Regulation of Hsp90 ATPase Activity by the Co-chaperone Cdc37p/p50 cdc37 *

In vivo activation of client proteins by Hsp90 depends on its ATPase-coupled conformational cycle and on interaction with a variety of co-chaperone proteins. For some client proteins the co-chaperone Sti1/Hop/p60 acts as a “scaffold,” recruiting Hsp70 and the bound client to Hsp90 early in the cycle and suppressing ATP turnover by Hsp90 during the loading phase. Recruitment of protein kinase clients to the Hsp90 complex appears to involve a specialized co-chaperone, Cdc37p/p50 cdc37 , whose binding to Hsp90 is mutually exclusive of Sti1/Hop/p60. We now show that Cdc37p/p50 cdc37 , like Sti1/Hop/p60, also suppresses ATP turnover by Hsp90 supporting the idea that client protein loading to Hsp90 requires a “relaxed” ADP-bound conformation. Like Sti1/Hop/p60, Cdc37p/p50 cdc37 binds to Hsp90 as a dimer, and the suppressed ATPase activity of Hsp90 is restored when Cdc37p/p50 cdc37 is displaced by the immunophilin co-chaperone Cpr6/Cyp40. However, unlike Sti1/Hop/p60, which can displace geldanamycin upon binding to Hsp90, Cdc37p/p50 cdc37 forms a stable complex with geldanamycin-bound Hsp90 and may be sequestered in geldanamycin-inhibited Hsp90 complexes in vivo.

The in vivo activity of heat shock protein 90 (Hsp90) 1 depends on its association with a variety of co-chaperones that are components in a series of Hsp90-based multiprotein complexes involved in folding of client proteins (reviewed in Ref. 1). Authentic Hsp90 function in vitro and in vivo is also dependent on the binding and hydrolysis of ATP (2)(3)(4)(5), and this is regulated by interaction with TPR domain co-chaperones such as Sti1/Hop/p60 and immunophilins (6).
Although not essential, mammalian Hop/p60 increases the efficiency of steroid hormone receptor activation by Hsp90 and Hsp70 in vitro (13), and sti1 Ϫ yeasts, although viable, are temperature-sensitive and display growth defects (14).
Whereas steroid hormone receptors have been the most studied in terms of their activation by Hsp90, protein kinases form the largest coherent class of Hsp90-dependent client proteins (1). Recruitment of many protein kinase clients to the Hsp90 system involves a specialized co-chaperone Cdc37p (in budding yeast) or its mammalian orthologue p50 cdc37 (15). Unlike Sti1/ Hop/p60, which recruits clients to Hsp90 via interaction with Hsp70, Cdc37p/p50 cdc37 can interact directly with client protein kinases via its N-terminal region (16) and with Hsp90 via its C terminus (17). However, it is far from clear whether all protein kinases that interact with Cdc37p/p50 cdc37 are Hsp90dependent or if all kinases that are Hsp90-dependent are recruited by Cdc37p/p50 cdc37 (18). Unlike Sti1, Cdc37p is essential for yeast viability probably due to its involvement in formation of Cdc28-cyclin complexes (19).
Sti1/Hop/p60 and other TPR domain co-chaperones bind with mutual exclusivity to Hsp90 (20) via the C-terminal MEEVD sequence (10,21), although this may not be the sole site of interaction (6,14,22). In contrast, Cdc37p/p50 cdc37 has no detectable TPR motifs and does not require the C-terminal MEEVD of Hsp90, however, its binding is mutually exclusive with Sti1/Hop/p60 (20,23), suggesting that the binding sites overlap or are at least topologically adjacent. The common ability of Sti1/Hop/p60 and Cdc37p/p50 cdc37 to interact (directly or indirectly) with Hsp90 client proteins, together with their mutual exclusivity of binding to Hsp90, suggests that they may act as alternative recruitment factors for different classes of client protein. To gain further insight into this possibility, we have characterized the interaction between Cdc37p/p50 cdc37 and Hsp90 and investigated the effect of Cdc37p/p50 cdc37 cochaperone binding on the inherent ATPase activity of Hsp90. The results of these studies suggest a common mechanism for recruitment of client proteins into the Hsp90 complex requiring suppression of ATP turnover by Hsp90 during the loading phase.

