The Mechanism of Hsp90 ATPase Stimulation by Aha1

Hsp90 is a dimeric molecular chaperone responsible for the folding, maturation, and activation of hundreds of substrate proteins called ‘clients’. Numerous co-chaperone proteins regulate progression through the ATP-dependent client activation cycle. The most potent stimulator of the Hsp90 ATPase activity is the co-chaperone Aha1p. Only one molecule of Aha1p is required to fully stimulate the Hsp90 dimer despite the existence of two, presumably identical, binding sites for this regulator. Using ATPase assays with Hsp90 heterodimers, we find that Aha1p stimulates ATPase activity by a three-step mechanism via the catalytic loop in the middle domain of Hsp90. Binding of the Aha1p N domain to the Hsp90 middle domain exerts a small stimulatory effect but also drives a separate conformational rearrangement in the Hsp90 N domains. This second event drives a rearrangement in the N domain of the opposite subunit and is required for the stimulatory action of the Aha1p C domain. Furthermore, the second event can be blocked by a mutation in one subunit of the Hsp90 dimer but not the other. This work provides a foundation for understanding how post-translational modifications regulate co-chaperone engagement with the Hsp90 dimer.

co-chaperones, both Hch1p and Aha1p N can stimulate the ATPase activity of Hsp90 to a similar degree 19,28,29 . Interestingly, though Hch1p has evolved to function as a single domain in vivo, fusion of Aha1p C to Hch1p enhances ATPase stimulation. ATPase stimulation by this Hch1p-Aha1p chimera (Fig. 1B) is lower than that achieved by full length Aha1p but consistent with both Hch1p and Aha1p N facilitating the activity of the Aha1p C domain. However, we have shown that a mutation in the catalytic loop of Hsp90, E381K, impairs ATPase stimulation by Aha1p and Aha1p N but not by Hch1p or the chimera 29 . Importantly, the catalytic loop (residues 370-390 in the Hsp90 middle domain) mediates communication with the N domain of Hsp90 30 . Binding of Aha1p N to the middle domain drives a conformational change in the catalytic loop that remodels residues in the ATP binding pocket. Thus, the Aha1p N domain has likely evolved to specifically facilitate the action of the Aha1p C domain by manipulating the catalytic loop in a way that Hch1p has not. Thus, comparing Hch1p and Aha1p can provide biological insight into the mechanics of Hsp90 regulation.
ATP hydrolysis by Hsp90 is intricately choreographed involving collaboration between numerous structural elements in the dimer structure 20,22,35 . Hsp90 dimerization is obligatory for function, and maintained by interaction between C terminal domains. Dimerization of the N terminal domains is required, albeit not sufficient, for ATP hydrolysis. Deletion of the N terminal 'strap' (residues  in one subunit of an Hsp90 dimer eliminates ATPase activity in the opposite protomer, demonstrating the importance of interdomain contact between the N domains 35,36 . Additionally, the ATPase activity of Hsp90 is auto-inhibited by a 'lid' that closes over bound nucleotide 37 . Deletion of the lid segment in both subunits of an Hsp90 dimer eliminates ATPase activity, however, deletion of this segment in only one subunit dramatically enhances ATPase activity in the opposite subunit in the absence of co-chaperones 37,38 . Moreover, progression through at least five different conformational states is accelerated in a similar manner upon deletion of the lid segment in one subunit or the addition of Aha1p, suggesting that this may be the mechanism by which Aha1p stimulates ATPase activity 38 .
In this report, we have studied the nature of asymmetric ATPase stimulation of Hsp90 by the intact Aha1p co-chaperone, and how the N terminal orthologue Hch1p differs from Aha1p N in this regard. Our work defines three discrete steps in Aha1p-mediated ATPase stimulation of the Hsp90 dimer and provides mechanistic insight into the different activities of Hch1p and Aha1p. In addition, we have found that the C-terminal domain of Aha1p is required for 'co-chaperone switching' in vitro.

Results
Regulation of Hsp90 lid dynamics by Aha1p and Hch1p. The ATPase activity of Hsp90 is auto-inhibited by a 'lid' segment (residues 98-121 in yeast Hsp82p) that functions as a gate to regulate nucleotide binding and commitment to hydrolysis 37 . The dramatic enhancement in ATPase activity that occurs upon deletion of this lid in one subunit of an Hsp90 dimer implies that efficiency of ATP hydrolysis is regulated by lid position. Interestingly, the progression through five different conformational states associated with the ATPase cycle, is similarly altered upon addition of Aha1p, or deletion of the lid segment in one protomer of Hsp90 in the context of a heterodimer consisting of one wildtype subunit and one lidless subunit (Hsp82p LL ) 38 . A straightforward explanation is that Aha1p promotes lid opening, driving N domain dimerization, and acceleration of ATP hydrolysis. In this case, Aha1p would be unable to stimulate an Hsp82p:Hsp82p LL heterodimer which has enhanced ATPase activity compared to a wildtype homodimer. Consistent with previous reports, the addition of Hsp82p LL , which like Hsp82p D79N and Hsp82p E33A , lacks ATPase activity ( Fig. 2A), potently stimulates the ATPase activity of wildtype Hsp82p (Fig. 2B) 37 . However, titration of Aha1p into reactions containing Hsp82p:Hsp82p LL heterodimers indicates that Aha1p further stimulates ATPase activity (Fig. 2C). As is the case with stimulation of wildtype Hsp82p, Aha1p N was able to stimulate the ATPase activity of Hsp82p:Hsp82p LL heterodimers but to a lesser degree than full length Aha1p (Fig. 2C). This suggests that while the lidless-stimulated Hsp82p:Hsp82 LL heterodimer relieves auto-inhibition of ATP hydrolysis to some extent, the N domain of Aha1p is important in subsequent remodelling of the Hsp90 N domains to facilitate catalysis.
