Specific Binding of Tetratricopeptide Repeat Proteins to the C-terminal 12-kDa Domain of hsp90*

The molecular chaperone hsp90 in the eukaryotic cytosol interacts with a variety of protein cofactors. Several of these cofactors have protein domains containing tetratricopeptide repeat (TPR) motifs, which mediate binding to hsp90. Using a yeast two-hybrid screen, the 12-kDa C-terminal domain of human hsp90α (C90) was found to mediate the interaction of hsp90 with TPR-containing sequences from the hsp90 cofactors FKBP51/54 and FKBP52. In addition, the mitochondrial outer membrane protein hTOM34p was identified as a TPR-containing putative partner protein of hsp90. In experiments with purified proteins, the TPR-containing cofactor p60 (Hop) was shown to form stable complexes with hsp90. A deletion mutant of hsp90 lacking the C90 domain was unable to bind p60, whereas deletion of the ∼25-kDa N-terminal domain of hsp90 did not affect complex formation. Both p60 and FKBP52 bound specifically to the C90 domain fused to glutathione S-transferase and competed with each other for binding. In reticulocyte lysate, the C90 fusion protein recognized the TPR proteins p60, FKBP52, and Cyp40. Thus, our results identify the C90 domain as the specific binding site for a set of hsp90 cofactors having TPR domains.

Several hsp90-associated cofactors contain multiple copies of the tetratricopeptide repeat (TPR) motif (5,10,12), a degenerate 34-amino acid consensus sequence that mediates proteinprotein interactions in diverse cellular pathways (13). The TPR-containing domains of FKBP52, Cyp40, and protein phosphatase pp5 mediate binding of these proteins to hsp90 (14 -16). The binding of p60 to hsp90 requires a central region of p60 that contains 5 TPR motifs (5,17). The crystallographic structure of the pp5 TPR domain has been solved (18), but the mode of interaction with hsp90 is as yet undetermined. Only a specific subset of TPR-containing proteins, with closely related TPR motifs, can bind hsp90. For example, the hsp70 cofactor protein Hip is not observed to bind directly to hsp90, although it contains 4 TPR motifs (19 -21). Other hsp90-binding cofactors such as Cdc37/p50 and p23 contain no identifiable TPR motifs.
In previous work, the hsp90 polypeptide was dissected by proteolysis into three independently folded domains (22). The ϳ25-kDa N-terminal domain (N90) was identified as the binding site for the benzoquinoid ansamycin drug geldanamycin (22) and for ATP (23). Both the N90 domain and 12-kDa Cterminal domain of hsp90 (C90) were shown recently to contain independent chaperone sites in vitro (4). In addition, the ϳ200 C-terminal residues of hsp90, including the C90 domain and part of the middle domain, is involved in hsp90 dimerization (24,25). Here, yeast two-hybrid screening and biochemical analysis was used to identify the C90 domain as the binding site for the TPR-containing cofactor proteins.

EXPERIMENTAL PROCEDURES
Plasmid Construction and Yeast Two-hybrid Experiments-Human hsp90␣ cDNA (StressGen, GenBank TM accession number X15183 (26)) was used as a template for PCR amplification. PCR products representing the full-length protein (FL90) and fragments encompassing residues 1-617 (⌬C90) and residues 629 -731 (C90) of hsp90 were inserted in frame into the pAS2-1 vector (CLONTECH). The construct pAS2-1-C90 was used as a bait in a two-hybrid screen (27) and was cotransformed into Saccharomyces cerevisiae (strain Y190) with a human liver cDNA library based on the pACT2 vector (CLONTECH). Transformed cells were plated on synthetic dropout (SD) medium/ϪHis,ϪTrp,ϪLeu plates supplemented with 25 mM 3-amino-1,2,4-triazole. After 8 days of incubation at 30°C, colonies were patched on filters, assayed for ␤-galactosidase activity, and the pACT2 plasmids harboring library cDNAs were isolated from positive clones. Isolated plasmids were cotransformed individually into S. cerevisiae (strain Y187) with the original pAS2-1-C90 bait or with the empty pAS2-1 vector serving as a negative control. After growth on SD/ϪTrp,ϪLeu plates for 5 days, colonies were re-assayed for ␤-galactosidase activity. Positive candidate cDNA inserts were further analyzed by nucleotide sequencing.
