|Home | About | Journals | Submit | Contact Us | Français|
Hsp90 is an essential molecular chaperone required for the folding and activation of many hundreds of cellular “client” proteins. The ATP-dependent chaperone cycle involves significant conformational rearrangements of the Hsp90 dimer and interaction with a network of cochaperone proteins. Little is known about the mechanism of client protein binding or how cochaperone interactions modulate Hsp90 conformational states. We have determined the cryo-EM structure of the human Hsp90:Hop complex that receives client proteins from the Hsp70 chaperone. Hop stabilizes an alternate Hsp90 open state, where hydrophobic client-binding surfaces have converged and the N-terminal domains have rotated and match the closed, ATP conformation. Hsp90 is thus simultaneously poised for client loading by Hsp70 and subsequent N-terminal dimerization and ATP hydrolysis. Upon binding of a single Hsp70, the Hsp90:Hop conformation remains essentially unchanged. These results identify distinct functions for the Hop cochaperone, revealing an asymmetric mechanism for Hsp90 regulation and client loading.
Heat shock protein 90 (Hsp90) is a ubiquitous and essential molecular chaperone required in eukaryotes for the folding and activation of a broad array of protein substrates (“clients”) including nuclear receptors (Picard et al., 1990), cell cycle kinases (Sato et al., 2000) and telomerase (Holt et al., 1999). Hsp90 activity is uniquely required during later stages of folding to stabilize near native-state forms of clients and promote rearrangements that lead to downstream protein-protein and protein-ligand interactions (Nathan et al., 1997; Picard, 2002). However, little is known where client proteins bind on Hsp90 or how ATP hydrolysis and conformational rearrangements contribute to productive folding and activation.
Hsp90 functions as a homodimer via high-affinity dimerization of the C-terminal domain (CTD) (Harris et al., 2004). N-terminal ATPase domain (NTD) and middle domain (MD) homologies make it a member of the GHKL protein family that includes topisomerase and MutL (Dutta and Inouye, 2000; Prodromou et al., 1997). ATP hydrolysis is critically important to Hsp90 function and ATPase inhibitor compounds, including geldanamycin and its derivatives have been shown in vivo to promote the degradation of client proteins, particularly cell-cycle kinases, and have led to Hsp90 becoming an attractive anti-cancer chemotherapeutic target (Neckers, 2002).
The Hsp90 ATPase cycle involves multiple conformational states involving significant rigid-body rearrangements about the domain interfaces. The apo full-length crystal structure of E. coli Hsp90 (HtpG) is in an open V-shaped form where the NTDs are separated by more than 80 Å (Shiau et al., 2006). An array of hydrophobic surfaces is exposed between the monomers and present plausible client and co-chaperone interaction sites (Harris et al., 2004; Meyer et al., 2003). Small angle x-ray scattering (SAXS) studies have shown that in solution the apo conformation is considerably more extended, suggesting that a combinatorial use of binding sites and apo conformations would allow accommodation of structurally diverse clients (Krukenberg et al., 2008). In contrast, the crystal structure of yeast Hsp90 bound to AMPPNP (nonhydrolyzable ATP analog) and the p23 cochaperone is in a closed, NTD dimerized conformation, revealing the ATP state of Hsp90 (Ali et al., 2006). A third, highly compact ADP conformation in which the exposed hydrophobic areas become buried was suggested to drive client release (Shiau et al., 2006). Single particle electron microscopy (EM) studies established that while all three of these states are conserved between E. coli, yeast and human Hsp90s, the conformational equilibria between states varies significantly for each species and correlates with the Hsp90 ATP hydrolysis rates from the different organisms (Southworth and Agard, 2008).
In eukaryotes, Hsp90 function requires interaction with a network of cochaperone proteins, forming dynamic multiprotein complexes that load specific clients on Hsp90 and stabilize nucleotide bound states during signaling and activation (Bose et al., 1996; Vaughan et al., 2006). Many cochaperones interact with Hsp90 via the tetratricopeptide repeat motif (TPR), a 7-member α-helix bundle that binds the C-terminal MEEVD pentapeptide on Hsp90 (Young et al., 1998). The Hsp90/Hsp70 organizing protein, Hop (Sti1 in yeast), is particularly critical, interacting with the apo form of Hsp90 during the client-loading phase of chaperone activity (Johnson et al., 1998). Hop has three TPR domains (TPR1, TPR2A, TPR2B), and is generally considered to be an adapter protein, facilitating client loading by Hsp70:ADP, which binds to the N-terminal TPR1 domain, while Hsp90 binds to the TPR2A domain (Figure 1A) (Brinker et al., 2002; Chen and Smith, 1998). However, studies show that Hop is a potent inhibitor of Hsp90 ATP hydrolysis – suggesting a more active role (Prodromou et al., 1999; Richter et al., 2003, Li et al., 2011). While interactions with Hop and Hsp70 are required for loading many essential client proteins, the conformational state of Hsp90, binding sites of the cochaperones and the molecular mechanism of the process are unknown.