Given the different biological activities of Hch1p and Aha1p 28,29 , it is of interest to probe the stimulatory mechanism of Hch1p, which shares 37% amino acid identity with the Aha1p N domain 28 . Unexpectedly, we found that Hch1p inhibits the ATPase activity of Hsp82p:Hsp82p LL heterodimers in a manner that is gradually overcome at higher co-chaperone concentrations (Fig. 2C). This suggested to us that the ATPase activation mechanism of Hch1p differs from that of Aha1p N . In principle, Hch1p can bind to either the ATPase-competent subunit, the lidless subunit, or both in these experiments. We observed the greatest inhibition (~16% of the intrinsic rate of the Hsp82p:Hsp82p LL heterodimer) when 2.5 μ M Hch1p was added to the reaction. When the Hch1p concentration ATPase rates are shown as a fold stimulation the intrinsic rate of Hsp82p:Hsp82p LL heterodimers (stippled line). (D) ATPase stimulation of wildtype Hsp82p by Aha1p, Hch1p and Aha1p N . The V max values for cochaperone stimulation by each of these co-chaperones are 3.8 ± 0.4 min −1 (Aha1p), 0.4 ± 0.1 min −1 (Hch1p), and 0.8 ± 0.1 min −1 . Reactions contained 2 μ M of Hsp82p with indicated concentrations of co-chaperones. All Aha1p, Hch1p, and Aha1p N titrations are shown as black triangles, blue circles, and blue squares, respectively. The ATPase rate is shown as a fold stimulation of the intrinsic Hsp82p rate (stippled line). reached 5 μ M, ATPase activity was only inhibited by ~11%. Since the total amount of Hsp82p was 6 μ M, inhibition was likely due to binding to one subunit of the heterodimer. ATPase stimulation of wildtype Hsp82p homodimers by each of these co-chaperones is shown for comparison (Fig. 2D). The maximal ATPase rates we observed when co-chaperones were added to wildtype Hsp82p were 3.8 ± 0.4 min −1 for Aha1p, 0.4 ± 0.1 min −1 for Hch1p, and 0.8 ± 0.1 min −1 for Aha1p N . Since Hch1p could bind to either subunit of Hsp82p:Hsp82p LL heterodimers we pursued a strategy to restrict co-chaperone binding to one subunit or the other.

Aha-type co-chaperones exert different effects from different subunits of Hsp90 heterodimers.
To determine how Hch1p influences ATPase activity from intact or lidless subunits, we employed the V391E mutation to diminish co-chaperone binding to the middle domain of Hsp82p 33 . We first confirmed that Hsp82p V391E homodimers have normal intrinsic ATPase activity ( Fig. 2A) but are not readily stimulated by Aha1p (Fig. 3A). We also verified that Hsp82p V391E ATPase activity is stimulated by the addition of Hsp82p LL (Fig. 3B). In Hsp82p V391E :Hsp82p LL heterodimers, co-chaperone binding will occur preferentially to the lidless protomer 37 . Aha1p and Aha1p N were less effective in stimulating the ATPase activity of Hsp82p V391E :Hsp82p LL heterodimers (Fig. 3C) than Hsp82p:Hsp82p LL heterodimers (compare to Fig. 2C). It is important to note that the V391E mutation does not block co-chaperone binding completely, but rather, reduces the affinity by roughly 10 fold 33 . Thus, the weak stimulation that we observe is likely due to weak binding to the Hsp82p V391E subunit of the heterodimer. These data suggest that neither Aha1p nor Aha1p N can stimulate ATPase activity in the Hsp82p V391E subunit when bound to the Hsp82p LL subunit. Furthermore, Hch1p was only inhibitory in these ATPase assays suggesting that binding to the Hsp82p LL protomer actually antagonizes the enzymatic activity of the Hsp82p V391E subunit of these heterodimers. We next tested Hsp82p:Hsp82p V391E/LL heterodimers for stimulation by our co-chaperone constructs. In this case, co-chaperones will preferentially bind to the wildtype subunit. As with our other lidless heterodimers, addition of Hsp82p V391E/LL stimulates ATPase activity in wildtype Hsp82p (Fig. 3D). In contrast to Hsp82p V391E :Hsp82p LL heterodimers, each of Aha1p, Aha1p N , and Hch1p stimulated the ATPase activity of Hsp82p:Hsp82p V391E/LL heterodimers (Fig. 3E).
