Structures of Hsp90α and Hsp90β bound to a purine-scaffold inhibitor reveal an exploitable residue for drug selectivity

Abstract

Hsp90α and Hsp90β are implicated in a number of cancers and neurodegenerative disorders but the lack of selective pharmacological probes confounds efforts to identify their individual roles. Here, we analyzed the binding of an Hsp90α-selective PU compound, PU-11-trans, to the two cytosolic paralogs. We determined the co-crystal structures of Hsp90α and Hsp90β bound to PU-11-trans, as well as the structure of the apo Hsp90β NTD. The two inhibitor-bound structures reveal that Ser52, a nonconserved residue in the ATP binding pocket in Hsp90α, provides additional stability to PU-11-trans through a water-mediated hydrogen-bonding network. Mutation of Ser52 to alanine, as found in Hsp90β, alters the dissociation constant of Hsp90α for PU-11-trans to match that of Hsp90β. Our results provide a structural explanation for the binding preference of PU inhibitors for Hsp90α and demonstrates that the single nonconserved residue in the ATP-binding pocket may be exploited for α/β selectivity.

Introduction

The hsp90s are a ubiquitous family of ATP-dependent molecular chaperones consisting of four paralogs: Hsp90α and Hsp90β in the cytosol, Grp94 in the endoplasmic reticulum, and Trap-1 in the mitochondria. Hundreds of client proteins, including growth factors, signaling kinases, transcription factors, and cell surface receptors depend on hsp90 family members for their conformational maturation, stabilization, and proper subcellular localization^1,2. This same chaperone function, however, is co-opted in cancer cells to protect a dysregulated proteome, stabilize client oncoproteins that would otherwise be degraded, and promote oncogenesis, making the hsp90s attractive therapeutic targets^3,4.

Most hsp90 inhibitors identified to date target the N-terminal domain (NTD) and block chaperone activity in an ATP-competitive fashion⁵, but they do so with nearly equal affinity for each paralog. While initially viewed as a virtue of targeting hsp90s, broad inhibition of a wide range of clients has also been associated in some cases with toxic side effects and up-regulation of pro-survival chaperone systems⁶. Because hsp90s have distinct sets of clients with widely diverse functions, the development of paralog-selective inhibitors may serve both as a way to more precisely target specific diseases and as chemical probes to differentiate the biological role of the individual paralogs⁷.

The two cytosolic Hsp90s differ in their expression profiles. Hsp90β is the constitutively expressed paralog while Hsp90α is the stress-inducible form that is often overexpressed in cancer ⁸. The extent to which the two cytoplasmic isoforms are functionally distinct has not been thoroughly addressed. Differences in the roles for Hsp90α and Hsp90β have been observed, however, in a limited number of cases including the p23-relative AARSD1⁹, eNOS¹⁰, the E protein of Japanese encephalitis virus¹¹, the KCNQ4 potassium channel¹², Serotonin transporter ¹³, the co-chaperone UNC45A¹⁴, the hERG channel¹⁵, and the response to acetic acid stress¹⁶. Hsp90α and Hsp90β are also implicated in a number of cancers and neurodegenerative disorders^3,8,17,18; however, the lack of selective pharmacological probes confounds efforts to identify their individual roles. While progress has been made toward selectively targeting Grp94^19–22, less is known about how the ATP-binding pocket can be exploited to promote Hsp90α/Hsp90β selectivity²³. Moreover, the high sequence identity between Hsp90α and Hsp90β (86% overall, 89% in the NTD, and ~95% in the ATP-binding pocket) (Figure 1A) further complicates the rational design of inhibitors that selectively target only one cytosolic paralog over the other (α- or β-selectivity).

Figure 1. PU-11-trans displays binding preference for Hsp90α over Hsp90β.

A) Sequence alignment between Hsp90α and Hsp90β NTDs. Identical residues are shaded black; homologous residues are shaded gray. Residues in the ATP binding pocket are indicated by colored circles or squares. B) Chemical structures of the PU scaffold and PU-11-trans. C,D) ITC analysis of PU-11-trans binding to Hsp90α and Hsp90β. Calculated dissociation constants are given on each thermogram.