The inherent flexibility of Hsp90 and the dynamics of the chaperone cycle have presented significant challenges in obtaining structural information on mammalian Hsp90 or cochaperone and client-bound complexes. Here we have achieved a 15Å resolution cryo-EM structure of the 300 kDa human Hsp90:Hop tetramer complex, identifying the client-loading conformation of Hsp90 and establishing an active role for Hop in the chaperone cycle. To overcome the transient nature of Hop:Hsp90 interactions, we developed critical crosslinking methods, including specific disulfide formation and low-level glutaraldehyde crosslinking that were essential in stabilizing this complex for structural analysis. In the structure, Hop stabilizes an Hsp90 apo conformation distinct from the solution state that is simultaneously poised for client loading and ATP binding. In this state the Hsp90 monomers remain accessible for client binding, but the NTDs have rotated to match the ATP closed state. Hop is positioned with a TPR domain extending from the Hsp90 dimer cleft, available for interaction with Hsp70. The interaction of Hop clearly blocks NTD dimerization, explaining how noncompetitive inhibition of ATP hydrolysis is achieved. Hsp70, in the ADP state, readily binds Hsp90:Hop, forming a client-loading complex with the Hsp90:Hop conformation essentially unchanged. Together these results present a molecular view of client protein delivery to Hsp90.
The interaction of Hop with Hsp90 is critically important for the delivery of many essential client proteins, including the glucocorticoid, estrogen and progesterone receptors by the Hsp70/Hsp40 system (Picard et al., 1990; Sanchez et al., 1985). Given what little is known about this mechanism we sought to in vitro reconstitute the human Hsp90:Hop complex for EM structural analysis.
To study Hsp90 and Hop association we used size exclusion chromatography in-line with a static multi-angle light scattering device (SEC-MALS) to separate complexes and accurately determine the molecular weight (mw). When wild type (wt) human Hsp90α and Hop are incubated and analyzed, a single peak elutes earlier than Hsp90 alone and was determined be approximately 230 kDa (Figure 1B). This corresponds to a 2:1 stoichiometry – a single Hop molecule per Hsp90 dimer. When Hsp90 and Hop are run alone the weights were determined to be 170 and 65 kDa, which match the calculated mw’s of an Hsp90 dimer (170 kDa) and Hop monomer (63 kDa), respectively. Previous experiments have shown that Hop dimerizes (Onuoha et al., 2008). However, we find only slight shifts in the elution volume and mass calculation when Hop alone is analyzed at very high concentrations (300 µM) indicating a very weak dimerization affinity and that Hop primarily exists as a monomer under our experimental conditions (Figure S1A).
A stoichiometry of 2:1 is counter to previous calorimetry binding studies (Prodromou et al., 1999), therefore, we surmised that a tetramer complex forms during the incubation but disassembles during SEC. When equimolar Hsp90 monomer and Hop (10 µM each) are incubated together the peak corresponding to free Hop is barely detectable, much reduced when compared to an equivalent amount of Hop protein alone (Figure S1B), suggesting that a tetramer complex is forming, but dissociates over the column resulting in a lower average mw. Indeed, substantial Hop protein is found in every fraction between the Hsp90:Hop and free Hop peaks. However, given that the mw consistently matches the mass of an asymmetric complex over a wide range of concentrations (1 µM to 50 µM, data not shown) we suspect that binding of the first Hop must decrease binding affinity at the second site favoring the formation of an asymmetric complex at lower concentrations.
