Two closed ATP- and ADP-dependent conformations in yeast Hsp90 chaperone detected by Mn(II) EPR spectroscopic techniques

Significance

Hsp90 is a homo-dimeric chaperone regulating clients involved in tumorigenesis and signal transduction. Its mechanism of action depends on ATP hydrolysis, which is coupled to a conformational cycle, including “closed” and “open” conformations involving the dimerization of the N-terminal domains, which comprise the ATP binding site. The ATP-bound closed conformation is important for chaperoning activity, while an ADP-bound closed conformation was suggested crucial for client release. We resolved 2 different closed ATP- and ADP-bound conformations for yeast Hsp90 in frozen solutions by measuring Mn(II)–Mn(II) distance distributions, with Mn(II) replacing the essential Mg(II) cofactor in the ATPase site. These results add to our understanding of the mechanism of Hsp90 and set the stage for studies in presence of cochaperones and substrates.

Abstract

Hsp90 plays a central role in cell homeostasis by assisting folding and maturation of a large variety of clients. It is a homo-dimer, which functions via hydrolysis of ATP-coupled to conformational changes. Hsp90’s conformational cycle in the absence of cochaperones is currently postulated as apo-Hsp90 being an ensemble of “open”/“closed” conformations. Upon ATP binding, Hsp90 adopts an active ATP-bound closed conformation where the N-terminal domains, which comprise the ATP binding site, are in close contact. However, there is no consensus regarding the conformation of the ADP-bound Hsp90, which is considered important for client release. In this work, we tracked the conformational states of yeast Hsp90 at various stages of ATP hydrolysis in frozen solutions employing electron paramagnetic resonance (EPR) techniques, particularly double electron–electron resonance (DEER) distance measurements. Using rigid Gd(III) spin labels, we found the C domains to be dimerized with same distance distribution at all hydrolysis states. Then, we substituted the ATPase Mg(II) cofactor with paramagnetic Mn(II) and followed the hydrolysis state using hyperfine spectroscopy and measured the inter–N-domain distance distributions via Mn(II)–Mn(II) DEER. The point character of the Mn(II) spin label allowed us resolve 2 different closed states: The ATP-bound (prehydrolysis) characterized by a distance distribution having a maximum of 4.3 nm, which broadened and shortened, shifting the mean to 3.8 nm at the ADP-bound state (posthydrolysis). This provides experimental evidence to a second closed conformational state of Hsp90 in solution, referred to as “compact.” Finally, the so-called high-energy state, trapped by addition of vanadate, was found structurally similar to the posthydrolysis state.

Hsp90 is an important, evolutionary conserved, molecular chaperone responsible for maintaining homeostasis of the cells upon heat and other stress conditions. Hsp90 is found in almost all cell compartments and is one of the most abundant proteins in the cytosol, where it constitutes 1 to 2% of the proteome (1–5). Even in nonstressed cells, Hsp90 is abundant and participates in many cellular processes, such as activation, folding, and maturation of proteins referred to as clients, which are involved in processes such as signal transduction, tumorigenesis, protein trafficking, and innate immunity (3, 5, 6). Inhibition of Hsp90’s activity by drugs has been found to abolish these processes (7–9) and, therefore, it is considered as a target for the therapy of cancer and of neurodegenerative diseases (3, 6, 10–12). Hsp90 is a homo-dimer (180 kDa) and works off-equilibrium utilizing energy from ATP hydrolysis, where Mg(II) ion is an essential cofactor. Each protomer binds one ATP molecule (13–15) and the ATP hydrolysis rate is slow, ∼0.1 to 1.0 ATP molecule per minute (15–19).

The different Hsp90 homologs (from bacteria, yeast, and mammals) have similar architecture; each protomer of Hsp90 consists of 3 consecutive domains: The highly conserved amino-terminal domain (N-domain), which comprises the ATP binding site (20), connected via a flexible linker to the middle domain (M-domain), followed by the carboxyl-terminal domain (C-domain) (3, 10). The C-domains are constitutively dimerized in all homologs (Fig. 1A) (21). The ATPase activity of Hsp90 is known to be weakly coupled to large global conformational changes, which alter the hydrophobicity of the exposed amino acids and subsequently the affinity to clients (19); the degree of coupling is different for the different Hsp90 homologs. For bacterial Hsp90 the coupling is significant, while for eukaryotes the coupling is strongly dependent on cochaperones (e.g., p23, Cdc23, and Aha1) (10, 17, 22, 23), which assist its chaperoning function, guiding the recognition of different clients (24–26). Many studies have aimed at resolving the conformational cycle of Hsp90, which is coupled to its function, as is briefly summarized next.

Fig. 1.

Structure and conformational cycle of yHsp90 with the spin labels used in our study and their locations indicated. (A) X-ray structure of yHsp90/AMP-PNP/p23 [PDB ID code 2CG9 (32)], corresponding to the closed conformation, showing only the yHsp90 part with the N-, M-, C-domains indicated. The residues before mutation to cysteines and spin labeling are shown in sphere representation as follows: D61C in blue, A152C in pink, Q385C in red, E517C in yellow, D560C in cyan, and K637C in orange. In the N-domains the nucleotides are shown as sticks, while the metal cofactor shown as a light orange sphere was added to the structure as a pseudoatom coordinated by the N₇, β-, and γ-phosphates of AMP–PNP. (B) Postulated conformational cycle constructed from literature data (in the spirit of Taipale et al., figure 1 in ref. 4 and Schulze et al., figure 6 in ref. 17). The arrows in the apo state denote a range of conformations the N-domains can adopt, while in the nucleotide-bound states Hsp90 is in an equilibrium between an open and a closed conformation. (C and D) Chemical structure of the spin labels used in the study after formation of a C-S (thio-sulfide) bond with a cysteine residue of the protein (in red).

