The Heat Shock Protein 90 Antagonist Novobiocin Interacts with a Previously Unrecognized ATP-binding Domain in the Carboxyl Terminus of the Chaperone*

Heat shock protein 90 (Hsp90), one of the most abundant chaperones in eukaryotes, participates in folding and stabilization of signal-transducing molecules including steroid hormone receptors and protein kinases. The amino terminus of Hsp90 contains a non-conventional nucleotide-binding site, related to the ATP-binding motif of bacterial DNA gyrase. The anti-tumor agents geldanamycin and radicicol bind specifically at this site and induce destabilization of Hsp90-dependent client proteins. We recently demonstrated that the gyrase inhibitor novobiocin also interacts with Hsp90, altering the affinity of the chaperone for geldanamycin and radicicol and causing in vitro and in vivo depletion of key regulatory Hsp90-dependent kinases including v-Src, Raf-1, and p185ErbB2. In the present study we used deletion/mutation analysis to identify the site of interaction of novobiocin with Hsp90, and we demonstrate that the novobiocin-binding site resides in the carboxyl terminus of the chaperone. Surprisingly, this motif also recognizes ATP, and ATP and novobiocin efficiently compete with each other for binding to this region of Hsp90. Novobiocin interferes with association of the co-chaperones Hsc70 and p23 with Hsp90. These results identify a second site on Hsp90 where the binding of small molecule inhibitors can significantly impact the function of this chaperone, and they support the hypothesis that both amino- and carboxyl-terminal domains of Hsp90 interact to modulate chaperone activity.

Functional analysis has revealed that Hsp90 is composed of well conserved amino-and carboxyl-terminal regions separated by a charged domain (13,14). The crystal structure of an amino-terminal fragment of Hsp90 complexed with ATP has recently been solved (15); ATP binding to this domain of Hsp90 has also been detected biochemically (13,16), and the functional importance of ATP hydrolysis for Hsp90 activity has been characterized (17,18). X-ray crystallographic and biochemical studies have also demonstrated that the natural products geldanamycin (GA) 1 and radicicol both interact with the amino-terminal nucleotide-binding site on Hsp90 (16,17,19,20), leading to alterations in the conformation and function of the protein. In contrast to the amino terminus, the crystal structure of the carboxyl-terminal region of Hsp90 remains undetermined, and biochemical characterization of this portion of the chaperone suggests that a complex interaction between this and the amino-terminal domain is a critical regulatory component of chaperone function (14,(21)(22)(23).
We recently reported that novobiocin, an antibiotic previously shown to bind adjacent to the ATP-binding site of bacterial gyrase B and to interfere with nucleotide binding (24), was also able to interact with Hsp90, albeit with lower affinity than with gyrase B, and to disrupt the chaperone activity of Hsp90 in a manner similar to GA and radicicol (25). Thus, cells exposed to novobiocin demonstrated rapid destabilization of various Hsp90 client proteins, including Raf-1, mutated p53, p60 v-Src , and p185 ErbB2 . Unexpectedly, however, although novobiocin antagonized Hsp90 binding to both GA and radicicol, the Hsp90binding domain of novobiocin appeared to be distinct from the amino-terminal GA/radicicol/nucleotide-binding domain (25).
In the current study, we have localized the Hsp90-binding site of novobiocin to a region overlapping the carboxyl-terminal dimerization domain of the chaperone. The current data, together with our previous study (25), emphasize the functional significance of novobiocin binding to Hsp90, support the possibility that the conformation of Hsp90's amino terminus can influence the behavior of the carboxyl-terminal portion of the molecule, and suggest that a previously unrecognized second nucleotide-binding domain exists in the carboxyl terminus of Hsp90.