Protein Binding Assays-p60 and the hsp90 deletion mutants ⌬N90 (residues 272-732 of human hsp90␣), ⌬C90 (residues 1-617), GST-N90 (residues 9 -236), GST-M90 (residues 272-617), GST-C90 (residues 629 -732) were expressed and purified as reported previously (4). hsp90␣ was purified from bovine brain (28). FKBP52 was expressed in Escherichia coli as a fusion protein with Intein/chitin-binding protein 2 and purified using chitin beads followed by dithiothreitol cleavage (New England Biolabs). * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  For gel filtration analysis, 25-l samples containing 10 M hsp90, ⌬N90, ⌬C90, or 5 M p60 were incubated for 10 min at room temperature in buffer F (4), then separated on a Superdex 200/3230 column equilibrated in the same buffer using a Smart chromatography system (Amersham Pharmacia Biotech). Also, 25-l samples containing 10 M hsp90, 5 M p60, or 50 M GST-C90 were incubated as above in buffer containing 100 mM NaCl, 20 mM Tris-Cl, pH 7.5, and 2 mM EDTA and then separated on the same column equilibrated in this buffer.
To assay binding to GST fusion proteins, 200-l reactions containing 2 M GST-N90, GST-M90, or GST-C90 and 2 M p60 or FKBP52 in buffer F were incubated at room temperature for 1 h in the presence of glutathione-agarose beads (Sigma). The beads were collected by centrifugation, washed with 2.4 ml of buffer containing 1% Nonidet P-40, 100 mM NaCl, 20 mM Tris-Cl, pH 7.5, and 2 mM EDTA, then with 1.2 ml of the same buffer without Nonidet P-40. Bound proteins were eluted with SDS-PAGE loading buffer. For competition experiments, reactions containing 2 M GST-C90 and 2 M FKBP52 were supplemented with 4, 10, and 20 M p60 and then analyzed as above.
Rabbit reticulocyte lysate (Green Hectares, Oregon, WI) was desalted, as reported previously (9). 300 g of GST-C90 was mixed with 500 l of lysate and incubated at 4°C for 2 h in the presence of glutathione-agarose. After washing as above, the beads were resuspended in buffer containing 500 mM NaCl, 20 mM Tris-Cl, pH 7.5, and 2 mM EDTA and incubated for 30 min at 4°C. The supernatant was removed and protein precipitated with trichloroacetic acid for SDS-PAGE analysis. Protein remaining on the beads was eluted with SDS-PAGE loading buffer. Immunoblots using antibodies against FKBP52, Cyp40, Hip (Affinity BioReagents) and p60 and p23 (StressGen) were visualized with the ECL detection kit (Amersham Pharmacia Biotech).

RESULTS AND DISCUSSION
The cDNA segment encoding amino acids 629 -731 of human hsp90␣ (C90) was used as the bait in a yeast two-hybrid screen with a rat liver cDNA library (Fig. 1A). Screening of ϳ10 6 cotransformants identified six positive clones that could be shown by rescreening to interact specifically with the C90 bait. One cloned cDNA coded for a segment of the immunophilin FKBP51/54 (Ref. 29; GenBank TM accession number U42031). Two other cDNAs contained identical inserts encoding another hsp90-interacting protein, FKBP52 (Ref. 30; GenBank TM accession number 88279). The remaining three inserts were also identical and coded for hTOM34p (GenBank TM accession number U58970), a mitochondrial outer membrane protein proposed to function in protein import into mitochondria. 3 Strikingly, all the partial cDNAs selected with C90 encode segments of hsp90-interacting proteins containing at least three TPR domains (Fig. 1B). As the immunophilins interact with hsp90 via their TPR motifs (14 -16), our results identify the C-terminal domain of hsp90 as a binding site for these sequences. Because the two-hybrid screen was not saturating, the identification of hTOM34p as a new, putative hsp90-binding protein suggests that the number of TPR proteins recognized by hsp90 may be larger than previously anticipated. Interestingly, the mitochondrial outer membrane protein TOM70p, involved in specific recognition of precursor proteins for import, is also able to bind hsp90 in vitro (16). Together, these findings point to a possible cellular role of hsp90 in mitochondrial protein import.
hsp90 itself was not selected by the C90 domain in the two-hybrid screen. The structural information required for homodimer formation has been localized to the C-terminal ϳ200 amino acid residues of hsp90 and is thus present only partially in the C90 domain (24).