These data may suggest that Aha1p enhances the conformational changes in the Hsp90 dimer that are driven by deletion of the lid segment. Alternatively, Aha1p interaction and lid deletion may cause discrete conformational changes in the Hsp90 dimer.
ATPase stimulation of Hsp90 ATP hydrolysis mutant heterodimers by Aha1p and Hch1p. In order to further probe conformational remodelling and ATPase stimulation by Aha1p and Hch1p, we used ATP binding deficient (D79N) and ATP hydrolysis deficient (E33A) Hsp82p mutants that are otherwise structurally intact 26,27,33,37 . A study of engineered Hsp82p heterodimers found that Hsp82p:Hsp82p E33A heterodimers support yeast viability whereas Hsp82p:Hsp82p D79N heterodimers do not 39 . To explore the nature of catalytic activity for Hsp82p heterodimers where only one subunit can hydrolyze ATP, we added either Hsp82p D79N or Hsp82p E33A (which can each bind Aha1p but lack ATPase activity) to ATPase reactions containing Aha1p and Hsp82p V391E (which can hydrolyze ATP but not bind Aha1p). Consistent with previous reports, we find that addition of Hsp82p D79N restores Aha1p-mediated ATPase stimulation of Hsp82p V391E (Fig. 4A) 33 . Intriguingly, addition of Hsp82p E33A does not restore Aha1p-mediated ATPase stimulation to Hsp82p V391E (Fig. 4A). This demonstrates that Aha1p can stimulate the ATPase activity of the ATPase-competent Hsp82p V391E subunit when it is bound to an Hsp82p D79N , but not an Hsp82p E33A subunit of a heterodimer. In other words, the E33A mutation ablates the ability of Aha1p to act asymmetrically from the ATPase-dead subunit. Importantly, neither the addition of Hsp82p V391E/D79N nor Hsp82p V391E/E33A (which cannot bind Aha1p or hydrolyze ATP) diminished Aha1p-stimulated ATPase activity of wildtype Hsp82p (Fig. 4B). This indicates that binding of Aha1p to the ATPase-competent subunit of a heterodimer can stimulate ATP hydrolysis regardless of whether the D79N or E33A mutation is present in the opposite subunit. Importantly, we confirmed that Hsp82p:Hsp82p E33A and Hsp82p:Hsp82p D79N heterodimers form equally well ( Figure S1) and that Aha1p interacted with each of our Hsp82p mutants except Hsp82p V391E ( Figure S2).
The importance of direct contact between the Aha1p N terminal domain and the middle domain of the catalytically active protomer of the Hsp90 dimer for ATPase stimulation by intact Aha1p raises the question of whether Hch1p can fulfill this role. We tested the Hch1p-Aha1p chimera for the ability to stimulate Hsp82p heterodimers. Similar to the results with Aha1p, chimera-mediated stimulation of Hsp82p V391E is restored with addition of Hsp82p D79N but not Hsp82p E33A (Fig. 4C). However, while the addition of Hsp82p V391E/D79N does not affect chimera-mediated ATPase stimulation of wildtype Hsp82p (Fig. 4D), the addition of Hsp82p V391E/E33A greatly diminishes stimulation (Fig. 4D). This suggests that the chimera cannot stimulate Hsp82p:Hsp82p E33A heterodimers from either subunit, perhaps because Hch1p cannot stimulate these heterodimers at all. To address this, we tested the Aha1p N domain and Hch1p for the ability to stimulate Hsp82p:Hsp82p E33A heterodimers. Interestingly, the addition of Hsp82p E33A promotes ATPase stimulation of Hsp82p V391E by both Aha1p N and Hch1p (Fig. 5A). Moreover, addition of Hsp82p V391E/E33A does not impair ATPase stimulation of Hsp82p by either co-chaperone construct (Fig. 5B). This suggests that the Aha1p C domain portion of the chimera is unable to participate in the ATPase stimulation of Hsp82p:Hsp82p E33A heterodimers. Moreover, the action of the Aha1p C domain in the stimulation of a heterodimer harbouring a single E33A mutation can only be restored by the interaction of the Aha1p N domain with the middle domain of the opposite subunit.

Hch1p interaction with the middle domain of Hsp90 drives N-M communication.
We have previously shown that overexpression in yeast of Hch1p, but not Aha1p, confers hypersensitivity to ATP-competitive inhibitors of Hsp90 28 . We have also shown that a mutation in the middle domain of Hsp82p (i.e. E381K in the catalytic loop) impairs Aha1p-mediated ATPase stimulation but not ATPase stimulation by Hch1p 29 . Taken together with the data described above, it is clear that Aha1p and Hch1p regulate Hsp90 differently. Given the Scientific RepoRts | 6:33179 | DOI: 10.1038/srep33179 paucity of available structures of Aha1p with intact Hsp90, we employed an NMR strategy to examine conformational changes that occur in Hsp90 upon co-chaperone interaction. To investigate the difference between Aha1p and Hch1p interaction with Hsp90, we added Aha1p N or Hch1p, with or without ATP, to the N-M fragment of Hsp82p (Hsp82p N-M ) and examined chemical shift changes in the Hsp82p N-M spectra. As the affinities of ATP and the co-chaperones for Hsp82p are in the μ M range 40,41 , the interaction is likely in the intermediate/slow-exchange regime on the NMR timescale. Titration of ligand would therefore result in a gradual appearance of peaks at their bound-state position, rather than a gradual movement towards that position. As such, the bound-state peaks cannot be easily assigned to particular residues. Thus, we adopted the approach in which the bound-state peaks were assigned to the same residue as the closest free-state peak 40 .