The current understanding of the full-length Hsp90 conformational changes coupled to its ATPase activity is based mainly on static (X-ray crystallography, cryogenic electron microscopy [cryo-EM]) and solution state (single molecule [sm] and bulk Förster resonance energy transfer [FRET], and small-angle X-ray scattering [SAXS]) data. In general, the conformational cycle of Hsp90 is described as follows: Before binding ATP the Hsp90 dimer is in the inactive “open” conformation (also referred as apo or resting state) with the 2 N-domains being separated (14, 18, 19, 23, 27, 28). The open structure of the bacterial Hsp90 homolog (HtpG) in the apo state, captured by X-ray crystallography, shows an opening of the N-domains of 7 nm between the inner residues and up to 13 nm between the outer residues (18). However, sm (29) and bulk (17) FRET studies, as well as EM data (27), showed that Hsp90 in the apo state is an ensemble of open and “closed” conformations featuring a variable N-domain opening, with the probability of each conformation being stochastic. The variability in the N-domain opening has been proposed to serve accommodating clients of variable sizes (4, 30, 31). Upon ATP binding the N- and M-domains undergo conformational changes and dimerize, forming an active closed conformation, as confirmed by the crystal structure of yeast Hsp90 (yHsp90) with bound nonhydrolysable ATP analog, AMP–PNP, and cochaperone p23/Sba1 (32) (Fig. 1A), as well as from the closed structure of human Hsp90 with cochaperone Cdc3, kinase Cdk4, molybdate ions, and nucleotide ATP or, more likely, ADP·molybdate (33). Agard and coworkers (18, 27) observed a closed structure of Hsp90 in the ATP-bound state for all Hsp90 homologs using negative-stained EM and X-ray crystallography. Even though the closed structure was captured for the ATP-bound state, it was reported that binding of ATP does not shift the open ⇄ closed equilibrium toward the closed state, but rather the states exist in dynamic equilibrium as found by EM (27), SAXS (28), bulk (17, 23), and sm FRET (14, 19) data for all homologs. Interestingly, EM showed that the open/closed ratio varies among homologs (27).

The ATP-bound state, referred as the prehydrolysis state, is biologically important because it is the active state upon which Hsp90 performs its chaperoning roles. After ATP is hydrolyzed, Hsp90 is found in the posthydrolysis state (before ADP release). Structural studies on the posthydrolysis state using different techniques do not give a coherent picture. Commonly, in proposed ATPase cycles this state is pictured as “even more closed” with respect to the ATP-bound state (17, 23, 34). However, to date only a combined EM/X-ray study by the Agard group has captured a “compact” (more closed) ADP-bound state for HtpG (18). A follow-up EM study of this work showed that all homologs can adopt this compact conformation, however, only after cross-linking of the 2 protomers (27). In contrast, bulk FRET data (17) report the solution ADP-bound state to be structurally similar to the apo open state. Other kinetic FRET data suggested the presence of a second closed ADP-bound related state that, however, has not been experimentally observed (19) or could not be probed by the specific set-up used (23). An sm FRET study reports that this state is comprised of an equilibrium between open and closed conformations (14). These diametrically opposite results for the ADP-bound conformation of Hsp90 are summed up in the conformational cycle shown in Fig. 1B, where we have adapted Agard’s notation and used the term “compact” (18, 27). Although little is known about the compact ADP-bound state, it is considered biologically important, representing an additional functional state of the ATPase cycle, which serves as a release step of the client from the catalytic cycle (5, 35, 36) or “presumably modulates client protein binding” (28). Once the client has been released from the cycle, the bound ADP is released as well and Hsp90 returns to the resting state, where it is ready for the next cycle. Evidently, identification of the structural changes of Hsp90 occurring upon the transition from the pre- (ATP-bound) to the post- (ADP-bound) hydrolysis, as well as in the intermediate high-energy state [HES, after hydrolysis but before the release of the γ-phosphate of ATP (35)] is crucial for a complete understanding of the Hsp90 conformational cycle and mechanism.

Here, we apply high-field electron paramagnetic resonance (EPR) techniques, not yet applied to Hsp90, to resolve yHsp90 conformations in frozen solutions in the apo state and in the presence of 1) ATP and AMP–PNP, 2) ATP+vanadate (herein vanadate is referred to as Vi) (35), and 3) ADP, or ATP after hydrolysis. These 3 states represent, respectively, the prehydrolysis, high-energy, and posthydrolysis states.

Our study includes 2 different EPR approaches. The first part follows the common method of probing protein conformations via double electron–electron resonance (DEER or PELDOR) distance measurements (37, 38) using Hsp90 protomers labeled with nitroxide and Gd(III) spin labels. All measurements are carried out on frozen solutions; therefore, the results include a superposition of all conformations present at the freezing temperature. We measured label-to-label reflecting protomer-to-protomer distance distributions at different hydrolysis states of the ATPase cycle. This approach generated rather broad distance distributions for spin labels placed in the N- and M-domains, with the results being inconclusive. For this reason, we turned to the second approach, where we avoided the use of standard spin labels and exploited the essential intrinsic Mg(II) ion located at the ATP binding “pocket” in the N-domain of each protomer (32), which we substituted with paramagnetic Mn(II), and it thus served as “single atom label.” Here, we 1) probed short-range structural rearrangements at the ATPase site at the different hydrolysis states by employing techniques that detect the coordination environment of Mn(II) via ¹⁴N and ³¹P hyperfine couplings, and 2) determined the distance between the 2 N-domains via Mn(II)–Mn(II) DEER.

For both parts we used the full-length protein without introducing any artificial linkers. This combination of techniques allowed us to concomitantly observe the state of the ATPase cycle and the global structural changes.

Results

DEER Measurements on Spin-Labeled Hsp90.

Six single mutants of yHsp90 were prepared (SI Appendix, Figs. S1–S3), chosen based on earlier sm FRET studies employing fluorescent labels (29, 39), designed to explore the relative positions of the 2 promoters for each of the Hsp90 domains: D61C and A152C in the N-domain, Q385C and E517C in the M-domain, and D560C and K637C in the C-domain (Fig. 1A). All mutants were labeled with 3-Maleimido-PROXYL (Fig. 1C) (herein abbreviated as “proxyl”). We have used proxyl and not the common MTSL [S-(1-oxyl-2,2,5,5-tetramethyl-2,5-dihydro-1H-pyrrol-3-yl)methyl methanesulfonothioate] label because the DEER modulation depth (λ) and signal-to-noise ratios (SNR) of the DEER measurements for the latter were significantly lower with respect to proxyl. Additionally, the C-domain mutants were labeled with BrPy-DO3A-Gd(III) (40) (Fig. 1D), herein abbreviated as Gd(III). The labeling efficiency of the other mutants with the Gd(III) label was low and degradation products were observed; therefore, we did not pursue measurements using Gd(III) for the M- and N-domain mutants. We verified that all of the spin-labeled mutants retained ATPase activity (SI Appendix, Figs. S4 and S5). We observed some variations in activity, depending on the labeling position and the label type, which may indicate a relation between the conformational landscape and ATPase activity. Models of the spin labels grafted on the crystal structure of the closed conformation for each mutant along with the predicted distance distribution are shown in SI Appendix, Fig. S6.

The W-band DEER measurements (see experimental set-up in SI Appendix, Fig. S7C) were carried out on protein concentrations of ∼100 μM (protomer) in 20 mM Tris·HCl, 20 mM KCl, pD 7.4, D₂O/glycerol-d₈ (8:2 vol/vol), and 3 types of samples, representing different hydrolysis states, were prepared: 1) the as-prepared purified labeled mutants, representing the apo state; 2) in the presence of excess Mg(II)·AMP–PNP (41) to mimic the prehydrolysis state; and 3) excess Mg(II)·ADP to mimic the posthydrolysis state.

