A. Synthesis of the small molecule FS23

As shown in Fig. 1, under reaction of lithium hexamethyldisilazide, Compound 1 was condensed with 4-methoxy phenylacetonitrile to give 2, followed by cyclization with hydroxylamine hydrochloride to obtain Compound 3, which reacted with cyclopropanecarbonyl chloride, and subsequently Suzuki reaction with isoquinoline-4-boronic acid took place to give Compound 5, which deprotected by boron trichloride to yield the target compounds FS23.

Fig. 1.Full-screen View

Synthesis of the small molecule FS23. (a) 4-methoxy phenylacetonitrile, lithium hexamethyldisilazide, tetrahydrofuran, -78 ℃, 1 h, 97.2%. (b) NH₂OH.HCl, pyridine, 100 ℃, 12 h, 54.7%. (c) Cyclopropanecarbonyl chloride, NEt₃, dichloromethane, 4-dimethylaminopyridine, 0 ℃, 1 h, 90%. (d) Isoquinoline-4-boronic acid, Pd(PPh₃)₄, K₂CO₃, CH₃OCH₂CH₂OCH₃/H₂O, 120 ℃, 1 h, 82.9%. (e) BCl₃, dichloromethane, rt, 1 h, 43.3%.

B. Protein purification and crystallization

The vector containing the gene of human Hsp90α N-terminal domain (residues 9-236) was cloned into a pET28a vector and the recombinant plasmid was transformed into E. coli Rosetta (DE3) pLysS for over-expression (Invitrogen, Carlsbad, USA). The bacteria were grown in 800 mL of LB (Luria-Bertani) broth at 37 ℃ to an OD₆₀₀ of 0.6–0.8 and then induced by 200 μm IPTG (isopropyl β-D-thiogalactopyranoside) for 3–5 h at 30 ℃. The bacteria were collected by centrifugation at 10000 g for 10 min (CF16RX, Hitachi). The precipitate was re-suspended in buffer A (100 mM Tris/Hcl buffer, pH 7.5, 300 mM NaCl and 5% glycerol) and then crushed by JNBIO 3000 plus (JNBI). The product was centrifuged at 30000 g for 30 min at 4 ℃ and the supernatant was loaded onto a 5 mL Ni-NTA (Ni²⁺-nitrilotriacetate) column (GE Healthcare) and protein was eluted with buffer B (100 mM Tris-Hcl buffer, pH 7.5, 300 mM NaCl, 100 mM imidazole, and 5% glycerol). The protein product was concentrated and buffer-exchanged to buffer C (100 mM Tris-Hcl buffer, pH 7.5, 300 mM NaCl and 5% glycerol) using an Amicon Ultra-15, 10000 Mr cut-off centrifugal concentrator (Millipore). For further purification, the impurities were removed by a Superdex 75 PG gel-filtration column (GE Healthcare) with buffer C. Positive fractions were analysed by a 15% SDS-PAGE to determine the purity. The protein was flash-frozen in liquid nitrogen and stored at -80 ℃.

For crystallization, the purified Hsp90 N-terminal domain was concentrated in Amicon Ultra-15, 10000 Mr cut-off centrifugal concentrator to a concentration of about 20 mg/mL. The small molecule FS23 was added to protein with a 5:1 molar ratio and the mixture was incubated for 1 h at 4 ℃. After incubation, the mixture was centrifuged for 10 min and the precipitation was removed from the system. The supernate was analysed by DLS (dynamic light scattering, Malvern) to determine whether the complex was suitable for crystallization. This was done following the standard protocols supplied by the manufacturer. Crystallization was carried out at 4 ℃ using the hanging drop vapor diffusion method and the cocrystals were grown under similar conditions previously described for native crystal [31], with a little difference, 100 mM Tris-HCl, pH 8.5, 20% PEG 4000, 200 mM MgCl₂.

C. Data collection, structure determination, and refinement

Crystals were mounted and flash-frozen in liquid nitrogen for diffraction test and data collection. All data sets were collected at 100 K on beamline BL17U1 at Shanghai Synchrotron Radiation Facility (SSRF) [32] and processed with the HKL2000 software package [33]. The structures were solved by molecular replacement, using the PHENIX [34]. The search model used for the crystal was the previously reported structure of the Hsp90^N (PDB code 3T0H) [31], the structures were refined using PHENIX [34]. With the aid of the program Coot [35], water molecules and others were fitted into to the initial Fo-Fc map. The complete statistics, as well as the quality of the solved structure, are shown in Table 1.

