2. Architecture of Hsp90

Hsp90 is highly conserved from bacteria to eukaryotes with ~50% sequence similarity between Escherichia coli and humans. There are four human Hsp90 isoforms: two cytosolic (Hsp90α, Hsp90β); one mitochondrial (Trap1); and one specific to the endoplasmic reticulum (Grp94). These isoforms all operate as obligate dimers consisting of three domains per monomer (Fig. 1). The C-terminal domain (CTD, brown) is the site of dimerization, the middle domain (MD, green) has been implicated in client binding, and ATP binds to the N-terminal domain (NTD, blue) (Pearl & Prodromou, Reference Pearl and Prodromou2006). The nucleotide-binding pocket in the NTD is highly conserved across species, and has been identified as related to the gyrase, Hsp90, histidine kinase, MutL (GHKL) superfamily of ATPases (Bergerat et al. Reference Bergerat, de Massy, Gadelle, Varoutas, Nicolas and Forterre1997; Dunbrack et al. Reference Dunbrack, Gerloff, Bower, Chen, Lichtarge and Cohen1997). As with other GHKL superfamily proteins, binding and hydrolysis of nucleotide leads to local and global conformational changes that result in closure by dimerization of the NTDs.

Fig. 1.

Hsp90 exists as a homodimer with three domains per monomer. The crystal structure of the bacterial homolog, HtpG (PDB code 2IOQ), is shown as both a cartoon representation and a surface representation. The domains of one monomer are colored blue (NTD), green (MD) and brown (CTD). The other monomer is colored gray. The NTD binds nucleotide and contains a variable charged loop (red) and the conserved lid region (purple), which is adjacent to the nucleotide-binding pocket. The MD contains the conserved catalytic loop (cyan) that is essential for ATP hydrolysis. The CTD provides the constitutive dimer interface.

Within the domains of Hsp90 there are conserved elements that are functionally important in the nucleotide cycle. These elements include the nucleotide lid (Fig. 1, purple region on NTD) and the catalytic loop (cyan region on MD), which will be discussed in detail in the ‘ATP-dependent conformational changes’ section. The three individual domains of Hsp90 show high levels of structural conservation but the domain boundaries are only weakly coupled, which allows for large structural changes to occur by domain–domain rearrangements.

Despite the conserved domain architecture and key structural elements, there are important differences in the overall domain organization of the different Hsp90s (Fig. 2). A repetitively charged region (Fig. 1, red) that varies in length is present at the C-terminal end of the NTD in yeast and cytosolic higher eukaryotic Hsp90s as well as in Grp94. This charged region is also present in the bacterial protein HtpG, but it is significantly shorter. Although the charged region is dispensable in yeast, it has been implicated in client binding (Scheibel et al. Reference Scheibel, Siegmund, Jaenicke, Ganz, Lilie and Buchner1999). In addition to the charged linker, the yeast and cytosolic higher eukaryotic Hsp90s also have a C-terminal motif (MEEVD) that facilitates co-chaperone binding. The unique regulatory role of Hsp90 in eukaryotes is likely due to the large number of co-chaperones that recruit specific classes of clients (e.g. kinases via p50/Cdc37), modulate the Hsp90 ATPase activity (e.g. activation by Aha1), aid in client folding (e.g. proline isomerization by cyclophilins) and maturation (p23). Many of these co-chaperones contain tetratricopeptide repeat (TPR) domains that bind Hsp90 via the C-terminal MEEVD motif. Later we will discuss the role of co-chaperones on Hsp90 conformational dynamics.

Fig. 2.

Domain architecture for different Hsp90 homologs. The domains are colored with the NTD in blue, the MD in green and the CTD in brown.

3. Apo state flexibility of Hsp90

The central importance of Hsp90 to both basic cellular processes and disease states makes Hsp90 an important target for structural studies. After the first crystal structures of the NTD were determined in 1997 by the Pearl and Pavletich labs (Prodromou et al. Reference Prodromou, Roe, Piper and Pearl1997b; Stebbins et al. Reference Stebbins, Russo, Schneider, Rosen, Hartl and Pavletich1997), it would take nearly another decade before full-length Hsp90 structures would be determined: in 2006 by the Pearl lab (AMPPNP conditions) (Ali et al. Reference Ali, Roe, Vaughan, Meyer, Panaretou, Piper, Prodromou and Pearl2006) and in 2007 by the Agard lab (apo and ADP conditions) (Shiau et al. Reference Shiau, Harris, Southworth and Agard2006). The significant difficulty in determining full-length Hsp90 crystal structures is likely due to its remarkable structural flexibility. For example, under apo conditions the Hsp90 from E. coli, HtpG, crystallized in a ‘V’-shaped conformation (Fig. 3a) (Shiau et al. Reference Shiau, Harris, Southworth and Agard2006) while in solution SAXS and electron microscopy measurements demonstrated that a highly open conformation and a more closed conformation are populated in a pH-dependent manner (Fig. 3b) (Krukenberg et al. Reference Krukenberg, Southworth, Street and Agard2009b). The more closed state is similar to a crystal structure of the Hsp90 homolog specific to the endoplasmic reticulum, Grp94 (Dollins et al. Reference Dollins, Warren, Immormino and Gewirth2007). The apo conformations largely differ by rigid-body rotation at the interface between the middle and CTDs, creating a variable-sized cleft between the monomer arms. These conformational changes are dramatic, with the NTDs undergoing 50 Å translations and 50° rotations upon the open/Grp94 transition.

Fig. 3.

Full-length structures of Hsp90. The NTD is shown in blue, the MD in green and the CTD in brown. In all structures, exposed hydrophobic surfaces are highlighted in purple. (a) Surface representations of the apo and ADP HtpG structure (Shiau et al. Reference Shiau, Harris, Southworth and Agard2006) and the yeast Hsp90 AMPPNP-bound structure (Ali et al. Reference Ali, Roe, Vaughan, Meyer, Panaretou, Piper, Prodromou and Pearl2006). (b) Surface representations of the apo HtpG solution state (Krukenberg et al. Reference Krukenberg, Forster, Rice, Sali and Agard2008) and the Grp94 crystal structure (Dollins et al. Reference Dollins, Warren, Immormino and Gewirth2007).

