The amino-terminal domain of heat shock protein 90 (hsp90) that binds geldanamycin is an ATP/ADP switch domain that regulates hsp90 conformation.

Many functions of the chaperone, heat shock protein 90 (hsp90), are inhibited by the drug geldanamycin that specifically binds hsp90. We have studied an amino-terminal domain of hsp90 whose crystal structure has recently been solved and determined to contain a geldanamycin-binding site. We demonstrate that, in solution, drug binding is exclusive to this domain. This domain also binds ATP linked to Sepharose through the gamma-phosphate. Binding is specific for ATP and ADP and is inhibited by geldanamycin. Mutation of four glycine residues within two proposed ATP binding motifs diminishes both geldanamycin binding and the ATP-dependent conversion of hsp90 to a conformation capable of binding the co-chaperone p23. Since p23 binding requires regions outside the 1-221 domain of hsp90, these results indicate a common site for nucleotides and geldanamycin that regulates the conformation of other hsp90 domains.

Heat shock protein 90 (hsp90) 1 is a cellular chaperone that participates in multiple signal transduction pathways. Recent studies have demonstrated a requirement for hsp90, or grp94, its homolog in the endoplasmic reticulum, for the proper function of 1) the mitogen-activated protein kinase pathway (1-6); 2) activity of several tyrosine kinases (Refs. 7-9 and references therein); 3) activity of several transcription factors, including p53 (10), retinoid receptors (11), steroid and aryl hydrocarbon receptors (Refs. 12 and 13 and references therein), and hypoxia-inducible factor ␣ (14); 4) activity of the cyclin-dependent kinase CDK4 (15) and the cell cycle-associated Wee1 tyrosine kinase (16); and even 5) activity of hepatitis B virus reverse transcriptase (17). Additionally, hsp90 has been shown to participate in the refolding of certain misfolded proteins (18 -20). hsp90 comprises the core of several multi-molecular chaperone complexes that interact with proteins at different stages of their maturation. The ability of hsp90 to participate in the assembly of multiple higher order chaperone complexes no doubt contributes to its involvement in diverse cellular pathways, although those factors regulating such participation remain unclear.
Until recently, yeast in which hsp90 is either mutated or conditionally suppressed has served as the only means by which to study the many functions of this chaperone in the cell. The recent observation that a class of drugs known as benzoquinone ansamycins, including herbimycin A and geldanamycin (GA), specifically bind and inhibit hsp90 and grp94 has provided a new tool for functional studies of these chaperones (9,21). Indeed, a study of structure-activity relationships has demonstrated a high correlation between the biologic effects of the benzoquinone ansamycins and their ability to bind hsp90 (22). These drugs have also been shown to possess anti-tumor activity in preclinical models, identifying the hsp90 chaperone family as a novel target for anticancer drug development (23).
For these reasons, it is of much interest to characterize the drug binding site in hsp90, both to understand how GA interferes with hsp90 function and to allow for the design of novel chaperone antagonists with greater activity. Toward this end, Stebbins et al. (24) have recently analyzed the crystal structure of an amino-terminal fragment of hsp90. This revealed a binding pocket for GA, and the authors have proposed that this hydrophobic pocket is a site for interaction with denatured protein substrates. Interestingly, this pocket also contains a unique glycine-rich ATP binding motif, first identified in a bacterial topoisomerase (25). A recent study of the binding of hsp90 to one of its partner proteins, p23, has implicated a relationship between the binding of GA and nucleotides to hsp90 (26). ATP is essential for the formation of an hsp90 state that is able to bind p23, and this ATP effect is blocked by GA and also by ADP. However, the proposal that hsp90 is an ATP-binding protein (27)(28)(29) has met with resistance (30,31). Several conventional methods for demonstrating ATP binding have failed, and the use of fluorescent or photoaffinity labeling analogs has not been successful. Furthermore, hsp90 has a passive chaperoning activity, to retard the aggregation of denatured proteins, that does not require ATP (32,33).
In the present study, we confirm that, in solution, GA binds to the amino-terminal portion of hsp90. In addition, we show that the same region of hsp90 does indeed bind to ATP when the nucleotide is linked to a solid support through its ␥-phosphate. This interaction is blocked by GA, indicating a common site of interaction. Finally, mutation of several glycine residues within the ATP binding motif markedly diminishes GA binding to full-length hsp90, further supporting a common site of interaction of GA and nucleotides in the amino-terminal domain of this chaperone.