The W-band DEER results for the C-domain D560C/Gd(III) and K637C/Gd(III) mutants are given in Fig. 2, and the primary DEER data are in SI Appendix, Fig. S8. The DEER traces revealed dipolar oscillations and distance distribution with maxima at ∼5.2 nm and ∼3.4 nm, respectively, for all hydrolysis states. The distance distributions are in good agreement with those predicted using the closed X-ray structure of yHsp90 (PDB ID code 2CG9) (32) (SI Appendix, Fig. S6) by the MtsslWizard software (42). The similar distance distributions for all hydrolysis states for the 2 mutants show that the C-domains are structured and dimerized, independent of the hydrolysis state, also in the absence of cochaperone binding.

Fig. 2.

W-band DEER distance measurements on the C-domain yHsp90 mutants (A) D560C/Gd(III) and (B) K637C/Gd(III). On the left are given the background-corrected DEER time traces with the fit shown in gray lines and on the right are the distance distributions with confidence intervals as analyzed in the DeerAnalysis software (63). Modeled distance distributions using the MtsslWizard software (42) and the closed X-ray structure [PDB ID code 2CG9 (32)] are shown in orange. For details of the data processing and modeling, see Methods. The apo-, pre-, and posthydrolysis states are in black, green, and blue, respectively. The background-colored areas in the distance distributions denote reliability regions as follows: green: shape reliable; yellow: mean and width reliable; orange: mean reliable; red: nonreliable. The data are shown with offset on the y axis. The DEER data were acquired with chirp pump pulses (74–76, 78) (for set-up, see Methods and SI Appendix, Methods and Fig. S7C) and the primary DEER data are given in SI Appendix, Fig. S8.

DEER measurements on the corresponding proxyl-labeled mutants (experimental set-up given in SI Appendix, Fig. S7A and data in SI Appendix, Fig. S9) revealed for D560C/proxyl a very broad distance distribution, with the maximum at ∼5.4 nm, similar to the maximum for D560C/Gd(III), and no difference in the distance distribution between the apo- and prehydrolysis states. The distance distribution obtained from K637C/proxyl revealed for the apo- and prehydrolysis states a superposition of a very broad distance distribution and a narrower one with a maximum at ∼3.5 nm, which agrees with the Gd(III) data and the crystal structure prediction for the closed state. This component was not detected for the posthydrolysis state.

The difference between the results obtained using proxyl and the Gd(III) label for both C-domain mutants is remarkable and unexpected and will be discussed later.

DEER data on the Q385C/proxyl and E517C/proxyl in the M-domain (SI Appendix, Fig. S10) yielded again very broad distance distributions, which did not reveal unambiguous changes upon the addition of nucleotides except for the disappearance of a short distance component in the posthydrolysis state. The results of the N-domain mutants, revealed again broad distance distributions and only subtle differences between the different hydrolysis states (SI Appendix, Fig. S11). For D61C the distance predicted for the closed state is significantly larger than experimentally observed. For both the N- and M-domain proxyl-labeled mutants we observed an increase in the modulation depth for the pre- and posthydrolysis states, which could be an indication for some general increase in the structural order of the broad distance distribution (43). We, also, performed a proxyl–proxyl DEER measurement at 34 GHz (Q-band) for A152C/proxyl in the apo state (set-up in SI Appendix, Fig. S7B), recovering the same broad distance distribution (SI Appendix, Fig. S12) as with the W-band measurement. Overall, the DEER data on the mutants report for the N- and M-domains a very wide range of protomer-to-protomer distances, practically insensitive to the presence of nucleotides.

The broadening of the distance distributions originates probably from both the protein multiple conformations and from effects related to the proxyl spin label. The contribution of the proxyl label can be apprehended by comparison of the distance distributions of the C-domain mutants with the proxyl and rigid Gd(III) labels (Fig. 2 and SI Appendix, Fig. S9). Although the proxyl label has a somewhat more flexible tether than the Gd(III), predicted distance distributions for the closed conformation, taking this into account, cannot explain the large difference in width. Additional broadening in the proxyl-labeled mutants can occur as a result of potential ring opening of the maleimide moiety due to hydrolysis (44, 45), which would add 1 more bond rotation, increasing the flexibility of the tether, however not to the extent observed. Another possibility that could explain the broad distance distributions is the presence of multiple labeling sites, which, however, we find unlikely as we kept the pH during labeling in the range of 7.2 to 7.4, where the specificity of the maleimide group to cysteine residues is high (46). Aggregation is another option, although we have no indications for it, such as decrease of the phase memory times or unexpected large modulation depth. All proxyl-labeled samples gave a similar phase-memory time of ∼2 μs (an example in SI Appendix, Fig. S13A), which is not unreasonably short. Finally, there exists the possibility that the label itself can perturb the structure, allowing for a higher motion amplitude between the 2 protomers. In silico labeling of A152C and Q385C, which are buried sites in the closed X-ray structure, with MMM (47) and MtsslWizard software (42), showed that these sites should be poorly labeled. Labeling was successful because it was done in the apo, open state, but the presence of the labels in these sites can perturb the closed state.

While using the Gd(III) spin-labeling approach we were able to establish the dimerization of the C-domains in apo-, pre-, and posthydrolysis states, with no indication for conformational changes, the spin-labeling approach was not useful for similar characterization of the N- and M-domains. Therefore, we turned to a different approach that can report on the protein conformational heterogeneity in the nucleotide-bound states, while minimizing the undesired spin-label contribution to the distance distribution. For this, we replaced the EPR silent Mg(II) cofactor, which is located at the ATPase site complexed with the nucleotide (20), with paramagnetic Mn(II). Mn(II) has the same size and charge as Mg(II) and has been often used as a paramagnetic probe for Mg(II) (48–50). Here, in addition to Mn(II)–Mn(II) DEER, hyperfine spectroscopic methods allowed us explore the hydrolysis-related changes in the Mn(II) coordination at the ATPase site itself (51).

Mn(II) EPR.

Sample composition.