Both SAXS and electron microscopy (EM) measurements have shown that Hsp90 conformations coexist in a delicate equilibrium. That is, for HtpG the open and Grp94 conformation have approximately the same free energy. The weak energetics associated with the conformational equilibrium has been illustrated by the use of a class of small molecules called osmolytes, small molecules that maintain protein homeostasis under harsh environments and stress conditions by their influence on protein folding and stability. SAXS measurements of Hsp90 in the presence of osmolytes demonstrated that the chaperone conformational equilibrium can be effectively dialed between different conformational states over a similar energetic range under which protein folding occurs (Street et al. Reference Street, Krukenberg, Rosgen, Bolen and Agard2010). This suggests that the Hsp90 conformation could be energetically linked to folding steps.

The extreme apo state conformational heterogeneity is universal to all homologs that have been examined (HtpG, Hsc82, Hsp90α, Grp94 and TRAP, Krukenberg et al. Reference Krukenberg, Bottcher, Southworth and Agard2009a and unpublished observations). This conservation implies that the flexibility of Hsp90 is functionally important, and that differences in conformational flexibility among Hsp90 homologs may relate to differences in how they interact with clients. If Hsp90 conformational changes are functionally linked to client interactions, two predictions can be made: (1) clients will play an active role in determining the chaperone conformation; and (2) different conformational states of Hsp90 will have different effects on client folding/misfolding.

The first prediction has been observed for three clients: the ligand-binding domain of GR; the Cdk4 kinase; and a model unfolded client Δ131Δ, all of which will be discussed in ‘Client protein binding affects Hsp90 conformation’. In summary, the ligand-binding domain of GR stimulates the human Hsp90 ATPase, which indicates that the Hsp90 conformation is being driven towards the closed catalytically active state. For the co-chaperone-recruited Cdk4, an electron microscopy reconstruction shows that binding occurs asymmetrically on Hsp90 and results in incomplete closure. Finally, the model unfolded protein Δ131Δ drives Hsp90 to partially close under apo conditions, and accelerates full closure and ATP hydrolysis under nucleotide conditions. As discussed later, Hsp90's differing response to these very different proteins highlights the remarkable capacity of this chaperone to accommodate structurally diverse clients.

The second prediction, that different Hsp90 conformations will have different effects on client folding/misfolding has also been observed for a model client. Specifically, the thermal aggregation of citrate synthase can be strongly suppressed by HtpG in the Grp94 conformation but only weakly suppressed in the open state (Krukenberg et al. Reference Krukenberg, Southworth, Street and Agard2009b). This observation was enabled by identifying a mutation pair (H446E/K) that selectively populates the open/Grp94 conformations independent of pH.

These observations together support the idea that the conformational equilibrium of Hsp90 plays a central role in client interactions, and is possibly tuned to the client requirements of each organism. The large structure of Hsp90 in combination with its conformational diversity suggests that the chaperone can adopt a combinatorial set of binding surfaces and chaperone geometries in order to interact with structurally diverse clients and large complexes that include co-chaperones. As discussed next, the effect of nucleotide binding and hydrolysis adds a layer of complexity, both structural and kinetic, to the Hsp90 conformational equilibrium.

4. ATP-dependent conformational changes

The transition to the closed ATP state involves three local structural changes that together coordinate large-scale closure of the chaperone. ATP binding (1) restructures an N-terminal helical lid region, (2) dramatically changes the NTD/MD orientation and (3) is associated with the release of a β-strand, which is swapped across monomers stabilizing N-terminal dimerization. These changes are discussed below.

The lid region (Fig. 1, purple), a helix–loop–helix motif adjacent to the nucleotide-binding pocket makes extensive NTD dimer contacts (Prodromou et al. Reference Prodromou, Roe, O'Brien, Ladbury, Piper and Pearl1997a; Stebbins et al. Reference Stebbins, Russo, Schneider, Rosen, Hartl and Pavletich1997). Crystal structures of isolated NTDs in the apo, AMPPNP and ADP bound states show that nucleotide binding induces large structural changes in the lid (Fig. 4). Biochemical studies highlight that the lid has a complex role in regulating NTD dimerization and ATP hydrolysis. Specifically, removal of the lid abolishes all ATPase activity, but heterodimers lacking the lid on only one NTD have substantially accelerated closure and ATPase hydrolysis rates (Hessling et al. Reference Hessling, Richter and Buchner2009; Richter et al. Reference Richter, Moser, Hagn, Friedrich, Hainzl, Heller, Schlee, Kessler, Reinstein and Buchner2006). The increased dimerization promoted by this heterodimer is coupled to the first 24 amino acids of the N-terminus that form a swapped β-strand that stabilizes NTD dimerization. Deletion of the N-terminal 24 residues confers flexibility to the lid suggesting that nucleotide binding, lid restructuring and β-strand swapping are all coupled (Richter et al. Reference Richter, Moser, Hagn, Friedrich, Hainzl, Heller, Schlee, Kessler, Reinstein and Buchner2006).

Fig. 4.

Variation in the NTD of Hsp90 occurs mostly in the lid. All structures are of the NTD of Hsp90 with the lid highlighted in red or blue. (a) Two conformations of the isolated human Hsp90 NTD (1YER and 1YES). In one conformation (blue) loop 2 (L2) extends into the nucleotide-binding pocket. Large motions in the lid can also be seen in (b) the NTD from the full-length yeast structure (2CG9) and (c) the isolated NTD of Grp94 bound to ADP (1TBW). (d–f) Three additional lid conformations are observed in crystals of HtpG (2IOR, 2IOQ and 2IOP).