When wt Hsp90:Hop is collected and further diluted for negative-stain EM analysis no large complexes are apparent and only extended Hsp90 particles are observed, preventing further characterization (Figure S2A). In order to increase the stability of the Hsp90:Hop complex, cysteine mutations were engineered in the TPR:MEEVD binding site with the aim of forming a covalent disulfide bond (S-S) between Hsp90 and Hop. Based on the crystal structure of the Hop TPR2A domain:MEEVD complex (Scheufler et al., 2000), two cysteine pairs were tested – M728C:E271C and 733C:T260C for Hsp90:Hop, respectively (Note: For 733C a cysteine was added to the end of the C-terminus). Both pairs specifically and efficiently crosslinked when incubated under non-reducing conditions, forming a larger complex observed as a single super-shifted band by non-reducing SDS-PAGE (Figure S1C). When homologous mutations were made in the TPR1 domain, reduced and non-reduced samples appeared identical and no additional bands were present indicating that the crosslink was specific for the TPR2A domain with no reactivity with cysteine mutants in TPR1 or with any of the native cysteines in Hsp90 and Hop. Only minor reactivity was seen between 733C and E271C, which is plausible given the proximity and potential flexibility of the site. The 733C:T260C pair was surprisingly stable under significant amounts of reducing agent (data not shown) and was thus chosen for further characterization. When 733C Hsp90 and T260C Hop are incubated together and analyzed by SEC-MALS, a symmetric peak with an average mw of 300 kDa is observed, corresponding to a 1:1 stoichiometry (Figure 1C). Furthermore, when this peak is collected and analyzed by SDS-PAGE a single super-shifted band predominates with very little observable free Hsp90 monomer (Figure 1D). Under reducing conditions, this band separates into equimolar amounts of Hsp90 and Hop. These results establish that both MEEVD peptides on the Hsp90 dimer can interact with Hop to form a tetrameric 300 kDa complex and that this complex is significantly stabilized by specific covalent disulfide crosslinks within the binding site (Figure 1E).
We next pursued EM analysis and structure determination of the S-S tetramer Hsp90:Hop complex characterized above. In typical negative-stain micrographs, S-S Hsp90:Hop forms a distinguishable globular complex that is more compact than apo Hsp90 alone (Figure S2B). Some extended Hsp90 molecules are observed (Figure 3, see below) as well. From reference-free (RF) class averages Hsp90:Hop has an apparent random orientation on the grid with an approximate diameter of 100 Å (Figure 2A). These results indicate that Hop binding stabilizes an Hsp90 conformation distinct from the extended apo form we previously observed.
Using the averages shown, a common-lines initial model was obtained with two-fold symmetry imposed (Figure 2A). Representative particles and class averages are shown matched to 3D views of the final model (Figure 2B). From the reconstruction a large, 80 Å diameter V-shaped arrangement is apparent with significant density spanning the middle (Figure 2C). By comparison to known structures of Hsp90 it was apparent that the V-shaped structure arises from an Hsp90 dimer conformation that closely matches the apo crystal structure (Shiau et al., 2006) and the globular density spanning the domains likely corresponds to the two Hop cochaperones (Figure 2D). By comparison of the front and side-views, additional changes in the Hsp90 conformation within the Hsp90:Hop complex seemed likely, given the differences in dimension. However, the 25 Å resolution of the map limited our ability to clearly determine the Hsp90 conformation or the Hop interaction sites. Therefore, we next attempted to improve resolution by collecting data from frozen, hydrated samples and achieve a cryo-EM reconstruction.
As noted above, extended Hsp90 molecules are clearly visible in typical images, indicating that despite the S-S crosslink and the purity of the 300 kDa complex, some conformational heterogeneity is present (Figure 3A). While the globular Hsp90:Hop structures are easily discernable by negative-stain and were specifically selected for that reconstruction, the low signal to noise of cryo-EM images presented additional challenges in selecting homogenous single particle data sets for cryo refinement. To overcome this we employed brief, low-level glutaraldehyde (0.005%) crosslinking prior to EM analysis. Previously we have shown this to be a highly successful strategy for trapping human and yeast Hsp90 in native, nucleotide-specific states (Southworth and Agard, 2008). When Hsp90:Hop is briefly crosslinked and analyzed by negative-stain EM the particles appear strikingly more homogeneous with no observable extended Hsp90 molecules (Figure 3B). Reference-free averages of uncrosslinked and crosslinked samples appear similar (Figure S2C) indicating that crosslinking is not generating structural artifacts. The quantified ratio of compact to extended molecules was determined to be approximately 70:30 for un-crosslinked (Figure 3A) and >90:10 for crosslinked (Figure 3B) samples by picking and comparing single particles from several micrographs (data not shown). Thus, for Hsp90:Hop there appears to be an equilibrium between an extended form with a minimally bound Hop and a well organized V-shaped form, likely stabilized by additional Hop contacts (Figure 3C). The addition of crosslinker efficiently traps the V-shaped structure ensuring a more homogeneous sample. Crosslinked human apo Hsp90 alone appears heterogeneous with primarily circular shaped particles that are distinct from the Hsp90:Hop complex (Southworth and Agard, 2008), indicating that crosslinker alone does not trap the observed Hsp90:Hop conformation.