Is the C-terminal domain of hsp90 the only segment of hsp90 that interacts with TPR proteins? To address this question, we tested the full-length hsp90 protein (FL90), a fragment of hsp90 comprising the N-terminal and middle domains (⌬C90), and the C-terminal domain (C90) (see Fig. 1A) in the twohybrid system for binding to the TPR protein fragments de-scribed above. The three hsp90 constructs were expressed at comparable levels in the yeast cells as determined by immunoblotting (not shown). Whereas FL90 and C90 exhibited strong binding to FKBP51/54, FKBP52, and hTOM34p, ⌬C90 did not (Fig. 1C). In conclusion, these two-hybrid data indicate that binding of TPR proteins to hsp90 is mediated by the C-terminal domain of hsp90.
Biochemical binding assays were performed to corroborate and extend the results of the two-hybrid screen. We first tested purified hsp90 and deletion mutants of hsp90 for their ability to interact with purified p60 protein. Unbound hsp90 and p60 were clearly resolved by gel filtration chromatography, with the peak of hsp90 eluting in fraction 11 ( Fig. 2A) and p60 eluting in fractions 16 and 17 (Fig. 2B). The apparent molecular size of the hsp90 homodimer of ϳ500 kDa is in agreement with earlier observations (4, 31). p60 fractionated as a mono-  (21) and of the hsp90 fragments used in the two-hybrid screen. The known ligands of hsp90 are indicated. GA, geldanamycin. B, schematic representation of the three TPR domain proteins selected as specific binders of C90. Arrows indicate the starting amino acid residue of the respective protein fragment as deduced from sequencing of the corresponding cDNAs. Hatched squares represent TPR motifs of FKBP51/54 and FKBP52 as described by Ratajczak et al. (15) and of hTOM34p as deduced by BLAST analysis. C, full-length hsp90 (FL90) and domain fragments of hsp90 in the pAS2-1 bait-construct (vertical) were cotransformed in S. cerevisiae (strain Y190) with the fragments of the TPR proteins in B inserted in the pACT2 target vector (horizontal) and grown on SD/ ϪTrp,ϪLeu,ϪHis plates containing 25 mM 3-amino-1,2,4-triazole for 7 days at 30°C. Growth was only observed for combinations of the TPR target proteins with FL90 or C90 but not with ⌬C90 or the vector controls without insert. mer. When a mixture containing equimolar amounts of hsp90 dimer and p60 monomer was analyzed by gel filtration, both hsp90 and the majority of p60 eluted in fractions 10 -11 (Fig.  2C), consistent with the formation of a stable complex between the two proteins. The observed binding of p60 to hsp90 dimer was saturable at an estimated stoichiometry of 1:1, was apparently ATP-independent (not shown), and was disrupted at a concentration of 500 mM NaCl (see Fig. 3).
⌬N90, a deletion mutant of hsp90 comprising the middle and C90 domains, contains the sequences implicated in hsp90 dimerization (24,25) and was expected to form homodimers. Similar to full-length hsp90, pure ⌬N90 eluted in a peak around fraction 12 with an anomalously large molecular size (not shown). When p60 and ⌬N90 were incubated together and then analyzed, p60 coeluted with ⌬N90 in a peak at fraction 12 (Fig. 2D). Purified ⌬C90 polypeptide comprising the N90 and middle domains of hsp90 eluted in a peak at fraction 16, consistent with the expected monomeric state of the deletion mutant (not shown). In the presence of ⌬C90, p60 eluted upon gel filtration in fractions 16 -17 (Fig. 2E), identical to the behavior of pure p60 alone (Fig. 2B). The elution profile of ⌬C90 was also unchanged in the presence of p60. These data suggest that p60 is unable to stably bind the ⌬C90 polypeptide, although it can form complexes with the ⌬N90 polypeptide and with full-length hsp90. (Binding of C90 to p60 is not shown in Fig. 2, because complexes containing C90 and p60 could not be resolved from either protein alone). Similar results were obtained using purified FKBP52 in place of p60 (not shown). Thus, the C90, but not the N90, domain is required for binding of these TPR proteins, in agreement with the results from the two-hybrid screen.
As an independent confirmation of this result, p60 and FKBP52 were assayed for binding to the N-terminal, middle, and C-terminal hsp90 domains fused separately to glutathione S-transferase (GST-N90, GST-M90, and GST-C90, respectively). Mixtures of p60 with each of the purified GST fusion proteins were allowed to bind, and proteins recovered with glutathione-agarose were analyzed by SDS-PAGE. As ex-pected, p60 was not recovered with GST-N90 or GST-M90 (Fig.  3A, lanes 3 and 4), but remained bound to GST-C90 (Fig. 3A,  lane 5). Similarly, FKBP52 was recovered only with the GST-C90 fusion protein (Fig. 3A, lane 6).