A second major structural change upon ATP binding is a repositioning of the NTD/MD interface, which orients a highly conserved arginine (R380, in yeast numbering) that resides on the MD catalytic loop (Fig. 1, cyan) to interact with the ATP γ-phosphate. In the AMPPNP-bound structure of Hsp90, R380 is the only residue contacting the γ-phosphate, highlighting its unique role in ATP binding and hydrolysis. The catalytic loop and arginine are both highly conserved and appear essential for hydrolysis by Hsp90 alone and for the acceleration of hydrolysis by the Aha1 co-chaperone (Meyer et al. Reference Meyer, Prodromou, Hu, Vaughan, Roe, Panaretou, Piper and Pearl2003). The NTD/MD orientation has been found to be variable in different crystal structures, suggesting that rotational freedom allows for the R380/γ-phosphate alignment to be regulated by co-chaperone and client binding. The R380 residue on the catalytic loop forms a hydrophobic network with a loop (T22, V23 and Y24, Hsp82 numbering) on the opposing monomer in the closed state, indicating cross-monomer coupling of closure and hydrolysis (Fig. 5). Indeed, cross-monomer regulation of hydrolysis was confirmed by examining the ATPase activity of a series of yeast Hsp90 heterodimers with mutations in this network (Cunningham et al. Reference Cunningham, Krukenberg and Agard2008). The catalytic role of the MD and the requirement for cross-monomer interactions in hydrolysis is further supported by truncation experiments where Hsp90 deletion constructs lacking the CTD retain basal ATPase activity if the two monomers are connected by a linker, but cross-linked NTDs lacking the MD exhibit 100-fold lower ATPase activity than WT (Wegele et al. Reference Wegele, Muschler, Bunck, Reinstein and Buchner2003).

Fig. 5.

A hydrophobic interaction network is formed between the MD catalytic loop and the NTD of opposing monomers. The NTD is shown in blue, the MD in green and the CTD in brown. ATP and the residues involved in the interaction network are shown as spheres with residues from the NTD shown in dark purple and residues from the MD shown in lavender. In the expanded view, the second monomer is shown in beige.

ATP binding by Hsp90 results in local conformational changes that drive closure leading to a state where the hydrophobic surfaces are less well exposed than in the apo state but still accessible on the outer edges of the dimer cleft (Fig. 3a, purple). The closed conformation of Hsp90 was determined crystallographically for the yeast homolog in complex with the co-chaperone p23, which binds the NTD. Subsequent SAXS and EM reconstructions demonstrated that the closed state is populated by the bacterial and human Hsp90 homologs as well, although the extent to which these homologs populate the closed state is quite variable (yeast>bacterial>human). This ranking matches the Hsp90 homolog ranking of ATP hydrolysis rates (Table 1), suggesting that closure is the rate-limiting step in hydrolysis. Indeed, yeast Hsp90 variants with decreased hydrolysis rates show decreased closure (Cunningham et al. Reference Cunningham, Krukenberg and Agard2008). However, the rate-limiting structural steps in closure are currently unknown.

The connection between the slow hydrolysis rate of Hsp90 and closure to the ATP state is also supported by FRET measurements showing slow closure kinetics similar to measured ATPase rates (Hessling et al. Reference Hessling, Richter and Buchner2009). The FRET-measured closure kinetics are multiphasic, indicating intermediates. In support of this, hydrogen–deuterium exchange mass spectrometry (HX-MS) experiments have demonstrated that an intermediate conformation, with increased catalytic loop protection, exists before the closed state in the hydrolysis reaction (Graf et al. Reference Graf, Stankiewicz, Kramer and Mayer2009). Mutation experiments also indicate that the structure of the ATP hydrolysis transition state must be distinct from the yeast AMPPMP:p23 crystal structure. Mutation of T22 to Y significantly accelerates ATP hydrolysis, but would be sterically precluded by the crystal structure (Cunningham et al. Reference Cunningham, Krukenberg and Agard2008). Closure kinetics have been measured upon addition of non-hydrolysable or slowly hydrolysable ATP by both SAXS (T. O. Street & D. A. Agard, unpublished results) and FRET (Hessling et al. Reference Hessling, Richter and Buchner2009), whereas in the presence of ATP an open, low FRET, conformation is maintained. This observation indicates that hydrolysis and subsequent reopening occur quickly compared to the slow, ~10 min scale, closure reaction. The large range of time scales associated with different steps in the Hsp90 reaction cycle demonstrates how kinetic regulation by co-chaperones can play a pivotal role in affecting client interactions.

In the presence of ADP, a compact structure was observed for HtpG (Fig. 3) (Shiau et al. Reference Shiau, Harris, Southworth and Agard2006). In this state the exposed hydrophobic surfaces contributed by each domain in the apo and ATP states are buried in the core of the protein suggesting a means for client protein release. Negative-stain EM and cross-linking studies demonstrated the presence of the compact ADP state in HtpG, yeast Hsp90 and human Hsp90 indicating that this state is conserved (Southworth & Agard, Reference Southworth and Agard2008). The ADP-state conformation is also consistent with cross-linking studies of Grp94 (Chu et al. Reference Chu, Maynard, Chiosis, Nicchitta and Burlingame2006), in which the observed cross-links are only compatible with the compact conformation. However, for human and yeast Hsp90 the ADP state is only observed in the presence of cross-linker, highlighting the transient nature of the conformation. This is also evident from SAXS measurements that show a minimal change in the bulk population under ADP conditions (Krukenberg et al. Reference Krukenberg, Forster, Rice, Sali and Agard2008). Together, the structural information on Hsp90's conformational flexibility and nucleotide dependence map out the basic steps of a nucleotide hydrolysis cycle (Fig. 6).

Fig. 6.

The structures of Hsp90 suggest a nucleotide-dependent conformational cycle. In this cycle, the binding of ATP shifts the conformation from an open structure (1) to an NTD-dimerized closed structure (3). The Grp94 crystal structure (2) may represent an intermediate along this path or non-catalytic conformation. The hydrolysis of ADP causes a further compaction of the protein (4) and the release of nucleotide restarts the conformational cycle.

Molecular dynamics simulations of full-length Hsp90 found an increase in communication efficiency between residues in the NTD and CTD upon the addition of nucleotide (Morra et al. Reference Morra, Verkhivker and Colombo2009). Interestingly, while both ATP- and ADP-bound Hsp90 show residues communicating over a distance of 80 Å, these studies suggest each nucleotide activates a different set of communicating residues between the NTD and CTD as shown in Fig. 7. These different pathways may determine the different conformations of Hsp90 in the presence of ATP or ADP; mutational studies could be used to elucidate each network's role in establishing Hsp90's conformation.

Fig. 7.

Molecular dynamics simulations have shown different long-range communication networks between the NTD and CTD in the presence of ATP or ADP. Residues that participate in long-range communication are shown as spheres on the yeast AMPPNP-bound crystal structure. The NTD is shown in blue, the MD in beige and the CTD in green with the second monomer shown in gray. Long-range interactions observed only in the presence of ATP are shown in purple. Interactions seen only in the presence of ADP are shown in orange. Interactions observed with both ATP and ADP are shown in dark gray.