EXPERIMENTAL PROCEDURES
Synthesis of ATP-Sepharose-␥-Phosphate-linked ATP-Sepharose (available from Upstate Biotechnology, Inc.) was synthesized according to the protocol of Haystead et al. (34), which links adenosine-5Ј-(␥-4aminophenyl) triphosphate to Sepharose 4B (Pharmacia Biotech Inc.) through a six-carbon spacer. The efficiency of this coupling was calculated to be 15 mol of ligand/ml of Sepharose resin, as determined by analytical reverse-phase high performance liquid chromatography of unincorporated adenosine-5Ј-(␥-4-aminophenyl) triphosphate.
Hsp90 Purification-Human hsp90␤ was overexpressed in insect Sf9 cells following the protocol of Alnemri and Litwack (35). Purification was carried out as described previously (26) with the following changes in chromatographic procedures. After DEAE-cellulose fractionation, the hydroxylapatite column was replaced with a heparin-agarose column. Elution from heparin-agarose was with a 50 -500 mM KCl gradient. Those fractions containing hsp90 (as determined by SDS-PAGE) were then dialyzed in 10 mM Tris-HCl, 0.1 mM EDTA, 1 mM DTT (dithiothreitol), pH 7.5, diluted 1:2 with water, and loaded onto a Mono-Q fast protein liquid chromatography column (Pharmacia) for elution with a 0 -1 M KCl gradient. Finally, peak fractions from the Mono-Q column were pooled and fractionated by size on a Superdex 200 column (Pharmacia). The Superdex peak fractions were pooled and dialyzed in 10 mM Tris-HCl, 0.1 mM EDTA, 1 mM DTT, 50 mM KCl, 10% glycerol, pH 7.4, for storage at Ϫ70°C. It was assayed at Ͼ98% purity by scanning densitometry of a stained SDS-polyacrylamide gel.
C507 Purification-The C507 fragment of chicken hsp90 (described below) was overexpressed in Escherichia coli BL21(DE3)pLysS cells (Novagen) following Novagen's pET system protocols. Purification followed that described previously for full-length hsp90 (26) except that peak fractions from the Mono-Q column were then fractionated on a Superdex 200 column. The peak fractions from this column were dialyzed in 10 mM Tris-HCl, 0.1 mM EDTA, 1 mM DTT, 50 mM KCl, pH 7.4, and stored at Ϫ70°C. The purity of this preparation was comparable to that of the hsp90 preparation above.
Site-directed Mutagenesis-Plasmid WT7.11 containing the cDNA sequence of chicken hsp90 subcloned in the SP6 orientation into the BamHI and SphI sites of pGEM-7Z (Promega) was the starting material for mutagenesis. Single amino acids were changed using the polymerase chain reaction-based protocol of splice overlap extension (36). The amplification product was digested with BamHI and EcoRI, and the resultant fragment was purified and subcloned into similarly prepared pGEM-7Z. Correct mutant synthesis was confirmed by dideoxy sequencing of the plasmid inserts. Nomenclature for the mutant constructs follows the practice of listing the residue to be changed (using single-letter amino acid codes), the position of that residue on chicken hsp90␣, and the amino acid to which it was changed. For example, G94D refers to a glycine to aspartic acid change at position 94. The sole exception to this system is the mutant 3G3V, which replaces three glycines at positions 131, 134, and 136 with valines.
Deletion Mutagenesis-Full-length chicken hsp90 cDNA and the N221, N303, N380, and N538 amino-terminal deletion mutants, all subcloned into pGEM-4Z (Promega), have been described previously (37). The C507 mutant was created using polymerase chain reaction to insert a SalI restriction site in the full-length hsp90 cDNA following the codon for amino acid 221. Taking advantage of the SalI site in the multiple cloning region of pGEM-4Z (Promega), the plasmid was then digested with SalI and re-ligated. Thus, the C507 deletion mutant encodes only the amino-terminal 221 amino acids of hsp90. For expression of C507 in E. coli, the insert from this plasmid was excised with BamHI and SalI and ligated into pET23a (Novagen) that had been cleaved with the same enzymes.