In Vitro Transcription and Translation-Recombinant proteins were expressed from 1 g of plasmids by in vitro transcription/translation using the TNT rabbit reticulocyte lysate kit (Promega Corp.) in the presence of translation grade [ 35 S]methionine (1458 Ci/mmol, ICN, Costa Mesa, CA), with the appropriate DNA polymerase and following the manufacturer's instructions. Aliquots of each reaction (1 l) were checked for translation efficiency, and 10 -20 l were diluted in the appropriate buffer and incubated with either novobiocin-Sepharose or ATP-Sepharose as described below.
Preparation of Novobiocin-Sepharose 6B and the Solid Phase Novobiocin Binding Assay-Novobiocin-Sepharose was prepared as described previously (25,27). 10 -20 l of in vitro translate or 4 g of purified GST-Hsp90 proteins were diluted in 300 l of buffer B (25 mM Hepes, pH 8, 1 mM EDTA, 10% ethylene glycol, 200 mM KCl) and incubated with novobiocin-Sepharose (100 l), with or without preincubation (60 min at 4°C) with various drugs, peptides, or ATP. Incubations were carried out with tumbling for 1 h at 4°C. The beads were then thoroughly washed with cold buffer B containing 0.6 M KCl, boiled for 5 min in reducing Laemli loading buffer, and analyzed by SDS-PAGE and Western blotting with appropriate antibodies or by autoradiography of dried gels.
ATP-Sepharose Binding Assay-␥-Phosphate-linked ATP-Sepharose (Upstate Biotechnology, Inc., Lake Placid, NY) was equilibrated in molybdate buffer (10 mM Tris-HCl, 5 mM MgCl 2 , 10 mM Na 2 MoO 4 , 0.2% Tween 20, pH 7.5). Aliquots of in vitro translates (10 -20 l) or 4 g of purified GST-Hsp90 proteins were diluted in 300 l of molybdate buffer and mixed, while tumbling, with ATP-Sepharose (100 l) for 2 h at 4°C. Samples were washed 3-4 times with cold molybdate buffer, and bound proteins were eluted and analyzed as described above. To study nucleotide competition of binding to ATP-Sepharose, we incubated 4 g of purified GST-C3 with increasing concentrations of the sodium salts of ADP, ATP, CTP, and GTP (all from Sigma) for 1 h at 4°C on a rotary shaker prior to assessment of ATP-Sepharose binding.
Co-immunoprecipitation of Hsc70 and p23 with Hsp90 -For investigation of the influence of novobiocin on Hsp90 association with the co-chaperones Hsc70 and p23, 1 mM novobiocin was added to reticulocyte lysate. After 30 min on ice, samples (25 l) were diluted in 400 l of buffer containing 10 mM Tris-HCl, 1 mM MgCl 2 , 0.2% Tween 20, 10 mM Na 2 MoO 4 , pH 7.5, and incubated for 2 h at 4°C with 2 g of an Hsp90 antibody (SPA-771, Stressgen) specific for the Hsp90 amino terminus. Protein A-Sepharose (Amersham Pharmacia Biotech) was added to all samples, which were then rotated for an additional 2 h. Sepharose beads were washed 3-4 times with dilution buffer and processed for SDS-PAGE as described above. Hsp90 immunoprecipitates were analyzed by Western blotting with antibodies specific for either Hsc70 or p23.
Antibodies-Hsp90 antibodies recognizing epitopes in either the

FIG. 1. Binding of Hsp90 point mutants and amino-terminal truncations to novobiocin-Sepharose. Fulllength (A) or amino-terminal (B) [ 35 S]
methionine-labeled chicken Hsp90 wild type (Wt) and point mutated proteins were translated in rabbit reticulocyte lysate as described under "Materials and Methods," and aliquots of protein (equal radioactivity) were incubated with novobiocin resin in buffer B for 1 h at 4°C and then thoroughly washed with buffer B containing 0.6 M KCl. Proteins bound on beads were analyzed by SDS-PAGE and autoradiography. Binding of the same lysates to GA-Sepharose beads is shown for comparison.