5. The role of co-chaperones in regulating the conformational cycle of Hsp90

In eukaryotes, Hsp90 is often found in large complexes with co-chaperones, some of which are listed in Table 2. Co-chaperones provide a method for the direct recruitment of different client proteins to Hsp90, a means for regulating the Hsp90 conformational and ATP hydrolysis cycle, and subsequently a level of control over client folding. A genomic comparison of Hsp90 co-chaperones shows that many organisms have a unique repertoire of Hsp90 co-chaperones, suggesting that specific co-chaperone combinations can tailor Hsp90's activity to best suit the client requirements for each particular organism (Johnson & Brown, Reference Johnson and Brown2009). In this section we discuss well-studied Hsp90 co-chaperones, their influence on Hsp90 conformational dynamics and their role in kinetic regulation.

Table 2.

Partial list of Hsp90 associated co-chaperones

For a list of additional co-chaperones see http://www.picard.ch/downloads/Hsp90interactors.pdf

* References can be found in the text.

A minimal set of necessary co-chaperones has been established for the progesterone receptor (PR), the GR and the kinase Chk1 providing information about the role of different co-chaperones and about the specific requirements among client proteins. In vitro, GR requires only Hsp70 and Hop in addition to Hsp90 for the reconstitution of steroid-binding activity (Arlander et al. Reference Arlander, Felts, Wagner, Stensgard, Toft and Karnitz2006; Dittmar & Pratt, Reference Dittmar and Pratt1997; Kosano et al. Reference Kosano, Stensgard, Charlesworth, McMahon and Toft1998). For the in vitro reconstitution of PR, Hsp70, Hsp40, Hop and p23 are required in addition to Hsp90 (Kosano et al. Reference Kosano, Stensgard, Charlesworth, McMahon and Toft1998). The kinase Chk1 also requires Hsp90, Hsp70, Hsp40 and Hop for maximal activation, but unlike PR, the kinase-specific co-chaperone p50 (Cdc37) is also required while p23 is not (Arlander et al. Reference Arlander, Felts, Wagner, Stensgard, Toft and Karnitz2006). These differences in co-chaperone requirements may provide a mechanism for tuning the Hsp90 response.

The reconstituted systems reveal that specific complexes are formed at different stages throughout the chaperone cycle (Fig. 8). Both Chk1 and PR bind Hsp40 to begin the cycle. Hsp40 delivers the client to Hsp70 and the initial Hsp90 complex forms by client transfer from Hsp70 facilitated by Hop which can simultaneously bind both Hsp70 and Hsp90 to stabilize a ternary complex (Felts et al. Reference Felts, Karnitz and Toft2007). In the case of Chk1, this initial complex also includes p50. Hsp70 dissociates to form late-stage complexes. For Chk1, this includes the kinase, Hsp90 and p50. The steroid hormone receptors are in complex with Hsp90 and p23 at this stage. Activated client proteins are then released from the late-stage Hsp90 complex (Felts et al. Reference Felts, Karnitz and Toft2007; Hernandez et al. Reference Hernandez, Chadli and Toft2002; Kosano et al. Reference Kosano, Stensgard, Charlesworth, McMahon and Toft1998; Smith, Reference Smith1993).

Fig. 8.

Proposed chaperone cycles for the activation of PR and the Chk1 kinase. PR or Chk1 is first bound by Hsp40 followed by recruitment to Hsp70. Next, a complex with Hsp90, Hop and Hsp70 forms. This intermediate complex also includes the kinase-specific p50 in the case of Chk1. The mature complex is then formed when Hsp70 and Hop dissociate. The mature complex for PR requires the binding of the additional factor, p23.

The diagrams in Fig. 7 detail the minimal systems required for activation, but numerous other co-chaperones have been identified and are believed to play important roles in the regulation of the chaperone cycle in vivo (Table 2). For example, the immunophilin FKBP52 is found in complex with Hsp90, GR/PR, and p23 in vivo; FKBP52 also increases the hormone-binding ability of GR and is essential for the correct activation of PR in vivo (Chadli et al. Reference Chadli, Bruinsma, Stensgard and Toft2008; Riggs et al. Reference Riggs, Roberts, Chirillo, Cheung-Flynn, Prapapanich, Ratajczak, Gaber, Picard and Smith2003). To understand the operation of these large Hsp90/client/co-chaperone complexes, the influence of each co-chaperone on Hsp90 needs to be known.

5.1 p23 arrests the ATPase cycle of Hsp90

Since it was first identified in complex with Hsp90 and unactivated PR (Smith et al. Reference Smith, Faber and Toft1990), p23 has been shown to have enhanced/diminished binding to Hsp90 under ATP/ADP conditions (Grenert et al. Reference Grenert, Johnson and Toft1999; Johnson & Toft, Reference Johnson and Toft1995; Sullivan et al. Reference Sullivan, Stensgard, Caucutt, Bartha, McMahon, Alnemri, Litwack and Toft1997). Binding of p23 occurs primarily at the NTD with minor MD contacts (Fig. 9) (Ali et al. Reference Ali, Roe, Vaughan, Meyer, Panaretou, Piper, Prodromou and Pearl2006; Martinez-Yamout et al. Reference Martinez-Yamout, Venkitakrishnan, Preece, Kroon, Wright and Dyson2006). The complex promotes and stabilizes a catalytically incompetent closed conformation (McLaughlin et al. Reference McLaughlin, Smith and Jackson2002, Reference McLaughlin, Sobott, Yao, Zhang, Nielsen, Grossmann, Laue, Robinson and Jackson2006; Richter et al. Reference Richter, Walter and Buchner2004; Siligardi et al. Reference Siligardi, Hu, Panaretou, Piper, Pearl and Prodromou2004). The affinity of Hsp90 for p23 can be tuned with Hsp90 mutants that either increase or decrease NTD dimerization (Richter et al. Reference Richter, Walter and Buchner2004; Siligardi et al. Reference Siligardi, Hu, Panaretou, Piper, Pearl and Prodromou2004). As expected, truncation of the NTD domain so that dimerization no longer occurs abolishes the binding of p23.