Prior to carrying out spectroscopic measurements on yHsp90 in the presence of various Mn(II)·nucleotides complexes, we ensured that the ATPase catalytic activity is maintained upon Mg(II) → Mn(II) substitution (SI Appendix, Figs. S4 and S5). Next, we evaluated the optimal distribution of Mn(II)·nucleotide species in solution such that the Hsp90-bound Mn(II)·nucleotide will comprise a significant part of the sample, taking into account the relatively low binding constant of Mg(II)·ATP and Mg(II)·ADP to Hsp90 (20) and DEER SNR requirements. It is not possible to use excess of Mn(II)·nucleotide, which ensures that all Hsp90 monomers will bind Mn(II)·nucleotide [as was previously done for the spin labeled mutants with Mg(II)], because the sample will be dominated by unbound Mn(II)·nucleotide. Since EPR detects only paramagnetic species, the unbound Mn(II)·nucleotide species or any free Mn(II) will contribute as an intense unwanted background in all EPR measurements. Therefore, we have to work under conditions of substoichiometric amounts of Mn(II)·nucleotide with respect to Hsp90 such that the Mn(II)·nucleotide·Hsp90 species comprises a significant part of the sample, minimizing background Mn(II) and Mn(II)·nucleotide. Together with the substoichiometric amounts of Mn(II)·nucleotide, a high concentration of protein is desirable since it will shift the equilibrium to the wanted Mn(II)·nucleotide·Hsp90 species. The different paramagnetic species in solution are free Mn(II), Mn(II)·nucleotide, and Mn(II)·nucleotide·Hsp90, the relative concentrations of which are determined by the chemical equilibria given in Eqs. 1 and 2.

$$\text{Mn}\left(\text{II}\right)+\hspace{0.17em}\text{nucleotide}\stackrel{K_{d,1}}{\rightleftarrows }\text{Mn}\left(\text{II}\right)·\text{nucleotide}$$ [1]

$$\text{Mn}\left(\text{II}\right)·\text{nucleotide}\hspace{0.17em}+\hspace{0.17em}\text{Hsp}90\stackrel{K_{d,2}}{\rightleftarrows }\text{Mn}\left(\text{II}\right)·\text{nucleotide}·\text{Hsp}90$$ [2]

We used literature values for the equilibrium constants K_d,1 and K_d,2 (20) for ADP and ATP (see values in Table 1) and estimated, using the software Tenua (52), the percentage of the various Mn(II) species that will exist at equilibrium (Table 2 and SI Appendix, Fig. S14) in our samples, having a composition of Mn(II):nucleotide:Hsp90 284:284:314 μM protomer. In principle, a third reaction, where free Mn(II) binds to Hsp90, may take place as well. This possibility is considered unlikely and ignored at this stage. Apo Hsp90 and free nucleotide are EPR silent and thus do not contribute to the EPR signal.

Table 1.

Equilibrium dissociation constants for the reactions shown in Eqs. 1 and 2

Dissociation constants	ATP	ADP
Kd,1*, M−1	10−5 (79)	4 × 10−5 (79)
Kd,2†, M−1	132 ± 47 × 10−6 (20)	9 ± 3 × 10−6 (20)

^*Refers to MnCl₂, at 25 ± 1 °C and pH 8.0 for ADP and 7.5 for ATP.

^†Refers to MgCl₂, at 25 °C using yHsp90 in 20 mM Tris buffer at pH 7.4.

Table 2.

Estimated percentage of free Mn(II), Mn(II)·nucleotide, Mn(II)·nucleotide·Hsp90 for a solution of Mn(II):nucleotide:Hsp90 284:284:314 μM protomer (SI Appendix, Fig. S14)

Species	ATP, %	ADP, %
Mn(II)	11.3	11.0
Mn(II)·nucleotide	38.7	8.7
Mn(II)·nucleotide·Hsp90	50.0 (25.0)	80.3 (64.4)

In parenthesis is the percentage corresponding to Hsp90 with both protomers having bound Mn(II)·nucleotide after squaring the percentage of Mn(II)·nucleotide·Hsp90 complexes. This value is important for the DEER measurements because these are the only species expected to produce a DEER effect.

It becomes evident that in the posthydrolysis state 80% of the total Mn(II) signal is expected from Mn(II)·ADP·Hsp90 complexes, while for the prehydrolysis state Mn(II)·ATP·Hsp90 is expected to be only 50%. The percentage of Hsp90 dimers with Mn(II)·nucleotide bound to both promoters is lower, 64% and 25%, for ADP and ATP, respectively. Accordingly, a lower than usual Mn(II)–Mn(II) modulation depth in the DEER measurements is expected.

Three types of samples, with the above-mentioned sample composition were prepared for hyperfine and DEER spectroscopic measurements. 1) The prehydrolysis state was mimicked by adding Mn(II)·ATP and freezing the sample rapidly (within ∼5 s) or by adding Mn(II)·AMP–PNP. 2) For the HES, Mn(II)·ATP/Vi (1:7 molar ratio) was added to Hsp90 followed by freezing within ∼5 s and at ∼2 h. 3) The posthydrolysis state was mimicked by adding Mn(II)·ADP or by adding Mn(II)·ATP and allowing long reaction time (∼2 h) to ensure complete hydrolysis before freezing the sample. SI Appendix, Table S1 summarizes the composition, reaction times, and names of all samples used.

Hyperfine spectroscopy measurements.

The electron-nuclear double resonance (ENDOR) experiment provides the coordination environment of the Mn(II) through measurement of the hyperfine coupling to nearby nuclear spins (53). We used W-band Davies ENDOR to detect the hydrolysis state of ATP through the ³¹P hyperfine coupling of nucleotide phosphates in yHsp90, as the ³¹P ENDOR spectra of the pre- and posthydrolysis states were previously demonstrated to be different (51, 54–57). All samples exhibited very similar echo-detected EPR spectra (see an example in SI Appendix, Fig. S7D) and all ENDOR measurements were performed having the magnetic field set to the fourth Mn(II) EPR line (blue arrow in SI Appendix, Fig. S7D). Fig. 3A shows the W-band ³¹P Davies ENDOR spectra in the pre-, HES, and posthydrolysis states.

Fig. 3.

The coordination environment of yHsp90. The sample names indicated on the figure are summarized in SI Appendix, Table S1. (A) ³¹P ENDOR spectra at different states of the Mn(II)·ATP·Hsp90 hydrolysis cycle with the pre-, HES, and posthydrolysis states in cyan, blue, and black, respectively. The gray arrows indicate the shoulder that comprises the major difference between the post- and prehydrolysis states. (B) EDNMR spectra of the different states as in A with the addition of samples prepared with ¹⁵N₅ATP. (C) As in B, with zoom on the nitrogen signals showing also the ¹⁵N hyperfine coupling.