EXPERIMENTAL PROCEDURES
Protein Production and Hsp90 ATPase Activity Assay-Expression and purification of His-tagged yeast Hsp90, His-tagged cSti11 (C-terminal Hsp90-binding domain of Sti1, residues 237-589), and Histagged Cpr6 were described previously (6). DNA sequences encoding a truncated N-terminal domain of human p50 cdc37 (sNp50, amino acid residues 30 -127) and C-terminal p50 cdc37 domain (Cp50, amino acid residues 128 -379) were cloned in-frame with the His tag of pRSETA (NheI-XhoI). Full-length p50 cdc37 cloned into pET16d was a kind gift from Nick Grammatikakis. Saccharomyces cerevisiae CDC37 was cloned in-frame with the His tag of pRSETA (NheI-XhoI). The various p50 cdc37 and Cdc37p constructs were expressed in Escherichia coli BL21(DE3) pLysS by induction at 20°C with 1 mM isopropyl-1-thio-␤-D-galactopyranoside. Cells were harvested and resuspended in 60 ml of 20 mM Tris (pH 8.0) containing 100 mM NaCl (Buffer A) and lysed by sonication. The cell lysate was centrifuged (20,000 ϫ g for 60 min at 4°C) and the supernatant applied to a Talon metal affinity column equilibrated in Buffer A. The column was washed with Buffer A containing 10 mM imidazole (pH 8.0) and finally p50 cdc37 /Cdc37p eluted with Buffer A containing 300 mM imidazole (pH 7.0). 10 mM EDTA and 1 mM dithiothreitol were added, and the eluent was concentrated with Vivaspin concentrators (Sartorius) with an appropriate molecular weight cut-off. Concentrated sample was applied to a gel filtration column (Superdex 75PG for Np50 and Cp50 and Superdex 200PG for p50 cdc37 and Cdc37p), equilibrated in 20 mM Tris (pH 7.5), 500 mM NaCl, and 1 mM EDTA. Fractions containing p50 cdc37 or Cdc37p were dialyzed (20 mM Tris (pH 7.5), 1 mM EDTA) and applied to a Q-Sepharose column equilibrated in the same buffer. The bound protein was eluted with a 0 -1 M NaCl gradient. Fractions containing pure p50 cdc37 or Cdc37p were dialyzed (20 mM Tris (pH 7.5), 1 mM EDTA) and subsequently concentrated using Vivaspin 5,000 and 30,000 molecular weight cutoff concentrators as appropriate. The protein concentrations reached are as follows: p50 cdc37 ϭ 99.7 mg ml Ϫ1 , Np50 ϭ 60 mg ml Ϫ1 , Cp50 ϭ 50 mg ml Ϫ1 , Cdc37p ϭ 131.6 mg ml Ϫ1 , Hsp90 ϭ 31.2 mg ml Ϫ1 , cSti1 33.2 mg ml Ϫ1 , and Cpr6 ϭ 92.7 mg ml Ϫ1 . Further treatment of proteins destined for ATPase assays and the regenerating enzymelinked ATPase assay itself were as described previously (3,6). Each assay was repeated between 3 and 6 times, and average activities were calculated. Average ATPase activities were plotted as a percentage of the maximum average activity for Hsp90 at 37°C.
Analytical Ultracentrifugation-Sedimentation equilibration experiments were conducted at 15°C in a Beckman-Optima XLI analytical centrifuge using an An-60 Ti rotor and 1.2-cm path length cells. p50 cdc37 , sNp50, and Cp50 were dialyzed against 20 mM Tris (pH 7.4) containing 1 mM EDTA and 25 mM NaCl. The proteins were used at concentrations of 23.6, 76.8, and 34.9 M, respectively. Experiments were performed on 100-l samples that were equilibrated for ϳ48 h at speeds ranging from 10,000 to 50,000 rpm. Sedimentation curves measured by absorbance at 280 nm were compared with a buffer reference cell. Multiple data sets were fitted simultaneously to yield the K d and molecular mass values quoted, using Beckman Optima XL-A/XL-I data analysis software (version 4.0). Data fitting and simulations used the following calculated parameters: buffer density at 15°C ϭ 1.00064 mg ml Ϫ1 ; monomeric molecular mass, p50 cdc37 ϭ 45,780 Da; sNp50 ϭ 13,508 Da, and Cp50 ϭ 30,817 Da; partial specific volumes, p50 cdc37 ϭ 0.7241; sNp50 ϭ 0.7161 and Cp50 ϭ 0.7271 and extinction coefficients (280 nm), p50 cdc37 ϭ 50,042, sNp50 ϭ 16,500, and Cp50 ϭ 26,930.
Circular Dichroism and K d Determinations-CD spectra were recorded on a nitrogen-flushed Jasco J720 spectropolarimeter. For an accurate determination of K d values, multiscanning was required for high data precision. At a scan speed of 20 nm min Ϫ1 , nine scans were demanded in the 245-335 nm region for the CD titrations of Hsp90 with several co-chaperones. The concentration of the proteins were determined spectroscopically using the following molar extinction coefficients: All CD spectra were reported in ⌬A ϭ (A L Ϫ A R ). The spectropolarimeter was calibrated with ammonium d-champor-10-sulfonate. The dissociation constant K d was determined by analyzing the CD data using a non-linear regression analysis as described previously (6,24).