Fig. 9.

Crystal structure of the p23:Hsp90 complex. (a) Residues 1–134 of yeast p23 were crystallized in complex with the full-length yeast Hsp90 (PDB code, 2CG9). The two monomers of Hsp90 are shown in brown and green whereas p23 is shown in red. (b) An expanded view of the ATP-binding pocket shows that the active site lid (shown in light blue) folds over the bound nucleotide when p23 is present. This conformation blocks access in and out of the ATP-binding pocket.

Promoting the closed state of Hsp90 and inhibiting the ATPase activity has the functional consequence of stabilizing client proteins in a complex with Hsp90 and p23. This has been observed for both GR and PR (Dittmar & Pratt, Reference Dittmar and Pratt1997; Kosano et al. Reference Kosano, Stensgard, Charlesworth, McMahon and Toft1998). p23 may serve as a molecular timer for the activation of client proteins; once clients are activated p23 is released, hydrolysis occurs and client proteins can dissociate. This timing regulation may occur by two mechanisms: (1) slowed hydrolysis of ATP in the presence of p23 or (2) dissociation of p23 driven by the binding of other co-chaperones. There is evidence for both mechanisms. In yeast, Δ8-Hsp90 (increased ATPase activity) hydrolyzes ATP even at saturating levels of bound p23 (Richter et al. Reference Richter, Walter and Buchner2004). Since the wild-type Hsp90 ATPase is strongly inhibited by p23, this indicates a close balance between the natural ATPase activity and p23's inhibition potency. Indeed, the half-life (T _1/2) of the human p23:Hsp90 complex is 40 s at 37°C, whereas the T _1/2 of hydrolysis is 7 min (McLaughlin et al. Reference McLaughlin, Sobott, Yao, Zhang, Nielsen, Grossmann, Laue, Robinson and Jackson2006), demonstrating that the kinetics are also comparable. Complete inhibition of human Hsp90's ATPase activity was observed in the presence of p23 but the binding of other co-chaperones can displace p23 and overcome inhibition (McLaughlin et al. Reference McLaughlin, Sobott, Yao, Zhang, Nielsen, Grossmann, Laue, Robinson and Jackson2006). The kinetic steps involved in closure, ATP hydrolysis and co-chaperone binding are clearly coupled and likely provide many ways to connect client activation to other cellular processes.

5.3 p50, a kinase specific co-chaperone, also traps Hsp90 in an open conformation

The kinase-specific co-chaperone p50 also inhibits the ATPase activity and affects the conformational dynamics of Hsp90, but its mechanism differs from both Hop and p23 (Grammatikakis et al. Reference Grammatikakis, Lin, Grammatikakis, Tsichlis and Cochran1999; Lee et al. Reference Lee, Rao, Fliss, Yang, Garrett and Caplan2002; Stepanova et al. Reference Stepanova, Leng, Parker and Harper1996). Domain mapping indicates that p50's NTD is critical for binding kinases whereas the middle and CTDs interact with Hsp90 (Grammatikakis et al. Reference Grammatikakis, Lin, Grammatikakis, Tsichlis and Cochran1999; Roe et al. Reference Roe, Ali, Meyer, Vaughan, Panaretou, Piper, Prodromou and Pearl2004; Shao et al. Reference Shao, Irwin, Hartson and Matts2003). Biochemical studies show that p50 binds to Hsp90 in a 1:1 ratio with the interaction occurring largely at the NTD of Hsp90. A crystal structure of the C-terminal region of p50 (residues 148–348) bound to the NTD of Hsp90 shows a dimer for p50 but in this configuration the dimer interface is small and may arise from crystal contacts (Roe et al. Reference Roe, Ali, Meyer, Vaughan, Panaretou, Piper, Prodromou and Pearl2004). Indeed, biochemical studies suggest that p50 binds Hsp90 as a monomer because the K _D for dimerization (80 μM) is higher than the K _D for complex formation (4 μM) (Zhang et al. Reference Zhang, Hirshberg, McLaughlin, Lazar, Grossmann, Nielsen, Sobott, Robinson, Jackson and Laue2004).

The crystal structure (CTD of p50 bound to NTD of Hsp90) indicates a mechanism for the inhibition of Hsp90's ATPase activity. Specifically, residue R167 of p50 points into the nucleotide-binding pocket of Hsp90 and forms a hydrogen bond with the catalytic glutamate of Hsp90 (E33). Nucleotide can still bind, but the catalytic water is no longer coordinated for hydrolysis (Roe et al. Reference Roe, Ali, Meyer, Vaughan, Panaretou, Piper, Prodromou and Pearl2004). The crystal structure also reveals another potential means of inhibition in which p50 blocks the conversion of Hsp90 from the open to the closed state. Specifically, when the p50:Hsp90 dimer is docked onto one NTD of the yeast Hsp90 AMPPNP-bound crystal structure, p50 sterically clashes with the second Hsp90 NTD (Fig. 10). When docked onto the full-length apo crystal form of Hsp90, the p50 dimer fits into the cleft of the Hsp90 dimer potentially stabilizing Hsp90 in an open conformation (Fig. 10c).

Fig. 10.

Crystal structure of p50 bound to Hsp90. (a) The C-terminal region of p50 (pink) and the NTD of Hsp90 (blue) form a heterotetramer with p50 forming a dimer between the two NTD domains of Hsp90. (b) p50 binding is incompatible with the dimerized NTDs of the Hsp90 closed state. The two NTDs of Hsp90 are shown in blue and beige and they are shown in the configuration seen in the yeast AMPPNP-bound crystal state. Docking of the p50:Hsp90 crystal dimer onto one NTD of the closed state reveals steric clashes between p50 (red) and the second Hsp90 NTD (beige). (c) The p50:Hsp90 heterotetramer from A is aligned with the two NTDs from the apo bacterial crystal structure suggesting a mechanism for blocking the closure of Hsp90. One monomer of the apo structure is shown in green and the other in beige. For p50 to bind to this structure, the NTDs of the apo state would have to move as seen by the blue Hsp90 NTDs.