FIG. 2. Map of the chicken Hsp90
constructs employed in the novobiocin and ATP binding studies, and their relative binding efficiency to novobiocin-Sepharose. Wild type (Wt) chicken Hsp90 consists of 728 amino acids. Various mutated amino acids are marked by an X; removal of amino acids 657-677 is symbolized by dots.
Peptide Synthesis-Selected peptides were synthesized by Research Genetics (Huntsville, AL) using solid phase peptide synthesis. Both wild type and mismatched sequences were checked by Edman chemistry in a peptide sequencer, and peptide purity was determined by mass spectroscopy. All peptides were soluble in water.
Western Blotting-Immunoprecipitates or 10 -20 l of translation lysate were separated on 10% SDS-polyacrylamide gels, transferred to nitrocellulose membranes by electroblotting, and blocked for 2 h with a solution containing 5% nonfat dry milk, 10 mM Tris-HCl, pH 7.5, 2.5 mM EDTA, pH 8, 50 mM NaCl, and 0.05% Tween 20. The membranes were probed with indicated primary antibodies, followed by secondary antibodies conjugated to horseradish peroxidase, and signals were detected using chemiluminescence reagents (Pierce).

Effect of Point Mutations within the Amino-terminal Hsp90
Nucleotide/GA/Radicicol-binding Domain on Hsp90 Binding to Novobiocin-Sepharose-Several amino-terminal point mutations known to abrogate GA and radicicol binding to Hsp90 were tested for their ability to affect binding to immobilized novobiocin. Previous analysis demonstrated that the mutants D92A, G94D, G113D, G131/4/6V, and G182D each displayed markedly impaired GA and radicicol binding (16,20). In contrast, with the exception of G113D, these mutants bound to novobiocin-Sepharose as well as or better than did wild type Hsp90 (Fig. 1A). Additionally, the amino-terminal Hsp90 fragment containing the GA/radicicol-binding site (amino acids 1-222) failed to bind to immobilized novobiocin (Fig. 1B, and (25)). Although the strongest binding to immobilized novobiocin was observed with the full-length point mutants K111A and G182D, the same mutations when made in the amino-terminal fragment of Hsp90 did not result in binding of this region to novobiocin (Fig. 1B).
Binding of Hsp90 Carboxyl-terminal Truncation Fragments to ATP-Sepharose-Since novobiocin competitively inhibits ATP binding to gyrase B, and because ATP inhibited Hsp90 binding to novobiocin-Sepharose, we tested whether the several carboxyl-terminal Hsp90 truncations that demonstrated novobiocin binding also bound to immobilized ATP in a manner competable by novobiocin. ⌬1 did not bind to ATP-Sepharose, as predicted from its failure to bind to novobiocin, but ⌬2, ⌬3, and ⌬4 did bind to immobilized ATP (Fig. 4A). Also predicted from its failure to bind to novobiocin, ⌬3⌬ did not bind to ATP-Sepharose. Finally, soluble ATP and novobiocin both com-petitively inhibited Hsp90 carboxyl-terminal fragment binding to ATP-Sepharose (Fig. 4, B and C, and data not shown).
Inhibition of Carboxyl-terminal Hsp90 Truncation Fragment Binding to Novobiocin-Sepharose by a Soluble Peptide-Because deletion of amino acids 657-677 markedly reduced the binding of ⌬3 and ⌬4 to immobilized novobiocin, we synthesized a peptide encompassing this sequence in order to determine if this region alone were sufficient to bind to novobiocin-Sepharose. A peptide of sequence YETALLSSGFSLED, corresponding to amino acids 663-676 of Hsp90, was able to inhibit novobiocin binding to ⌬3, ⌬4, and the full-length G182D point mutant (Fig. 5). In contrast, a mismatched peptide, of sequence YTEFLSASLEGLSD, had minimal effect even at the highest concentrations used (Fig. 5).