We next determined the cryo-EM structure of Hsp90:Hop. Following purification and analysis of the S-S Hsp90:Hop complexes by SEC-MALS, samples were briefly treated with glutaraldehyde and quenched to stabilize the V-shaped conformation prior to freezing. The small size of the complex required the use of moderately high defocus values (3.5 to 6 µm) and very thin vitreous ice to readily visualize the compact Hsp90:Hop particles (Figure S3A). The negative-stain structure (Figure 2), low-pass filtered to 40 Å, was used as a starting point for the cryo refinement. For comparison, the very good agreement between reference-free class averages generated from an initial cryo-EM data set and projections from the final model is shown (Figure 4A). This equivalence combined with the notable differences between the initial and final models indicates minimal model bias from the negative-stain starting structure. The final model was achieved following 15 rounds of refinement and estimated as having 15 Å resolution by Fourier shell correlation (Figure 4B). We found near-complete angular distribution of the single particles, indicating all views were represented, however some views predominate, suggesting either preferred orientation within the ice or some bias in particle-picking (Figure S3B).
The final model is similar to the negative-stain structure but with significant improvement in the Hsp90 domain features and the definition of the Hop cochaperone (Figure 4C). Hop is clearly making distinct contacts, bridging the Hsp90 structure with significant density spanning across and out from the V-shaped monomers. To determine the Hsp90 conformational state several full-length crystal structures and SAXS models were fit into the EM density. The CTD and MD from the crystal structure of apo E. coli Hsp90 (HtpG) fit accurately into the density and precisely match the open angle of the structure (Figure 4D). This conformation is significantly more closed then the solution apo state, with a 30° decrease in the open angle compared to HtpG (Krukenberg et al., 2008) and an even greater change in comparison to the extended apo state of yeast and human Hsp90s modeled from SAXS analysis (Krukenberg et al., 2009). In the crystal structure, the V-shaped open angle that matches the Hop-bound conformation appears to be stabilized by interactions with symmetry-related molecules in the unit cell. Given the significant flexibility of apo Hsp90, it is surprising that these E. coli and Human apo structures, achieved under different conditions, are in an identical MD-CTD open arrangement. This suggests that the open conformation stabilized by Hop binding observed here is a conserved and functionally important part of the Hsp90 conformational repertoire across organisms.
While the MD and CTD from the apo crystal structure fit well into the Hsp90:Hop map, the NTDs sit outside the EM density in this arrangement, at near right angles to large globular domains that are likely the NTDs of Hsp90 (Figure 4D, panel 2). To determine the NTD arrangement in the Hsp90:Hop structure we fit all other known NTD-MD conformations into the EM density. Remarkably, we found that the NTD-MD arrangement of the ATP, closed structure fits precisely (Figure 4E and Movie S1). Thus, Hop binding simultaneously promotes a reduced MD-CTD open angle and a significant, near 90° rotation of the NTD compared to the apo solution state. Importantly, this fit identifies an alternate conformation of Hsp90, revealing that Hop promotes a significant conformational change at both domain interfaces, likely positioning Hsp90 for the subsequent client loading, ATP binding and hydrolysis stages of the chaperone cycle.
By accurately fitting Hsp90 into the reconstruction we are able to identify density corresponding to the two Hop cochaperones (Figure 5A). Hop binds Hsp90 in an elongated arrangement, spanning more than 70 Å, from below the CTD extending across the interdimer cleft, making interactions with the adjacent Hsp90 monomer. No interaction between the two Hop molecules is observed, indicating that each Hop binds the Hsp90 dimer independently. The dominant Hop interactions with Hsp90 appear to be at the MD-CTD junction. Given proximity of this extra density to the CTD, the MEEVD:TPR interaction that is required for Hop binding likely resides in this region. In the yeast Hsp90 crystal structure 33 C-terminal residues are missing, preventing localization of the MEEVD. Although the final ordered residue in the structure (N677) is at the base of the CTD, the additional 30 residues from this point would be sufficient to position the MEEVD at the MD-CTD junction where Hop appears to bind. The NTD-MD bridging interaction across the interdimer cleft is not seen at increased contour levels (Figure S3C), indicating that this interaction is somewhat disordered or tenuous. However, interaction with the NTD is also observed in the negative-stain reconstruction (Figure 2) in the absence of glutaraldehyde, supporting the validity of this connection and previous biochemical studies identified NTD residues that are required for complete Hop binding as well (Richter et al., 2003).