Two additional lines of evidence support the idea that p50 affects Hsp90 conformational dynamics. First, biochemical studies using the Hsp90 mutants T22I and T101I support the preference of p50 for the open conformation of Hsp90 (Millson et al. Reference Millson, Truman, Wolfram, King, Panaretou, Prodromou, Pearl and Piper2004). Second, for p50 to bind to the full-length apo conformation a structural rearrangement must occur between the NTD and MD of Hsp90 (Fig. 10c). Indeed, SAXS measurements of the p50:Hsp90 complex indicate that complex formation causes a contraction in apo Hsp90 (Zhang et al. Reference Zhang, Hirshberg, McLaughlin, Lazar, Grossmann, Nielsen, Sobott, Robinson, Jackson and Laue2004). HX-MS also shows changes in the protection of human Hsp90 in the presence of p50 consistent with a conformational change (Phillips et al. Reference Phillips, Yao, Zhang, McLaughlin, Laue, Robinson and Jackson2007).

5.4 Aha1 activates the ATPase of Hsp90

Unlike the other co-chaperones discussed, Aha1 stimulates the ATPase activity of Hsp90 (Lotz et al. Reference Lotz, Lin, Harst and Obermann2003; Meyer et al. Reference Meyer, Prodromou, Liao, Hu, Mark Roe, Vaughan, Vlasic, Panaretou, Piper and Pearl2004; Panaretou et al. Reference Panaretou, Siligardi, Meyer, Maloney, Sullivan, Singh, Millson, Clarke, Naaby-Hansen, Stein, Cramer, Mollapour, Workman, Piper, Pearl and Prodromou2002). Aha1 binds mainly to the MD of Hsp90 and this interaction is mediated by the N-terminal portion of Aha1 (N-Aha1) (Lotz et al. Reference Lotz, Lin, Harst and Obermann2003; Meyer et al. Reference Meyer, Prodromou, Liao, Hu, Mark Roe, Vaughan, Vlasic, Panaretou, Piper and Pearl2004; Siligardi et al. Reference Siligardi, Hu, Panaretou, Piper, Pearl and Prodromou2004). Recent NMR studies suggest that Aha1 also interacts with the NTD of Hsp90 (Retzlaff et al. Reference Retzlaff, Hagn, Mitschke, Hessling, Gugel, Kessler, Richter and Buchner2010). N-Aha1 has been crystallized in complex with the MD of Hsp90 (Meyer et al. Reference Meyer, Prodromou, Liao, Hu, Mark Roe, Vaughan, Vlasic, Panaretou, Piper and Pearl2004) and when aligned with either the closed structure or the open structure, Aha1 binds to Hsp90 at the edge of the dimer cleft approximately 90° from the dimer interface (Fig. 11), and NMR experiments with the full-length proteins suggest that Aha1 binds to both NTDs of the Hsp90 dimer in the closed conformation (Retzlaff et al. Reference Retzlaff, Hagn, Mitschke, Hessling, Gugel, Kessler, Richter and Buchner2010). Gel filtration and NMR measurements show that Aha1 competes with both Hop and p23 for binding to Hsp90 (Harst et al. Reference Harst, Lin and Obermann2005; Martinez-Yamout et al. Reference Martinez-Yamout, Venkitakrishnan, Preece, Kroon, Wright and Dyson2006), indicating that Aha1 can act at multiple stages of the chaperone cycle to disrupt inhibition by other co-chaperones. Aha1 may displace Hop from early complexes and p23 from mature complexes stimulating the ATPase activity and allowing for the continuation of the chaperone cycle.

Fig. 11.

Aha1 binds Hsp90 on the edge of the cleft. By docking the N-Aha1:MD-Hsp90 complex on the full-length apo (a) and ATP (b) crystal structures, the approximate orientation of Aha1 can be determined. Aha1 is colored in gray whereas one monomer of Hsp90 is green and the other monomer is beige.

Aha1 stimulates the ATPase activity of Hsp90 by an asymmetric interaction, and Aha1 bound to one Hsp90 monomer is able to activate both NTDs of the Hsp90 dimer (Retzlaff et al. Reference Retzlaff, Hagn, Mitschke, Hessling, Gugel, Kessler, Richter and Buchner2010). This activation most likely occurs by the interaction of Aha1 with the Hsp90 NM domain optimally positioning the MD catalytic loop for catalysis (Fig. 11). As discussed earlier, in the apo state, the catalytic loop (and the critical R380) is positioned away from the NTD (Fig. 12b), whereas in the closed state R380 contacts the γ-phosphate of ATP (Figs 5 and 12 c). Even though in the Aha1:Hsp90 complex parts of the catalytic loop are disordered, the orientation is much more consistent with the loop conformation found in the closed state (Fig. 12). Furthermore, the R380A mutant, which has a decreased ATPase rate, is not activated by Aha1 (Meyer et al. Reference Meyer, Prodromou, Hu, Vaughan, Roe, Panaretou, Piper and Pearl2003) implying that R380 is critical in the Aha1-mediated activation of Hsp90. Data showing that the Hsp90 F349A mutant, located at the NM interface and potentially involved in interdomain communication, is hypersensitive to Aha1 also support this conclusion (Meyer et al. Reference Meyer, Prodromou, Hu, Vaughan, Roe, Panaretou, Piper and Pearl2003; Siligardi et al. Reference Siligardi, Hu, Panaretou, Piper, Pearl and Prodromou2004).

Fig. 12.

Aha1 binding to Hsp90 causes a rearrangement in the MD catalytic loop. (a) N-Aha1 (gray) was crystallized in complex with the MD of Hsp90 (green). The catalytic loop, shown in red, is partially disordered in the crystal structure. (b) The catalytic loop in the apo full-length structure (beige) has more α-helical structure than in the Aha1:Hsp90 complex, and the catalytically required R380 (shown in sticks) forms contacts with the MD in the apo structure. (c) In the closed AMPPNP-bound crystal structure the catalytic loop, shown in purple, orients R380 towards the NTD where it interacts with the γ-phosphate of ATP. The loop structure in the Aha1:Hsp90 complex is closely related to the loop structure in the closed state.