Purified GST-Hsp90 Proteins Bind to Novobiocin-and ATP-Sepharose-In order to exclude spurious results due to possible ]methionine-labeled ⌬3 was incubated as described under "Materials and Methods" with increasing concentrations of soluble novobiocin prior to addition of novobiocin-Sepharose. The percent of ⌬3 that bound to novobiocin-Sepharose was plotted against the concentration of soluble novobiocin competitor. The proteins were synthesized, incubated with the novobiocin resin, processed, and analyzed as above.
dimerization between the various in vitro translated chicken Hsp90 constructs used in this study and the endogenous Hsp90 present in rabbit reticulocyte lysate (since such heterodimerization could produce binding artifacts mediated by rabbit Hsp90), we repeated several experiments using purified GST-Hsp90 proteins. First, we demonstrated that GST-C3 (a carboxyl-terminal Hsp90 fusion protein) bound to ATP-Sepharose and that binding was competed by excess ATP (Fig. 6A). As expected, GST-C1 (an amino-terminal Hsp90 fusion protein) also bound to ATP-Sepharose. However, GST-C3 binding to ATP-Sepharose was specifically competed by the peptide derived from amino acids 663-676 of Hsp90, whereas GST-C1 binding was not (Fig. 6B). Next, we demonstrated that GST-Wt Hsp90 and GST-C3, but not GST-C1, bound to novobiocin-Sepharose in a manner competable by excess novobiocin (Fig.  6C). GST-C3 binding to immobilized novobiocin was also specifically competed by the soluble peptide derived from the Hsp90 carboxyl terminus (Fig. 6D). Finally, we examined the efficiency of competition of soluble ADP, ATP, CTP, and GTP with ATP-Sepharose binding of GST-C3 (Fig. 6E). The order of efficiency was found to be ATP Ͼ ADP Ͼ GTP Ͼ CTP. These results are in contrast to the nucleotide affinity of the aminoterminal domain, which shows a distinct preference for ADP over ATP (16).
Novobiocin Interferes with Hsp90-Hsc70 and Hsp90-p23 Association-Hsp90 participates in at least two multichaperone complexes. One complex contains Hsc70, whose binding site overlaps the carboxyl-terminal dimerization domain identified in this study to contain the novobiocin-binding site (28,29). An alternative Hsp90 multichaperone complex contains an acidic protein, termed p23, that binds to Hsp90 in an ATP-dependent fashion (30). The binding site of p23 is not fully characterized, but it is not confined to the isolated amino terminus (amino acids 1-221), although mutations in this region certainly affect p23 binding to full-length Hsp90 (16). Residues in both aminoterminal and carboxyl-terminal portions of Hsp90 were found necessary for binding of SBA1, a yeast homolog of p23 (31). Because novobiocin binds to a unique site on Hsp90, we wished to determine whether the drug might disrupt both Hsc70-and p23-containing multichaperone complexes. To perform this ex- periment, we made use of the fact that rabbit reticulocyte lysate contains copious amounts of both Hsp90-containing multichaperone complexes. We added novobiocin (1 mM) to reticulocyte lysate for 30 min on ice, and then we immunoprecipitated Hsp90 using an amino-terminally directed antibody (SPA-771). Following SDS-PAGE and electrotransfer, resultant blots were probed for p23 and Hsc70. Whereas p23 and Hsc70 were both readily co-immunoprecipitated with Hsp90 from untreated reticulocyte lysate, novobiocin preincubation caused a marked decrease in the amounts of both p23 and Hsc70 co-precipitated with Hsp90 (Fig. 7). DISCUSSION Novobiocin, a coumarin-type antibiotic, antagonizes Hsp90 function in vitro and in vivo in a manner similar to GA and radicicol, two structurally different small molecules that both bind to an amino-terminal nucleotide-binding pocket on the chaperone (15,16,19,20,25). However, this study demonstrates that the structural requirements for novobiocin binding to Hsp90 are unique. First, several point mutations in the amino terminus of the chaperone that abrogate both GA and radicicol binding either did not affect or actually augmented novobiocin binding. Second, an amino-terminal fragment comprising the GA/radicicol-binding domain failed to bind to immobilized novobiocin. In contrast, analysis of progressively smaller carboxyl-terminal Hsp90 peptides revealed the novobiocin-binding site to be contained within amino acids 538 -728. Within this region, the removal of amino acids 657-677 severely compromised novobiocin binding, and a synthetic peptide composed of amino acids 663-676 efficiently competed the binding of Hsp90 to immobilized novobiocin. Thus, these data localize the novobiocin-binding site to a region in Hsp90 known to be important both for its dimerization and for association of other co-chaperones.