The CD titrations were conducted in a stepwise manner adding small aliquots of 5-10 l of ligand stock solution directly into the cuvette of 1 cm path length containing an initial volume of 520 l of Hsp90 in 20 mM Tris-HCl. Each titration was terminated upon ligand saturation that never reached a final volume greater than 15% of the initial volume, and each CD spectrum was subsequently corrected for its dilution. To achieve the highest accuracy and precision, Finnpipette PCR with a volume range of 2-20 and 20 -200 l were used to add aliquots of ligand solution in a stepwise manner to reach the desired molar ratio for the solution of Hsp90 of the appropriate concentration.
For a single binding site at equilibrium, the CD, expressed in ⌬A, at any wavelength is proportional to the concentration of the bound and unbound species of the host (H) and ligand (L) components (as shown in Equation 1) (24).
Based on the Beer's Law (⌬A ϭ ⌬⑀.c.l, where differential molar extinc- , and the simulated spectrum (SIM) for the mixture. Simulated spectrum is derived using spectra of the individual components and is dissimilar to the OBS, which indicates that Hsp90 and Cdc37p form a complex. B, near-UV difference spectra for titration of Cdc37p into Hsp90 (145.1 M), obtained by subtracting the spectra of the free Cdc37p at n molar equivalents, from Hsp90-Cdc37p mixtures at Hsp90:Cdc37p molar ratios of 1:n. The spectrum for Hsp90 is indicated, and curves 1-5 are from Hsp90-Cdc37p mixtures in the molar ratio of 1:n (n ϭ 0.5, 1.0, 2.0, 3.0, and 4.0). Changes in the difference spectra are consistent with changes in the environment of aromatic residues due to molecular interaction. C, estimation of the K d for the binding of Cdc37p to Hsp90. The solid line represents the simulated curve that would be obtained if the titration were continued to obtain saturation of Cdc37p-binding sites and yields a K d of 100 M. This is in close agreement to the estimation of 113 Ϯ 4 M using the Levenberg-Marquardt equation (dotted line). In the titration, the total concentration of the host and the ligand is as shown in Equations 6 and 7, Rearranging Equations 6 and 7 (see Equations 8 and 9), Substituting Equations 8 and 9 into 5 leads to Equations 10 -13,

FIG. 2. Analysis of Hsp90-p50 cdc37 interactions by near-UV CD.
A, near-UV difference spectra for titration of p50 cdc37 into Hsp90. The spectrum of Hsp90 (96.7 M) alone is indicated, and curves 1-6 are from Hsp90-p50 cdc37 mixtures in the molar ratio 1:n (n ϭ 0.4, 0.7, 1.0, 1.4, 2.0, and 3.2). The saturable changes seen at a molar ratio close to 1:1 (Hsp90:p50 cdc37 ) suggests that there are two p50 cdc37 -binding sites on Hsp90. Subsequent changes (dotted line spectra), seen above a molar ratio of 1:4, indicate that the excess p50 cdc37 can homodimerize or that a second p50 cdc37 -binding site exists. B, near-UV difference spectra for p50 cdc37 . The spectrum of p50 cdc37 (1.62 M) is indicated, and curves 1-6 represent further addition of p50 cdc37 in the molar ratios 1:n (n ϭ 1, 3, 7, 15, 31, and 63). Spectral changes in the near-UV of the difference spectrum are consistent to those observed when p50 cdc37 is in excess over Hsp90 (Fig. 1A) and are consistent with homodimerization. C, the K d for the interaction of p50 cdc37 to Hsp90 is estimated as 2.5 M, and for D, the interaction of p50 cdc37 with itself is 9.1 M and 9.8 Ϯ 1.9 M with the Levenberg-Marquardt equation (dotted line).

Substituting Equations 16 into 12 leads to the final Equation 17
, Of the two solutions, only the one with Ϫ2 fits the observed CD data. Equation 17 can be simplified if the ligand has no CD (⌬⑀ L ϭ 0) either because it has no absorbing chromophore in the UV region studied (25) or because it is non-chiral (26) (see Equation 18).
The plot of the titration CD data at single wavelength versus the molar concentration of the ligand was analyzed by non-linear regression to Equation 18 as a function of ligand concentration to determine the dissociation constant K d ϭ 1/K. This equation is not only valid when only one component of the binary complex has intrinsic CD (one "active" CD component with bound and unbound species). It can also be applied to systems where both components show intrinsic CD, providing the data collected are transformed back to one active CD component system. This can be achieved by transforming the CD data to differential CD data (⌬(⌬A)) calculated by subtracting from each observed spectrum of the hostligand mixture the equivalent CD contribution of the ligand. The ⌬⑀ L [L T ] component of Equation 17, which is the CD of the ligand added to the host, is subtracted, and ⌬⑀ HL Ϫ ⌬⑀ H becomes the new ⌬⑀ HL . In this way the fitted or simulated Equation 18 resembles a Michaelis-Menten saturation curve rather than a change in slope of the curve generated by Equation 17.