A. Purification of Hsp90^N

Hsp90^N was purified to apparent homogeneity using metal-chelating column, followed by gel filtration chromatography. The elution volume peak corresponds to monomeric Hsp90^N with an apparent molecular weight of 25 kDa and showed a good purity (the purity of protein was assessed by SDS-PAGE to be 98%), as shown in Fig. 2(a). The protein and FS23 were mixed at 1:5 molar ratio, and the result of DLS showed that there was only one peak which abided by the Gauss distribution, indicating that the diameter of particles was almost uniform. The PDI (Polydispersity Index) was 13.7%, so it is conceivable that nearly all the complex existed in monomer, as shown in Fig. 2(b). All these data demonstrated that this mixture was suitable for crystallization.

Fig. 2.Full-screen View

(Color online) Purification of Hsp90^N. (a) Size-exclusion chromatography and SDS-PAGE analysis of purified Hsp90^N. The flow rate was 1 mL/min and a single peak was observed for the purified protein solution. Lane M contains molecular weight marker (labeled in kDa). (b) Distribution of DLS. Hsp90^N and FS23 were mixed at 1:5 molar ratio and the concentration of the protein is 2 mg/mL. There is only one peak in the diagram which shows that the complex is almost monodisperse and would be ideally suited for crystal growth.

B. Crystallization and structure determination

After 3–4 days, crystals were grown, with an average dimension of approximately 200 μm×100 μm×50 μm (Fig. 3). Diffraction data were collected to 1.70 Å resolution (Fig. 4) and indexed in space groups I222. The unit-cell parameters were a=70.06 Å, b=88.27 Å, c=97.00 Å; =90°, =90° and =90°.

Fig. 3.Full-screen View

(Color online) Crystals of the Hsp90^N-FS23 complex obtained by the hanging-drop method at 277 K. The average dimensions of the crystals were approximately 200 μm×100 μm×50 μm.

Fig. 4.Full-screen View

X-ray diffraction from Hsp90^N-FS23 complex crystal. The diffraction pattern was recorded on a ADSC Q315r detector with 0.9792 Å wavelength X-rays on beamline BL17U1 at SSRF, 0.5 oscillation, 0.3 s exposure and 200 mm sample-to-detector distance.

The structure of Hsp90^N-FS23 was solved by molecular replacement using the known 3-D structure of Hsp90^N as a search model (PDB code 3T0H), determined at 1.70 Å resolution (PDB code 5CF0) [31]. In the crystal structure, FS23 binds to the ATP-binding pocket completely and its intact electron density has been captured, as shown in Fig. 5(a). So we have reason to believe that FS23 was fully bound into Hsp90^N in the complex structure.

Fig. 5.Full-screen View

(Color online) Structure of FS23. (a) Stereo view of the FS23 (2Fo-Fc) electron density before any FS23 atoms were built into the Hsp90 model. The blue mesh, contoured at 1.0, is the refined FS23 model in a stick representation. Carbon, nitrogen and oxygen atoms are shown in green, blue and red, respectively. (b) Chemical structure of FS23.

The refined model contains residues Val17-Lys224 and no electron density was observed for residues Asp9-Glu16 and Glu225-Glu236 in the structure. The missing residues are at the N- and C-termini of the model and are presumed to be disordered.

C. Analysis of crystal structure of Hsp90^N-FS23 complexes

Although the refined structure of Hsp90^N-FS23 differs from that of Hsp90^N (PDB code 3T0H) in resolution level, no significant conformational changes were observed between the two protein structures, as shown in Fig. 6(a). But there are some tiny differences from Asp102 to Tyr139 in the two structures. The complex structure of Hsp90^N-FS23 provides the structural basis to further investigate the activated conformational change of Hsp90 after FS23 binding. Superimposition of Hsp90^N-FS23 complex with the native Hsp90^N structures shows that the H4 (amino acids from Thr99) to Leu107 and H5 (amino acids from Gly114 to Ala124) of Hsp90^N undergo a conformational rearrangement in response to the bound FS23 (Fig. 6(b)).

Fig. 6.Full-screen View

(Color online) Comparison of crystal structures of Hsp90^N-FS23 (magenta) with Hsp90^N (green) and Hsp90^N-ATP (cyan) complexes: ribbon (a) and cartoon (b) superimpositions of crystal structures of Hsp90^N-FS23 and Hsp90^N, and ribbon (c) and cartoon (d) superimpositions of crystal structures of Hsp90^N-FS23 and Hsp90^N-ATP complexes.