Serving as a proof of principle, Khandelwal et al. recently described a modified resorcinol-based compound, KUNB31, with a 50-fold preference in competition binding assays for Hsp90β over Hsp90α²⁴. Co-crystal structures revealed that this scaffold exploits differences in one of the two non-conserved residues of the ATP-binding pocket, predicting a steric clash with the hydroxyl of Ser52 in Hsp90α but not with the smaller side chain of Ala47 at the equivalent position in Hsp90β. While this steric interference strategy has proven successful in producing ligands with Hsp90β selectivity, it is not clear how these differences could be exploited to generate compounds that exhibit Hsp90α selectivity.

The purine-based (PU) scaffold was the first fully synthetic family of compounds to target the ATP-binding pocket of hsp90s²⁵. PU compounds consist of a central purine ring connected to an aromatic moiety from the C8 position by a single C or S linker, and an alkyl-tail originating from the N3 or N9 positions (Figure 1B)²⁶. A screen of more than 130 unique PU compounds identified distinct chemical spaces that promote either Hsp90- or Grp94- selectivity²¹. The Hsp90-preferring subclass consists of compounds containing a tri-methoxy phenyl C-linked 8-aryl moiety. Remarkably, compounds within this subclass exhibited a 3-5-fold preference in competition binding assays for Hsp90α over Hsp90β. This suggested that these inhibitors exploit subtle differences within the ATP-binding pocket of the two cytosolic paralogs. Unlike the resorcinylic compounds described above, which exploit steric differences to yield a preference for Hsp90β over Hsp90α, the basis for the inverse Hsp90 α/β-selectivity in the PU compounds has not been structurally characterized. The protein determinants of α-selectivity are thus unknown.

Here, we analyzed the binding of PU-11-trans, a representative α-selective PU compound, to the two cytosolic paralogs. To determine the structural origins of α-selectivity with the PU-scaffold, we solved the co-crystal structures of Hsp90α and Hsp90β NTDs bound to PU-11-trans, as well as the first structure of the apo Hsp90β NTD. The two inhibitor-bound structures reveal that a nonconserved pocket residue (Ser52 in Hsp90α and Ala47 in Hsp90β) provides additional stability to PU-11-trans in the ATP-binding pocket of Hsp90α through a water-mediated hydrogen-bonding network. In agreement with the structural observations, we showed that mutation of Ser52 to alanine alters the dissociation constant of Hsp90α to match that of Hsp90β). Taken together, our results provide a structural explanation for the binding preference of some PU inhibitors for Hsp90α and shows that the same nonconserved region of the ATP-binding pocket that yields Hsp90β selectivity can also be exploited for Hsp90α selectivity.

Materials and Methods

Reagents

PU-11-trans was synthesized, purified, and characterized as reported previously²¹.

Construct generation, protein expression, and purification

Human Hsp90α (1-236) and Hsp90β (1-221 and 1-231) were cloned into pET15b and expressed as N-terminal His-tagged fusion proteins in E. coli Rosetta (DE3). The S52A mutant of Hsp90α (1-236) in pET15b was made using the QuikChange II mutagenesis kit (Agilent Technologies). Bacterial cultures were grown at 37°C to an A₆₀₀ of 0.8 and induced with isopropyl-1-thio-β-D-thiogalactopyranoside (IPTG) to a final concentration of 1 mM. After 3 hours of induction, cells were harvested by centrifugation at 10,000 × g for 10 minutes. Cell pellets were resuspended in a lysis buffer consisting of 50 mM Tris-HCl (pH 8.0), 350 mM NaCl, 20 mM imidazole, and 1 mM β-mercaptoethanol (β-me) and lysed by multiple passes at 10K - 12K psi through a M110L microfluidizer (Microfluidics). Lysis debris was removed by centrifugation at 34,000 × g for 45 minutes. The supernatant was loaded onto a Ni-NTA agarose affinity column (Qiagen) equilibrated with lysis buffer. After washing with 1 L of lysis buffer, the His-tagged proteins were eluted using a linear imidazole gradient (20 - 300 mM) in 25 mM Tris-HCl (pH 8.0), 200 mM NaCl, and 1 mM β-me. Eluted protein fractions were pooled, concentrated by spin filtration (Amicon), and further purified on a Superdex 75 16/60 size exclusion column (GE Healthcare) equilibrated with 10 mM Tris-HCl (pH 7.6), 100 mM NaCl, and 1 mM DTT. Pure protein fractions were pooled, concentrated to 20 mg/mL, flash frozen in liquid nitrogen, and stored at −80°C prior to crystallization.

