2.1 Synthesis of a small molecule

The compound FS5 was prepared according to the methods reported in our previous work [43]. N-(3-(5-chloro-2,4-dihydroxyphenyl)-4-(4-methoxyphenyl)isoxazol-5-yl)cyclopropanecarboxamide (9). 1H NMR (400 MHz, CD3OD) δ 7.13 (d, J=8.4 Hz, 2H), 7.04 (s, 1H), 6.90 (d, J=8.4 Hz, 2H), 6.44 (s, 1H), 3.78 (s, 3H), 1.76 (m, 1H), 0.94-0.85 (m, 4H); MS(ESI): 400.9[M+H]+ (Fig. 1).

Fig. 1Full-screen View

Synthesis of the small molecule FS5. (a) CH3COOH, BF3·OEt2, 90 °C, 3.5 h; (b) PhCH2Br, K2CO3, MeCN, 100 °C, 12 h; (c) Br2, NaOH, 1,4-dioxane/H2O, 12 h; (d) CH3I, K2CO3, MeCN, 100 °C, 2 h; (e) 4-methoxy phenylacetonitrile, LHMDS, THF, -78 °C, 1 h; (f) NH2OH·HCl, pyridine, 100 °C, 12 h; (g) Cyclopropanecarbonyl Chloride, NEt3, DCM, DMAP, 0 °C, 1 h; (h) BCl3, DCM, rt, Room Temperature, 1 h.

2.2 Protein purification and crystallization

The Hsp90 N-terminal domain residues (9-236) were cloned into the vector pET28a to obtain the final plasmid for the overexpression. This plasmid was transformed into competent E. coli cells, Rosetta DE3. The bacteria were grown in 800 mL of LB (Luria-Bertani) broth at 37 °C. Upon reaching OD600 value of 0.6–0.8, 200 µM IPTG was added to induce the expression of N-terminal Hsp90 and further incubated for 3–5 h at 30 °C. We then centrifuged the cells at 10,000 g for 10 min at 4 °C and resuspended the precipitate in buffer A (100 mM Tris/HCl buffer, pH 7.5, 300 mM NaCl, and 5% glycerol) for rupturing the cells with the JNBIO 3000 plus (JNBI). The supernatant containing protein of interest was acquired by centrifugation at 30,000 g and 4 °C for 30 min. The supernatant was then transferred into a 5-ml Ni-NTA (Ni²⁺-nitrilotriacetate) column (GE Healthcare) together with buffer B (100 mM Tris/HCl buffer, pH 7.5, 300 mM NaCl, 100 mM imidazole, and 5% glycerol) for the recovery of Hsp90 protein using imidazole. Superdex 75 PG gel-filtration column (GE Healthcare) allows a high yield of the overexpressed protein to be obtained and the removal of imidazole by changing the buffer system to buffer C (100 mM Tris-HCl buffer, pH 7.5, 300 mM NaCl, and 5% glycerol). The positive peak protein was collected and verified by SDS-PAGE. Finally, the protein was flash-frozen in liquid nitrogen and stored at -80 °C.

For the experiments, the protein was thawed and adjusted to the concentration of 20 mg/ml in Amicon Ultra-15,10000Mr cut-off centrifugal concentrator (Millipore). Subsequently, the solution of small molecules was mixed with the protein in a ratio of 5:1 and incubated at 4 °C for 1 h. The hanging drop vapor diffusion method was used to gain crystals at 4 °C using previously described modifications for crystallization (100 mM Tris-HCl buffer, pH 8.5, 20% PEG 4000, 200 mM MgCl₂)[44].

