The Human Cdc37·Hsp90 Complex Studied by Heteronuclear NMR Spectroscopy*

The cell division cycle protein 37 (Cdc37) and the 90-kDa heat shock protein (Hsp90) are molecular chaperones, which are crucial elements in the protein signaling pathway. The largest class of client proteins for Cdc37 and Hsp90 are protein kinases. The catalytic domains of these kinases are stabilized by Cdc37, and their proper folding and functioning is dependent on Hsp90. Here, we present the x-ray crystal structure of the 16-kDa middle domain of human Cdc37 at 1.88 Å resolution and the structure of this domain in complex with the 23-kDa N-terminal domain of human Hsp90 based on heteronuclear solution state NMR data and docking. Our results demonstrate that the middle domain of Cdc37 exists as a monomer. NMR and mutagenesis experiments reveal Leu-205 in Cdc37 as a key residue enabling complex formation. These findings can be very useful in the development of small molecule inhibitors against cancer.

The 90-kDa heat shock protein (Hsp90) 3 is an abundant and essential protein, accounting for 1-2% of the total protein amount in unstressed eukaryotic cells. A 2-to 10-fold higher expression level of Hsp90 is found in cancer cells and virally transformed cells (1)(2)(3), suggesting a crucial role of Hsp90 for growth and/or survival of tumor cells. Hsp90 is a protein chaperone that helps a variety of client proteins to adopt their native conformation. Several co-chaperones such as Hop/Sti1, FKBP52, cyclophilin 40, and p50/Cdc37 are identified to be associated with Hsp90. These co-chaperones are required for binding to diverse client proteins and controlling the ATPase cycle of Hsp90 (4 -6). Hsp90 is composed of three domains: an N-terminal domain of ϳ25-kDa, a middle domain of ϳ35-kDa, and a C-terminal domain of ϳ10-kDa (7). The ATP-binding site of Hsp90 is located in the N-terminal domain. Low molecular weight compounds that interfere with ATP-binding such as geldanamycin or radicicol inhibit the Hsp90 function by competing for the interaction site.
Many of the client proteins of Hsp90 are crucial elements of the cellular signal transduction pathway. Protein kinases represent the largest class of Hsp90 client proteins, and many have been shown to cause cancer when deregulated. These protein kinases depend on Hsp90 for their proper folding and functioning suggesting that Hsp90 itself is a valid anti-cancer target of pharmaceutical interest (8).
Hsp90 does not directly bind to protein kinases, but this interaction is mediated by the co-chaperone Cdc37/p50. Hence, Cdc37 has also been dubbed as a kinase targeting subunit of the Hsp90 machinery (9). Interestingly, Cdc37 is also able to act as a chaperone by itself independent of Hsp90 (10).
The cell division cycle protein 37 (Cdc37) was first identified in Saccharomyces cerevisiae as a protein required for the "Start" event in the cell division cycle (11). In yeast, cdc37 mutations cause synthetic lethality (as defined by Kaelin (56)) in combination with thermosensitive cdc28 or kin28 mutations (12,13). The mammalian homolog of Cdc37, also referred to as p50 cdc37 , was first identified within a complex consisting of the Rous sarcoma virus encoded oncogene pp60 v-scr and Hsp90 (14,15). Further studies confirmed that protein kinases are the most favored targets of Cdc37 (16). A plethora of other oncogenic kinases, including Raf isoforms, Akt, and ErbB2, have been identified as potential clients of Cdc37 (17)(18)(19)(20). Expression of Cdc37 is up-regulated in cancer cells and tissues and can promote tumorigenesis when overexpressed (21,22). In addition, Cdc37 can inhibit the ATPase activity of Hsp90 implied in promoting the assembly of kinase clients with both chaperone proteins (23,24).
Structurally, the 44.5-kDa protein Cdc37 can be dissected into three domains (10,25). Proteolytic fingerprinting studies indicate that it is comprised of an N-terminal domain (residues 1-127 (Cdc37 N ), 15.5 kDa), a middle domain (residues 147-276 (Cdc37 M ), 16 kDa), and a C-terminal domain (residues 283-378 (Cdc37 C ), 10.5 kDa). The middle domain Cdc37 M is highly resistant to proteolytic digestion and was found to be the most stable domain of Cdc37 (26). Several studies (19,(27)(28)(29)(30) have shown that Cdc37 N binds to the client proteins (e.g. kinases). A highly conserved region within Cdc37 N contains Ser-13 as a unique phosphorylation site for the protein kinase CKII. Phosphorylation seems to be an important mechanism for controlling the binding affinity to the client protein kinases and as a consequence thereof regulating the activity of multiple protein kinases (31). The Hsp90 binding site is located in the middle domain of Cdc37, which also comprises a putative dimerization region (32). An extended ␣-helix connects the Hsp90 binding site to the C-terminal domain, which might also be required for dimer formation (24,32). The structure of the N-terminal domain is yet unknown but sequence analysis predicts a high content of ␣-helices (32).