Chemical shift changes are observed upon addition of ATP ( The interaction, or conformational change in the N domain of Hsp82p N-M upon addition of Hch1p is also supported by changes in peak intensity for the NMR spectra upon addition of the various ligands. Lower intensity   Examination of the NMR data shows that peak shifts within the Hsp82p middle domain are not the same upon Hch1p or Aha1p N interaction, which is consistent with their different biological activities 28,29 . The peaks for nineteen residues within the Hsp82p M domain shift by more than one standard deviation (1 σ ) from the average shift upon Aha1p N interaction whereas ten residues shift to this extent upon Hch1p binding. Moreover, only two residue-specific chemical shift changes in the M domain are shared between the two co-chaperones. Considering only Hsp82p N domain NMR spectra, the chemical shift changes are striking. Only two peaks from the Hsp82p N domain shift by > 1 σ upon Aha1p N interaction; K178 and L207. It is noteworthy that the largest shift (> 2 σ ) occurs for K178, the site of SUMOylation that recruits Aha1p to the Hsp90 dimer 42 . In contrast, twelve different peaks in the Hsp82p N domain shift by > 1 σ upon Hch1p binding to the N-M construct. These residues localize to two different regions; the face of the Hsp82p N domain that is oriented towards the middle domain, and the nucleotide binding pocket. The former chemical shift changes suggest that Hch1p interacts directly with the Hsp82p N domain and the latter changes suggest that Hch1p binding influences the nucleotide-binding pocket in a way that Aha1p cannot. This is consistent with our previous work, where we have shown that Hch1p regulates the ability of Hsp90 inhibitors to bind in vivo 28,29 .
The C terminal domain of Aha1p is required for 'co-chaperone switching'. The Hsp90 functional cycle is thought to involve the progression through numerous co-chaperone-bound states 43 . How late-acting co-chaperones like Aha1p displace early-acting co-chaperones like Sti1p is only beginning to be revealed 43,44 . Sti1p is a potent inhibitor of Hsp90 ATPase activity but can be displaced by the cooperative action of Aha1p and the TPR-containing co-chaperone Cpr6p 44 . However, the mechanics of how this happens are not fully understood. Sti1p and Cpr6p bind to the MEEVD motif at the C terminus of each subunit of the Hsp90 dimer. Consistent with the presence of two MEEVD motifs per dimer, these two co-chaperones can exist in a ternary complex with Hsp90 44 . We wondered how each domain of Aha1p might contribute to displacement of Sti1p from the Cpr6p-Hsp90-Sti1p ternary complex to restore ATPase activity. We hypothesized that the Aha1p C domain is required for this 'co-chaperone switching' . We tested Aha1p, Aha1p N , Hch1p, and the Hch1p-Aha1p chimera for the ability to stimulate the ATPase activity of the Cpr6p-Hsp90-Sti1p ternary complex. Consistent with previous reports we observed almost complete inhibition of Hsp90 ATPase activity by Sti1p regardless of the presence of Aha1p or Hch1p co-chaperone constructs (Fig. 7A). We also observed that Cpr6p had a mild stimulatory effect on the intrinsic Hsp90 ATPase rate as well as the stimulated rates mediated by our Aha1p and Hch1p co-chaperone constructs (Fig. 7A). These data show that ATPase stimulation mediated by all of our Aha1p and Hch1p constructs are overcome by Sti1p and augmented by Cpr6p. If the Aha1p C domain is required for co-chaperone switching (i.e. the displacement of Sti1p in the presence of Cpr6p) then only full length Aha1p and the Hch1p-Aha1p chimera would be able to stimulate the ATPase rate of reactions containing both Cpr6p and Sti1p. Consistent with this hypothesis we observed a statistically significant increase in ATPase activity in reactions containing both Sti1p and Cpr6p upon the addition of Aha1p or the Hch1p-Aha1p chimera (Fig. 7B). However, when Hch1p or Aha1p N was added to reactions containing both Cpr6p and Sti1p, ATPase activity actually decreased, albeit not to a significant degree (Fig. 7B). This suggested to us that neither Hch1p nor Aha1p N were able to displace Sti1p and stimulate ATPase activity. To further investigate this we employed a more direct measure of physical displacement of Hsp90 from Sti1p. We expressed and purified an N-terminally myc-tagged version of Sti1p for co-immunoprecipitation experiments. We incubated equimolar amounts of Hsp82p or myc-tagged Sti1p (5 μ M final) together in order to co-immunoprecipitate Hsp82p with Sti1p. Hsp82p is readily recovered in complex with myc-tagged Sti1p (lane 3 - Fig. 7C). Addition of equimolar amounts of Cpr6p (5 μ M) resulted in a small decrease (~20% of the total - Fig. 7D) in Hsp82p co-immunoprecipitated with myc-Sti1p (lane 4 - Fig. 7C). This is consistent with previous studies showing that Cpr6p and Sti1p can form a ternary complex with Hsp90 but also that these two co-chaperones compete for binding to the MEEVD motif at the C terminus of Hsp82p 44 . In contrast, addition of Aha1p (10 μ M final) resulted in a negligible displacement of Hsp82p from myc-Sti1p (lane 5 - Fig. 7C). However, the addition of both Cpr6p (5 μ M final) and Aha1p (10 μ M final) displaced approximately 50% of Hsp82p from myc-Sti1p (lane 6 - Fig. 7C,D). Consistent with our ATPase data, the addition of a large excess (50 μ M final) of Aha1p N did not result in displacement of Hsp82p from myc-Sti1p on its own (lane 7 - Fig. 7C). Furthermore, addition of Aha1p N together with Cpr6p did not result in any further displacement of Hsp82p compared to Cpr6p alone (lane 8 - Fig. 7C,D). Taken together, these data show that it is the C terminal domain of Aha1p that is required for the displacement of Sti1p from Hsp82p in cooperation with Cpr6p.