2.3 Data collection, structure determination and refinement.

The crystals were tailored with cryo-loop (Hampton research, USA) and then flash-frozen in liquid nitrogen for collecting better X-ray data. All datasets were generated at 100 K on macromolecular crystallography beamline 17U1 (BL17U1) at Shanghai Synchrotron Radiation Facility (SSRF, Shanghai, China) through an ADSC Quantum 315r CCD detector. The obtained results were processed by the HKL2000 software package. The structures of complex Hsp90^N-FS5 were determined by molecular replacement with PHENIX software. The apo structure of Hsp90^N (PDB code 3T0H) was considered as a model for the analysis of Hsp90^N-FS5 complex [45]. The Coot software was used to rebuild the initial model. The final models were refined to resolution limit of 1.65Å by using the PHENIX software. The superimposed data were analyzed with PyMOL software package. The complete data collection and statistics of refinement are shown in Table 1. The complex Hsp90^N-FS5 structure has been deposited in PDB (PDB code 5XRB).

3.1 Purification, crystallization, and structure determination of Hsp90^N-FS5

The intensively purified protein was collected using metal-chelating chromatography and gel filtration chromatography. The ordinate value of the gel filtration chromatography reached 800 mAu, and a single sharp peak was achieved indicating the high overall purity of the target protein. The collected protein was subjected to SDS-PAGE analysis, which was clearly consistent with the expected band at the molecular weight of 25 kDa, and the purity reached 98%.

After 3-5 days, the complex crystals of Hsp90^N-FS5 were obtained by the hanging-drop method at 277 K. The form of the crystals is shown in Fig. 2. The mean dimension of the crystals was roughly 250 μm × 150 μm × 50 μm.

Fig. 2.Full-screen View

(Color online) Crystals of Hsp90^N-FS5 complex. The crystals were obtained by the hanging-drop method at 277 K. The average dimension of the crystals was 250 μm × 150 μm × 50 μm.

The complete information on the crystals was collected for the statistical analysis, which is shown in Table 1. The structure of complex Hsp90^N-FS5 has been deposited in PDB (PDB code 5XRB). The diffraction data of crystals were collected at 1.65 Å resolution and indexed in space groups I222. The unit cell parameters were as follows: a = 67.45 Å, b = 90.70 Å, c = 98.28 Å; α = β = γ = 90.00°. The resolved structure of Hsp90^N-FS5 included only Val17-Glu223, while no electron density was obtained for residues Asp9-Glu16 and Val224-Glu236 in complex Hsp90^N-FS5, which is may be due to the disordered residues in the electron density.

3.2 Analysis of crystal structure and interaction of Hsp90^N-FS5

The structure of the Hsp90 N-domain includes a high conservative area for ATP binding, which comprises H2, H4, H7, and L1. This pocket is about 15 Å in depth and has an α/β sandwich-like structure. The pocket base consists of eight uninterrupted antiparallel β-sheets. According to the crystal structure resolved by X-ray, compound FS5 completely binds to the ATP pocket, which is shown in Fig. 3(a) and Fig. 3(b) in two different directions, and the intact electron density of the compound FS5 has been captured and is shown in Fig. 3(c).

Fig. 3.Full-screen View

(Color online) Structure of the complex and the interactions between Hsp90^N and FS5. The complex Hsp90^N-FS5 is demonstrated in two different directions: (a) and in 90 degrees clockwise rotation (b). Hsp90^N is shown in polychrome and FS5 in green (a) and cyan (b). (c) Electron density map of FS5. The map was contoured at 1/σand the FS5 model is shown in sticks. The green, blue, and red express atoms carbon, nitrogen, and oxygen, respectively. (d) Interactions between Hsp90^N and inhibitor FS5. FS5 is shown in sticks, and the color of the main atoms of FS5 like carbon, nitrogen, and oxygen are expressed in green, blue, and red, respectively. Water is shown as a red small ball, hydrogen bonds in black dashed lines, and the residues that have interactions with Hsp90^N are shown in lines in various colors.

There is a crucial structure in Hsp90^N termed ATP lid, which comprises H4-L2-H5-H6 (residues 99-134). This structure can adjust the size of the ligand-binding pocket by flipping its loop in order to regulate the conformation switch between the opened and closed states. Only at the opened conformation can the ligand bind to Hsp90. Subsequently, Hsp90 can activate target proteins and accomplish its function. Thus, ATP lid plays an indispensable role in facilitating the efficient binding of ATP in vivo.