The crystal structure of Cdc37 MC (middle and C-terminal domain) bound to Hsp90 N (from yeast) is a heterotetrameric complex, comprising Hsp90 bound to a Cdc37 dimer (32). The function of Cdc37 to inhibit the ATPase activity of Hsp90 (24) is due to the extended side chain of Arg-167 that binds to Glu-33 of Hsp90, thereby inhibiting the ATPase reaction.
Additional studies indicate that Cdc37 M exists as a monomer (26). Low resolution cryoelectron microscopy images of the Hsp90⅐Cdc37⅐Cdk4 complex show that Cdc37 is a monomer in this interaction (33). However, these studies utilized yeast Hsp90, and it was realized that it is necessary to investigate the interaction with human Hsp90 (26) to resolve the differences.
Here, we report the structure determination of the human Hsp90 N ⅐Cdc37 M complex based on the NMR backbone assignments of Cdc37 M (34) and Hsp90 N (35). The interface of the complex between human Cdc37 M and Hsp90 N could be mapped based on NMR chemical shift perturbation (CSP) studies, and cross-saturation transfer (CST) techniques. Subsequently, we solved the x-ray crystal structure of human Cdc37 M at 1.88 Å in its monomeric form. The structure of the Cdc37 M ⅐Hsp90 N complex was generated by incorporating NMR data (CSPs and CSTs as ambiguous distance restraints and RDCs as orientational restraints) in the HADDOCK (36) docking protocol.
The supernatant was applied to a nickel-nitrilotriacetic acid FastFlow column (Qiagen) following the manufacturer's recommendations. After Tev-protease cleavage to remove the His tag ProtGB1, the protein was applied again on a nickel-nitrilotriacetic acid column. The pooled protein fractions were further purified by gel filtration on a Hiload 26/60 Superdex 75 preparative grade column (Amersham Pharmacia Biosciences) in Hepes 50 mM (pH 7.5), NaCl 100 mM, and 1 mM DTT. The protein-containing fractions were pooled, concentrated, and either immediately used for the experiments or stored at Ϫ20°C. The 15 N-labeled protein was produced by using M9 minimal media with 15 NH 4 Cl (1 g/liter) as the sole nitrogen source. Uniformly triple ( 2 H, 13 C, and 15 N) or double ( 2 H and 15 N) labeled protein was produced using Silantes OD2 media. The degree of isotopic enrichment for the Silantes media was Ͼ98% for all isotopes (manufacturer's data). For selective labeling with 15 N-labeled amino acids, the following 15 N-labeled amino acids were used: alanine, arginine, lysine, methionine, isoleucine, leucine, valine, phenylalanine, tyrosine, and tryptophan (Spectra Stable Isotopes, Germany). The procedure applied for selective amino acid labeling is essentially the same as described in the literature (37).
Hsp90 N Cloning, Expression, and Purification-From the cDNA (Stratagene) the base pairs corresponding to amino acids 18 -223 (N-terminal domain) of human Hsp90 were amplified by PCR using 5Ј-AAACCATGGGGGAGACGTTCGCCTTT-CAGG-3Ј and 5Ј-GCAGCCGGATCCTCACTCCACAAAAA-GAG-3Ј as primers. The PCR product was double digested with NcoI and BamHI and was cloned into the derivatized pKM263, with a Protein G B1 domain (ProtGB1) fusion protein located C-terminally behind a stretch of six histidines and a tev cleavage site under the control of a T7-promoter (see above). The protein was expressed and purified in the same way as Cdc37, except for the post induction temperature, which was 25°C. In addition, tetra-selective labeling with 15 N-glycine, 15 N-alanine, 15 N-phenylalanine, and 15 N-leucine was carried out as described above.
Due to the design of the expression vectors, Cdc37 and Hsp90 N constructs carry two (GA) or four (GAMG) additional amino acids at the N terminus. High expression levels were achieved after induction with 1 mM isopropyl 1-thio-␤-D-galactopyranoside at 37°C for Cdc37 and 25°C for Hsp90 N and ϳ95% of the fused protein was found in the soluble fraction.
Typically, 30 mg/liter of pure (95%) proteins was obtained after cleavage of protein G. The final samples were dialyzed against the NMR buffer (Hepes 50 mM (pH 7.5), NaCl 100 mM, and 1 mM DTT).
Mutagenesis-Mutants of Cdc37 M were created with Quik-Change kit (Stratagene), and sequenced to confirm the incorporation of the correct mutation. The mutant proteins were purified following the same protocol as the wild-type Cdc37 M .