Crystallization of Hsp90-PU-11-trans complexes and apo Hsp90 β

Hsp90α (1-236), or the S52A mutant, at 20 mg/mL was co-crystallized at 4°C with 3-fold molar excess of PU-11-trans by hanging drop vapor diffusion. Crystallization reservoirs contained 100 mM Na cacodylate pH 6.5, 160-220 mM MgCl₂, and 4-6 % PEG 2000 monomethyl ether. Crystals typically reached full size in 1-2 days. For data collection, crystals were cryo-protected by a quick transfer through the reservoir solution followed by transfer through a second solution containing 100 mM Na cacodylate pH 6.5, 160-220 mM MgCl₂, and 30% PEG 2000 monomethyl ether. Crystals were immediately flash frozen in liquid nitrogen following cryo-stabilization. Hsp90β (1-221) bound to PU-11-trans was crystallized in a similar manner as Hsp90α, except that the reservoir solution contained 100 mM MES pH 5.6, 300 mM MgCl₂, and 5% PEG 2000 monomethyl ether, and crystals were cryo-protected using reservoir solution followed by a solution containing 100 mM MES pH 5.6, 300 mM MgCl₂, and 30% PEG 2000 monomethyl ether. Apo Hsp90β (1-231) crystallized at 18°C in 80 mM Tris-HCl pH 8.5, 160 mM MgCl₂, 22.5% PEG 4000, and 20% glycerol by hanging drop vapor diffusion. Crystals were removed directly from the drop and frozen in liquid nitrogen.

Data collection, structure determination, and refinement

X-ray data were collected at Advanced Photon Source (APS) beamline 23-ID-B at 100K using a MARCCD and processed using XDS ²⁷ (Hsp90β:PU-11-trans and apo Hsp90β) or HKL-2000²⁸ (Hsp90α:PU-11-trans). Data for the Hsp90α S52A:PU-11-trans complex was collected at APS beamline 17-ID at 100K using a Pilatus detector and processed using Autoproc ²⁹. The structures of Hsp90α and Hsp90β in complex with PU-11-trans were solved by molecular replacement with Phaser³⁰ using PDB entry 3T0H as the search model. PDB entry 2FWZ, with ligand and waters removed, was used as the search model for the Hsp90α S52A:PU-11-trans complex. Model building was done using COOT ³¹ and the structure was refined iteratively using PHENIX ³². The coordinate and restraints file for PU-11-trans was generated using the PRODRG server³³ and the compound was modeled into strong positive difference density in the ATP-binding pocket. The final model of Hsp90α:PU-11-trans consists of residues 16-225, 387 water molecules, and PU-11-trans. The final model of Hsp90β:PU-11-trans consists of residues 11-220, 345 water molecules, and PU-11-trans. The final model of Hsp90α S52A:PU-11-trans consists of residues 16-224, 350 water molecules, and PU-11-trans. Apo Hsp90β was solved by molecular replacement using the protein component of the Hsp90β:PU-11-trans structure as the search model. The solution contains four molecules in the asymmetric unit and the final model consists of residues 1-217 in chain A, residues 1-218 in chain B, residues 1-222 in chain C, and residues 1-217 in chain D, with 53 water molecules and 33 glycerol molecules in total.

Structure validation was performed using MolProbity³⁴ and molecular graphics were generated using PyMol. Data collection and refinement statistics are given in Table 1.

Table 1.