All spectra are characterized by a doublet centered around the ³¹P Larmor frequency at 58.3 MHz. The ³¹P-Mn(II) hyperfine splitting, as determined by the peak-to-peak frequency separation, of the posthydrolysis state (black traces in Fig. 3A) is ∼3.4 MHz, indicated by a black arrow in Fig. 3A, while for the prehydrolysis state (cyan traces in Fig. 3A) it is ∼3.9 MHz, indicated by a cyan arrow. We estimated that in the posthydrolysis state 80% of the Mn(II) species are in the form of Mn(II)·ADP·Hsp90 complexes (Table 2); therefore, the observed hyperfine splitting reflects mostly the ³¹P-Mn(II) coordination from nucleotides bound to Hsp90. For the prehydrolysis state, however, we anticipate the sample to consist of ∼40% Mn·ATP and ∼50% Mn·ATP·Hsp90 species; therefore, the observed signal is a convolution of both. The spectra of the pre- and posthydrolysis states are different, as reported previously (51, 55, 56, 58, 59), with the major difference being the inner “shoulder” corresponding to a spliting of 1.8 MHz, marked with gray arrows and dashed lines in Fig. 3, in addition to the shift of the maxima. These shoulders were shown to be present also in the free Mn(II)·ATP and Mn(II)·AMP–PNP complexes (58, 59). This coupling was previously observed for other Mn(II) ATPase sites (51, 55) and has been used as a signature of ATP coordination to Mn(II). This signal is absent from the spectra of posthydrolysis samples (black traces in Fig. 3A), therefore indicating that hydrolysis has occurred and the γ-phospate of the nucleotide is no longer coordinated to Mn(II). With regard to the HES state (long time, blue trace in Fig. 3A), the ³¹P ENDOR spectrum is similar to that of the posthydrolysis state both in terms of the peak-to-peak splitting and absence of the inner shoulder, indicating that the coordination of Mn(II) ions is the same in the posthydrolysis state and HES and that the γ-phospate is no longer bound to the nucleotide.

After having observed the coupled ³¹P nuclei from the nucleotides to the Mn(II) and distinguished the pre- and posthydrolysis states by the ENDOR experiment, we proceeded on to probing direct Mn(II)–protein interaction via ¹⁴N couplings using electron–electron double resonance (ELDOR) -detected NMR (EDNMR) (60, 61), as was done previously for the MsbA and BmrCD transporters (51). We used EDNMR spectroscopy because it is particularly useful for low-γ nuclei having low NMR frequencies, like ¹⁴N with a Larmor frequency of ∼10 MHz at the W-band. Potential ¹⁴N–Mn(II) interactions are expected from nitrogen atoms of the nucleotide (51) and nitrogen atoms of yHsp90 amino acid residues. To distinguish between the 2 contributions we used ¹⁵N-enriched ATP, ¹⁵N₅ATP, as reported earlier (51).

For EDNMR measurements we set the field to the fourth Mn(II) line, similar to the ENDOR measurements, and depending on the sample composition we observed, together with ¹⁴N (nuclear spin I = 1) signals, signals from ¹⁵N (I = 1/2), ^35/37Cl (I = 3/2), ⁵¹V (I = 7/2), ¹H (I = 1/2), ¹³C (I = 1/2), ³¹P (I = 1/2), and ⁵⁵Mn (I = 5/2) indicated in Fig. 3B (the full-range spectra are given in SI Appendix, Fig. S15). The Vi signal, appearing at the ⁵¹V Larmor frequency of ∼38 MHz, is present only in the samples where it was deliberately added and confirms that Vi is located in the ATPase site and Hsp90 is indeed trapped in HES (35). The strong ¹⁴N signal at 10.0 MHz corresponds to ¹⁴N single-quantum transition indicating the presence of Mn(II)–nitrogen interaction for all states. Next, we focused on the samples prepared with ¹⁵N₅ATP. In all samples we detected ¹⁵N signals, originating from coordination of the nucleotide to Mn(II), probably nonbound to protein (62), along with an intense ¹⁴N signal, which originates from the protein (SI Appendix, Fig. S16). In general, we observe a systematically larger ¹⁴N intensity in the posthydrolysis state than for the prehydrolysis state (Fig. 3C and SI Appendix, Fig. S16). The same is observed for the short- vs. long-time HES samples. The EDNMR data show that not only the ³¹P environment is the same in the ADP-bound and HES as reported by ENDOR (Fig. 3A), but the same holds true for the nitrogen atoms around the Mn(II) cofactor. Control experiments were carried out to ensure that the ¹⁴N coordination does not originate from the His₆ tag and the Tris·HCl buffer (SI Appendix, Figs. S17–S19). Here, we also found that in presence of nucleotides there exists no unspecific binding of Mn(II) to the protein (SI Appendix, Fig. S19).

Mn(II)–Mn(II) DEER measurements to probe inter–N-domain interactions.

After having determined the coordination of Mn(II) in the ATPase site in the various hydrolysis states, we proceeded to measure the distance between the 2 Mn(II) cofactors at the different states of the ATPase cycle using DEER spectroscopy. The background corrected DEER data and the derived distance distributions obtained using the DeerAnalysis software (63) are shown in Fig. 4 A and B, respectively. For DEER we used the same samples as in the ENDOR and EDNMR experiments to allow direct comparison of the different datasets. Therefore, all Mn(II)–Mn(II) DEER measurements are recorded in protonated buffers, in contrast to the common approach of performing DEER in deuterated buffers that allows prolonged phase-memory times. We avoided deuterated buffer in EDNMR because of the strong ²H signal (expected to be at 22.2 MHz at the W-band), which would overlap with ¹⁴N/¹⁵N signals. In addition, due to the low modulation depth, λ, of the DEER traces (indicated with red arrow in Fig. 4A), we ensured reproducibility by preparing each sample several times, with the number of preparations (×n) given in Fig. 4B. Additionally, for some of the samples we repeated the measurement on a different day. All primary data of the various samples and repeats are given in SI Appendix, Figs. S20–S22.

Fig. 4.

Mn(II)–Mn(II) DEER measurements of yHsp90. The sample names indicated on the figure are summarized in SI Appendix, Table S1. (A) Mn(II)–Mn(II) DEER time traces at different states of the ATP cycle in WT yHsp90. The gray lines indicate the fit to the data after removing the background signal using the DeerAnalysis software (63) (primary data and validation of the background for the individual DEER traces are in SI Appendix, Figs. S20–S22). The DEER data were acquired with chirp pump pulses (74–76, 78) (for set-up see Methods and SI Appendix, Methods and Fig. S7D). (B) Average Mn(II)–Mn(II) distance distributions obtained upon multiple repeats of sample preparation and/or measurements with the number of different sample preparations indicated (×n). The gray lines at ∼4.3 and 3.8 nm indicate the Mn(II)–Mn(II) maximum/center of the distance distribution for pre- and posthydrolysis states, the red dotted line indicates the distance between the Mg(II) ions in the closed cryo-EM structure of human Hsp90/Cdc37/Cdk4/MoO₄²⁻ complex [PDB ID code 5FWPB (33)], and the red dashed line represents the distance between 2 metal pseudoatoms coordinated by the nucleotide in the closed X-ray structure of yHsp90 (32) (see also Fig. 1A). (C) λ Values at different nucleotide-bound states; the error bars represent the SD among different samples and repeats.