Although it is clear from these studies that NTD dimerization and rearrangement of the NM domain are important for Aha1's effect on Hsp90, it is not clear as to whether Aha1 simply stabilizes the catalytic conformation once it forms or if binding of Aha1 results in a conformational change in Hsp90 leading to catalysis. A recent FRET study suggests that Aha1 directly induces structural changes in yeast Hsp90. Aha1-binding results in the rapid acceleration of conformational changes between the NTD and MD similar to the acceleration seen in the lidless Hsp90 mutant, which shows increased NTD dimerization. The FRET results also suggest that Aha1 binding allows Hsp90 to bypass the first conformational intermediate after nucleotide binding leading to an accumulation of the second NTD dimerized intermediate instead (Hessling et al. Reference Hessling, Richter and Buchner2009).

6. Client protein binding affects Hsp90 conformation

The conformational dynamics of Hsp90 are likely involved in the folding and activation of client proteins, and changes in Hsp90 likely confer structural changes upon the client protein itself. As another level of control over the Hsp90 cycle, client proteins themselves may also affect the conformation of Hsp90. Early indirect evidence for client proteins altering the conformation of Hsp90 came from in vitro studies showing that the ligand-binding domain of GR stimulates the human Hsp90 ATPase rate (McLaughlin et al. Reference McLaughlin, Smith and Jackson2002) suggesting that GR shifts the population of human Hsp90 towards the closed catalytic state. This correlates well with the observation that human Hsp90 has a low basal ATPase rate because it does not significantly populate the closed state. Furthermore, both Hop and p23 inhibit the GR-stimulated rate (McLaughlin et al. Reference McLaughlin, Smith and Jackson2002), as expected based on their effect on conformational dynamics as discussed above.

The clearest structural evidence for client binding influencing the conformation of Hsp90 comes from a 3D negative-stain EM reconstruction of the p50:Hsp90:Cdk4 complex (Vaughan et al. Reference Vaughan, Gohlke, Sobott, Good, Ali, Prodromou, Robinson, Saibil and Pearl2006). The reconstruction shows an asymmetric complex with two Hsp90 monomers, one p50 monomer and one Cdk4 kinase monomer. The Hsp90 dimer is in a conformation similar to the closed AMPPNP-bound state but the NTDs are not dimerized. One NTD hinges backwards away from the other and p50 binds between the two NTDs. Cdk4 interacts with the outer edge of the complex. The kinase likely contacts the Hsp90 MD with the C-lobe, and the N-lobe contacts one Hsp90 NTD and p50. Previous studies, discussed above, indicate that p50 binds as a 1:1 complex with Hsp90 and this complex is locked in a conformation more closely related to the apo crystal structure. It has been suggested that the kinase initially binds a dimer of p50 and this complex is recruited to Hsp90 forming a symmetric complex. A structural rearrangement then occurs displacing one p50 monomer and leading to the asymmetric complex observed by EM (Fig. 13) (Vaughan et al. Reference Vaughan, Gohlke, Sobott, Good, Ali, Prodromou, Robinson, Saibil and Pearl2006).

Fig. 13.

Proposed model for Hsp90:p50:Cdk4 complex formation. Initially, the kinase Cdk4 binds to a dimer of p50. This complex is then recruited to a dimer of Hsp90 to form a symmetric complex. A conformational rearrangement then occurs and one monomer of p50 is displaced leading to the asymmetric complex observed by EM.

Both GR and kinase clients are actively recruited to Hsp90 by co-chaperones, which adds a layer of complexity to the chaperone/client interactions. Understanding how Hsp90 interacts with just a client is crucial for building more complex models involving the additional involvement of co-chaperones. However, trapping Hsp90 with a mostly unfolded protein presents significant technical challenges. As a solution to this problem, Hsp90/client interactions have been measured with a constitutively unfolded protein Δ131Δ, a well-studied fragment of staphylococcal nuclease. While Δ131Δ is largely unfolded it does not aggregate even at high concentrations. These properties have made it an ideal tool for investigating structural properties of unfolded proteins (Shortle, Reference Shortle2002). Binding of this unfolded protein to Hsp90 drives large-scale conformational changes. Specifically, under apo conditions Hsp90 adapts its conformation by partially closing around Δ131Δ. In the presence of AMPPNP, Δ131Δ binds with increased affinity to Hsp90's fully closed state, accelerates the open/closed transition and stimulates ATP hydrolysis (Street and Agard, in preparation). Further studies will be required to see how Hsp90 responds to different unfolded proteins, and how co-chaperones modulate the Hsp90 response.

A central unanswered question concerns the physical state of Hsp90-bound clients; currently there is conflicting evidence. For example, studies of thermally denatured citrate synthase suggest that Hsp90 interacts in early unfolding or the late stages of folding (Jakob et al. Reference Jakob, Lilie, Meyer and Buchner1995). In contrast, NMR studies of a p53 mutant suggest that Hsp90 binds after complete client unfolding (Rudiger et al. Reference Rudiger, Freund, Veprintsev and Fersht2002). Additionally, Δ131Δ has significant residual structure despite being globally unfolded (Alexandrescu et al. Reference Alexandrescu, Abeygunawardana and Shortle1994; Alexandrescu & Shortle, Reference Alexandrescu and Shortle1994). More studies will be required to determine the folding state of Hsp90-bound clients and which structural features are involved in the binding.

7. The effect of post-translational modifications on the conformational dynamics of Hsp90

Emerging evidence indicates that post-translational modifications regulate specific client protein interactions with Hsp90. Several modifications have been identified including S-nitrosylation, phosphorylation, ubiqutination and acetylation (Blank et al. Reference Blank, Mandel, Keisari, Meruelo and Lavie2003; Wandinger et al. Reference Wandinger, Richster and Buchner2008). As discussed below, these modifications likely affect multiple aspects of the Hsp90 conformational cycle.

8. Small molecules also shift the conformation of Hsp90

The conformational dynamics of Hsp90 are essential for proper function and multiple layers of regulation affect the conformation, suggesting that small molecules that affect Hsp90's conformational dynamics may have clinical value. Currently, most Hsp90 inhibitors compete for binding with ATP and were not designed to influence the conformation of Hsp90. There are, however, indications that some inhibitors shift the conformation of Hsp90 to states that are incapable of client protein binding.