The Hop cochaperone contains three TPR domains, comprising more than 70% of the molecule (Figure 1A). While the full-length structure is unknown, individual TPR1 and TPR2A domains bound to the respective Hsp70 and Hsp90 C-terminal peptides have been crystallized (Scheufler et al., 2000) and by homology, are likely similar in structure TPR2B. Given the arrangement of the Hop density that is apparent from the Hsp90 fit and the known interaction between TPR2A and the MEEVD, Hop is likely positioned such that TPR2A is in the main Hsp90-interacting density, packed against the MD-CTD junction. Depending on the N to C orientation, the density extending between the Hsp90 monomers could correspond to either TPR1 or TPR2B. Hsp70 specifically binds TPR1 during client delivery and interaction in this region would optimally position Hsp70 for client release into the interdimer cleft and proposed hydrophobic client-binding surfaces (Fang et al., 2006). To study this further we labeled an N-terminal His-tag on Hop with Ni-NTA gold and imaged particles by negative-stain EM. For the vast majority of single particles the gold is centrally located, and in particles whose shape matches front views of the reconstruction, the gold label appears localized near the Hsp90 NTDs, most consistent with the adjacent Hop density being the N-terminal TPR1 domain (Figure S4). Further 2D and 3D analysis was impaired due to flexibility of the label and preferential alignment to the dense gold particle. Thus, given this flexibility, we cannot rule out the possibility of a reverse arrangement.
We accurately fit two individual TPR structures into density corresponding to Hop (Figure 5B). These fits identify significant interaction between TPR2A and the MD and CTDs of Hsp90 while our proposed position for TPR1 domain presents accessible surfaces between the Hsp90 monomers for Hsp70 binding. While there is no unambiguous way to fit the third TPR domain into the structure, there is significant unaccounted for density near the CTD, neighboring the proposed TPR2A location that could contain all or part of TPR2B. Alternatively, TPR2B could be unbound and not observed in the reconstruction due to flexibility. Weak density extending into space from the TPR2A region is not apparent even at very low threshold values, suggesting that it is more likely bound but not well resolved. Previous studies have shown a requirement for both TPR2A and TPR2B for sufficient Hsp90 interaction in vivo, supporting a role for both domains in binding to Hsp90 (Flom et al., 2007).
The cryo-EM Hsp90:Hop structure determined here presents a highly plausible interface for client loading by Hsp70. If valid, it should be competent to bind Hsp70. Using in vitro immuno pull-down methods previous studies have shown that Hsp70 binds Hsp90:Hop in the absence of a client protein (Hernandez et al., 2002). This interaction was shown to be dependent on the presence of Hsp40 to drive ATP hydrolysis by Hsp70, likely converting Hsp70 to the ADP substrate bound conformation. We pursued similar methods to in vitro reconstitute a human Hsp90:Hop:Hsp70 complex.
Wt Hsp90, Hop, Hsp70 and ATP were incubated in the presence of substoichiometric amounts of Hsp40 (yeast YDJ1) to enzymatically convert Hsp70 to the ADP state. When His-tagged Hsp90 is bound to TALON magnetic beads, roughly equal amounts of Hop and Hsp70 are pulled-down and are dependent on the presence of ATP and Hsp40 (Figure S5A). When samples are analyzed by SEC-MALS, a single peak elutes early with an average mw of 310 kDa (green curve, Figure 6A), corresponding to the addition of a single Hsp70 molecule to the wt 230 kDa Hsp90:Hop complex. SDS-PAGE confirmed the presence of Hsp90, Hop and Hsp70, while Hsp40 was not present in the complex (data not shown). While association of additional Hop and Hsp70 molecules may be occurring, the calculated mw accurately matches that of an asymmetric complex of two Hsp90 monomers and a single Hop and Hsp70 molecule.
Results from the wt complex indicate that Hop and Hsp70 may interact asymmetrically on Hsp90 during client loading. To determine whether dissociation of Hop is affecting the stoichiometry and preventing binding of a second Hsp70 molecule we next tested Hsp70 interaction with the S-S Hsp90:Hop tetramer complex. The S-S complex was pre-formed and purified by gel filtration and incubated, as before, with Hsp70, ATP and Hsp40. When analyzed by SEC-MALS a single high mw peak is observed (red curve, Figure 6A), calculated to be 370 kDa, precisely corresponding to two Hsp90 monomers, two Hop molecules and a single Hsp70 molecule. The reduced peak size is a consequence of the slightly lower protein concentration (ca. 5 µM) achievable for the pre-formed, purified Hsp90:Hop complex, but is 20-fold above the reported 250nM Kd for Hsp70 (Hernandez et al., 2002). These results indicate that, despite the presence of a second Hop binding site, only a single Hsp70 interacts with the Hsp90:Hop complex supporting an asymmetric Hsp70 binding interaction. Importantly, these results serve as a critical functional test of the S-S complex, establishing that the Hsp90:Hop structure is in a native conformation, competent to bind Hsp70.
Negative-stain EM analysis was performed next in order to determine the overall conformation of the Hsp90:Hop:Hsp70 complex. Glutaraldehyde crosslinking was performed, as before to maximize stability. From single particles and reference-free averages (Figure S5B and 6B) the dimensions of Hsp90:Hop:Hsp70 appear generally similar to those of Hsp90:Hop with an approximate diameter of 120 Å. The particles appear more globular with additional central density and less definition in the V-shaped Hsp90 arrangement (compare Figure 6B to to2A),2A), indicating that the Hsp90 conformation remains relatively unchanged following Hsp70 binding. Density is not observed extending from the globular particles, but is likely more central, arguing that Hsp70 likely binds between the extended Hsp90 monomers.