The Mn(II)–Mn(II) DEER data of the prehydrolysis state, Mn(II)·ATP, 5 s and Mn(II)·AMP–PNP, show clear modulation yielding a distance distribution with a maximum at 4.3 nm, and some low intensity at shorter distances. At the HES the distance distribution is almost double in width and centered at ∼3.8 nm. Finally, the posthydrolysis state features a broader distance distribution, similar to that of the HES. Additionally, Mn(II)–Mn(II) distance measurement with small excess of Mn(II)·nucleotide [i.e., Hsp90:Mn(II)·ATP (2 h reaction time) 1:1.3:1.3] did not lead to a noticeable shift in the distance distribution (SI Appendix, Fig. S23). Overall, the data suggest that the Mn(II)–Mn(II) distance distribution in the ATP-bound state is centered at 4.3 nm and broadens toward shorter distances centered around 3.8 nm after ATP hydrolysis, indicating the presence of 2 different nucleotide-dependent conformational ensembles of yHsp90.

The differences between the pre- and posthydrolysis samples was not limited only to the distance distribution but also to the modulation depth; λ was ∼1.7% for the posthydrolysis state, while for the prehydrolysis state it was only ∼0.6%, a factor of 2.8 lower (Fig. 4C). Based on the dissociation constants of Mn(II)·nucleotide and nucleotide·Hsp90 complexes (Table 2 and SI Appendix, Fig. S14), we expect the prehydrolysis state to have a higher percentage of free Mn(II)·ATP and Hsp90 dimers with only 1 bound Mn(II)·ATP, as compared to free Mn(II)·ADP and Hsp90 dimers with only 1 bound Mn(II)·ADP in the posthydrolysis samples. Accordingly, λ is expected to be larger for the posthydrolysis by a factor of ∼2.6, in good agreement with the experimental observation. For comparison, the modulation depths of Mn(II)–Mn(II) DEER of a DnaB helicase with bound Mn(II)·AMP–PNP and of the BmrA transporter with bound Mn(II)·ATP/Vi, were 1% and 0.12%, respectively (49).

The dissociation constant of dimeric Hsp90 is in the range of few nanomolars (64, 65); therefore, the low modulation depths of the Mn(II)–Mn(II) DEER measurements (Fig. 4 A and C) are not expected to originate from the presence of monomeric Hsp90. Additionally, DEER distance measurements between the D560C/Gd(III) sites in the C-domain, which is known to be the dimerization site (21), resolved a well-defined distance distribution at the pre- and posthydrolysis states that matched well the distribution predicted by the X-ray structure (Fig. 2A), with modulation depth values indicative of quantitative dimers with Gd(III) tags (40, 66). Accordingly, Hsp90 is in its dimeric form in our hands and conditions (pH, buffer, salt and protein concentration), as also evident from size-exclusion chromatography data (SI Appendix, Fig. S2). Finally, we performed a control DEER measurement on free Mn(II) ions in solution in order to exclude that the observed DEER signals are artifacts of our experimental set-up (SI Appendix, Fig. S24). Indeed, no significant pairwise DEER effect could be detected.

Discussion

Prior to discussing the implications of the results described above, we summarize our major experimental observations as follows: Distance distributions between the C-domains reported by a rigid Gd(III) spin label were the same at all hydrolysis states, showing the C-domains of yHsp90 to be dimeric with well-defined distance distribution. In contrast, proxyl-based DEER gave inconclusive results due to very broad C-to-C, N-to-N, and M-to-M interdomain distance distributions, suggesting a heterogeneous distribution of contacts between the 2 protomers for apo-Hsp90 and pre- and posthydrolysis states. The broad distance distributions include contributions from both the protein flexibility, as well as from the proxyl label-specific effects. We circumvented this difficulty for the N-domain by using Mn(II) as a paramagnetic probe, substituting the intrinsic Mg(II) ion in the ATPase site. This approach is free of complications related to spin-label effects and targets an active site of the protein. Here, we identified the hydrolysis state via the Mn(II) coordination environment using ENDOR and EDNMR techniques. We detected, in addition to the coordination of the nucleotide phosphates to the Mn(II), a Mn(II)–nitrogen interaction with a yHsp90 residue at all states. For the HES state we detected the presence of the vanadate ion in the ATPase site and a Mn(II)–phosphate coordination similar to that of the posthydrolysis state. Additionally, Mn(II)–Mn(II) DEER gave well-defined distance distribution for the prehydrolysis state, which shifted to a shorter and broader distance distribution after hydrolysis. The distance distributions in the HES and ADP-bound states were found similar.

The observation of dimerized C-domains in apo- and nucleotide-bound yHsp90 is in agreement with biochemical studies on human (Hsp90α and Hsp90β) homologs (21, 67, 68). Additionally, the C-domains were found dimerized in the closed AMP–PNP-bound yHsp90 X-ray crystal structure in the presence of cochaperone p23/Sba1 (32). Earlier sm FRET measurements on the D560C mutant labeled with an Atto dye in the absence of cochaperones/clients reported that the C-domains underwent open/close transitions (label-to-label distances of 6.42 vs. 4.42 nm, respectively) in an anticorrelated mode with respect to the N-domain opening and closing (39). Because DEER in frozen solutions reports a superposition of all conformations this difference of 2 nm between the 2 conformations should have been resolved in the Gd(III)–Gd(III) DEER measurement, considering the width of the Gd(III)–Gd(III) distance distribution. Its absence suggests that under our conditions no transitions between open/closed conformations of similar probabilities of the C-domains occur.

Tracking the Mn(II)–phosphate coordination via ENDOR allowed us to follow the ATP hydrolysis in yHsp90 and relate it to global conformational changes. In addition, via EDNMR, we detected the proximity of nitrogen atoms from protein residues to the Mn(II) cofactor at both pre- and posthydrolysis states. The increased ¹⁴N signal in the posthydrolysis state with respect to the prehydrolysis can be well understood by the larger percentage of Mn(II)·ADP bound to Hsp90 with respect to Mn(II)·ATP (Table 2), and does not necessarily reflect a different ¹⁴N coordination sphere around the Mn(II). A close look at the closed X-ray structure of yHsp90 (32), as well as of the isolated N-domain (20), shows residues Asn37 (N37) and Phe124 (F124) to interact with the α- and β-phosphates of the nucleotide via hydrogen bonding. Additionally, N37 makes a hydrogen bond with a water molecule that binds the N₇ of the nucleotide; therefore, N37 is the most likely candidate for the observed ¹⁴N signals (69).