NMR Spectroscopy-NMR experiments were performed at 298 K, on Bruker 600-, 700-, 800-, 900-, and 950-MHz spectrometers equipped with cryogenic triple-resonance probes. For the Cdc37 M ⅐Hsp90 N complex a ratio of 1:1 was used for all measurements. The spectrometer was locked on D 2 O. The spectra were processed using Topspin 2.0 (Bruker Biospin) and analyzed using either SPARKY 3.113 4 or CARA 1.8.3 (39). The resonances for the free form of Cdc37 M at pH 5.5 and Hsp90 N at pH 7.5 have been assigned before (34). Spectra of Cdc37 M at pH 7.4 have been assigned by performing a pH series (5.0, 5.5, 6.0, 6.5, 7.0, 7.5, and 8.0) with 2,2-dimethyl-2-silapentane-5-sulfonic acid as an internal standard for calibration. Sequential backbone assignment of 15 N-labeled Cdc37 M in complex with unlabeled Hsp90 N and vice versa was achieved by overlaying the 1 H, 15 N-HSQC spectra of free Cdc37 M (or Hsp90 N ) and Cdc37 M (or Hsp90) in complex and verified by the NOE patterns from a high resolution three-dimensional 1 H, 15 N-NOESY HSQC spectrum and selective 15 N-amino acid labeling (alanine, arginine, glycine, isoleucine, leucine, lysine, methionine, phenylalanine, tryptophan, tyrosine, and valine). Additionally, a three-dimensional 1 H, 15 N-NOESY HSQC spectrum for Hsp90 N with a tetra-selectively labeled sample in which only glycine, alanine, phenylalanine, and leucine were 15 N-labeled was used to assist the assignment. 1 D(H,N) RDCs were measured for the Cdc37 M ⅐Hsp90 N complex (in which both proteins were 15 N-labeled) in pf1 phage (5 g/liter, Profos) and in 4% polyacrylamide (PAA) gel alignment media at 700 MHz. The 1 D (H,N) were extracted from IPAP-( 1 H, 15 N)-HSQC spectra (40,41). Signals that could be traced back reliably and determined unambiguously were analyzed using MODULE (42) and PALES (43). CST experiments were carried out at 25°C according to the procedure of Takahashi et al. (57). The sample contained 1 mM of 2 H, 15 N, 13 C-labeled Cdc37 M protein with the unlabeled Hsp90 N protein in 1 H 2 O/ 2 H 2 O (1:9, v/v) containing 50 mM Hepes (pH 7.5), 1 mM DTT, and 100 mM NaCl. The Gaussian pulse scheme was used for saturation of aliphatic protons with irradiation frequency set to 1.1 ppm. A saturation time of 2.0 s was employed.
X-ray Crystallography-Initial screening for crystallization conditions of Cdc37 M was performed using screening kits purchased from JB (Jena Bioscience). These trials were successful, with five promising conditions producing crystals. The protein samples that did not give crystals were analyzed by mass spectrometry, which identified that they contained small fractions of proteins with an increased molecular mass of ϳ75, 150, and 228 Da. We attributed this observation to the formation of mixed disulfides of ␤-mercaptoethanol with the four free cys-teines of Cdc37 M . Therefore, ␤-mercaptoethanol was omitted from the further protein preparations. The final diffracting quality crystals were grown from Cdc37 M (Hepes, 50 mM, NaCl, 100 mM, pH 7.5, "protein buffer") at a final concentration of 1.1 mM, in solution containing Hepes, 100 mM, pH 7.5, sodium acetate, 100 mM, 22% polyethylene glycol 4000, ("reservoir buffer"). Crystal drops were setup using the sitting drop vapor diffusion method at 18°C. Crystals were transferred in a stepwise manner into solutions of increasing glycerol concentrations of 5, 10, 15, 20, and 25%, respectively, based on one volume of protein buffer and two volumes of reservoir buffer, and flash-cooled in an X-Stream 2000 nitrogen stream at 140 K. Diffraction data were collected using a Rigaku R-Axis IV image plate detector at a wavelength of 1.5418 Å (CuK␣ radiation from a MicroMax-007 Microfocus generator). In addition to a high exposure data set consisting of 360 images of 1.0°rotations, a low exposure (one-third of the exposure time per image of that for high exposure) data set was collected consisting of 240 images of 1.5°rotations. The oscillation images were processed with the HKL/DENZO program package (44). The crystals belong to the hexagonal space group P6 1 with unit cell dimensions a ϭ b ϭ 48.67 Å and c ϭ 104.26 Å. Scaling and merging of the oscillation images was performed with SCALE-PACK (44). Data reduction and approximate absolute scaling of the data were performed with TRUNCATE from the CCP4 program suite (45). The phase problem was solved by molecular replacement using the program AMoRe (46) from the CCP4 suite, the diffraction data between 15 Å and 3 Å resolution, and a Cdc37 search model constructed from PDB entry 1US7 (32), extending from residue His-148 to residue Lys-276 and consisting of 1086 non-hydrogen atoms. The top solution with rotation angles ␣ ϭ 45.52°, ␤ ϭ 150.55°, ␥ ϭ 311.52°and translations T x ϭ Ϫ0.3117 and T y ϭ 0.0954 (fractional coordinates) was associated with a correlation coefficient of 58.2% and an R-factor of 40.4%. After rigid body refinement with Refmac5 (47), the R-factor for the data between 42.14 Å and 1.88 Å was 43.8% (R free T ϭ 43.9%). After several cycles of restrained refinement with Refmac5, the modeling of tightly bound water molecules, and manual rebuilding with COOT (48), the final structure was obtained.