Discussion
The observation that Aha1p can stimulate the ATPase activity of an Hsp90 dimer from either subunit defines a model for the asymmetric action of Aha1p but does not provide mechanistic insight into how Aha1p drives the process 33 . We report here that the E33A mutation blocks the ability of Aha1p to stimulate ATP hydrolysis in the Scientific RepoRts | 6:33179 | DOI: 10.1038/srep33179 opposite subunit of a heterodimer. However, Aha1p N can exert a small stimulatory effect from either subunit of a heterodimer harbouring one Hsp82p E33A subunit. This suggests that the E33A mutation blocks the rearrangement of the Hsp90 N domains that is required for the action of the Aha1p C domain. How can the E33A mutation eliminate the action of the Aha1p C domain only in cis to the Aha1p N domain interaction with the Hsp82p M domain (Fig. 8A)?
We propose a model for Aha1p-mediated ATPase stimulation that involves three different steps (Fig. 8B). Interaction of the Aha1p N domain with the M domain of one subunit of the Hsp90 dimer drives two different conformational events. The first results in weak ATPase stimulation of the dimer regardless of what subunit is capable of hydrolyzing ATP. Hch1p and the Aha1p N domain are capable of driving this first event in an asymmetric manner, that is, from either subunit, as both stimulate the ATPase activity of a heterodimer harbouring one E33A mutation, regardless of which protomer the V391E mutation resides on. The second conformational change does not result in further ATPase stimulation, but induces a conformational change in the cis N domain that is critical for the action of the Aha1p C domain. This second event ultimately drives a final rearrangement of both N domains that allows Aha1p C to exert its effect. In this third step, Aha1p C interacts with the rearranged Hsp90 N domains to further simulate ATPase activity.
The precise structural basis for each step of ATPase stimulation is currently unknown as a result of a lack of structural data for the Aha1p-Hsp82p complex. Typically, structural inferences are derived from the structure of the Hsp82p-Sba1p complex 32,33,45 . However, it is important to note that this structure is an ATPase-inhibited conformation of Hsp82p. Thus, it is challenging to propose a structural model for Aha1p-mediated ATPase stimulation, however, several mechanistic possibilities can be ruled out. We initially hypothesized that Aha1p acts by alleviating lid-mediated inhibition of ATPase activity. However, Aha1p is capable of robustly stimulating Hsp82p:Hsp82p LL heterodimers suggesting that the mechanism for ATPase stimulation is unrelated to alleviation of lid-inhibition. None of the co-chaperone constructs we tested could stimulate ATPase activity when bound to the lidless subunit suggesting that lid deletion blocks the earliest intra-protomer conformational events that drive ATPase stimulation. Moreover, Hch1p inhibited ATPase activity in Hsp82p:Hsp82p LL heterodimers when bound to the lidless subunit. This may result from antagonizing local or global conformational changes that are induced by lid deletion, as a result of the N-M communication (observed by NMR) upon addition of Hch1p. Our analysis also allows us to rule out acceleration of global conformational rearrangements in Hsp90 as the basic mechanism for Aha1p action. Transition between the conformational states associated with the intrinsic ATPase cycle are similarly accelerated upon deletion of one lid segment or the addition of Aha1p 38 . However, both lid deletion and Aha1p binding contribute independently to ATPase stimulation in our experiments. This suggests that the action of Aha1p is at least partly independent of the five global conformational rearrangements identified in previous studies 38 .