It is crucial to terminate the titration upon reaching saturation. As a rule of thumb, to achieve saturation within the ligand excess used in the CD titration, the total host concentration [H T ] should be Ն30 to 50 times the value of K d . Obviously, for dissociation constant K d in the order of 10 Ϫ4 M, these conditions might not be achieved and hence saturation will not be reached, but as long as the plot ⌬A versus ligand concentration is of parabolic type, the K d can be still determined by Equation 18. For K d in the region between 10 Ϫ7 and 10 Ϫ6 M, the plot is of Michaelis-Menten type of shape with saturation that can be achieved at 1:1 molar ratio stoichiometry. In the case of a ligand with no CD, the CD changes associated with the host plateau upon ligand saturation. In the case of a ligand with CD, the saturation is normally achieved with to fit again the experimental CD data in order to determine a more accurate value of K and hence K d . This method has been used to calculate the K d value for 29 titrations of Hsp90 and several of its mutants with different ligands, such as ADP, ATP, AMP-PNP, Sti1, Cpr6, and geldanamycin (6). 2 The K d values determined from differential CD data were in very good agreement with those obtained from calorimetry. A similar approach has been used to determine the K d values of binary complexes by a non-linear regression method (27), using the difference CD obtained by subtracting the CD of the total added concentration of both host and ligand from the observed CD of the host-ligand complex.
Simulation of CD Spectra-To address the question of whether X and Y form a binary complex, three CD spectra were measured ((X), (Y), and (X ϩ Y)). Spectrum (X ϩ Y) is the observed spectrum of X ϩ Y, whereas spectrum ((X) ϩ (Y)) is the simulated spectrum calculated by adding the observed spectrum of X to that of Y. If the observed and simulated spectra are identical, one can readily infer that there is no detectable interaction between X and Y, whereas if the spectra are different, there is unambiguously a binding interaction and hence a binary complex.
To address the question whether X, Y, and Z form a ternary complex, eight CD spectra were measured ((X), is the observed spectrum of X ϩ Y, whereas spectrum ((X ϩ Y) ϩ (Z)) is the simulated spectrum calculated by adding the observed spectrum of the binary complex X ϩ Y to the observed spectrum of Z. Difference CD spectra {(X ϩ Y) Ϫ (X)} and {(X ϩ Z) Ϫ (X)} were calculated by subtracting the spectrum of the total concentration of (X) from the observed spectrum of the binary complexes (X ϩ Y) and (X ϩ Z), respectively. The observed spectra of (X ϩ Y ϩ Z) and (X ϩ Z ϩ Y) were compared with the simulated spectra ((X) ϩ (Y) ϩ (Z)), The simulated spectrum ((X) ϩ (Y) ϩ (Z)) is the sum of the spectra of each component as if no binding interactions were present among the three components X, Y, and Z. The simulated spectra ((X ϩ Y) ϩ (Z)), ((X ϩ Z) ϩ (Y)), ((X) ϩ (Y ϩ Z)) represent each of the three independent combinations of binary complexes together with the non-interacting third component. The simulated spectrum ((X ϩ Y) ϩ (Z)) can be seen as the spectrum obtained when component Z does not bind to the binary complex (X ϩ Y). The simulated spectra ((X ϩ Z) ϩ (Y)) and ((X) ϩ (Y ϩ Z)) represent the displacement of Y and X, respectively, from their binary (X ϩ Y) complex by ligand Z. Of course, if ligand Y and Z do not interact with each other, the simulated spectrum represents the spectrum as if a ternary complex was formed but without interactions between ligands Y and Z, namely without direct contacts. The difference CD components represent the CD changes associated with each binary complex. If none of the simulated spectra superimposes the observed spectra, it excludes all the possibilities described above. There is only one possibility left, which cannot be simulated, that all three components X, Y, and Z interact with each other upon forming the ternary complex. recruitment of client proteins to the Hsp90 complex prompted us to characterize the binding of Cdc37p and p50 cdc37 to Hsp90 using CD spectroscopy. As observed previously with Sti1/Hop/ p60 (6), the near-UV CD spectra obtained for mixtures of Hsp90 and Cdc37p cannot be simulated by linear combination of the individual spectra of Hsp90 and Cdc37p in isolation, indicating that Hsp90 and Cdc37p interact to form a complex (Fig. 1A). Titration of Cdc37p into Hsp90 produces dose-dependent perturbations in the near-UV region (240 -280 nM), and difference spectra (Fig. 1B) were obtained by subtracting the spectrum of the isolated co-chaperone (n molar equivalent) from the spectrum of Hsp90 ϩ co-chaperone (1:n) mixtures. Although saturable changes in the near-UV region were not achieved due to the relatively weak binding of Cdc37p to Hsp90, the observed signals are consistent with changes in the environment of aromatic residues due to molecular interaction. The K d value for the interaction of Cdc37p with Hsp90 was estimated as 113 Ϯ 4 M (Fig. 1C).