ATP-Sepharose Column Elution-An ATP-Sepharose column was made by loading 0.5 ml of ␥-phosphate-linked ATP-Sepharose (equilibrated in 10 mM Tris-HCl, 50 mM KCl, 5 mM MgCl 2 , pH 7.5) into a 1-ml tuberculin syringe. The resin was then washed with 10 ml of elution buffer (10 mM Tris-HCl, 50 mM KCl, 20 mM MgCl 2 , 2 mM DTT, 20 mM Na 2 MoO 4 , 0.01% Nonidet P-40, Sigma, pH 7.5). One hundred g of purified hsp90 was loaded onto the column. Collection of 1-ml fractions began at this point. Nonbinding protein was washed from the column with 5 ml of elution buffer. Following this wash, an ATP gradient (0 -20 mM) was used to elute hsp90 from the ATP-Sepharose resin. All chromatography was performed at room temperature (22°C). Fractions were resolved using SDS-PAGE and analyzed for protein content by scanning densitometry of the stained gel. The ATP concentration of each fraction was determined by the measuring absorption at 259 nm.
ATP-Sepharose Binding Assay-The typical ATP-Sepharose binding assay was performed in 200 l of incubation buffer consisting of 10 mM Tris-HCl, 50 mM KCl, 5 mM MgCl 2 , 2 mM DTT, 20 mM Na 2 MoO 4 , 0.01% Nonidet P-40, pH 7.5. The assay tubes contained 25 l of pre-equilibrated ATP-Sepharose plus 5 g of the assay protein. Tubes were incubated at 30°C for 30 min, with mixing every 5 min to resuspend the resin. In competition assays, free nucleotide was included in the incubation buffer prior to addition of protein. Following incubation, samples were chilled on ice, and the resin was pelleted. The supernatant containing unbound protein was removed, and the ATP-Sepharose pellet was washed four times with 1-ml volume of ice-cold incubation buffer. Bound protein was eluted from the resin by boiling in SDS sample buffer for subsequent SDS-PAGE.
Use of GA-Affi-Gel Beads-Full-length chicken hsp90 (wild type), deletion constructs, and point mutants were transcribed/translated using TnT rabbit reticulocyte lysate (Promega Corp.) in the presence of translation grade [ 35 S]methionine (1458 Ci/mmol, ICN) using SP6 polymerase and following manufacturer's instructions. Following translation, 1 l was removed for analysis of translation efficiency, and the remaining lysates were diluted 1:20 with TNESV buffer (50 mM Tris-HCl, pH 7.4, 1% Nonidet P-40, 2 mM EDTA, 100 mM NaCl, 1 mM orthovanadate, 1 mM phenylmethylsulfonyl fluoride, 20 g/ml leupeptin, 20 g/ml aprotinin). The solid-phase GA binding assay was performed as described previously (21), using either GA-derivatized Affi-Gel beads or control Affi-Gel beads (blocked with glycine). In some cases, variable amounts of unlabeled GA or geldampicin were added to the lysate prior to addition of the GA-derivatized beads. Geldampicin is a GA derivative that is at least 2 logs less biologically active than GA and binds hsp90 very poorly (21). Both total translation extract (1 l) and the material bound to GA (or control) beads was subjected to polyacrylamide gel electrophoresis (7% gels). Gels were fixed, enhanced, dried, and exposed at Ϫ70°C for 1-24 h. Films were scanned into a Macintosh computer, and band densities were obtained using Adobe Photoshop and NIH Image software. GA bead binding data were corrected for the translation efficiency of each construct. The amount of binding to control beads was uniformly less than 1% the amount bound to GA-derivatized beads. In the data presented in Figs. 1 and 2, hsp90 binding to GA beads was quantified as follows. Material remaining bound to the beads after four washes in TNESV buffer was analyzed by SDS-PAGE. Gels were fixed, placed in plastic wrap, and exposed to film at 4°C for 24 -48 h. The developed film was used as a template to overlay the gel. Radioactive hsp90 bands were excised with a razor blade, minced, and placed in 10 ml of Aquasol scintillation fluid (NEN Life Science Products). Radioactivity was determined by scintillation counting and femtomoles of hsp90 bound to the beads was calculated by correcting for the specific activities of the individual hsp90 fragments translated. Representative specific activities of each construct, expressed as cpm ϫ 10 Ϫ3 /pmol, were as follows: wild type hsp90, 28 I] DMGA-Ten micrograms of hsp90 purified from bovine brain (Stressgen Biotechnology Corp.) was reacted in the dark with 2 Ci (29 nM) of [ 125 I]DMGA in 10 mM Tris-HCl, pH 7.5, for 10 min and then exposed to 254 nm ultraviolet light for 20 min, chilled on ice, and heat-inactivated at 95°C for 3 min. After SDS-PAGE the gel was autoradiographed, and using the radiograph as template, the radioactive hsp90 band was excised and subjected to limited proteolysis with V8 protease as described previously (21). After SDS-PAGE and transfer to a polyvinylidene difluoride membrane, the smallest labeled fragment also staining with Coomassie Blue dye (approximately 10 -12 kDa, estimated to contain approximately 20 -25 pmol of protein) was cut out and processed for protein micro-sequence analysis. Briefly, the photoaffinity labeled protein was reduced, alkylated, and digested with trypsin, and the resulting peptides were separated by reverse-phase high performance liquid chromatography using a C-8 column and a trifluoroacetic acid/water/acetonitrile gradient. The resulting pools were then analyzed for label, and the pool with activity was subjected to sequence analysis on a Beckman 2090 protein sequencer to determine the identity of the labeled peptide.