A large body of research regarding the structure and function of Hsp90 suggests that it is a highly dynamic, allosteric machine that is regulated by numerous co-chaperones and post-translational modifications (PTMs) 8,19,28,30,32,33,[42][43][44][45][46][47][48][49][50][51][52][53][54][55][56][57] . However, a complete view of the catalytic cycle, and mechanism of action for stabilizing client proteins remains elusive. An emerging theme from recent work suggests that the in vivo regulation of Hsp90 is dependent on asymmetry in its activation, and has the potential to clarify many aspects underlying the molecular basis of Hsp90 function. A recent example is the observation that Aha1p is recruited to asymmetrically SUMOylated Hsp90 42 . It is interesting that Aha1p N binding to the Hsp82p M domain elicits a significant chemical shift change for K178, the site of SUMOylation. Given the identification of multiple PTMs that are required for Aha1p recruitment to Hsp90, it seems likely that there are numerous species of Hsp90 dimers, characterized by different asymmetric modifications, that may be regulated differently. Two well-characterized sites of phosphorylation within Hsp90 are T22 and Y24 52,53 . Similar to SUMOylation of K178, modification of these sites is required for recruitment of Aha1p. In the case of T22, yeast strains that express either non-phosphorylatable (Hsp82p T22A ) or phospho-mimetic (Hsp82p T22E ) versions of Hsp82p have similar phenotypes 53 . Moreover, while phosphorylation of T22 is required for Aha1p recruitment, Aha1p is not recovered in complex with either Hsp82p T22A or Hsp82p T22E from yeast lysates. In the case of Y24, a phospho-mimetic version of Hsp82p (Hsp82p Y24E ) lacks ATPase activity and does not support viability in yeast 52 . A possible explanation for these observations is that T22 and Y24 phosphorylation occurs asymmetrically in a manner similar to that for SUMOylation of K178, raising the possibility that such modifications may enhance or restrict the action of one or both Aha1p domains in a subunit-specific fashion. However, more work is required to understand the relationship between post-translational modifications of Hsp90 and co-chaperone function.
It is important to note that while ATPase activity can be attributed to Hsp82p heterodimers in our assays, measurements regarding the affinity of co-chaperones for these heterodimers are complicated by the presence of the excess, but inactive, pool of Hsp82p homodimers. However, some inferences can be made from the results with the lidless variant. In particular, our experiments with Hsp82p:Hsp82p LL heterodimers (Fig. 2C) stand out from other ATPase assays involving titration of Aha1p. Figure 2C shows a lag in ATPase stimulation, with robust stimulation not occurring until addition of 5 μ M Aha1p. This suggests that Aha1p is binding preferentially to the excess, but inactive, pool of Hsp82p LL homodimers present in these experiments. We calculated the apparent affinity of Aha1p for wildtype Hsp82p as ~1.9 μ M (Fig. 2D). However, the apparent affinity of Aha1p for the ATPase-competent Hsp82p:Hsp82p V391E/LL heterodimers is ~0.8 μ M (Fig. 3E), suggesting that Aha1p has higher affinity for these heterodimers than wildtype Hsp82p homodimers. This is consistent with previous reports showing that Aha1p binds more strongly to N-terminally dimerized Hsp82p, which is expected to be caused by lid deletion in one subunit 38 . However, calculation of the exact affinity (in Fig. 3E) is not possible owing to the presence of the excess pool of Hsp82p V391E/LL homodimers. Such calculations would require a different strategy than that employed in the current study.
This study demonstrates the key role of subunit-specific regulation of Hsp90 in the context of co-chaperone regulation. Our model provides a framework for interrogating the effect of post-translational modifications on Aha1p action that can be applied to understand the function of other co-chaperones that interact with the Hsp90 dimer at different sites. Sba1p binds to the dimerized N terminal domains of Hsp90 and can inhibit Aha1p-mediated ATPase activity 45,[58][59][60] . Sti1p utilizes one of its three tetratricopeptide repeat (TPR) domains to bind to the MEEVD motif at the C terminus of Hsp90 61,62 . While two Sti1p molecules can bind to the Hsp90 dimer simultaneously, binding to one subunit reduces the affinity at the second site 63,64 . This suggests that binding of Sti1p induces asymmetry in the Hsp90 dimer. Asymmetry in binding may also be intrinsic to client:Hsp90 complexes in general. Recent work shows that two molecules of the glucocorticoid hormone receptor (GR) can interact with an Hsp90 dimer, but addition of Sba1p or Sti1p drives formation of an asymmetric complex consisting of only one co-chaperone and one GR 65 .
Functionalization of Hsp90 protomers by asymmetric modification may be a mechanism for co-chaperone recruitment in vivo. It remains to be determined whether these modifications occur on one or both subunits, and to what degree they are coordinated. In addition, the nature by which asymmetric modifications alter the function of co-chaperones with respect to client activation and ATPase activity, remains to be resolved. Our three-step model is an important step towards this goal for Aha1p.