Our earlier SEC-MALS results indicating that an asymmetric trimer complex predominates for wt Hsp90:Hop along with the observation that a single Hsp70 binds the tetramer complex suggests that an asymmetric arrangement of Hsp90:Hop may be functionally relevant for client-loading. To investigate this further we sought to establish whether the trimer complex was indeed the dominant species under our wt assembly conditions. We hypothesized that intact tetramer and trimer complexes may be separable by SDS-PAGE following low-level glutaraldehyde crosslinking. Hsp90 alone and wt and S-S Hsp90:Hop were purified by SEC-MALS and crosslinked as before and analyzed by SDS-PAGE (Figure 6C, lanes 5–7). Remarkably, crosslinked Hsp90 dimer and Hsp90:Hop trimer and tetramer are clearly distinguishable, migrating as large molecular weight species that approximately correspond to the correct molecular weight standard. For wt Hsp90:Hop the predominant band is clearly a trimer, migrating between Hsp90 alone and the S-S Hsp90:Hop tetramer. This establishes that the trimer complex is the stable form of wt Hsp90:Hop under our conditions and that the binding of the first Hop negatively impacts binding at the second site.
Next we wanted to investigate the conformation of the wt Hsp90:Hop trimer complex by EM. In micrographs and single particle views, wt Hsp90:Hop is clearly compact, homogeneous and remarkably similar to the S-S complex (Figure S5C). To better determine the conformation, single particles were collected and reference-free averages were generated and aligned to reference-free averages of the S-S complex (Figure 6C). These averages appeared nearly identical, establishing that the tetramer and trimer complexes are in a similar conformation. Next we performed a 3D refinement using the cryo-EM structure, filtered to 30 Å, as a starting model with no symmetry imposed (Figure 6D). This was compared to an identical, unsymmetrized refinement using the S-S Hsp90:Hop data. For the wt Hsp90:Hop complex one of the arms corresponding to Hop by previous analysis immediately disappears after two rounds of refinement while Hsp90 remains unchanged. In the control reconstruction of the tetramer complex both Hop molecules remain fixed during the same rounds of refinement. These results demonstrate that the wt Hsp90:Hop trimer complex is in the same arrangement as the tetramer and that the binding of a single Hop is sufficient to promote conformational changes in both Hsp90 monomers.
Recent structural studies have determined that Hsp90 undergoes significant conformational rearrangements during ATP binding and hydrolysis. In eukaryotes, cochaperones interact with Hsp90 during the chaperone cycle, regulating client protein binding and hydrolysis-dependent folding and activation. However, the mechanism by which cochaperones facilitate client loading and modulate the Hsp90 cycle is unknown.
Our cryo-EM structure reveals that Hop does not simply act as a passive linker to co-localize Hsp70 in the vicinity of Hsp90, but instead promotes two distinct conformational changes in Hsp90 compared to the apo solution state. First, a rotation about the MD-CTD junction results in a substantially smaller open angle and second, the NTDs rotate nearly 90° such that the NTD-MD angle matches the closed ATP conformation (Figure 4 and Movie S1). The resultant conformation shares structural features of both the apo state and the closed ATP state: putative client interaction surfaces along the interdimer cleft are accessible, while the NTDs are on-path to dimerize and hydrolyze ATP, but held in an hydrolysis-incompetent state.
The apo Hsp90 crystal structure reveals an extensive array of exposed hydrophobic surfaces in all three domains. When the same hydrophobic surfaces are mapped onto the dominant solution conformation of Hsp90 (Krukenberg et al., 2008) they appear significantly separated (Figure 7A). Remarkably, the conformational state stabilized by Hop results in a convergence of the hydrophobic surfaces along the interdimer cleft, effectively pre-organizing the surfaces for efficient client binding (right panel, Figure 7A). In particular, the rotation of the NTD connects the hydrophobic surfaces between helix 1 (the lid domain) in the NTD and residues in the MD implicated in v-SRC activation (Meyer et al., 2003). The Hsp90 conformation observed here is remarkably similar to the conformation observed for the E. coli Hsp90 incubated with a model non-native substrate protein (Street et al., 2011).