Docking-Calculation of the Cdc37 M ⅐Hsp90 N complex was achieved using a high ambiguity driven docking (HADDOCK 2.0) approach with CNS 1.1 (49) essentially as described in the literature (36). The ambiguous interaction restraints have been defined for the residues that exhibited significant NMR amide chemical shift changes and/or strong saturation transfer upon interaction with its partner protein. For 189,192,196,197,200,201,203,207,211,242, and 249 as passive. The loop region 164 -168 was left fully flexible during the docking. One intermolecular NOE (Ala-117-HN to Ala-204-HB), as obtained from NOESY experiments, fitted very well with the initial docking results and was included in the final calculation. To enhance the convergence, 1 D(H,N) RDCs obtained both in 4% PAA gel and in 5 g/liter pf1 phage alignment media, were included in the calculation and the axial (D a ) and rhombic (D r ) components of the alignment tensor were estimated using PALES (43). For the initial rigid-body docking, only the RDCs in the stable second-ary structure elements were used, whereas in the further refinement steps, all RDCs were used. The final values for D a and D r were Ϫ8.21 and 0.62 for the phage alignment and 9.39 and 0.45 for the alignment in the PAA gel. The dockings were performed using our Cdc37 M crystal structure and the Hsp90 crystal structure (pdb: 1YES) (7). The protein allhdg 5.3 force field (50) was used for the calculations. The 200 final water-refined docking results are clustered at the interface within a threshold of 3.0 Å pairwise backbone r.m.s.d. The top-ranked ensemble, according to the average interaction energy and buried surface area, was accepted as the best representative of the complex.

RESULTS
Comparing Different Constructs of Cdc37 and Hsp90-The study of the rather large complex formed between the 45-kDa Cdc37 protein and the 23-kDa N-terminal domain of Hsp90 is a challenge for NMR spectroscopic studies. To study this complex, we reduced the boundaries of Cdc37 to obtain a construct that retains its structural and functional properties and is suitable for NMR studies. Beside the Hsp90 N binding domain of Cdc37 (147-276, Cdc37 M ), for which we recently published the NMR assignments (34), also the following Cdc37 constructs were investigated: fulllength Cdc37 (1-378), the kinase binding domain (1-126, Cdc37 N ), and both kinase and Hsp90 N binding domains (1-283, Cdc37 NM ). The suitability of the Cdc37 constructs for NMR was investigated by recording 1 H, 15 N-TROSY spectra (Fig. 1). The spectrum of the fulllength Cdc37 revealed considerable signal overlap and, in addition, signals that indicate unfolded regions (Fig. 1A). For Cdc37 N , the spectrum shows again signal overlap and evidence for a high content of partially unfolded and/or helical regions (Fig. 1C). Cdc37 M shows a nicely resolved high resolution 1 H, 15 N-TROSY spectrum (with 131 out of the 133 expected amide resonances), which indicates that this Hsp90binding domain of Cdc37 is properly folded (34) (Fig. 1D). It was suspected that the Cdc37 N might become more structured when it is together with the middle domain. Nevertheless, for the corresponding Cdc37 NM construct the 1 H, 15 N-TROSY spectra indicated insufficient chemical shift dispersion (Fig.  1B). The amide signals originating from Cdc37 M (shown specifically for three peaks as in the insert in Fig. 1D) have the same chemical shifts as in the full-length Cdc37 and Cdc37 NM (Fig. 1,  A and B), indicating that the fold of the Cdc37 M is the same as in the full-length protein. Hence, the 16-kDa Cdc37 M domain was used for all further NMR experiments. For Hsp90, the 23-kDa construct comprising the N-terminal ATP-binding domain (18 -223) known to interact with Cdc37 was used throughout this study (35).
Structure of Human Cdc37 M -In the x-ray structure of the complex between yeast Hsp90 and human Cdc37, the middle domain of the Cdc37 protein forms a homodimeric interaction with a symmetry-related molecule (32). So far, the structure of Cdc37 is only available in the complex with Hsp90. However, by NMR, we do not see any evidence for dimer formation of Cdc37 M . To elucidate whether Cdc37 M also forms dimers when crystallized without Hsp90, we solved the x-ray structure of free Cdc37 M . The obtained resolution after refinement was 1.88 Å with the characteristics summarized in Table 1. The structure of Cdc37 M is entirely helical consisting of a 6-helix bundle (148 -243) connected to a single long helix (245-276) as shown in Fig. 2. In the crystals, Cdc37 M is a monomer, which is in excellent agreement with NMR diffusion-based measurements showing that Cdc37 M exists as a monomer in solution with an R g exp ϭ 21.3 Å and a theoretical (51) predicted R g theo ϭ 22.2 Å. These findings agree very well with previous findings based on SAXS (26) and electron microscopy (33) data. The overall structure of this free-form monomeric Cdc37 M is very similar to the x-ray structure of the dimeric Cdc37 M when in complex with the yeast Hsp90 (32). The average r.m.s.d.
between the structures for all heavy atoms is 0.8 Å for the 6-helix bundle (148 -243) and 1.3 Å for the full domain (148 -276). The largest differences are observed in the loop connection (242-251) to the single long ␣ 7 helix, which is exactly at the previously proposed dimerization interface (32). Interestingly, this region is also near to Hsp90 and may thus expose a potential additional interaction site. Furthermore, we found two possible conformations for the side chains of Arg-167 and Gln-208 in the Cdc37 M crystals, which are also slightly different from the conformation found in the previously reported structure. Both residues reside in and near the interaction interface with Hsp90.