Methods
Expression plasmid construction. Bacterial expression vectors encoding Hsp82p, Sti1p, Aha1p, Hch1p, Hch1p-Aha1p chimera, and Aha1p N were constructed as previously described (Armstrong 2012, Horvat 2014). The CPR6 coding sequencing was amplified by PCR with primers designed to introduce Nde1 and BamH1 restriction sites at the 5′ and 3′ ends, respectively. The PCR product was digested with NdeI and BamHI for ligation into similarly cut pET11dHis. The lidless-mutant of Hsp82 was constructed and kindly provided by Johannes Buchner in a pET28a vector 37 . We digested the lidless Hsp82 (Hsp82p LL ) construct with Nde1 and BamH1 for ligation into similarly cut pET11dHis. Site directed mutagenesis was carried out to construct Hsp82p variants (Hsp82p V391E , Hsp82p D79N , Hsp82p E33A , Hsp82p LL , Hsp82p V391E/LL , Hsp82p V391E/D79N , and Hsp82p V391E/E33A ) using QuikChange mutagenesis (Agilent). The coding sequences contained in all mutagenized plasmids were verified by sequencing. The myc and the HA epitope was fused in-frame with the 6xHis-tag sequence upstream of the NdeI site of the pET11dHis vector. The coding sequences for Aha1p, Hsp82p and Hsp82p V391E were cloned into this pET11dHis-Myc vector, and the coding sequences for Hsp82p D79N and Hsp82p E33A were cloned into the pET11dHisHA vector, as described above.
For NMR studies, 2 H, 15 N-Hsp82p N-M domain (Hsp82p NM ), as well as unlabeled Aha1p N-domain (Aha1p N ), full-length Aha1p, and Hch1p were produced for high-resolution nuclear magnetic resonance (NMR) spectroscopy studies. Aha1p N and Aha1p were expressed and purified as described above. Hch1p was expressed in 2× TY media. Protein expression was induced at an OD 600 ~0.6 with 0.4 mM IPTG and incubation overnight at 25 °C. 2 H, 15 N-Hsp82p NM was expressed using the method of Marley et al. 66 ; E. coli were grown in 2 L of LB media to an OD 600 ~0.6, pelleted, and transferred to 0.5 L of M9 minimal media containing 99% D 2 O, 0.5 g ( 15 NH 4 ) 2 SO 4 and 4 g unlabeled glucose. The cells were incubated for 2 hr, subsequently induced with 0.4 mM IPTG, and grown overnight at 25 °C. Cells expressing Hch1p and Hsp82p NM were lysed by sonication and purified using FPLC with HisPrep 16/10 and Superdex75 26/60 columns. ATPase Assays. ATPase assays were carried out using the enzyme coupled assay as previously described where the regeneration of ATP is coupled to the oxidation of NADH [27][28][29]67 . All reactions were carried out in triplicate, three times in 100 μ L volumes using a 96-well plate. Activity was detected as a decrease in absorbance at 340 nm which was measured every minute for 90 minutes using a BioTek Synergy 4 and the path-length correction function. Average values of the experiments are shown with error expressed as standard error of the mean. The decrease in NADH absorbance at 340 nm was converted to micromoles of ATP using Beer's Law Scientific RepoRts | 6:33179 | DOI: 10.1038/srep33179 and then expressed as a function of time 68 . The final conditions of all the reactions are 25 mM Hepes (pH 7.2), between 1-25mM NaCl, 5 mM MgCl 2 , 1 mM DTT, 0.6 mM NADH, 2 mM ATP, 1 mM phosphoenol pyruvate (PEP), 2.5 μ L of pyruvate kinase/lactate dehydrogenase (PK/LDH) (Sigma), and 0.5% DMSO. Identical reactions were quenched with 100 μ M NVP-AUY922 and subtracted from unquenched reactions to correct for contaminating ATPase activity. All ATPase assays were carried out at 30 °C with recombinant proteins harboring N-terminal 6xHis tags.
Heterodimer ATPase Assays were performed as described earlier in Retzlaf et al. 33 . We allowed equilibration of heterodimers by mixing two samples of Hsp82p (one being a functional, ATPase-competent Hsp82p and the other, an ATPase-dead Hsp82p variant) at a specific concentration ratio for 15 minutes at room temperature. In Figs 2 and 3, Hsp82p heterodimers were formed by mixing functional, ATPase competent Hsp82p (wildtype Hsp82p and Hsp82p V391E ) with the ATPase dead Hsp82p variant (Hsp82p LL or Hsp82p V391E/LL ). Hsp82p LL was titrated into 2 μ M wildtype Hsp82p or Hsp82p V391E and Hsp82p V391E/LL was titrated into 2 μ M Hsp82p to determine at which ratio we have to mix the ATPase competent Hsp82p with the ATPase dead Hsp82p to ensure > 80% heterodimer formation (based on resulting ATPase rate). We determined that a 2:10 ratio (or 1:5 ratio of ATPase competent to ATPase dead) results in > 80% heterodimer formation, and used this 1:5 ratio in the following co-chaperone titration experiments. Heterodimers were first formed by incubating them separately for 15 minutes prior to adding them to specific wells. Co-chaperones were then added to the wells containing the heterodimer mix. In Figs 4 and 5, the ATPase-dead variants (Hsp82p E33A , Hsp82p D79N , Hsp82p V391E/E33A or Hsp82p V391E/D79N ) were titrated into reactions containing functional, ATPase-competent, Hsp82p (wildtype Hsp82p or Hsp82p V391E ) and a defined concentration of co-chaperone. First, the ATPase competent Hsp82p was added to wells, followed by the addition of the ATPase-dead variants of a specific concentration. We allowed equilibration of heterodimers to occur by waiting 15 minutes, and then added the co-chaperones to the wells.