As would be expected for small molecules occupying the nucleotide-binding pocket, inhibitor binding causes local changes in the lid of Hsp90 (Immormino et al. Reference Immormino, Dollins, Shaffer, Soldano, Walker and Gewirth2004; Stebbins et al. Reference Stebbins, Russo, Schneider, Rosen, Hartl and Pavletich1997). Binding of the inhibitor GA leads to a lid conformation of Hsp90 (Stebbins et al. Reference Stebbins, Russo, Schneider, Rosen, Hartl and Pavletich1997) similar to the conformation seen in the isolated apo human Hsp90 NTD (Fig. 3a, red). This conformation is unlike the lid state seen in the nucleotide-bound complexes where hydrophobic surfaces are accessible for NTD dimerization. The local changes in the NTD induced by the inhibitor most likely correlate with changes in the global structure of the Hsp90 dimer. The GA-bound lid conformation suggests that GA and nucleotide have very different effects on the overall structure of Hsp90.

In vivo, the binding of GA to Hsp90 leads to the destabilization and degradation of client proteins (Mimnaugh et al. Reference Mimnaugh, Chavany and Neckers1996; Schulte et al. Reference Schulte, Blagosklonny, Ingui and Neckers1995; Whitesell et al. Reference Whitesell, Mimnaugh, De Costa, Myers and Neckers1994). In most cases, destabilization is due to the disruption of Hsp90:client complexes as observed for the kinases v-Src and Bcr-Abl (An et al. Reference An, Schulte and Neckers2000). GA also disrupts the binding of p23 (An et al. Reference An, Schulte and Neckers2000; Sullivan et al. Reference Sullivan, Stensgard, Caucutt, Bartha, McMahon, Alnemri, Litwack and Toft1997), indicating that GA also inhibits closure. GA increases the presence of Hop and Hsp70 in pull-downs of Hsp90 (An et al. Reference An, Schulte and Neckers2000; Sullivan et al. Reference Sullivan, Stensgard, Caucutt, Bartha, McMahon, Alnemri, Litwack and Toft1997). Although GA binds to an open conformation of Hsp90, this conformation is distinct from the conformation of apo Hsp90. As measured by SAXS, GA causes compaction of Hsp90 but not closure (Zhang et al. Reference Zhang, Hirshberg, McLaughlin, Lazar, Grossmann, Nielsen, Sobott, Robinson, Jackson and Laue2004). Binding of inhibitors clearly affects the local structure of Hsp90, as HX-MS experiments have shown that DMAG, a derivative of GA and radicicol, an inhibitor similar to GA, causes changes in the protection of amide protons in all three domains of Hsp90 (Phillips et al. Reference Phillips, Yao, Zhang, McLaughlin, Laue, Robinson and Jackson2007). Radicicol also increases the protease sensitivity of Hsp90 as compared to the apo protein (Nishiya et al. Reference Nishiya, Shibata, Saito, Yano, Oneyama, Nakano and Sharma2009).

Other small molecule inhibitors that function via alternative mechanisms may also cause Hsp90 conformational changes. Novobiocin is a CTD-binding inhibitor (Marcu et al. Reference Marcu, Chadli, Bouhouche, Catelli and Neckers2000a, Reference Marcu, Schulte and Neckersb; Soti et al. Reference Soti, Racz and Csermely2002) which, like GA and radicicol, inhibits the formation of Hsp90 complexes with client or p23, but unlike GA it also reduces the affinity of Hsp90 for TPR-containing co-chaperones that bind to the CTD (Allan et al. Reference Allan, Mok, Ward and Ratajczak2006; Yun et al. Reference Yun, Huang, Leach, Hartson and Matts2004). Epigallocatechin gallate (EGCG) the active ingredient found in green tea also directly binds to Hsp90 and acts as an AhR antagonist. As opposed to disrupting the interaction of Hsp90 with AhR, EGCG functions by stabilizing the interaction between Hsp90 and AhR (Palermo et al. Reference Palermo, Westlake and Gasiewicz2005). A cyclic peptide, Sansalvamide A, appears to bind at the NM domain interface, has no effect on ATPase activity, and yet differentially alters client protein binding in vitro and client stabilization in vivo (Vasko et al. Reference Vasko, Rodriguez, Cunningham, Ardi, Agard and McAlpine2010). This demonstrates the potential of non-ATP directed inhibitors to show unique client selectivity, presumably by allosteric modulation of Hsp90 conformational dynamics.

9. Concluding remarks

Structural dynamics play an important role in the function of Hsp90, and understanding the complete conformational ensemble is imperative for defining Hsp90's molecular mechanism. Large conformational changes occur throughout the ATPase hydrolysis cycle, and the conformational state is not rigorously determined by the bound nucleotide. Instead nucleotide binding shifts a conformational equilibrium between states that are also sampled in the absence of nucleotide. In vivo, Hsp90 will also sample nucleotide-bound and -unbound states simultaneously due to the weak interaction with ATP. For the cytosolic eukaryotic Hsp90s, co-chaperones regulate the conformational state of Hsp90 by shifting the equilibrium, and the finely tuned regulation is essential for proper chaperone function. Other levels of regulation exist in the form of post-translational modifications such as S-nitrosylation, phosphorylation and acetylation, but the exact nature of these effects remains poorly understood. Other modifications will also require investigation to determine their importance in the regulation of Hsp90. Disrupting the conformational dynamics of Hsp90 provides an attractive target for new therapeutics for the treatment of cancer and potentially vascular disease and Alzheimer's, and inhibitors currently in clinical trials show signs of altering the conformation of Hsp90 and thereby disrupting the binding of client proteins.

Although a much clearer picture of the mechanism of Hsp90 has emerged over the past few years, many outstanding questions require investigation. To begin, the nature of the interaction of client proteins with Hsp90 is largely uncharacterized, and it is essential to understand how client protein binding affects the conformation of Hsp90 and vice versa. This information is key to the understanding of how Hsp90's conformational changes relate to the activation of client proteins. Another open question is the origin of the energy required for the remodeling of client protein. Because the ATPase activity is low and not strongly coupled to the conformational changes that occur, it is unclear from where the force originates to drive the conformational changes in Hsp90 and any consequent changes that occur in the client protein. Also, important differences exist between different homologs and each homolog has a uniquely determined balance between conformational states and transition kinetics. Understanding how the equilibria are established and the functional consequences for shifting the equilibria will provide critical information about the molecular mechanism of Hsp90. Ultimately, a more detailed understanding of Hsp90, its dynamics and interactions will pave the way for the development of more targeted therapeutics.