Assignment of Human Cdc37 M and Human Hsp90 N -The 1 H, 15 N and 13 C backbone and side-chain resonance assignment of the free-form Cdc37 M has been reported previously at pH 5.5 (34). Because the interaction studies with Hsp90 N have to be conducted at pH 7.5, we reassigned the amide resonances of Cdc37 M at this different condition. Hence, we performed a series of NMR experiments ranging from pH 5.0 through pH 8.0 with a 0.5 pH unit increment, and used 2,2-dimethyl-2silapentane-5-sulfonic acid as an internal standard for calibration. Using the 1 H, 15 N-HSQC spectra from the pH series we   15 N and 13 C backbone and side-chain resonances of the human Hsp90 N (18 -223) at pH 7.5 in its free form have been assigned previously (35). The amide backbone assignment was complete with the exception of four stretches (Gln-23 to Glu-25, Leu-107 to Lys-116, Ser-165 to Gly-168, and Phe-213 to Tyr-216) and the amino acids Phe-37, Asn-40, Leu-45, Thr-65, Lys-74, Leu-76, Lys-84, Gly-132, Phe-138, Ser-140, Asp-156, Thr-176, Glu-199, and Gln-212.
Mapping of the Binding Interface of the Cdc37 M /Hsp90 N Complex by NMR-To map the Cdc37 M binding site for Hsp90 N , NMR titrations were performed with 15 N-labeled Cdc37 M and unlabeled Hsp90 N at 25°C and pH 7.5. An overlay of the 1 H, 15 N-HSQC spectra of free Cdc37 M and the Cdc37 M ⅐Hsp90 N complex is shown in Fig. 3A. These spectra reveal significant changes in the combined amide chemical shifts ( 1 H, 15 N) of the Cdc37 M upon addition of Hsp90 N (up to a molar ratio of 1:1). Many peaks in the two-dimensional 1 H, 15 N-HSQC spectrum of Cdc37 M are either significantly shifted or disappear (exchanged broadened) from the spectra indicating complex formation. The assignment of the shifted peaks was confirmed by using a combination of 15 N amino acid-selective labeling and three-dimensional 1 H, 15 N-NOESY-HSQC spectra. To facilitate the assignment in the complex, Cdc37 M was selectively labeled with ten different 15 (Fig. 4B). When we combine CSPs, CSTs, and the decrease of signal intensity and map the binding site onto the Cdc37 M crystal structure, two distinct regions appear to be involved in complex formation. The first region is composed of the loop (residues 164 -167), and the second region is defined by the loop connecting ␣-helices 3 and 4 and the beginning region of the ␣-helix 4 (residues 203-208).
The reciprocal experiments using 15 N-labeled Hsp90 N and unlabeled Cdc37 M to identify the residues of human Hsp90 N involved in the interaction with the Cdc37 M were carried out as well. An overlay of the 1 H, 15 N-HSQC spectra of free Hsp90 N and the Hsp90 N ⅐Cdc37 M (molar ratio 1:1) complex is shown in Fig. 3B. The shifted amide resonances of Hsp90 N were assigned using a combination of selectively labeled samples and three-dimensional 1 H, 15 N-NOESY-HSQC spectrum. Hsp90 N was selectively labeled with six different 15 Nlabeled amino acids (Ala, Gly, Ile, Leu, Met, and Phe). An additional three-dimensional 1 H, 15 N-NOESY-HSQC spectrum was recorded for a sample in which four different amino acids (Ala, Gly, Phe, and Leu) were selectively 15 N-labeled. The Hsp90 N amide CSPs upon binding to Cdc37 M were analyzed to characterize the amino acids potentially involved in the interaction. The largest combined ( 1 H, 15 N) chemical shift changes (Ͼ0.10 ppm) are observed for a few residues in the first, second, and fifth ␣-helix (residues 29, 55, and 103, respectively), and many residues in the sixth ␣-helix (residues 117-119 and 121-126) as well as in the seventh ␣-helix (residues 128, 129, and 131). These residues define two distinct Hsp90 N interacting sites for Cdc37 M , which are located on one side of the molecule and are localized around a hydrophobic patch (Fig. 4C).
