤1 interact with each other. We determined that the hsp90␤ M domain and the PAS domain of sGC␤1 are the primary mediators of interaction between the two proteins. Our data place sGC␤1 in context relative to other selected hsp90 clients that contain PAS domains, including the aryl hydrocarbon receptor (AhR) and the hypoxia-inducing factor-1␣ (HIF1␣) (58 -61). There are notable parallels and differences in how the hsp90-PAS interaction regulates the biological functions of apo-sGC␤1, AhR, and HIF1␣. For example, the hsp90-bound form of all three proteins is inactive, because hsp90 replaces a partner subunit (the sGC␣1 subunit for apo-sGC␤1) or partner protein (ARNT for AhR; HIF1␤, P300, or RACK1 for HIF1␣) that is otherwise essential for biological function (27, 58 -62). The hsp90-PAS interaction influences the binding of a small molecule ligand, the heme cofactor, to apo-sGC␤1 (28); similarly, hsp90 regulates binding of aromatic hydrocarbon ligands to AhR (60,61,63). Both ligand-binding events also require hsp90 ATPase activity. However, in the PAS domain of AhR, the binding sites for hsp90 and aromatic hydrocarbons overlap, and it has been suggested that hsp90 binding and ligand binding directly regulate each other (59,63,64). A similar mechanism may hold for ligand-binding domains of other hsp90 clients, including the glucocorticoid hormone receptor (10,11) and multiple hsp90-chaperoned protein kinases (14,65). In these cases, the hsp90-binding interface overlaps with the steroid ligand-binding site and with the ATPbinding site of the respective clients.
The PAS domain of sGC␤1 has a more densely packed hydrophobic core than typical PAS domains (such as that of AhR). The sGC␤1 PAS domain does not contain a discernible small molecule-or ligand-binding pocket (modeling data not shown). Rather, the heme-binding site of sGC␤1 is located in the adjacent H-NOX domain and therefore is separated from the PAS. For this client, therefore, hsp90 binds to a PAS domain and thereby may modulate a distinct ligand-binding domain. These considerations suggest that hsp90 operates by a blend of similar and distinct mechanisms to modulate sGC␤1 maturation and function, as compared with other PAS-containing clients or even other small molecule ligandbinding clients.
Our data show that sGC␤1(1-359)-H105F, and presumably full-length apo-sGC␤1, primarily interact with the hsp90␤ M domain. For the two M domain regions that displayed increased Hsp90 Interaction with Heme-free Soluble Guanylate Cyclase AUGUST 28, 2015 • VOLUME 290 • NUMBER 35 protection at 3000 s primarily due to their having a slower exchange after initial complex formation, the protection could arise from distant allosteric effects or could also result from a slow binding event that directly involves these protein regions. In either case, these regions of the M domain only partly coincide with binding sites proposed for other client proteins. For example, the binding sites for the DNA-binding domain of p53 (50) and sGC␤1-PAS overlap relatively little (Fig. 10A), and p53 binding involves an M domain groove formed by one side of ␤9 and elements from helices ␣6, ␣7, and ␣8 that are not protected upon sGC␤1-PAS binding. Binding sites for Tau (an intrinsically disordered protein) and sGC␤1 partly overlap within the M domain (66). However, Tau also interacts with the hsp90␤ N domain, and, unlike Tau, sGC␤1 engages the C-terminal portion of the M domain (Fig. 10B). The putative sGC␤1-PAS binding interface of hsp90␤ also differs from those of Akt/PKB (65), kinases such as CDK4 that are co-chaperoned by cdc37 (55,56), and the glucocorticoid receptor ligand binding domain (Fig. 10, C and D) (10,11). In particular, a flexible basic loop that connects strands ␤10 and ␤11 of the hsp90 M domain has been suggested to interact directly with kinases and ligand-binding domains (small molecule-binding PAS domains and steroid hormone receptor ligand-binding domains), but it is unaffected by the sGC␤1 PAS domain. Finally, the sGC␤1-PASbinding site also differs from the M domain and C-terminal domain regions identified as binding sites for model client proteins (fragments of staphylococcal nuclease or ribosomal protein L2) of the E. coli hsp90 orthologue HtpG (Fig. 10E) (13,53).
The respective PAS domains of sGC␣1 and sGC␤1 interact with each other in a stable heterodimeric sGC (37,38). The sGC␤1 PAS domain also homodimerizes, and several representative crystal structures have been solved, including a homodimer of the PAS domain of M. sexta sGC␣ (38). However, a homology model of a human sGC␤1 PAS homodimer shows that the two regions we found to be heavily protected against proton exchange by the hsp90␤ M domain (sGC␤1 residues 266 -276 and 332-351) lie outside the putative PAS dimer interface (Fig. 11, A and B). In addition, among the PAS domain peptides that became deprotected (showed increased exchange) due to complex formation with the hsp90␤ M domain, one peptide (residues 204 -244) lies within the PAS homodimer interface in a region that contains residues that are essential for sGC heterodimerization (Fig. 11) (37). Thus, the binding of the hsp90␤ M domain at specific C-terminal regions in the sGC␤1 PAS domain appears to destabilize upstream regions in the PAS domain, including one needed in sGC heterodimer stabilization. This might represent a mechanism by which hsp90␤ prevents apo-sGC␤1 binding to its sGC␣1 partner, an important outcome of chaperoning that would prevent formation of futile sGC heterodimers that do not respond to NO.