Based on the Hsp90 conformation identified in the Hsp90:Hop structure, we propose a model for how Hop functions to stabilized a hydrolysis inactive, client-loading Hsp90 conformation (Figure S6). While it has been previously shown that Hop inhibits ATP hydrolysis (Richter et al., 2003), how this was accomplished has been unclear. In the cryo-EM structure the primary binding interaction between Hsp90 and Hop is located at the MD-CTD junction. This interaction likely dictates the open angle and prevents the complete rotation of the MD-CTD interface required for NTD dimerization in the closed, hydrolysis-active conformation. Secondly, the position of the TPR1 domain sterically blocks NTD dimerization by being situated between the Hsp90 monomers and interacting with the adjacent NTD-MD. Furthermore, this cross-monomer interaction and the overall similar conformation of the asymmetric trimer complex explain how the binding of a single Hop is sufficient to block NTD dimerization and ATP hydrolysis (Li et al., 2011).
Our SEC-MALS and SDS-PAGE analysis of native and crosslinked wt Hsp90:Hop (Figure 1B and and6C)6C) indicate that the asymmetric complex with a singly-bound Hop is more stable than the tetramer form, suggesting the binding of one Hop results in a reduced affinity for the second Hop. This is likely indirect and the result of the significantly different Hsp90 conformational state compared to the extended solution form. However, the second Hop binding site remains accessible given that the S-S complex readily forms a tetramer to nearly 100% occupancy when incubated under the same conditions. Thus, the shift in Kd for the second site is likely subtle. While it is unknown what the predominant in vivo stoichiometry of the Hsp90:Hop complex is, these results combined with our analysis of the Hsp90:Hop:Hsp70 complex indicate that an asymmetric form may be favored.
Finally, nucleotide binding experiments also revealed that while ATP affinity is unchanged in the Hsp90:Hop complex, both the association and dissociation kinetics are significantly increased (Richter et al., 2003). Considering that in the apo crystal structure the nucleotide-binding pocket is inaccessible to ATP, requiring a conformational change, the NTD rotation induced by Hop identified here likely increases accessibility of the nucleotide-binding pocket. While ATP should readily bind in this state, the presence of Hop would then prevent closure of Hsp90:ATP obviating nonproductive and premature hydrolysis (Figure S6). Consequently, the Hsp90:Hop complex is kinetically poised in the catalytic cycle to receive client proteins while bound to ATP and immediately advance to the closed state following release of Hop and Hsp70.
We propose a model for Hsp90 client protein binding where Hop is initially bound to Hsp90 stabilizing a client loading conformation that favors Hsp70 interaction (Figure 7B). While Hop can bind Hsp70:ADP alone via the TPR1 domain, the affinity increases nearly 5-fold, to 250 nM in the presence of Hsp90 (Hernandez et al., 2002), suggesting that Hsp90 binding either makes the TPR1 domain more accessible or, more likely presents additional contacts for Hsp70 that increase the affinity. Our fit of two TPR domains into the Hsp90:Hop structure and gold-labeling data suggests that TPR1 is accessible between the Hsp90 monomers while our 2D EM analysis of the Hsp90:Hop:Hsp70 complex indicates that Hsp70 binds within the Hsp90 monomers. Together these results support a model where Hsp70 binds TPR1 via its C-terminal EEVD peptide but makes additional contacts with Hsp90, likely with the NTD and MD, in order to deliver a client protein to the Hop-stabilized, V-shaped interdimer cleft. This likely occurs asymmetrically given that even when two Hop molecules are present only a single Hsp70 stably binds. This would be expected if Hsp70 projects into the cleft, sterically hindering the binding of a second Hsp70. Importantly, asymmetric cochaperone interactions are emerging as a critical mechanism for Hsp90 function and have been identified for cdc37/p50 (Vaughan et al., 2006), Aha1 (Retzlaff et al.) and PPIase cochaperones (Li et al., 2011). Such an asymmetric mechanism would be advantageous for regulatory functions in vivo, potentially allowing other TPR-containing cochaperones to simultaneously bind the second MEEVD site and promote cochaperone release and reorganization of the Hsp90:client complex.
Human Hsp90α, human Hop and yeast YDJ1 (Hsp40) DNA (David Toft lab) were cloned into the pET151 vector with a 6x His-tag and TEV cleavage site (Invitrogen). Protein was purified by Ni-NTA affinity, ion exchange (Q Sepharose, GE Healthcare), and SEC (Sephacryl S-200 column, GE Healthcare). TEV cleavage of the His-tag was performed prior to the SEC step. Cysteine mutagenesis was performed using QuickChange (Stratagene). Baculovirus-infected Sf9 cell pellets expressing human Hsp70 were produced at the Baculovirus/Monoclonal Antibody Facility at Baylor College of Medicine and protein was purified as described (Schumacher et al., 1996).