Determination of the Human Cdc37 M ⅐Hsp90 N Complex Structure-Based on the interaction interface as determined from CSPs and CSTs (Fig. 5, A and B) we calculated the structure of the binary Cdc37 M ⅐Hsp90 N complex using HADDOCK (36). The calculations were performed using the crystal structures of human Cdc37 M determined here and Hsp90 N in the open conformation (pdb: 1YES (7)). The initial docking model proposed a number of possible solutions that differed in the relative orientation of the proteins (Fig. 5C). To resolve this ambiguity, we measured two sets of RDCs of the complex. RDCs could be obtained using either 5 g/liter phage pf1 phage alignment medium (Profos) or alignment in 4% polyacrylamide gels. Both sets of 1 D(N,H) RDC values were extracted from IPAP-( 1 H, 15 N)-HSQC spectra and yielded a linearly independent overall alignment tensor (Fig. 5C) and thus provided additional information on the relative orientation of the two proteins. The RDCs have been measured on a sample containing both 15 Nlabeled Cdc37 M and 15 N-labeled Hsp90 N in a 1:1 ratio.  FEBRUARY 6, 2009 • VOLUME 284 • NUMBER 6

The Human Cdc37⅐Hsp90 Complex Studied by NMR
Although some amide signals of both proteins overlap, there is sufficient dispersion to identify many residues from Cdc37 M or Hsp90 N . In the phage alignment media, we could reliably identify 42 1 (Fig. 5D) indicating that their overall structure will not change much when in complex. In addition to the RDCs, we measured intermolecular NOEs. Two-dimensional 15 N-edited NOESY spectra were recorded for the free 15 N, 2 H Cdc37 M and Hsp90 N proteins as well as two-dimensional and threedimensional 15 N-filtered NOESY spectra for the 15 N, 2 H-labeled protein when in complex with the 14 N, 1 H (unlabeled) partner. In these experiments, we could identify several intermolecular cross-peaks. However, because no full side-chain assignment is available, in most cases we could not unambiguously assign them. Based on dockings using the interaction interface either with or without the RDCs, we included one strong intermolecular NOE as observed for Hsp90 N Ala-117-HN to Cdc37 M Ala-204-HB in the final docking calculation.
In the final docking calculations, the active and passive residues in the interaction site and the flexible interfaces have been included as stated in under "Experimental Procedures" as well as both sets of RDCs and the strong intermolecular NOE. The top-ranked docking model for the Cdc37 M ⅐Hsp90 N complex (Fig. 5C) shows the preferential binding mode and includes 185 out of 200 final structures when clustered within the 3.0 Å pairwise backbone r.m.s.d. These final structures have been analyzed whether they satisfy the biochemical and structural restraints. The active residue 168 of Cdc37 M causes violations at a 2 Å ambiguous interaction restraint cut-off, but is still acceptable at a distance of 3 Å from Hsp90 N .
Cdc37 M -Hsp90 N Interface and Structural Reorganization upon Complex Formation-The interface between Cdc37 M and Hsp90 N buries a total surface area of 1579 Ϯ 92 Å 2 , which is highly comparable to the interface for the average protein⅐protein complex (1600 Ϯ 400 Å 2 ) (52). There are two main Cdc37 M ⅐Hsp90 N interaction regions that together form the interface. The hydrophobic core is mainly formed by the interaction of Cdc37 M residues in the ␣ 4 and ␣ 5 helix with the residues in the Hsp90 N ␣ 6 helix and adjacent Left, the molecular surface is color-coded by electrostatic potential, as calculated with DELPHI (38). Right, CSPs from Fig. 4C (Ͼ0.1 ppm) in purple mapped onto the structure. C, bundle of the Cdc37 M ⅐Hsp90 N complex as obtained using HADDOCK, with and without using RDCs. For the calculation without RDCs (left), the ten best docking results for each of the two major HADDOCK clusters are shown, and for the calculation with RDCs (right) the ten best docking results are shown. The active/passive restraints were used as defined under "Experimental Procedures" and are colored, respectively, in red and yellow for Hsp90 N or blue and dark-green for Cdc37 M . The RDC tensors for alignment in either polyacrylamide (PAA) gel or pf1 Phages are indicated below. D, correlation of the experimental versus back calculated RDCs for the free-form x-ray structures of Cdc37 M (blue) and Hsp90 N (red). Protons were added to the x-ray structures using CNS 1.1 (49) without using the RDCs. RDCs were back-calculated using the program PALES (43) with the bestFit option. Left: correlation plot in PAA gel alignment medium with Cdc37 in blue (25 RDCs, correlation coefficient of 0.822) and Hsp90 N in red (52 RDCs, correlation coefficient of 0.903). Right: correlation plot in pf1 phages alignment medium with Cdc37 in blue (42 RDCs, correlation coefficient of 0.816) and Hsp90 N in red (79 RDCs, correlation coefficient of 0.964).
residues. Additionally, the Cdc37 M ␣ 1 , ␣ 2 helices and connecting loop interact with the Hsp90 N ␣ 7 helix and adjacent loop regions, involving many charged residues. Our docking results show an ensemble with many electrostatic and hydrophobic contacts as shown in of the molecule. When superimposing the Cdc37 M ⅐Hsp90 N complex onto the uncomplexed molecules (coordinates from our Cdc37 M crystal structure and 1YES for Hsp90 N ), the r.m.s.d. on the C ␣ atoms is 0.7 Å for Cdc37 M and 0.8 Å for Hsp90 N (0.5 Å for Hsp90 N when excluding residues 106 -127). The difference in the helical region (residues 106 -127) of Hsp90 N is due to the involvement of this region as a hydrophobic core in the complex structure.