It has been proposed by several investigators that Hsp90 contains two distinct chaperone sites (14,22,32), one in the amino terminus and one in the carboxyl terminus of the protein, and that these sites are linked by a charged domain whose function may be to modulate the interaction of both chaperone domains (23). Although possible effects of the charged domain on the chaperone activity/conformation of the carboxyl terminus are not known, our data suggest that the presence of the charged domain in the absence of the amino terminus clearly inhibits novobiocin binding to the carboxyl terminus. It is not certain if both chaperone domains operate independently of each other, but ATP (or GA) occupancy of the amino-terminal nucleotide-binding site certainly affects the binding of co-chaperones to the carboxyl terminus, whereas the binding of specific co-chaperone proteins to the carboxyl terminus negatively impacts the ability of the amino terminus to bind nucleotide or GA (16,33). Our current data support the existence of cross-talk between the amino and carboxyl ends of Hsp90, since point mutations that disrupt ATP/GA binding to the amino terminus markedly enhanced novobiocin binding to the carboxyl terminus. Additionally, novobiocin binding to the carboxyl terminus of Hsp90 antagonizes the binding of GA and radicicol to the amino terminus (25). By further supporting functional interaction between amino-and carboxyl-terminal domains of Hsp90, Hartson et al. (21) have shown that molybdate ion interacts with the carboxyl terminus of the chaperone in a region that, in light of our results, overlaps the novobiocin-binding site and that such interaction influences the conformation of the amino terminus. Prior association of GA with Hsp90 prevented the chaperone from attaining the molybdate-favored conformation (21). Taken together these data support a model in which the conformational state of either end of Hsp90 may affect the conformation/function of the opposite end, with the middle charged domain perhaps mediating such interactions.
Surprisingly, binding of carboxyl-terminal Hsp90 fragments to novobiocin-Sepharose was competed not only by excess novobiocin but also by ATP. Additionally, the smallest carboxylterminal Hsp90 fragment, ⌬4, that bound to immobilized novobiocin also bound to ATP-Sepharose, and this binding was competed by both novobiocin and ATP. Similar to its effects on FIG. 6. Purified GST-Hsp90 fusion proteins bind to novobiocin-and ATP-Sepharose. A, the Hsp90 carboxyl-terminal fusion protein GST-C3 or amino-terminal fusion protein GST-C1 (4 g each) were incubated with ATP-Sepharose beads in the presence or absence of excess ATP. Beads were processed as described under "Materials and Methods" and analyzed by SDS-PAGE followed by silver staining. B, GST-C3 and GST-C1 (4 g each) were preincubated with either 10 mM sense peptide (PP) or "mismatch" peptide (MP) prior to analysis of ATP-Sepharose binding as in A. C, GST-Wt Hsp90, GST-C3, or GST-C1 (4 g each) were incubated with increasing concentrations of soluble novobiocin prior to analysis of binding to novobiocin-Sepharose. Samples were analyzed by SDS-PAGE followed by silver staining. D, GST-C3 (4 g) was incubated with 10 mM sense peptide (PP) or mismatch peptide (MP) prior to analysis of novobiocin-Sepharose binding. E, GST-C3 (4 g) was incubated with increasing concentrations of various nucleotides prior to analysis of ATP-Sepharose binding. novobiocin binding, deletion of amino acids 657-677 from either ⌬3 or ⌬4 carboxyl-terminal Hsp90 fragments reduced binding to ATP-Sepharose by greater than 90%, whereas the synthetic peptide duplicating the amino acid sequence from residues 663 to 676 effectively blocked Hsp90 binding to immobilized ATP.