Stoichiometry of
As in the case for Cdc37p, titration of human p50 cdc37 into Hsp90 also produced dose-dependent perturbations in the near-UV region (240 -280 nM). The difference spectra ( Fig. 2A) showed changes that initially peaked at a molar ratio between 1:1 and 1:1.4 (Hsp90:p50 cdc37 ) and subsequently decreased at higher p50 cdc37 concentrations. The behavior of these spectra is consistent with two different association processes occurring. The first phase results from binding of p50 cdc37 to Hsp90 that saturates at Ϸ1:1, whereas the second phase results from selfassociation of excess p50 cdc37 . This interpretation is supported by the observation that the CD spectrum of p50 cdc37 is concentration-dependent, consistent with self-association (Fig. 2B). The K d value calculated for the interaction of p50 cdc37 with Hsp90 was 2.5 M (Fig. 2C) and that for p50 cdc37 dimerization was 9.8 Ϯ 1.9 M (Fig. 2D), assuming saturation of Hsp90 by p50 cdc37 at a 1:1 molar ratio. Because binding of the mammalian p50 cdc37 to Hsp90 was significantly tighter than that of its yeast orthologue Cdc37p, further analysis of interactions with Hsp90 was conducted mainly with p50 cdc37 .
Analytical Ultracentrifugation-To confirm and quantitatively analyze the self-association of p50 cdc37 observed in the CD studies, p50 cdc37 oligomerization was analyzed by equilibrium sedimentation in an analytical ultracentrifuge. Equilibrium data sets obtained were intermediate between theoretical traces calculated assuming ideal single species with molecular weights equal to one or two p50 cdc37 units. These data could be fitted with high confidence to an equation describing a monomer-dimer equilibrium mixture (Fig. 3) with a K d for the mon- It has been demonstrated previously (16,17) that Cdc37p/ p50 cdc37 possesses two separable domains as follows: an Nterminal domain capable of binding Hsp90-dependent protein kinases and a C-terminal domain that binds Hsp90 itself. To determine which domain was responsible for the inherent dimerization of p50 cdc37 , the separate N-and C-terminal domains of p50 cdc37 were subjected to analytical centrifugation in a similar way to the full-length protein. For both domains the data could only be fitted with high confidence to an equation describing monomer-dimer equilibrium mixtures (results not shown) indicating that homodimerization is a consequence of both N to N-terminal domain and C to C-terminal domain interactions. However, dimerization of the N-terminal kinasebinding domain was very weak with a K d estimated as 971 M (95% confidence limits ϭ 812-1160 M) and that for the Cterminal domain was 167 M (95% confidence limits ϭ 145-192), indicating that the main dimerization interface is between the C-terminal domains. However, it should be noted that the p50 cdc37 N-terminal domain used in the current stud-ies lacks the first 29 N-terminal amino acids which, although relatively poorly conserved, could contribute to dimerization.
Modulation of the Hsp90 ATPase Activity by Cdc37p/ p50 cdc37 -The similar recruiting roles that Cdc37p/p50 cdc37 and Sti1/Hop/p60 play in the Hsp90 system, and the fact that both are capable of dimerization and of binding to Hsp90 as dimers, led us to question whether, like Sti1/Hop/p60, Cdc37p/ p50 cdc37 could also inhibit the inherent ATPase activity of Hsp90. By using a regenerating enzyme-linked ATPase assay as described previously (3), we observed that both Cdc37p and p50 cdc37 can inhibit the ATPase activity of Hsp90 (Fig. 4, A and  B). At a given co-chaperone concentration, p50 cdc37 is a more potent inhibitor of the Hsp90 ATPase than Cdc37p, consistent with the higher affinity of p50 cdc37 for Hsp90 observed in the CD spectroscopy. As observed previously (6) with Sti1/Hop/p60, inhibition of ATPase activity by Cdc37p or p50 cdc37 could be reversed by competition with the TPR domain-binding co-chaperone Cpr6/ Cyp40 (Fig. 4, C and D). When the isolated N-or C-terminal domains of p50 cdc37 were added to Hsp90 ATPase assays, the N-terminal domain showed no inhibitory effect, whereas the Cterminal domain of p50 cdc37 was as potent as the full-length protein (Fig. 4, E and F). Localization of inhibitory activity in the  1 molar ratio), and the simulated spectrum for the mixture (SIM). The simulated spectrum is derived using the spectra of the individual components and is dissimilar to OBS, which indicates that Hsp90 and geldanamycin form a complex. Of particular interest are the spectral changes above 310 nm that are due to