P23 Binding Assay-35 S-Labeled hsp90 point mutants were expressed in rabbit reticulocyte lysate using the TnT combined transcription/translation system (Promega). A 25-l aliquot was taken out and added to 175 l of buffer containing 10 mM Tris-Cl, 50 mM KCl, 2 mM DTT, 5 mM MgCl 2 , 0.01% Nonidet P-40, 20 mM Na 2 MoO 4 , pH 7.5, with or without 5 g/ml geldanamycin (3.6 M). The mixture was incubated for 5 min at 30°C. Then 5 mM ATP and an ATP regeneration system (10 mM phosphocreatine and 7 units of creatine phosphokinase) were added. To enhance hsp90 recovery, the reticulocyte lysate was supplemented with 1-2 g of purified, bacterially expressed human p23, prepared as described previously (26). This mixture was then incubated at 30°C for 30 min. After incubation, the reactions were chilled on ice and added to anti-p23 antibody JJ3 (39) bound to protein A-Sepharose for a 1-h immunoprecipitation on ice with frequent mixing. The resin pellets were then washed four times with 1-ml volume of 10 mM Tris-HCl, 50 mM KCl, 5 mM MgCl 2 , 1 mM DTT, pH 7.5. Bound proteins were eluted by boiling 2 min in SDS sample buffer and resolved by SDS-PAGE for subsequent transfer to nitrocellulose and autoradiography.

Analysis of Hsp90 Full-length and Deletion Mutant
Binding to GA Beads-Examination of the binding to GA beads of in vitro translated full-length hsp90 and several amino-terminal deletion mutants revealed that binding was entirely dependent on the amino-terminal portion of the chaperone (Fig. 1). In two independent experiments, we observed that deletion of the amino-terminal 221 amino acids from hsp90 abrogated GA binding. Additional amino-terminal deletions did not restore binding. When an hsp90 fragment containing only the aminoterminal 221 amino acids was tested (C507), this construct bound to GA-derivatized beads as efficiently as the full-length protein. These data thus localize the GA-binding site in solution to the amino-terminal 221 amino acids of hsp90 and confirm the recent structural studies of Stebbins et al. (24).
To characterize further the binding of wild type hsp90 and the C507 deletion mutant, we examined the ability of soluble GA or soluble geldampicin to block binding to solid phase GA. The data in Fig. 2A demonstrate that soluble GA is able to block the binding of wild type hsp90 to GA-derivatized beads with an IC 50  FIG . 2. The specificity of hsp90 binding to geldanamycin. Radiolabeled hsp90 was prepared by in vitro translation and aliquoted to several tubes containing GA beads. A shows the competitive effect of increasing concentrations of soluble GA or geldampicin (GM) on the binding of full-length hsp90 to GA beads. B, shows a similar experiment using the C507 hsp90 fragment. The quantitation of GA binding was performed as described in Fig. 1.
potencies of these two ansamycins. 2 When the C507 hsp90 fragment was tested in a similar fashion (using a different preparation of GA-derivatized beads), soluble GA inhibited binding at an IC 50 of between 0.4 and 0.5 M, whereas soluble geldampicin at 10 M had no inhibitory effect (Fig. 2B).