Binding of ⌬4 to ATP-Sepharose was competitively inhibited by both soluble ATP and novobiocin, suggesting that the novobiocin-binding domain which we have mapped to the carboxyl terminus of Hsp90 may overlap a second nucleotide-binding site on the chaperone. An Hsp90 fragment lacking the aminoterminal 222 amino acids but containing the charged domain failed to bind ATP-Sepharose just as it failed to bind to novobiocin-Sepharose, suggesting that the charged domain may regulate nucleotide access to the carboxyl-terminal site. In support of this hypothesis, removal of the charged domain restored ATP binding to carboxyl-terminal Hsp90 peptides.
Although we found no evidence that our results with in vitro translated Hsp90 carboxyl-terminal fragments were due to interference by rabbit Hsp90 present in the reticulocyte lysate (data not shown), we confirmed our findings using purified GST-Hsp90 fusion proteins. Thus a carboxyl-terminal GST-Hsp90 fusion protein (GST-C3) behaved exactly as the in vitro translated ⌬4 carboxyl-terminal fragment, not only with respect to binding to novobiocin-and ATP-Sepharose, but also with respect to the ability of the soluble peptide to compete such binding. Analysis of the efficiency of various nucleotides as competitors of GST-C3 binding to ATP-Sepharose revealed that ATP was more efficient than ADP in this regard. In contrast, previous data demonstrated that the amino-terminal nucleotide-binding site has a 10-fold greater affinity for ADP as compared with ATP (16). The fact that GTP was only minimally able and CTP completely unable to reduce the binding of GST-C3 to ATP-Sepharose lends further support to the specificity of our data.
Although controversial, a precedent exists for a second ATPbinding site in the carboxyl terminus of Hsp90, analogous to the two nucleotide-binding sites found in the Hsp110 chaperone family (34). Indeed, an early study by Csermely and Kahn (35) identified the presence of an ATP-binding consensus sequence in the carboxyl-terminal half of Hsp90. Although the recent identification and extensive characterization of the amino-terminal nucleotide-binding site, together with the reported lack of ATP dependence of the chaperone activity mediated by the Hsp90 carboxyl terminus, have focused attention away from possible nucleotide interactions with other portions of this chaperone, careful kinetic analysis of ATP-dependent Hsp90 activities suggests the cooperativity of two nucleotide-binding sites (36). Whether these sites represent two amino-terminal binding domains of an Hsp90 dimer, or perhaps one carboxyland one amino-terminal domain, remains to be determined.
The physiologic significance of novobiocin binding to the Hsp90 carboxyl terminus is amply demonstrated by the ability of this antibiotic to destabilize and deplete a series of Hsp90 client proteins, similar to the effects of both GA and radicicol (25). Further evidence of the biologic importance of novobiocin binding to Hsp90 is supplied by the data in this study showing that the drug interferes with both Hsc70 and p23 association. Since Hsc70 and p23 are components of two distinct multichaperone complexes formed around Hsp90, these results suggest that novobiocin binding disrupts both Hsp90-Hsc70-p60 Hop and Hsp90-p50 (or immunophilin)-p23 complexes. This is in contrast to the effects of GA and radicicol that disrupt only p23containing Hsp90 complexes (1,20,37). Thus, novobiocin, besides mimicking the biologic effects of GA and radicicol, may have an additional negative impact on Hsp90 function not seen with the other drugs, although this hypothesis remains to be examined.