The crystal packing patterns of Hsp90^N-FS23 (I222) and Hsp90^N-ATP (P21) is different, and the Met98 to Tyr139 in the two structures are different, too (Fig. 6(c)). In the complex of Hsp90^N-ATP, a loop containing amino acids from Leu107 to Thr115, called L6 (Fig. 6(d)). The ATP lid consisting of H4-L6-H5-H6 secondary structure elements has two different conformations in Hsp90^N which is essential for the function of Hsp90 in regulating the entrance and width of the pocket. In free Hsp90^N the lid is in open conformation and the pocket is about 12 Å in diameter near its entrance. When the ATP binds to the pocket, H4 and H5 undergo helix-to-coil and coil-to-helix transition which impel L6 moving into the pocket to replace the H4 as one of the pocket walls. Then the ATP lid changes into closed conformation. The functional consequence is that the L6 loop acts as a gate to constrict the pocket entrance from a width of 12 Å to 8 Å [9]. Only when the ATP lid is in open conformation, Hps90 can interact with its client proteins to accomplish conformational maturation and activation. But there is only a big helix from Thr99 to Ala124 in Hsp90^N-FS23, that is the L6 loop was substituted by helix, which means that even though there is substrate bind to Hsp90, conformational change would not happen, resulting in its interaction with client proteins failed (Fig. 6(d)). The disappearance of L6 in complex Hsp90^N-FS23 plays an important role in explaining why FS23 could block up the function of Hsp90.

Electrostatic potential surface surrounding the active site of Hsp90N-FS23 and Hsp90N-ATP are shown in Fig. 7. In common with the structure of Hsp90N-ATP, FS23 binding site of Hsp90 is located in the ATP binding site adopts the shape of a cleft (Fig. 7(a)). It is interesting that the binding of inhibitor induce structural rearrangements of L6 and produced a hole in the structure (Fig. 7(a)).

Fig. 7.Full-screen View

(Color online) Electrostatic potential surface distribution of Hsp90. (a) Electrostatic potential surface near the FS23-binding site. (b) Electrostatic potential surface near the ATP-binding site. The protein surface is colored to reflect the electrostatic potential, red for negative charge and blue for positive charge, FS23 and ATP in stick representation.

D. Interactions between Hsp90^N and FS23

There are two direct hydrogen bonds between Hsp90^N and FS23. One is the interaction between FS23 and Asp93, with the distance of 2.7 Å. The other is the interaction between FS23 and Thr184, with the distance of 3.5 Å. A π-π stacking interactions were formed between the side chain of the residue Phe138 and aromatic rings of FS23. FS23 forms water-mediated HBs with residues Leu48, Ser52, Ile96, Asp102 and Gly135. The hydrogen bond network between Hsp90^N and FS23 is the most critical portion of the interface and they make FS23 combine Hsp90^N effectively (Fig. 8). What’s more, there is extensive, though not complete, surface complementarity between FS23 and the pocket, and this allows for a high density of van der Waals contacts.

Fig. 8.Full-screen View

(Color online) Interactions between Hsp90^N and FS23. (a) Surface representation of FS23 bound in the pocket of Hsp90^N, protein and FS23 present in cartoon and stick, respectively. (b) Interactions between Hsp90^N and FS23. Hydrogen bonds are shown by black dashed lines with distances between two atoms labeled. Water molecules are shown as red spheres.

Superimposition of Hsp90^N-FS23 (PDB code 5CF0) with Hsp90^N-ATP (PDB code 3T0Z) and Hsp90^N-GDM (PDB code 1YET) structures revealed that the FS23, ATP and GDM occupied the same location, with some subtle difference in the chemical group, though. In the complex of Hsp90^N-GDM (PDB code 1YET) for example, geldanamycin nearly occupies the whole pocket and there is no much room for modification. So it is hard to see significant changes in solubility, bioavailability and hepatotoxicity in derivatives of geldanamycin. But the situation is different for FS23 binding and there is enough room for redesigning and reform of FS23 to obtain derivatives with better qualities (Fig. 9).

Fig. 9.Full-screen View

(Color online) Comparison of the crystal structures of Hsp90^N-FS23 with Hsp90^N-GDM complexes. Superimposition of Hsp90^N-FS23 (magenta, PDB code 5CF0) and Hsp90^N-GDM (cyan, PDB code 1YET).

The pocket is mostly polar at its entrance, so hydrophilic groups which can interact with polar amino acids could be added to FS23 to increase the solubility of small molecule. While it becomes predominantly hydrophobic near the pocket bottom, derivatives with increased affinity for Hsp90 may be yielded if groups with non-hydrogen atoms are added to FS23 and hydrogen bond complementarity is increased. The most interesting portion of the interface in the complex Hsp90^N-FS23 is the hole produced by structural rearrangements of L6. There is neither a direct nor an indirect interaction between residues in the hole and FS23. So groups can be added to FS23 to fill the cavity and interact with residues in the hole to increase the activity of derivatives.