Residue Leu-205 of Cdc37 Is Critical for the Complex Formation-Our structure of the Hsp90 N ⅐ Cdc37 M complex indicates the important residues involved at the protein-protein interface. These residues can be divided into two categories: one forms typical salt bridge interactions, and the other forms the hydrophobic core of the complex as described previously. NMR cross-saturation transfer data show that Leu-205 in Cdc37 M experiences the largest decrease in signal intensity. To assess the role of Leu-205, we mutated this residue to phenylalanine, isoleucine, alanine, glycine, aspartic acid, and glutamic acid. The two-dimensional 1 H, 15 N-HSQC spectra of all mutants typically retain the spectra of the wild-type, except for the region around the mutation, indicating that there are no large conformational changes brought about by the mutation. Typically, NMR can identify protein-protein interactions even with as low affinity as ϳ5 mM (K D ). The wild-type Cdc37 M binds to Hsp90 N with a K D of 4.0 M (26, 32) and shows strong changes in the 1 H, 15 N-HSQC spectrum upon complex formation. For mutants L205D and L205Q, changes in the 1 H, 15 N-HSQC cannot be observed ([Cdc37 M ]:[Hsp90 N ] ϭ 1:4), indicating that those Cdc37 M mutants do not bind anymore to Hsp90 N (Fig. 7). Control experiments showed that the L205F and L205I mutants retained the interaction, whereas there was a significant decrease in the affinity for mutants L205A and L205G. The observation supports that residue Leu-205 of Cdc37 M drives the complex formation by its hydrophobic nature.
Comparison with Previously Published Yeast Hsp90 N -Human Cdc37 Crystal Structure-In our study we used human Hsp90 N , which has 69% sequence identity (Fig. 8) with that of the yeast homologue (32). However, the residues of Hsp90 in contact with Cdc37 are well conserved to a large extent, with the exception of Gln-123, which is serine in the case of yeast homologue. A superposition of our structure of the human Cdc37 M ⅐Hsp90 N complex with the Cdc37 (human)-Hsp90 (yeast) crystal structure (pdb: 1US7) (32) reveals a similar overall conformation, with an r.m.s.d. of 1.0 Å, when fitted onto the C ␣ atoms of the Cdc37. The interface of our NMR-based solution structure has a buried surface area of ϳ1600 Å 2 , which is distinctly larger than the ϳ1100 Å 2 of the crystal structure. Our human Cdc37 M ⅐Hsp90 N structure is thus more compact than the previously reported heterogeneous structure. The most   FEBRUARY 6, 2009 • VOLUME 284 • NUMBER 6 eminent differences in Cdc37 M are observed in the ␣ 6 -␣ 7 -loop region. This difference already observed in the free-form crystal structure and very likely to arise from the fact that Cdc37 exists as a dimer in the crystal structure. Furthermore, changes are observed for the orientation of some side chains (Arg-167 and Gln-208) in the ␣ 1 -␣ 2 region of Cdc37 M .

DISCUSSION
Studying the protein-protein interaction of the full-length human Cdc37 (ϳ45 kDa) with human Hsp90 N (ϳ23 kDa) at molecular detail by NMR is most challenging due to the relative large size of this complex. Because the best quality NMR spectra was obtained with the Cdc37 M domain, which is known to interact with Hsp90, this study was focused on the examination of the Cdc37 M structure and its interaction with Hsp90 N . An x-ray structure of human Cdc37 M in complex with yeast Hsp90 N was already available (32). In this study, an albeit longer Cdc37 construct (amino acids 148 -347) was used. Interestingly, this construct was found to form a dimer. The core of the dimer interface has been reported to be around Gln-247 and Tyr-248 and the prolonged C terminus, which forms a small three-helix bundle (292-347), is not at all involved in dimer formation (32). However, we did not observe any evidence for dimer formation with the constructs used in our study. Also our experiments on the Cdc37 M ⅐Hsp90 N complex did not show signals that indicate formation of larger complexes. The existence of Cdc37 M as a monomer was strongly supported by the high resolution x-ray structure of uncomplexed Cdc37 M (Fig. 2). It should be noted, however, that the overall structure of monomeric and dimeric Cdc37 M remains the same. The largest differences between the structures are found in the previously reported dimer interface, which is most probably caused by crystal packing and which is interestingly near to Hsp90 in the complex. Our findings for monomeric Cdc37 are also consistent with recent observations from Vaughan et al. (33) who analyzed a complex of Hsp90⅐Cdc37⅐Cdk4 by cryoelectron microscopy.