geldanamycin alone. B, near-UV difference spectra for titration of cSti1 into Hsp90 (2.42 M), obtained by subtracting the spectra of the free cSti1 at n molar equivalents, from Hsp90/cSti1 mixtures at Hsp90:cSti1 molar ratios of 1:n. The spectrum for Hsp90 is indicated, and curves 1-7 are from Hsp90/cSti1 mixtures in the molar ratio of 1:n (n ϭ 0.3, 0.8, 1.4, 2.0, 2.6, 3.1, and 5.0). Changes in the difference spectra are consistent with changes in the environment of aromatic residues due to molecular interaction. The saturable changes seen at a molar ratio close to 1:1 (curve 2 represents 1:0.8 and curve 3 represents 1:1.4 molar ratios of Hs90:cSti1) suggests that there are two cSti1-binding sites on Hsp90. Subsequent changes (dotted line spectra), seen above a molar ratio of 1:4, indicate that the excess cSti1 can homodimerize or that a second cSti1-binding site exists. C, estimation of the K d for the binding of cSti1 to Hsp90. The solid line represents the simulated curve that would be obtained if the titration were continued to obtain saturation of cSti1-binding sites and yields a K d of 0.1 M. D, near-UV spectra for Hsp90 (10 M), cSti1 (30 M), and geldanamycin (10 M), and spectra for mixtures of Hsp90 and geldanamycin (Hsp ϩ GELD, 1:1 molar ratio), Hsp90 and cSti1 (Hsp ϩ cSti1, 1:3 molar ratio), cSti1 and geldanamycin (cSti1 ϩ GELD, 3:1 molar ratio), and Hsp90, geldanamycin and cSti1 (OBS, 1:1:3 molar ratio). The simulated spectrum for the mixture of all three components (SIM1) is also shown and was derived from the spectra of the individual components and their spectra resulting from appropriate interactions. SIM2 represents the simulated spectrum of a cSti1 and geldanamycin mixture (3:1 molar ratio), and this spectrum exactly predicts the observed spectrum for this mixture (cSti1 ϩ GELD). Changes above 310 nm are due to geldanamycin, and a reduction in signal above 310 nm indicates that geldanamycin is expelled from Hsp90, and this is observed to occur in the presence of cSti1.
C-terminal domain of p50 cdc37 is consistent with its known function as the domain that interacts with Hsp90 (16).
Co-chaperone Interaction with ATP-binding Domain-Binding of geldanamycin to the nucleotide-binding site in the Nterminal domain of Hsp90 produces changes in the near-UV CD spectrum above 300 nm, where contributions from protein are negligible, so that the intensity of the difference CD in this region indicates the degree of saturation of the geldanamycinbinding site in the protein (6) (Fig. 5A). Geldanamycin CD thus provides a sensitive probe for measuring the accessibility of the nucleotide-binding pocket and for detecting interactions with co-chaperones or client proteins close to or in that pocket. By using this system we have shown previously that binding of Sti1/Hop/p60 displaces Hsp90-bound geldanamycin (6). As a C-terminal fragment of Sti1 (cSti1, residues 237-589) can reproduce suppression of Hsp90-ATPase by full-length Sti1/Hop/ p60, we investigated the effect of this fragment on geldanamycin binding. Titration of cSti1 into Hsp90 produced dosedependent perturbations in the near-UV region (240 -280 nm), and difference spectra (Fig. 5B) showed changes that initially peak close to a molar ratio of 1:1 (Hsp90:cSti1), as in the binding of full-length Sti1. However, unlike the full-length protein, titration with cSti1 results in further perturbations in the difference spectra above molar ratios of 1:1.4 (Hsp90:cSti1) that are consistent with cSti1 undergoing a second binding interaction (Fig. 5C) either by self-association, as seen with p50 cdc37 (Fig. 2B), or by binding to a second site on Hsp90. Because a second binding interaction was not observed previously (6) with Sti1, it is most likely that cSti1 is undergoing homodimerization at these concentrations. This situation is analogous to that observed for p50 cdc37 and suggests that the observed homodimerization of Sti1 also involves contributions from N-to N-terminal domain and C to C-terminal domain interactions. Changes in the difference CD spectra due to the dimerization of cSti1 were taken into account in subsequent experiments. Addition of cSti1 to geldanamycin-bound Hsp90 produces a decrease in the CD above 300 nm indicative of displacement of geldanamycin from the complex (Fig. 5D), as observed previously (6) with the full-length Sti1. This suggests a possible correlation between the ability of the Sti1 constructs to suppress Hsp90 ATPase activity and to interact with the nucleotide-binding pocket.