We compared the crystal structures of Hsp90^N-FS5 (PDB code 5XRB) with other available structures for Hsp90^N (PDB code 3T0H), Hsp90^N-ATP (PDB code 3T0Z), Hsp90^N-GDM (PDB code 1YET) by using PyMOL software [5, 45]. It was clearly observed that the refined crystal structure of Hsp90^N-FS5 is different from Hsp90^N, Hsp90^N-ATP, Hsp90^N-GDM, and this difference is mainly focused on H4 and H5. The crystal packing patterns are also distinct, such as Hsp90^N-FS5 (I222), Hsp90^N (I222), Hsp90^N-ATP (P21), and Hsp90^N-GDM (P12₁1).

The identified differences between Hsp90^N-FS5 and Hsp90^N, Hsp90^N-ATP, and Hsp90^N-GDM are described below. As shown in Fig. 4a, we can detect the difference spanning Leu103-Ile128 between Hsp90^N-FS5 and Hsp90^N, which is a not-overlapping area in superimposition. Meanwhile, as demonstrated in Fig. 4b, we observed a conformation change directly in H4 and H5 following the binding of Hsp90 with compound FS5. At the same time, the L2 loop is rearranged. Because H4, L2, and H5 are parts of the ATP pocket, changes in L2 loop directly lead to the shape alterations of the binding pocket. Specifically, binding of FS5 to Hsp90 induces narrowing of the ligand-binding pocket and its entrance is closed, which makes Hsp90 unreachable for ATP and results in the complete dysfunction of Hsp90.. In the same way, as shown in Fig. 4c and Fig. 4e, Met98-Gly137 in Hsp90^N-FS5 is different from that in Hsp90^N-GDM, and Met98-Met130 in Hsp90^N-FS5 is different from that in Hsp90^N-ATP. These results illustrate that the conformation changes after Hsp90 binding with the ligand are different between compound FS5 and other classic ligands. As shown in Fig. 4d and Fig. 4f, the most differences between these three complex structures were observed for H4, H5, and L2 loop.

Fig. 4.Full-screen View

(Color online) Comparison of the crystal structure of Hsp90^N-FS5 with classic structures. The crystal structure of Hsp90^N-FS5 is compared with Hsp90^N, Hsp90^N-GDM, and Hsp90^N-ATP, respectively. The comparison is shown both in ribbon and cartoons. In the ribbon, residues of both ends of the not-overlap area are demonstrated in the sticks and described, while in the cartoon, the difference between the two is ringed. a. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N (blue) in the ribbon. b. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N (blue) in the cartoon. c. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N-GDM (green) in the ribbon. d. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N-GDM (green) in the cartoon. e. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N-ATP (cyan) in the ribbon. f. Comparison of Hsp90^N-FS5 (orange) and Hsp90^N-ATP (cyan) in the cartoon.

The electrostatic potential surface distribution of complexes Hsp90^N, Hsp90^N-ATP, Hsp90^N-GDM, and Hsp90^N-FS5 with the ligands is shown in Fig. 5. It was revealed that the electrostatic potential surface surrounding the active pocket in Hsp90^N (Fig. 5a) is different from that of Hsp90^N-GDM (Fig. 5b), Hsp90^N-ATP (Fig. 5c), and Hsp90^N-FS5 (Fig. 5d). According to the position of GDM and ATP, we can conclude that compound FS5 binds Hsp90^N in the ligand-binding pocket. There is enough space for optimal FS5 binding.

Fig. 5.Full-screen View

(Color online) Electrostatic potential surface distribution of Hsp90^N-ligands: (a), Hsp90^N-GDM, (b), Hsp90^N-ATP, (c), Hsp90^N-FS5 (d) in the same direction. The color of the surface denotes the electrostatic potential, while red signifies negative charge and blue signifies positive charge. The ligands GDM, ATP, FS5 are shown in sticks (green).