Do the different TPR proteins that associate with hsp90 compete for binding to the C-terminal hsp90 domain? This question was addressed by analyzing the binding of FKBP52 to GST-C90 in the presence of increasing amounts of p60. Binding reactions contained p60 in molar ratios of 2:1, 5:1, and 10:1 relative to FKBP52 (Fig. 3A, lanes 7-9). When the molar excess of p60 increased, more p60 and less FKBP52 were recovered with GST-C90 (Fig. 3A, compare lanes 6 and 9). Densitometry of Coomassie-stained gels showed that in the presence of a 10-fold excess of p60, binding of FKBP52 was ϳ45% of that observed in the absence of p60. The efficiency of competition by p60 is underestimated in these experiments because the recovery of the TPR proteins is always less than stoichiometric (see Fig. 3A, lanes 5 and 6), most likely because of loss of material during the washing steps. Nevertheless, it is clear that two different TPR domain proteins quantifiably compete for binding to the C-terminal domain of hsp90. Thus, hsp90 most likely will interact with only one specific TPR protein at a time.
If the interaction between the C90 domain and TPR proteins is specific, then GST-C90 should compete with full-length hsp90 for binding to TPR proteins. To test this, equimolar amounts of hsp90 dimer and p60 monomer were mixed with a 5-fold excess of GST-C90 and then analyzed by gel filtration chromatography as in Fig. 2. The elution profile of p60 in this experiment is shown in Fig. 3B (solid circles). The majority of p60 eluted in a peak around fraction 14, slightly earlier than the elution peak of GST-C90 (marked with arrow) and thus consistent with a complex containing both p60 and GST-C90. A minor p60 peak eluted at fraction 11, where hsp90-p60 complexes formed in the absence of GST-C90 are known to elute (Fig. 3B, open squares, and Fig. 2C). No p60 eluted in fractions where monomeric p60 would be expected (Fig. 3B, open circles, and Fig. 2B). These data indicate that GST-C90 and hsp90 compete for the same binding site on p60. Therefore, the isolated C-terminal domain of hsp90 provides the same specific TPR interaction site as the full-length hsp90 protein.
To further test the specificity of protein-protein interactions mediated by the C90 domain, GST-C90 was incubated with a rabbit reticulocyte lysate and recovered with glutathione-agarose. Because complexes between hsp90 and p60 or FKBP52 could be dissociated in 500 mM NaCl, lysate proteins bound to GST-C90 were eluted from the agarose matrix with a high-salt buffer. A limited set of proteins were observed in the eluted fraction (Fig. 3C, lane 3), clearly distinct from the total protein profile of the lysate (Fig. 3C, lane 1). Prominent bands with apparent molecular sizes of 62, 54, and 42 kDa (Fig. 3C, lane 3, marked bands) were identified by immunoblotting as p60, FKBP52, and Cyp40, respectively (Fig. 3C, right panel). Thus, the C90 domain specifically recognizes in a total cytosol extract the major TPR-containing cofactors associated with the cellular function of hsp90. Except for GST-C90 itself and a lower molecular weight band corresponding to a GST degradation product, few proteins remained bound to the glutathione-agarose beads after the high-salt wash and were eluted with SDS (Fig.  3C, lane 2). Significantly, the TPR-containing protein Hip, which binds to the ATPase domain of hsp70 (19,21,32), could not be detected by immunoblotting in the eluted or tightly bound fractions. Thus, GST-C90 apparently maintains specificity for interactions with a subset of TPR-containing hsp90 cohorts. Moreover, immunoblotting revealed that p23 was not bound to GST-C90, suggesting that only the TPR-containing cofactors are recognized by the C-terminal hsp90 domain. In agreement with our data, interaction of hsp90 with p23 has been shown to be independent of the C90 domain (33). Other unidentified bands detected in the salt-eluted fraction (Fig. 3C, lane 3) may represent novel TPR proteins that cooperate with hsp90.
The results presented in this study identify the C-terminal domain of hsp90 as necessary and probably sufficient for interaction of hsp90 with TPR-containing proteins (Fig. 1A). This domain is surprisingly small, but the minimal sequence required for TPR binding may even be smaller. Recently the isolated C90 domain has also been found to act as a chaperone in vitro by preventing the aggregation of unfolded proteins such as rhodanese and by binding the antigenic peptide VSV8 (4). Because TPR protein had no effect on the ability of C90 to act as a molecular chaperone (not shown) (4), the C90 domain may bind simultaneously to a specific TPR protein and to an unfolded substrate polypeptide.