To determine whether the ATPase inhibitory properties of p50 cdc37 also involved interaction with the nucleotide-binding pocket, we examined the interaction of p50 cdc37 with geldanamycin and Hsp90. Whereas geldanamycin shows no interaction with Sti1 (Fig. 5D), initial experiments indicated that geldanamycin has a weak affinity for p50 cdc37 (Fig. 6A), and this interaction was therefore taken into account in subsequent experiments. Unlike the observation with Sti1 binding, the addition of p50 cdc37 to an Hsp90-geldanamycin complex did not diminish the difference CD signal above 300 nm, indicating that geldanamycin was not displaced (Fig. 6B). Furthermore, and in contrast to Sti1, the same spectrum was obtained whether geldanamycin was added to Hsp90 prior to or following addition of p50 cdc37 . The observed spectrum for the Hsp90/p50 cdc37 / geldanamycin mixture could not be simulated by any combination of spectra for a binary complex and one unbound component, nor could it be simulated by combination of spectra for binary interactions. This implies the presence of a ternary Hsp90-p50 cdc37 -geldanamycin complex in which additional changes have occurred above those resulting from binary interactions of the components.

DISCUSSION
Cdc37p/p50 cdc37 and Sti1/Hop/p60 have been widely implicated in recruitment of client proteins, such as protein kinases and steroid hormone receptors, respectively, into the Hsp90 chaperone system. Cdc37p/p50 cdc37 forms a direct structural bridge between Hsp90 and protein kinases (23, 28 -30), whereas Sti1/Hop/p60 couples Hsp90 to Hsp70 that is in turn bound to a steroid hormone receptor (7)(8)(9). Both co-chaperones bind at or near the C terminus of Hsp90, and their binding is mutually exclusive and exclusive of other TPR domain cochaperones such as Cpr6/Cyp40. The precise location of the Cdc37p/p50 cdc37 -binding site on Hsp90 is not yet known, but it does not involve the C-terminal MEEVD peptide that is essential , and a p50 cdc37 and geldanamycin mixture (OBS, 1:1 molar ratio). The simulated spectrum (SIM) for the mixture is derived from the spectra of the individual components. The simulated spectrum is dissimilar to OBS, which indicates that p50 cdc37 and geldanamycin (GELD) form a complex. Of particular interest are the spectral changes above 310 nm which are due to geldanamycin alone. B, near-UV CD spectra for mixtures of Hsp90 (90.3 M), p50 cdc37 (90.3 M), and geldanamycin (180.6 M). Spectra for a mixture of Hsp90, p50 cdc37 , and geldanamycin (OBS) could be superimposed irrespectively of the order of addition of the components to Hsp90. The simulated spectra for the mixture when one component remains unbound are represented by curves 1-3 (1, geldanamycin is unbound; 2, p50 cdc37 is unbound; 3, Hsp90 is unbound). Curve 4 represents no interactions between all the components, and curve 5 represents interactions between Hsp90 and geldanamycin as well as Hsp90 and p50 cdc37 . The observed spectrum for the mixture shows that geldanamycin remains bound to Hsp90 in the presence of p50 cdc37 (indicative by the absolute value at 310 nm). Furthermore, none of the simulated spectra match the observed spectra which indicates that a three-way complex is formed by these components that cannot be simulated by addition of spectra derived from two-component systems.
for binding the TPR domains of other co-chaperones (10,11,23,31). However, the mutual exclusivity of Cdc37p/p50 cdc37 and TPR domain co-chaperone binding certainly suggests a significant three-dimensional overlap between the binding sites.
Both Sti1/Hop/p60 and Cdc37p/p50 cdc37 bind to Hsp90 with a 1:1 molar stoichiometry, so that two molecules of the co-chaperone bind to the Hsp90 dimer. We had shown previously (6) that isolated Sti1 is a stable homodimer at a sub-micromolar concentration and binds to the Hsp90 dimer as such, simultaneously occupying both Hsp90 C-terminal TPR domain-binding sites. We have shown here that Cdc37p/p50 cdc37 also homodimerizes, so that it too binds as a dimer to the Hsp90 dimer. For both co-chaperones dimerization involves extensive homomeric interactions, so that the isolated Hsp90-interacting C-terminal fragments of both are also dimeric. Binding of Sti1/ Hop/p60 to Hsp90 had been shown previously (6) to inhibit the ATPase activity of Hsp90. We have shown here that Cdc37p/ p50 cdc37 also has this property. However, unlike Sti1/Hop/p60, inhibitory binding of Cdc37p/p50 cdc37 to Hsp90 does not cause displacement of bound geldanamycin from Hsp90, suggesting that there is no direct interaction between Cdc37p/p50 cdc37 and the N-terminal nucleotide-binding pocket. As with Sti1/Hop/ p60, the inhibitory activity of Cdc37p/p50 cdc37 is localized and can be reproduced by a C-terminal Hsp90-binding fragment of the co-chaperone, which also retains the ability to homodimerize. It is possible that the common ability of these two cochaperones to occupy both binding sites on the Hsp90 dimer simultaneously, and to homodimerize, plays a key role in their inhibitory activity. The ATPase cycle of Hsp90 is directly coupled to changes in the juxtaposition of the monomers within the dimer (12). Effective "cross-linking" of the Hsp90 monomers due to simultaneous binding of a dimerized co-chaperone would restrict these conformational changes and inhibit the coupled ATP turnover. In other words the binding of two monomeric units to each Hsp90 site would not restrict its conformational state. Support for this model comes from the behavior of the monomeric TPR domain immunophilin Cpr6. The binding of Cpr6 monomers and Sti1/Hop/p60 or Cdc37p/p50 cdc37 dimers to the C terminus of Hsp90 is competitive. However unlike the dimeric Sti1/Hop/p60 and Cdc37p/p50 cdc37 , Cpr6 shows no ability to dimerize and does not inhibit the ATPase activity of Hsp90 (6).