Furthermore, we compared structures of Hsp90^N-FS5 complex with other Hsp90^N complexes available for some promising inhibitors together with the comparison of the inhibitor structures. Thus, there is a big difference between Hsp90^N-Ganetespib (PDB code 3TUH), Hsp90^N, and Hsp90^N-FS5. The H5 and L2 loops in all three structures demonstrate high discrepancy. As shown in Fig. 6b, the L2 loop in Hsp90^N is turned inward the pocket, and indicating the minimal size of the pocket between these three structures. However, the L2 loop in Hsp90^N-FS5 is neither reversed nor pronated, and is almost flat with H4 and H5. The conformational change of Hsp90^N upon binding of FS5 increases the space in the ligand-binding pocket. The Hsp90^N-Ganetespib has different structure compared with the two described above. The first half of its L2 loop is turned outward, whereas the second half is turned inward, leading to the significant difference of the shape of the ligand-binding pocket. At the same time, the C-terminal part of H5 of Hsp90^N-Ganetespib also has some differences with Hsp90^N and Hsp90^N-FS5. With regard to its position, the C-terminal of H5 of Hsp90^N is above, C-terminal of H5 of Hsp90^N-FS5 is in the middle, and C-terminal of H5 of Hsp90^N-Ganetespib is at the bottom of the complexes. The major parts of Ganetespib and FS5 structures are similar, showing differences only in individual residues (Fig. 6a).

Fig. 6Full-screen View

(Color online) Comparison of the orientation of FS5 in Hsp90^N-FS5 with classic inhibitors. Hsp90^N, Hsp90^N-FS5, Hsp90^N-Ganetespib, Hsp90^N-XL888, Hsp90^N-NMS-E973, and Hsp90^N-17-DMAG are shown in red, yellow, blue, green, magenta, and wheat, respectively. The superimposition of complex structure is shown in the cartoon, and the corresponding inhibitors are shown as sticks. (a) Structure comparison of inhibitors FS5 and Ganetespib. (b) Structure comparison of complexes Hsp90^N, Hsp90^N-FS5, and Hsp90^N-Ganetespib. (c) Structure comparison of inhibitors FS5, Ganetespib, and XL888. (d) Structure comparison of complexes Hsp90^N, Hsp90^N-FS5, Hsp90^N-Ganetespib, and Hsp90^N-XL888. (e) Structure comparison of inhibitors FS5, Ganetespib, and NMS-E973. (f) Structure comparison of complex Hsp90^N, Hsp90^N-FS5, and Hsp90^N-NMS-E973. (g) Structure comparison of inhibitor FS5, Ganetespib, NMS-E973, and 17-DMAG. (h) Structure comparison of complex Hsp90^N, Hsp90^N-FS5, Hsp90^N-NMS-E973, and Hsp90^N-17-DMAG.

Comparison of Hsp90^N-XL888 complex (PDB code 4AWQ) with Hsp90^N and Hsp90^N-FS5 also demonstrated essential difference between all three structures [46]. At the same time, comparing inhibitor XL888 with Ganetespib similarly revealed high structural discrepancy.. Notably, the L2 loop of Hsp90^N-XL888 is turned more inward than in Hsp90^N, which further reduces the space in the ligand-binding pocket. As shown in Fig. 6d, C-terminal of H5 is Hsp90^N, Hsp90^N-FS5, Hsp90^N-Ganetespib and Hsp90^N-XL888 are different. Moreover, the inhibitor XL888 and FS5 vary significantly. The structure of XL888 inhibitor has a long, almost a bar shape, and the framework is very different from FS5. Indeed, the structure of XL888 differs essentially from all the other inhibitors. (Fig. 6c)

Hsp90^N-NMS-E973 (PDB code 4B7P) is different when compared with Hsp90^N and Hsp90^N-FS5 (Fig. 6f) [47]. The L2 loop of Hsp90^N-NMS-E973 scrolls completely down that is different from Hsp90^N and Hsp90^N-FS5 as depicted before. This particular conformation leads to the largest space in the ligand-binding pocket between all the conformations that we discussed in this paper. Likewise, the C-terminal of H5 rotates down as well, which position is different in Hsp90^N, Hsp90^N-FS5, Hsp90^N-Ganetespib, and Hsp90^N-XL888. Because L2 loop and C-terminal of H5 are both downward, the shape of H4-H5 of Hsp90^N-XL888 is inverse-triangular. As can be observed in Fig. 6e, the structures of these three inhibitors, NMS-E973, Ganetespib, and FS5 are quite different.