Affinity Labeling Analysis of the Hsp90 GA-binding Site-To confirm and extend the data obtained with hsp90 deletion mutants, we utilized a previously described iodinated, photoaffinity label derivative of GA (38). After incubating purified hsp90 with this reagent and subjecting the labeled hsp90 to limited proteolysis with V8 protease, we obtained a 10-kDa labeled peptide. Microsequence analysis of the first 10 amino acids of this hsp90 fragment yielded a predicted 10,032-dalton peptide of the sequence, TLTDPSKLDSGKELHINL-IPNKQDRTLTIVDTGIGMTKADLINNLGTIAKSGTKAFM-EALQAGADISMIGQFGVGFYSAYLVAEKVTVITKHND. This sequence lies within the amino-terminal fragment previously identified, corresponding to amino acids 62-155 of chicken hsp90. It is also 80% homologous to amino acids 137-218 of human GRP94.
Binding of hsp90 to ATP-Sepharose-A previous study examining the possibility of ATP binding by hsp90 demonstrated that, unlike the known ATP-binding protein hsc70, hsp90 did not bind specifically to ATP immobilized on agarose (30). However, linkage of ATP to the agarose substrate was via C-8 of adenine, creating a steric barrier to any proteins that require the availability of this moiety for ATP binding. Negative results were also obtained using ATP-Sepharose with linkage from the ribose. Crystallographic data show that some protein kinases, including cyclic A kinase, bind ATP through adenine, leaving the phosphate groups exposed (40,41). We have previously described the synthesis of ␥-phosphate-linked ATP-Sepharose, a resin that permits this type of binding (34).
Here, we sought to determine if a column formed of ␥-phosphate-linked ATP-Sepharose is capable of specifically retaining hsp90. As is shown in Fig. 3, hsp90 remained bound to the column through a wash of 10 resin bed volumes. It was then eluted as a discrete peak by an increasing ATP gradient. Elu-tion began at an ATP concentration of 0.5 mM and was complete by 5 mM. A similar binding and elution profile was observed for the C507 fragment of hsp90 (not shown). Thus, hsp90 is capable of binding to immobilized ATP. Our results suggest that the negative result seen in the earlier study with C-8-linked ATPagarose was due to steric interference caused by the adenine linkage.
Previous work on the formation of hsp90⅐p23 complexes from purified proteins showed that this process requires ATP, Mg 2ϩ , and elevated temperature (26). In addition, complex formation is greatly enhanced by the addition of molybdate ions and the nonionic detergent Nonidet P-40. Inhibitors of this association include ADP and GA. Using these conditions as a starting point, we examined hsp90 binding to the ␥-phosphate-linked ATP-Sepharose resin.
First, resin binding of purified hsp90 and the GA-binding C507 fragment were compared with that of ovalbumin, a protein without known ATP binding activity (Fig. 4A). Under conditions optimal for hsp90⅐p23 complex formation (except for the absence of free ATP), hsp90 and the C507 fragment bound quantitatively to the ATP resin (compare bound protein in lanes 1 and 2 to the protein load in lanes 9 and 10). Ovalbumin, on the other hand, showed no binding under these conditions (lane 3 versus lane 11). Similar to what was seen with hsp90 binding to p23, binding to the ATP-resin required Mg 2ϩ . No binding was observed in magnesium-free buffer (lane 7), and increasing the Mg 2ϩ from 5 to 20 mM had no effect (lane 8). However, in contrast to hsp90 binding to p23 that requires elevated temperature, hsp90 binding to ATP-Sepharose occurred at both 4 and 30°C (lanes 1 and 5).
The concentration of hsp90 in these experiments was 0.3 M, and the ATP concentration of the resin was 15 mM (which is then diluted 9-fold by the incubation buffer). If hsp90 were binding to the resin through an ATP-binding site, then free ATP should be able to compete with the resin-bound ATP for hsp90. This is shown in lane 4, where the addition of 10 mM free ATP was able to prevent binding to the resin. Ten g/ml GA (17.9 M), which inhibits hsp90⅐p23 complex formation, also was effective in preventing hsp90 from binding to the resin (lane 6). Similarly, prior binding of C507 fragment to GA blocked its binding to ATP-Sepharose (not shown).  3. The binding of hsp90 to ATP-Sepharose. Purified hsp90 (100 g) was loaded on a 0.5-ml column of ATP-Sepharose, and 1-ml fractions were collected. After washing, the hsp90 was eluted using a gradient of ATP. The protein was detected by SDS-PAGE and quantitated by densitometry.