Sti1/Hop/p60 and Cdc37p/p50 cdc37 show no detectable homology but have nonetheless evolved a common ability to recruit client proteins to Hsp90 and to inhibit progress through the ATP-coupled chaperone cycle. The stages in Hsp90-dependent activation of steroid hormone receptors have been well characterized (32). Sti1/Hop/p60 is associated with the early phase of this process, recruiting Hsp70 and the bound client to Hsp90, and is replaced in subsequent stages by TPR domain immunophilins (33). Whereas the activation of protein kinase clients is far less well described, Cdc37p/p50 cdc37 clearly has an analogous function to Sti1/Hop/p60 in the early "loading" phase of client activation. The common ability of these recruitment factors to prevent ATP turnover suggests that efficient client protein loading requires Hsp90 to be in the "relaxed" conformation in which the N-terminal domains are unassociated, rather than the N-terminally associated conformation promoted by ATP binding (12) (Fig. 7).
At the present time the stoichiometry of binding between Cdc37p/p50 cdc37 and an Hsp90 protein kinase "client" has not been experimentally determined. However the simultaneous binding of two Cdc37p/p50 cdc37 molecules to Hsp90 raises the tantalizing prospect that two different protein kinases could theoretically be bound simultaneously to the Hsp90 complex, via independent interaction with the N termini of the two Cdc37p/p50 cdc37 co-chaperones. Thus in the activation of phosphorylation cascades, the Hsp90-Cdc37p/p50 cdc37 complex could simultaneously bind the client protein kinase requiring activation and the upstream kinase responsible for carrying out the phosphorylation that achieves that activation. There are in fact several examples of sequential pairs of protein kinases, which are implicated in interactions with Hsp90 and/or Cdc37p/p50 cdc37 . For example, in yeast cell cycle regulation the cyclin-dependent kinase Cdc28 and its activating kinase Cak1 both depend on Cdc37p for stability (34); in insulin and cell survival signaling both protein kinase B (PKB) and its activating kinase PDK1 are Hsp90-dependent (35,36); and in mitogenic signaling, mitogen-activated protein kinase kinase (MAPKK or Mek), its activating kinase Raf, and its activating kinase c-Src (29,(37)(38)(39) are all dependent on interaction with p50 cdc37 and Hsp90 for their function. Thus, the Hsp90-Cdc37p/ p50 cdc37 complex could act as a "scaffold" protein, co-localizing sequential kinases and thereby improving the efficiency of the phosphorylation reaction between an upstream kinase and its specific substrate kinase in vivo. As kinases from different pathways would also be brought into proximity by simultaneous binding to the Hsp90-Cdc37p/p50 cdc37 complex, such a mechanism could generate promiscuity and produce undesirable cross-talk between disparate signaling pathways. However, it is unlikely that scaffolding would overcome the inherent specificity of these phosphorylation reactions which is overwhelmingly governed by the specific amino acid sequences in the vicinity of the target serine, threonine, and tyrosine residues, and by direct specific interactions between docking sites on the sequential kinases. FIG. 7. Loading of client proteins into Hsp90 complex. Steroid hormone receptors (SHR) are delivered to Hsp90 bound to Hsp70 via interaction of Hsp70 with the Hsp90-Sti1/Hop/p60 complex. A variety of Hsp70 co-chaperones (including Ydj1, Hip, and Bag1) are consequently involved in coordinating the release of the SHR to Hsp90. In contrast, protein kinases bind directly to Cdc37p/p50 cdc37 . It is not presently known whether each Cdc37p/p50 cdc37 molecule in the functional dimer can recruit a protein kinase domain independently.