For the Sti1p displacement ATPase assay in Fig. 7, 4 μ M Aha1p, Chimera, Hch1p, or Aha1p N were added to the designated wells first, followed by 4 μ M Sti1p and Cpr6p. Lastly, 2 μ M wildtype Hsp82p was added to these reactions. Statistical significance was measured using one way ANOVA analysis.
All ATPase assay was started by the addition of the regenerating system consisting of MgCl 2 , DTT, NADH, ATP, PEP, and PK/LDH. Fit lines were calculated according to the following equation (Y = ((Bmax*X)/(Kapp + X)) + X 0 ) 33 . The ATPase rates are either shown as μ M ATP hydrolyzed per minute per μ M of Hsp82p (1/min) or as a fold ATPase stimulated rate of the starting Hsp82p or heterodimer intrinsic rate.
V max (Fig. 2D) was calculated using the Michaelis-Menten non-linear fit in GraphPad Prism.
In vitro immunoprecipitation. In vitro immunoprecipitation assays were conducted using Ultralink Protein G beads (Pierce Thermo Fisher) that had been coupled to anti-myc monoclonal antibodies (clone 9E10) at a concentration of 5 μ g antibody per 1 μ L of beads. All recombinant proteins used in these assays harbour N-terminal 6xHis tags, and where specified, the additional myc-tag (6xHis-myc) or HA-tag (6xHis-HA).
To assess Aha1p binding, 5 μ M Hsp82p (wildtype or mutant -Hsp82p V391E , Hsp82p D79N , Hsp82p E33A , Hsp82p LL ) was mixed with 5 μ M 6xHis-myc-tagged Aha1p. To assess Sti1p displacement from Hsp82p, 5 μ M Hsp82p was incubated with 5 μ M 6xHis-myc-tagged Sti1p in the presence of 5 μ M Cpr6p, with and without 10 μ M Aha1p or 50 μ M of Aha1p N , and 5 mM AMPPNP. Band intensities of Hsp82p and myc-Sti1p from Coomassie stained gels were measured using the multiplex band analysis function of AlphaView software (FluorochemQ, Protein Simple). The ratio of Hsp82p to myc-Sti1p was used to measure the efficiency of recovery in each condition. The percent displacement represents the average relative reduction in this ratio from three independent experiments. Statistical significance was measured using one way ANOVA analysis.
To assess heterodimer formation, 5 μ M 6xHis-myc-tagged Hsp82p or Hsp82p V391E was incubated with 5 μ M 6xHis-HA-tagged Hsp82p D79N or Hsp82p E33A for 15 minutes. We added 10 μ L of Ultralink Protein G beads coupled to anti-myc monoclonal antibodies to all these 50 μ L reactions and then incubated them on a rotator at room temperature for 90 minutes. Beads were then pelleted, washed once in 250 μ L of binding buffer, resuspended in 50 μ l SDS sample buffer, and run on SDS-PAGE. Complexes were analyzed by Coomassie blue staining or western blotting. Myc-tagged proteins were detected with mouse anti-myc monoclonal antibody (4A6, Millipore), His-tagged proteins were detected with mouse anti-Tetra-His monoclonal antibody (34670, Qiagen), and HA-tagged proteins were detected with rat anti-HA monoclonal antibody (3F10, Roche). Final buffer conditions of all reactions were 25 mM Hepes (pH 7.2), 10-15 mM NaCl, 5 mM MgCl 2, and 0.1% tween-20. NMR spectroscopy. BCA analysis was used to determine the concentration of purified proteins.
Samples for NMR contained ~0.25 mM 2 H, 15 N-Hsp90-NM, 20 mM sodium phosphate, 50 mM NaCl, 5% D 2 O, 1 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid as a chemical shift reference, pH 7.5. Samples containing co-chaperone included 0.5 mM unlabeled Aha1p N or Hch1p. Samples containing ATP included 2 mM ATP and 4 mM MgCl 2 . 2D 1 H-15 N TROSY-HSQC NMR spectra were acquired on a 600 MHz spectrometer at 25 °C. Spectra were processed using NMRPipe 69 and analyzed using NMRViewJ 70 . The spectrum of Hsp82p NM in the free state was assigned using chemical shifts deposited in the Biological Magnetic Resonance Bank database for the yeast Hsp90-N 71 and Hsp90-M 72 domains. Peaks in the spectra of Hsp82p NM bound to co-chaperone and/or ATP were assigned mainly based on the closest peak to the free-state spectrum 40 . Chemical shift changes were calculated using Δ δ = [(Δ δ 15 N/5) 2 + (Δ δ 1 H N ) 2 ] 1/2 .