Regarding binding stoichiometry, our hsp90␤ M domain is monomeric, whereas our full-length hsp90␤ is dimeric (by gel filtration analysis). Because we observed no significant difference in the apparent binding affinity between hsp90␤ and the The dashed oval indicates a putative binding groove for several loops of the p53 DNA-binding domain (50) formed by one side of ␤9 and elements from helices ␣6, ␣7, and ␣8. B, distributed interaction surface of Tau protein on N and M domains of hsp90 identified by mutagenesis, NMR, and small angle x-ray scattering data (66). C, putative interaction surface of glucocorticoid receptor ligand-binding domain identified by NMR, mutagenesis, and cryo-EM data (10,11). D, putative kinase-interacting epitopes of hsp90. Dashed oval indicates location of an Akt/PKB-interacting loop (65) connecting strands ␤10 and ␤11. Residues Trp-312 and Leu-343, shown as spheres, are implicated in binding of kinases such as CDK4, which are cochaperoned by cdc37 (55). E, residues on C-terminal half of M domain and C domain corresponding to client-binding residues on E. coli HptG (13,53). Residues are shown as spheres.
M domain, the hsp90␤ dimeric structure and the included Nand C-terminal domains do not appear to be too important. This result is consistent with our other data indicating that little or no sGC␤1 binding takes place with the individual hsp90␤ Nor C-terminal domains, and the binding interaction appears to involve the solvent-facing regions of the M domain. For our sGC␤1(1-359)-H105F protein, we utilized the longest construct that can be successfully expressed in bacteria, which in our hands was primarily dimeric (with some higher oligomers also present). It is presently unclear whether complex formation with hsp90␤ or with the M domain weakened or dissociated the sGC␤1(1-359)-H105F homodimeric structure, but as noted such weakening may not be necessary for complex formation to occur because the putative interacting surfaces are solvent-exposed. In any case, our current results can now guide investigation into the hsp90␤ interactions that take place within cells during the course of sGC␤1 maturation.
A prior HxDMS study provided strong evidence that the sGC␤1 PAS domain may also engage in intramolecular interactions with the (heme-loaded) H-NOX domain. These results were obtained using an sGC␤1 construct spanning residues 1-385 (67), which is similar to our sGC␤1(1-359) construct. The PAS domain interacts with an H-NOX interface defined by helices ␣E, ␣F, and ␣G, which flank the heme-binding pocket and, in the case of ␣F, respond conformationally when NO binds to the heme. The prior study did not identify which elements of the PAS domain are involved in the interaction, and we did not obtain HxDMS data from the H-NOX peptides located in the putative interface. However, it is conceivable that the segments of the sGC␤1 PAS domain that become deprotected upon hsp90␤ M binding (and are outside the PAS dimerization interface) may be PAS surfaces that interact with the H-NOX domain. Such surfaces could thus become exposed upon binding to hsp90␤. Our modeling data (Figs. 7 and 11) also suggest that interactions between the hsp90␤ M domain and the sGC␤1 PAS-coiled-coil linker segment could modulate the tertiary arrangement of the sGC␤1 domains (Figs. 7 and 11). Such interaction could displace intramolecular contacts between the linker segment and the H-NOX or PAS domains, indirectly affecting their interactions with each other as well as their stability and dynamics. By abstracting intramolecular interaction elements of the PAS domain and PAS-coiled-coil linker, hsp90 may indirectly drive a heme-loading competent state of the H-NOX domain.
Intriguingly, the two structural elements in sGC␤1 PAS that were indicated by our study to interact with hsp90␤ M domain may also help to orchestrate sGC activation by NO and to enable sGC heterodimer formation. A recent study (54) showed that NO binding to heme-containing sGC␤1 increased the degree of exchange, and thus deprotected the same two regions in the PAS domain (residues 266 -276 and 332-351) that become strongly protected upon binding of the hsp90␤ M domain. One of these regions (residues 332-351) was shown by deletion analysis to be essential for the sGC␣/␤ heterodimerization (68). Likewise, NO binding to the sGC␤1 heme was reported to stabilize the N-terminal helices of the H-NOX domain (54), including the same H-NOX segment that we find becomes deprotected by complex formation with hsp90␤ (helix ␣A). Because this segment lies adjacent to the heme pocket in H-NOX, it conceivably might influence the kinetics or energetics of heme binding. That the same few structural elements in sGC␤1 might provide a basis for it to alternate between an inactive heme-free form that is associated with hsp90␤ and an active heme-containing form that is associated with sGC␣1 (27,28) is an interesting concept emerging from our data, and it deserves further study.
A current model for hsp90 function in heme insertion is based on its role in steroid or aromatic hydrocarbon binding (18). In these cases, hsp90 may directly regulate the conformation of an "open cleft" in a client protein domain to help it accept a small ligand. Thus, one might have predicted that hsp90␤ would predominantly interact with the H-NOX domain of apo-sGC␤1 to stabilize and shape it for heme insertion, but our results suggest that the regulation is not so straightforward. Instead, hsp90␤ interacts through its M domain with the apo-sGC␤1 PAS domain, and this appears to cause some specific H-NOX structural relaxation and at the same time helps to block premature association with sGC␣1. Further studies may now investigate how its anchoring of the PAS domain and accompanying structural perturbations enable hsp90␤ to direct heme insertion into the H-NOX domain of apo-sGC␤1. Author Contributions-A. S. and Y. D. designed and performed the experiments, analyzed data, and helped with manuscript preparation. M. M. H. designed and made the sGC and hsp90 variants and helped with manuscript preparation. E. D. provided parent sGC constructs and purification protocols and helped with manuscript preparation. F. S. generated the parent sGC constructs. A. G. analyzed the data and prepared Figs. 7 and 8. W. R. M. provided structural models and helped analyze the data. S. L. H. provided HxDMS instrumentation and advice. S. M. helped design and interpret experiments, helped with manuscript preparation, and performed computer modeling and protein docking simulations. D. J. S. directed the research, analyzed the data, and helped in manuscript preparation. All authors performed a critical reading of the paper prior to submission and approved the final version of the paper.