The interaction interface between human Cdc37 M and Hsp90 N was mapped based on NMR studies. Mapping of the CSPs and CSTs onto the structures reveals two eminent surface areas corresponding to a small area with charged residues and a large hydrophobic area. Subsequently, the docking program HADDOCK was used to calculate the NMR-based structure of the complex. When only using the ambiguous distance restraints that are based on all the existing data (CSPs, CSTs, and NOE) for the docking calculation, the final solutions did not converge and yielded two possible clusters with a different orientation. A single solution was obtained when we incorporated RDCs in our calculations. The overall conformation of the final NMR-based structure of the human complex is very similar to the known crystal structure of the heterogeneous complex. However, there are some clear differences observed for specific intermolecular contacts in the distinctly larger interaction interface, most probably causing this structure to be slightly more compact.
The Hsp90 residues in the interface of the complex are very well conserved, except for Gln-123, which is serine in case of the yeast homologue. In the crystal structure with the yeast homologue of Hsp90, this serine residue forms an intermolecular hydrogen bond with the Lys-202 side chain of Cdc37. In our model of the human complex, however, we find that the Gln-123 side chain is just outside the binding interface, and makes no significant contribution to the interaction. Instead, in this case, the nearby side chain of the Glu-120 residue of the human Hsp90 generally forms a hydrogen bond with the Lys-202 side chain of Cdc37. An important hydrogen bond is formed between Glu-47 (human sequence numbering) of Hsp90 and Arg-167 of Cdc37, which is considered to be the key interaction that arrests the ATPase action of Hsp90 (32). In the best structure from our docking bundle, we indeed also observe this hydrogen bond. However, we do not observe this hydrogen bond in all the docking structures, which is most probably a result of the two possible conformations for the side chain of Arg-167 that we observed in our free-form Cdc37 M x-ray structure, that we use for our docking. We suppose that this side chain is flexible in the free-form Cdc37 and stabilizes the complex by forming an intermolecular hydrogen bond with Hsp90. Furthermore, in the heterogeneous crystal structure of the complex, the backbone oxygen of Hsp90 residue Gln-133 (human sequence numbering) accepts an intermolecular hydrogen bond from the side-chain Gln-208 of Cdc37, and its side-chain oxygen forms intermolecular hydrogen bonds with the backbone Arg-166 and Arg-167 amides of Cdc37. In our human complex we did not observe the interaction between the Hsp90 residue Gln-133 and the Cdc37 residue Gln-208. This could again be a result of the two possible conformations observed for this Gln-208 side chain we observed in the freeform x-ray structure, but on the other hand for the different orientation of this side chain in our human model, we observed that the amino moiety is rather close to the side-chain hydroxy of the Ser-113 residue of Hsp90, with which it may possibly form an intermolecular hydrogen bond. The hydrogen bonds between the side chain of Hsp90 residue Gln-133 and the backbone Arg-166 and Arg-167 amides of Cdc37 is very well maintained in our human structure of the complex. Furthermore, the hydrophobic core in our NMR-based structure of the com- plex includes the proximity of Ala-124 of Hsp90 N and Trp-193 of Cdc37 M , which are slightly further apart from each other in the heterogeneous crystal structure.
In recent years, the Hsp90 protein became attractive as a pharmaceutical anti-cancer target. Many protein kinases, of which some might be causative for cancer when deregulated or overexpressed, require Hsp90 to be maintained in a functional state. Targeting of Hsp90 thus seems to be attractive, because several protein kinases can be targeted at the same time. The Hsp90-specific inhibitor, 17-allylamino-17-demethoxy geldanamycin is already undergoing clinical trials (53). Hsp90-specific inhibitors known so far are directed against the ATP-binding pocket of Hsp90. Because Hsp90 needs co-chaperones to fulfill its function and one of these, Cdc37, is mainly found in combination with regulation of protein kinases, the whole Hsp90⅐Cdc37 complex could be considered as an interesting drug target. In addition, increased levels of Cdc37 have shown to enhance cancer cell growth (54). A drug that inhibits the formation of this complex might be a very valuable compound. Knowing the interaction interface of the Cdc37⅐ Hsp90 is a prerequisite for designing substances inhibiting formation of the Hsp90⅐Cdc37. In our study we analyzed the human Cdc37 M ⅐Hsp90 N complex in great detail, and identified Leu-205 of Cdc37 as a key residue. This residue resides in the interface and makes many contacts with different residues of Hsp90 to stabilize the complex. Mutational analysis of Leu-205 indicates that the hydrophobic nature of this critical residue promotes the complex formation.
Very recently, a compound was identified that disrupts the Hsp90⅐Cdc37 complex in cancer cells (55) supporting the idea to target the Hsp90-Cdc37 interaction. We expect that the structure of the human Cdc37 M ⅐Hsp90 N complex solved in this study can be used to assist further studies in the development of small molecule inhibitors.