With respect to the global conformational changes, we found with Mn(II)–Mn(II) DEER the N-domains to be dimerized (closed) in solution in the pre-, HES, and posthydrolysis states in the absence of cochaperones. Furthermore, the conformational ensemble of the closed, prehydrolysis state is different from that of the HES and posthydrolysis states. Here, we stress that it is the use of the Mn(II) single-atom label, which does not introduce label-dependent contributions to the distance distribution, that allowed differentiating 2 spatially proximate, however structurally different, closed conformations. We term the ADP-bound (posthydrolysis) closed conformation as compact in line with the Agard group (18, 27), although it is more “wobbly” than the ATP (prehydrolysis) closed conformation. An alternative picture could be that there is an equilibrium between closed ⇄ compact conformations, with the closed conformation ensemble featuring a distance distribution maximum at 4.3 nm and the compact one having a maximum at 3.3 nm. With ATP-bound the closed conformation dominates, whereas with ADP-bound the 2 have comparable populations such that they are no longer resolved in the distance distribution, showing a mean distance of 3.8 nm. In this picture ATP hydrolysis shifts part of the population to the compact form.

It was reported that Hsp90 adopts more than 1 conformation when nucleotides are present. Specifically, EM showed that in the presence of saturating amounts of AMP–PNP and ADP, the bacterial homolog HtpG is in a mixture of open/closed and open/compact conformations, respectively (27). The same was observed by SAXS data on AMP–PNP-bound HtpG, which was found to be in a mixture of open/closed conformations (28). Additionally, sm FRET data showed an open ⇄ closed equilibrium in the apo state, which was shifted toward the closed with AMP–PNP (14, 19). In contrast, both open and closed conformations were detected with ATP under turnover conditions (19). Similar results were reported by another 3-color sm FRET study, where in the presence of ATP under turnover conditions or with ADP, the conformational transitions were found in thermal equilibrium and both nucleotides could bind to both open and closed conformations in yHsp90 (14). With the exception of FRET studies where AMP–PNP shifted completely the equilibrium toward the closed conformation (14, 19), the literature suggests that Hsp90 is found in the nucleotide-bound states in 2 distinct open and closed conformations. The question that arises is why we do not observe the open conformation? If the open state of Hsp90 has an N-to-N domain opening of 7 nm between the inner residues and up to 13 nm between the outer residues, as found from X-ray on HtpG (18) and from FRET and subsequent structure modeling on yHsp90 (65), such long distances are beyond the detectable DEER distance range under our experimental conditions, due to the limited phase memory time of ∼1.2 μs of our samples (SI Appendix, Fig. S13B). This can explain 1) why we observe only the closed states and 2) why the λ values of the Mn(II)–Mn(II) DEER were found low for the pre- and posthydrolysis states (even though their relative ratio was found as expected from K_Ds).

Observation of the ADP-bound compact conformation, which differs from the closed AMP–PNP/ATP-bound conformation, in frozen solution, without cross-linking is reported here. This compact conformation has been previously observed by the Agard group for HtpG with EM (18), while this conformation could not be observed for the yeast homolog if no interdomain cross-linking was present (27). This compact ADP-bound was also not observed by solution SAXS measurements on HtpG presumably owing to sample composition used for SAXS (5, 28). In contrast to our observations, bulk FRET studies reported the ADP-bound state of yHsp90 to be in an open conformation with an inter–N-domain distance of 7 to 13 nm (65). Previous sm FRET measurements showed no differences in the FRET trajectories under ATP or ADP conditions, suggesting similar open/closed conformation kinetics, and the authors concluded that “we did not observe a distinct conformation of an ADP bound state. If this state does also exist in yeast Hsp90, it must be short lived and little populated” (19). This result was recently supported by 3-color FRET data showing that yHsp90 open/closed conformations are driven thermally, and that ATP and ADP nucleotides bind to both open and closed conformations of Hsp90, with different kinetics, without affecting the open/closed conformations kinetics (14).

We measured a Mn(II)–Mn(II) distance distribution with a maximum at 4.3 nm in the prehydrolysis, ATP-bound, state, which broadened and yielded a broad maximum at the ∼3.8 nm region in the posthydrolysis, ADP-bound state, as well as in the HES. This indicates a larger inter–N-domain motion amplitude. Our experimental observations are summarized in Fig. 5 in the spirit of Fig. 1B. The range of the Mn(II)–Mn(II) distances measured with DEER are in the range of values reported for closed Hsp90 structures. More specifically, in the mitochondrial Hsp90 (Trap1) crystalized with AMP–PNP, ADP–BeF₃⁻, and ADP–AlF₄⁻ (the latter 2 reflect a trapped HES state), the Mg(II)–Mg(II) distance was found to be 2.79 nm (35). Crystallographic data on AMP–PNP-bound yHsp90 in the presence of cochaperone p23 revealed a distance between the α-phosphates of the nucleotide to be 2.6 nm (in this structure the metal ion was omitted). In silico engineering a metal pseudoatom at each ATPase site yields a metal–metal distance of 3.0 nm (Fig. 1A). In a cryo-EM structure of human Hsp90 with Cdc37/Cdk4/MoO₄^2- in an ATP state or, more likely, ADP·molybdate, the Mg(II)–Mg(II) distance was found to be 3.2 nm (33). The distances observed for both structures (X-ray and cryo-EM) are in the same range as our Mn(II)-based DEER data and seem to agree better with the average distance of the ADP-bound state as found from DEER rather than the somewhat more open ATP-bound state in solution. This would not be surprising since the presence of cochaperones could indeed stabilize the structure to the more compact client-driven conformation (10). This suggests that the current closed structures agree more with our ADP-bound conformation rather than the ATP-bound, even though AMP–PNP and ATP were used for crystallization for the yeast and human homologs, respectively. Interestingly, the ambiguous determination of the type of nucleotide in the human Hsp90 closed X-ray structure was pointed out by Verba et al. (33).

Fig. 5.

Summary of our DEER results. (A) Apo state, showing the dimerization of the C-domains with the distances obtained from the Gd(III)–Gd(III) DEER on 2 mutants indicated. (B) Prehydrolysis closed conformation as found from the Mn(II)–Mn(II) DEER. (C) Posthydrolysis and HES compact conformation as found from the Mn(II)–Mn(II) DEER. The schematic structure of yHsp90 is shown in the spirit of Schulze et al. (figure 6 in ref. 17). All maxima of the distance distributions are noted on the figure. To illustrate the broader distance distribution for the ADP-bound state in C, 2 positions were drawn for the N-domains.