It has been shown that nucleotide effects on hsp90 binding to p23 and to hydrophobic resins are specific for adenine di-and trinucleotides (26,42). We sought to further characterize this specificity by comparing how well different nucleotides are able to compete with ATP-Sepharose for hsp90 binding (Fig. 4B). Of the four nucleotides tested, only ATP and ADP were effective in reducing hsp90 binding to the ATP resin. AMP and GTP had only minimal effects at 5 mM, the highest concentration used. In our earlier work, ADP was demonstrated to be a strong inhibitor of ATP-dependent p23 binding at 1/10th of the ATP concentration (26). This correlates well with the current finding that ADP is approximately 10 times more effective than ATP at competing for hsp90 binding.
In related experiments we have shown that ␥-phosphatelinked ATP-Sepharose can be utilized to affinity purify native hsp90 from crude cell extracts. Following application of extract to the column a pure preparation of hsp90 can be obtained by selective elution with Mg/ADP. Elution of the resin with any other nucleotide (other than ATP) fails to elute the native form of hsp90. 3 Effect of Point Mutations within and Near the Hsp90 GAbinding Domain on Binding to GA Beads-Several point mutants within and near the GA-binding domain were analyzed for their ability to bind to GA beads (Fig. 5). In two independent experiments, replacement of glycine at position 94 with aspartic acid (G94D) and replacement of the three glycine residues at positions 131, 134, and 136 with valine (3G3V) led to markedly diminished binding to GA beads. Replacement of glycine at position 182 with aspartic acid (G182D) also significantly reduced binding. Interestingly, replacement of just the glycine at position 131 with valine had no effect on binding to GA beads nor did mutation of glycine at position 113, arginine at position 181, or lysine at position 190. Replacement of lysine at position 111 with alanine (K111A) resulted in an intermediate but reproducible reduction in binding.
Binding of Hsp90 Mutants to p23-When the hsp90 mutants described above were tested for binding to ATP-Sepharose, all of the mutants bound to an extent comparable to wild type hsp90 (results not shown). To pursue this further, we tested the effects of mutation on an ATP-dependent function of hsp90.
We have previously analyzed the effect of hsp90 deletions on its binding to the progesterone receptor (37). One of the other components in activated progesterone receptor complexes is p23, and this has been shown to form a complex with hsp90 in an ATP-dependent fashion (26). Analysis of the effect of these deletions on p23 binding has shown that the amino-terminal region of hsp90 is essential for binding, but binding to p23 is not observed using the C507 fragment of hsp90. 4 To better define the p23 binding requirements and the role of ATP, p23 binding was studied using the point mutants tested above. These were transcribed and translated in rabbit reticulocyte lysate, and their binding to p23 (endogenous reticulocyte p23 supplemented with purified, bacterially expressed human p23) was analyzed by p23 immunoprecipitation. As is shown in Fig.  6, wild type hsp90 coprecipitates with p23 but not in the presence of GA (lanes 1 and 2). Point mutants G113D and K190A (lanes 4 and 8) both show p23 binding comparable (although diminished, Fig. 6B) to that of wild type hsp90, whereas mutants G94D, G131D, 3G3V, and G182D (lanes 3, 5, 6, and 7) all show negligible binding. Mutants K111A and R181Q showed no effect on p23 binding. The conservative nature of the 3G3V substitutions (three valines for three glycines) indicates that a very specific interaction is taking place at this location. Also, the loss of p23 binding with one glycine to aspartic acid change (G94D or G131D) but not another at a nearby site (G113D) demonstrates that these effects are unlikely to be the result of gross changes in overall protein structure.
An assessment of the mutation results is presented in Fig. 7. This shows the sequence of chicken hsp90␣ that is 96% identical to human hsp90␣ in this region. The major residues involved in forming the GA binding pocket and the adjacent surface groove described by Stebbins et al. (24) are indicated, as well as two ATP binding motifs relating to topoisomerase II (25). Two of four mutations that strongly reduce GA binding (asterisks) lie within proposed ATP-binding regions and in the surface groove or GA binding pocket. Similarly, three of four mutations affecting p23 binding (number symbols) are located in these described domains. Of the three mutations not affecting either GA or p23 binding, none lie in the ATP-binding region or the groove and pocket domain. DISCUSSION Although the crystal structure of hsp90 has not yet been solved, two laboratories have prepared crystals of the aminoterminal domain of human hsp90 (24) and yeast hsp90 (43). General agreement was obtained in the structural analysis from these two species even though the yeast preparation was dimeric and the human was monomeric. The study by Stebbins et al. (24) revealed a 15-Å deep pocket in this domain that bound GA. The present study confirms the localization of GA binding to the amino-terminal domain of hsp90 in solution and shows a lack of GA binding to other domains of the protein. The binding of both full-length, wild type hsp90 and the C507truncated amino-terminal fragment to immobilized GA is readily competed by soluble GA, but not by soluble geldampicin, an inactive benzoquinone ansamycin. Furthermore, using a GA derivative containing a photoaffinity label, we identified a 10-kDa hsp90 peptide that, by partial microsequence analysis, maps to a 94-amino acid region within the amino-terminal fragment and that contains the hydrophobic GA binding pocket identified by Stebbins et al. (24). Our data also identify several amino acid residues within the GA-binding domain of hsp90 that appear critical for its binding to immobilized GA.