Interestingly, the comparison of the crystal structures of Hsp90^N, Hsp90^N-FS5, Hsp90^N-NMS-E973, and Hsp90^N-17-DMAG (PDB code 1OSF) indicated that the L2 loop of Hsp90^N-NMS-E973 and Hsp90^N-17-DMAG are very similar (Fig. 6h) [48]. The orientation, range, and position were almost overlapped. However, there are some subtle differences in C-terminal of H5 between the two complexes so that Hsp90^N-17-DMAG drops less than Hsp90^N-NMS-E973. C-terminal of H5 in the structures of Hsp90^N, Hsp90^N-FS5, Hsp90^N-17-DMAG, Hsp90^N-NMS-E973 are different, and the structure of inhibitor 17-DMAG varies greatly from the both NMS-E973 and FS5. (Fig. 6g)

According to these comparisons, it is noticeable that the binding structures with the inhibitor show remarkable changes in only the L2 and H5 loops, while the other parts are almost similar, suggesting that L2 and H5 are significant for Hsp90 activity. At the same time, being a part of a ligand-binding pocket, the alterations of L2 and H5 directly lead to a change of space in the pocket. Therefore, the design of an appropriate inhibitor should be particularly focused on L2 and H5.

The binding of diverse ligands to Hsp90 induces different conformational changes in the ligand-binding pocket, as shown in Fig. 7. We added another complex structure of Hsp90^N-BIIB021 (PDB code 3QDD) to show more changes of the ligand-binding pocket [49]. It is apparent that different ligand structure leads to the corresponding alterations in the shape of the ligand-binding pocket. According to the structure of the pocket and the ligand, and their relative position, we can optimize these inhibitors by modifying functional group to enhance performance and reduce side-effects. For better resolution of the inhibitor structures, the 2D diagram of inhibitors is shown in the lower right corner of the corresponding ligand-binding pocket.

Fig. 7Full-screen View

(Color online) The relative position of inhibitors in different structures. The ligand-binding pocket of Hsp90^N is shown in white. (a) GDM-cyan; (b) ATP-salmon; (c) FS5-yellow; (d) Ganetespib-blue; (e) XL888-green; (f) NMS-E973-magenta; (g) 17-DMAG-wheat; (h) BIIB021-red.

As it is shown in Fig. 3(d), there are two benzenes in compound FS5, and one of the two benzenes has two hydroxyls. One hydroxyl forms four hydrogen bonds mediated by water with residues Leu48, Ser52, Asp93, and Thr184 (2.98 Å, 2.89 Å, 2.92 Å, and 3.12 Å, respectively). The other hydroxyl forms three hydrogen bonds mediated by water with residues Asp93, Thr184, Gly97 (2.76 Å, 2.87 Å, 3.01 Å, respectively), and a direct hydrogen bond with residue Asp93 (2.66 Å). Meanwhile, the nitrogen in heterocycle forms not only three hydrogen bonds mediated by water with residues Thr184, Asp93, Gly97 (2.87 Å, 2.76 Å, 3.01 Å, respectively), but also forms a direct hydrogen bond with residue Thr184 (3.33 Å). Besides, the other nitrogen in FS5 forms two hydrogen bonds mediated by water with residues Gly97 and Asp102 (2.76 Å and 2.76 Å, respectively). One oxygen forms one hydrogen bond mediated by water with residue Lys58 (2.89 Å). These hydrogen bonds assure tight binding of FS5 inhibitor to Hsp90. Furthermore, there are pervasive van der Waals forces between compound FS5 surface and Hsp90N ligand-binding pocket. According to the above reasons, inhibitor FS5 can effectively bind to Hsp90.