Finally, we discuss the DEER results obtained with the spin labels for the N- and M- domains in light of the above discussion. Considering that in all hydrolysis states contributions from conformational ensembles characteristic of distinct open and closed states are present, one would expect to see 2 resolved peaks (even if they are broad) in the distance distribution, or 1 if the second is outside the range accessible by the method as observed for the Mn(II) DEER. In some cases, this was indeed observed. For example, mutant A152C, which is found in the vicinity of the ATPase site (Fig. 1A and SI Appendix, Fig. S11), yielded 2 major populations, one of which (∼3.6 nm) matches well the range of Mn(II)–Mn(II) distances, in addition to a second population with a very broad distribution. The distance distribution centered around ∼3.6 nm can be assigned to the closed conformations of the chaperone based on the X-ray closed structure and our Mn(II) DEER data (32). However, the distance resolution was too low to differentiate 2 different ADP- and ATP-bound closed states, as was done with the Mn(II) DEER. We cannot unambiguously assign the broader distance distribution to the open state. We note that in the case of DEER with spin labels, the spin-label flexibility adds to the distance distribution, thus compromising resolution. However, the excessive broadening observed for the proxyl spin label compared to the Gd(III) results on the C-domain suggests that other spin-label effects may contribute to the width, as we discussed earlier. The broad distance distribution obtained with the proxyl spin labels is in line with a thesis from the Kay group (70) reporting on DEER on MTSL-labeled yHsp90 mutants, where a large ensemble of conformational states that did not change upon nucleotide addition was also found.

Conclusions

First, by applying DEER and a rigid Gd(III) spin label we established the tight dimerization of the C-domains in apo- and nucleotide-bound yHsp90, representing the pre- and posthydrolysis states, in frozen solution in the absence of an artificial linker that ensures “holding” the yHsp90 protomers together. Second, by employing Mn(II) cofactor as a paramagnetic probe situated at the ATPase sites, we detected the hydrolysis state of the nucleotide-bound protein and the Mn(II) coordination and concomitantly probed the N-to-N domain distance distribution at the pre-, HES, and posthydrolysis states in frozen solution. Specifically, we detected closed and compact conformations for the ATP-bound, prehydrolysis state and ADP-bound, posthydrolysis state, respectively. The conformation of the HES was found similar to that of the posthydrolysis state. This suggests that the presence of the inorganic phosphate prior to its release does not affect the inter–N-domain contacts in the ATPase region. Although the ATP-bound state has been extensively studied, little was known about the HES and ADP-bound states, which are highly relevant to Hsp90’s functional conformational cycle. It has been proposed that the ADP-bound state represents a second functional state upon which clients are displaced from the hydrophobic surface areas of Hsp90, allowing completion of the cycle. Our results, apart from probing a biologically important solution state of yHsp90, open the route to studying via EPR Hsp90 in the presence of cochaperones and clients to further elucidate the mechanism of action of this important protein.

Last but not least, our approach can be applied to all molecular machines that use ATP/Mg(II) as fuel independently of the size of the protein.

Methods

Protein Expression and Purification.

Full-length yHsp90 was expressed and purified according to the literature (64) and detailed in SI Appendix.

Spin Labeling and Sample Preparation.

1) For Mn(II)-labeling, a premixed solution of MnCl₂·6H₂O/nucleotide 1:1 was added to yHsp90. The system was frozen at various times (SI Appendix, Table S1). The HES was trapped by adding sodium ortho-vanadate and MnCl₂·6H₂O/ATP. 2) BrPy-DO3A-Gd(III) labeling was done as reported in ref. 40. 3) Proxyl-labeling was done as in 2, except that TCEP was removed before proxyl was added and the labeling was done at pH 7.2 to 7.4.

For samples in the apo state, the labeled mutants were used as prepared (after dilution with deuterated buffer at ∼100 μM as protomer), while for samples in the pre- and posthydrolysis states saturating amounts MgCl₂·6H₂O (in D₂O, pD 7.5) and nucleotide (AMP–PNP or ADP, in D₂O, pD 7.5) were added to the labeled mutants to final concentrations of 10 mM MgCl₂·6H₂O and 10 mM nucleotide, ∼100 μM protein. See SI Appendix for details.

ATPase Activity Measurements.

The enzymatic activity of yHsp90 was measured at 37 °C by using a standard ATPase activity assay (71). See SI Appendix for details.

EPR Spectroscopy and Data Analysis.

Unless otherwise stated, all data were recorded on a home-built W-band (∼94.9 GHz) spectrometer (72). All Mn(II)-tagged or Gd(III)-labeled samples were recorded at 10 K, while proxyl-labeled samples at 25 K and the temperature was stabilized with a cryo-free cooling system. Echo-detected EPR (ED-EPR) spectra were recorded using a Hahn echo sequence (π/2 – τ – π – echo) and sweeping the magnetic field. Echo-decay traces were recorded at the maximum of the EPR spectrum monitoring the echo intensity of a Hahn echo sequence by varying the τ interval of the Hahn echo sequence. ³¹P Davies ENDOR data were recorded using the standard Davies ENDOR sequence (π_mw – T – t_rf – T – π/2_mw – τ – π_mw – τ – echo) (53) combined with a Carr Purcell Meiboom Gill (CPMG) sequence for enhanced sensitivity (73) and monitoring the echo intensity at the observed microwave (mw) frequency as a function of the frequency of the radiofrequency (rf) pulse. EDNMR data were recorded using the standard EDNMR sequence (HTA_mw2 – T – π/2_mw1 – τ – π/2_mw1 – τ – echo) (60, 61) combined with a CPMG sequence (73). W-band DEER data were recorded using the 4-pulse reversed [π/2_νobs – τ₁ – π_νobs – (τ₁ − t) – π_νpump – (τ₂ + t) – π_νobs – τ₂ – echo] DEER sequence (37, 38) using chirp pump pulses (74–76) monitoring the echo intensity with increasing t of the pump pulse π_νpump, while Q-band DEER data were recorded at 50 K on a Bruker E580 spectrometer (∼34.0 GHz) spectrometer equipped with a Bruker EN5107D2 resonator at 50 K using the standard DEER sequence (37, 38). All experimental parameters are given in SI Appendix.

The DEER time domain data for Gd(III) and Mn(II) samples were transformed into distance distributions using Tikhonov regularization implemented in the DeerAnalysis software (63). Due to the absence of visible dipolar oscillations, the proxyl DEER data were analyzed using Gaussian fits within the DD software (77).

Modeling of the Spin-Label Rotamers.

The position of spin labels was modeled using the PyMOL plugin MtsslWizard (42) using the closed crystal structure of yHsp90 [PDB ID code 2CG9 (32)]. Details are provided in SI Appendix.

Data Availability Statement.

All data discussed in the paper are available in the main text or SI Appendix.