Because of the hydrophobic nature of the GA binding pocket, Stebbins et al. (24) proposed that it is a binding site for unfolded proteins. GA has been shown to inhibit numerous systems that involve the binding of hsp90 to protein substrates. For example, it destabilizes the complexes between hsp90 and mutant p53 (44), Raf-1 (5,45), the tyrosine kinases pp60 v-src (21) and p56 lck (46), steroid receptors (42,47,48), and the hepatitis B virus reverse transcriptase (17). Although these results suggest that GA binds to a site of protein-protein interaction, based on our data, it now appears more likely that the GA binding pocket is actually a site for binding nucleotides. This raises an important concern regarding the specificity of GA action. Most existing studies suggest that GA is highly specific, and its effects can be rationally linked to hsp90 func-FIG. 6. The binding of hsp90 mutants to p23. Radiolabeled hsp90 products were prepared by in vitro translation. These were incubated further in supplemented reticulocyte lysate to allow binding to p23. After immune precipitation using an antibody to p23, the bound proteins were resolved on SDS-PAGE and quantitated by densitometry. The upper panel shows the total loads and bound protein bands from one experiment. The results from two independent experiments are shown in the lower panel. Note that mutant R181Q was only tested once. tion. However, our data strongly suggest that GA interacts with hsp90 at a nucleotide-binding site that is not only highly conserved in members of the hsp90 family but also in type II DNA topoisomerases, and the product of the hereditary nonpolyposis colon cancer gene (MLH1), a protein homologous to the bacterial mismatch repair protein MutL (49,50). Thus, hsp90 may be only one of several targets for this drug.
A relationship between GA and nucleotide interactions with hsp90 was demonstrated in an earlier report that proposed a two-state model of hsp90 regulated by the binding of ATP or ADP (26). GA was shown to antagonize ATP-dependent hsp90 function but not the ADP-dependent activity of the chaperone. However, it was not known whether GA and nucleotide bound to a common site on hsp90. Bergerat et al. (25) have noted sequence similarity between the amino terminus of hsp90 and three regions involved in ATP binding in topoisomerase II. Two of these regions lie within the hydrophobic pocket identified by Stebbins et al. (24). Gerloff et al. (51) also noted a topoisomerase (DNA gyrase B) relationship using structural predictions for the amino terminus of hsp90.
The binding of nucleotides to hsp90 has been an area of much controversy. Earlier claims (27)(28)(29) that hsp90 could bind ATP affinity resins and the photoaffinity analog, 8-azido-ATP, and that hsp90 contained an ATPase activity have not been reproduced in other laboratories, and a systematic study by Jakob et al. (30) failed to detect such activities. On the other hand, the binding of hsp90 to p23 is clearly modulated by nucleotides (26,42). p23 is a small, ubiquitous phosphoprotein that is bound to hsp90 in cell lysates. Although its function is unknown, p23 may be a co-chaperone with hsp90, and it clearly potentiates complex formation between hsp90 and steroid receptors (52,53), between hsp90 and hepatitis B virus reverse transcriptase (54), and between hsp90 and mutant p53 (44). ATP is required for p23 binding to hsp90, and this effect is inhibited by both ADP and GA.
In the current experiments, we immobilized ATP by linking its ␥-phosphate to Sepharose via a six-carbon spacer, thus allowing free access to the ribose and adenine moieties. We believe that this orientation is critical for permitting hsp90 binding, since we and others 4 (30) have failed to detect hsp90 binding to ATP immobilized via either its ribose or adenine constituents. If the ribose and base are primary sites of interaction with the hsp90-binding site and thus cannot be modified substantially, these data would not be unexpected. This would also explain why hsp90 binding could not be detected using several fluorescent or photoaffinity labeling analogs (30). Such characteristics of an ATP-binding site are not unusual and have been observed with many protein kinases (34).