Wt and S-S Hsp90:Hop complexes were formed by incubating protein (10 µM and 20 µM Hsp90 and Hop, respectively) in binding buffer (20 mM, Tris pH 7.5, 50 mM KCl and 0.05% n-Octyl-β-D-Glucopyranoside (β-OG, Calbiochem)) at 37° for 60 minutes. The Hsp90:Hop:Hsp70 complex was formed by incubating Hsp90 (10 µM), Hop (20 µM), Hsp70 (20 µM) and Hsp40 (200 nM) in the binding buffer with 200 µM ATP and 5 mM MgCl2. Purification and mw determination were achieved by SEC (Shodex KW-804 column) with an Ettan LC (GE Healthcare) and an in-line DAWN HELEOS MALS and Optilab rEX differential refractive index detectors (Wyatt Technology Corporation). Data were analyzed by the ASTRA V software package (Wyatt Technology Corporation). SEC was performed in 20 mM HEPES, pH 7.5, 50 mM KCl and 0.01% β-OG. 200 µM ADP and 5 mM MgCl2 were included for the Hsp90:Hop:Hsp70 complex. Chemical crosslinking was performed with 0.005% glutaraldehyde as previously described (Southworth and Agard, 2008).
Hsp90 protein complexes were negatively-stained with uranyl formate (pH 5.5–6.0) on thin carbon-layered (40–50 Å thick) 400 mesh copper grids (Pelco) as described (Ohi et al., 2004). Samples were imaged using a G2 Spirit TEM (FEI) operated at 120 keV. Micrograph images, with a 2.2 Å pixel size, were recorded using a 4k × 4k CCD camera (Gatan). Gold-labeling was performed by incubating N-terminal His-tagged Hop with 1.8 nm Ni-NTA-Nanogold (Nanoprobes) (two-fold molar excess) during Hsp90:Hop assembly, followed by SEC purification prior to negative-staining. Samples for cryo-EM (1 µM) were prepared on C-FLAT holey carbon grids (Protochips) in SEC buffer with a Vitrobot (FEI Co.). Initial cryo-EM data sets not used in the final reconstruction were collected at the National Resource for Automated Molecular Microscopy, supported by the NIH though the National Center for Research Resources' P41 program (RR17573). Cryo-EM data in the reconstruction were collected on a Tecnai F20 operated at 200 keV with an 8k × 8k CCD TemCam-F816 CCD camera (TVIPS) with a 1.19 Å pixel size and a dose of ca. 30 e/Å2.
The negative-stain 3D reconstruction of Hsp90:Hop was determined using EMAN (Ludtke et al., 1999). Reference-free class averages were created with startnrclasses and a subset was used to generate an initial model by the cross common lines method (Van Heel, 1987) with two-fold symmetry imposed (Figure 2A). Defocus values ranged between 1.2 and 1.5 µm defocus, calculated by ctfit. For the final model, 10 rounds of refinement were applied with two-fold symmetry and the resolution was measured to be 25 Å by the even-odd test in EMAN where the Fourier shell correlation method is used to compare two models generated from a random split of the data. A total of 2,428 (2,133 for the final model) particles were picked from 53 micrographs.
The cryo-EM Hsp90:Hop reconstruction was determined using SPIDER (Frank et al., 1996). 45,722 particles (27,509 in the final model) were picked from 600 images with a 3.5–6.0 µm defocus range. Defocus values were determined using CTFFIND (Mindell and Grigorieff, 2003) and corrected using a Wiener filter. The negative-stain reconstruction, filtered to 40 Å resolution, was used as an initial model. For the final model, 15 rounds of refinement were applied with two-fold symmetry and 6° angular increments. The resolution was measured to be 15 Å by the Fourier shell correlation method using SPIDER (RF 3).
Images were produced using Chimera (Pettersen et al., 2004). Docking the Hsp90 crystal structures into the EM map was performed using the ‘fit model in map’ command, which locally optimizes a fit in the EM map by maximizing the sum of the density values of the selected atoms. For the final fit shown (Figure 4E), the alternate Hsp90 conformation was produced in PyMOL by aligning residues in the MD (346–366) of the HtpG structure (Shiau et al., 2006) to corresponding residues (390–410) in the yeast Hsp82 closed structure (Ali et al., 2006), then truncating and separately aligning the NTD (residues 1–230) to the NTD of the Hsp82 structure. This structure fit with more than 75% of atoms within the EM density contoured to the molecular weight of Hsp90:Hop. The fits of the Hop TPR domains were performed manually.
Funding was provided by the Howard Hughes Medical Institute and an American Cancer Society postdoctoral fellowship. Special thanks to B. Cararragher, C. Potter and J. Quispe at the National Resource for Automated Molecular Microscopy for their generous assistance with data collection. Thanks to T. Street, U. Boettcher, Y. Chang, J. Kollman, , T. Street, and L. Lavery for helpful discussions and comments.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.