Based on our data, we propose that ATP binds in the pocket that has been described for GA with the more hydrophobic adenine moiety binding within the pocket and the more polar sugar and phosphate groups lying closer to the pocket surface. This is consistent with a space-filling model of adenylic acid (AMP), which adopts a "C"-like configuration (55) highly analogous to that reported for hsp90-bound GA (24). In this case, the adenine base is equivalent to the ansa ring of GA, forming the bottom and stem of the C, whereas the ribose and phosphate are equivalent to the benzoquinone moiety of GA and form the top of the C. At the very bottom of the C is the amine linked to the C-8 carbon of the adenine ring. Stebbins et al. (24) have identified the carbamate moiety on the ansa bridge of GA as making critical hydrogen bond contacts deep within the primarily hydrophobic pocket. Under physiologic conditions, this carbamate probably exists in a protonated form, and the bottom of the pocket contains a buried cavity holding 3 water molecules. Stebbins and colleagues propose that the natural occupant of this site, unlike GA, might fill the cavity completely, eliminating the water. Our data suggest that, on the contrary, ATP fills the cavity in a fashion quite similar to GA, with the protonated amine of the adenine forming critical hydrogen bond contacts at the base of the pocket. Structural analysis of the ATP-binding site on the DNA gyrase B protein of E. coli has identified a similar mostly hydrophobic adenine binding pocket (56). The existence of a GXXGXG motif (one of the ATP binding motifs homologous to that described by Bergerat et al. (25) and also found in DNA gyrase B) near the opening of the pocket allows for close contact of the phosphates with the ATP-binding site (56).
Mutation of the three glycine residues in the GXXGXG motif (corresponding to Gly-132, Gly-135, and Gly-137 of human hsp90) markedly diminished the binding of the triple point mutant (3G3V) to GA beads. Although these amino acids are either within the hydrophobic pocket or at its surface (see Fig.  7), their importance for GA binding to hsp90 was not predicted by structural analysis. However, the inability of the triple mutant to bind p23 in the presence of ATP emphasizes the importance of this domain within the hydrophobic pocket in modulating the conformation of distal domains of hsp90 (see below) and further supports co-localization of GA-and ATPbinding sites. p23 binding appears to be more sensitive to alterations in the glycine-rich motif since the single mutant G131D inhibited the ATP-dependent binding of p23 but not GA binding. The fact that mutant 3G3V retains the ability to bind to ATP-Sepharose (data not shown) is surprising. This might be explained by a major role of adenosine interactions, rather than phosphate interactions in the ATP binding. Also, because of the mode of ATP attachment, the binding of hsp90 to ATP-Sepharose may differ significantly from that to soluble ATP, with a lesser participation of phosphate binding on the affinity resin. Mutation of glycine in positions 94 or 182 (G94D and G182D, corresponding to Gly-95 and Gly-183, respectively, of human hsp90) is predicted from structural analysis (24) to affect GA binding to the chaperone. Our data confirm this prediction and further demonstrate loss of ATP-dependent p23 binding to these mutants.
These data, together with those presented in a recent study (26), suggest that the ATP/GA-binding site acts as a conformational switch to regulate the assembly of hsp90-containing multi-chaperone complexes. For example, in its ATP-bound state, hsp90 can interact with p23, whereas in its ADP-bound state it cannot do so. GA blocks p23 binding to hsp90 but not by directly competing for a p23-binding site (which does not exist in the amino-terminal portion of hsp90). 4 Instead, we propose that GA binding to hsp90 locks the chaperone into its ADP-dependent configuration. This hypothesis predicts that hsp90 binding phenomena that are not ATP-dependent, or that are stimulated by ADP, would not be affected by GA. Indeed, hsp90 binding to hydrophobic resins, which is inhibited by ATP, is favored by the presence of either ADP or GA (26). This hydrophobic binding activity, which has been suggested to identify a site for protein substrates (57), also exists outside of the aminoterminal domain. 4 Thus, the nucleotide/GA-binding site in the amino-terminal domain of hsp90 is capable of modifying the properties of other regions of the molecule. An essential task for future research will be to identify these other regions of hsp90 interaction and to provide a structural basis for their regulation.