Data Collection and Refinement Statistics

Data Collection	Hsp90α:PU-11-trans	Hsp90β:PU-11-trans	Hsp90β (apo)	Hsp90aS52A:PU-
PDB ID	6N8X	6N8Y	6N8W	6OLX
Source	APS 23-ID-B	APS 23-ID-B	APS 23-ID-B	APS 17-ID
Dataset	jh3b2	jh11b1	q371d6b	jh155a2d
Wavelength (Å)	1.03320	1.03320	1.03320	1.00000
Space group	1222	1222	C21	1222
a, b, c (Å)	66.78, 91.25, 97.86	67.17, 90.95, 96.29	175.25, 70.25, 101.61	66.79, 91.38, 98.13
α, β,γ (°)	90, 90, 90	90, 90, 90	90, 123.45, 90	90, 90, 90
Resolution (Å)	50.00 – 1.49	28.96 – 1.42	29.20 – 3.09	66.88 – 1.44
Rmerge (last shell)	0.061 (0.378)	0.048 (0.623)	0.094 (0.417)	0.043 (0.526)
Average I/σI	42.4 (3.4)	20.4 (1.8)	8.1 (1.8)	20.0 (2.3)
Completeness (%)	99.4 (98.7)	96.1 (64.3)	97.6 (89.0)	99.2 (100)
Unique reflections	48469	53866	18636	54417
Multiplicity	7.1 (6.4)	7.0 (4.6)	2.5 (2.4)	6.5 (6.2)
CC(1/2)	N/D	0.998 (0.752)	0.991 (0.686)	0.999 (0.895)
Refinement
Resolution (Å)	22.81 – 1.50	22.74 – 1.55	28.89 – 3.09	33.44 – 1.44
Rwork / Rfree	0.1594 / 0.1849	0.1642 / 0.1830	0.2250 / 0.2859	0.1630 / 0.1809
Nonhydrogen atoms	2092	2006	6878	2026
Water molecules	387	345	53	350
Rmsd in bond lengths (Å)	0.006	0.007	0.002	0.007
Rmsd in bond angles (°)	1.08	1.09	0.510	1.10
Ramachandran favored (%)	99	99	93	99
Ramachandran outliers (%)	0	0	0.46	0
Clashscore	0.29	0.31	5.75	0.30
Average B-factor (Å2)	26.00	26.60	63.50	25.70
protein	23.50	24.60	63.50	23.50
ligands	16.90	17.90	66.10	16.80
solvent	37.40	36.90	56.90	36.80

Isothermal Titration Calorimetry

To measure PU-11-trans binding, ITC experiments were performed using a MicroCal VP-ITC instrument at 25°C with the ligand in the sample cell and protein in the injection syringe. Titrations were carried out in matching buffers consisting of 40 mM HEPES-KOH pH 7.4, 100 mM NaCl, and 1% DMSO. Two replicate experiments were performed for Hsp90α (wild-type and S52A) and Hsp90β. For wild-type Hsp90α, 290 μL of 503 μM and 336 μM protein were injected into 2.4 mL of 50.7 μM and 30.1 μM PU-11-trans, respectively. For Hsp90α S52A, 290 μL of 255 μM and 393 μM protein were injected into 2.4 mL of 33.0 μM and 41.8 μM PU-11-trans, respectively. For Hsp90β, 290 μL of 891 μM and 885 μM protein were injected into 2.4 mL of 93.3 μM and 97.8 μM PU-11-trans, respectively.

For the ATP and ADP binding measurements, two replicate experiments were performed with protein in the sample cell and ligand in the injection syringe. The ligand concentrations in the syringe were 2.01 and 3.06 mM, and 0.642 and 1.50 mM for ATP and ADP, respectively. For ATP binding measurements, protein concentrations in the cell were 219 and 322 μM (Hsp90α), 301 and 309 μM (Hsp90α S52A), and 300 and 309 μM (Hsp90β). For ADP binding measurements, protein concentrations in the cell were 70.6 and 97.3 μM (Hsp90α), 91.7 and 157 μM (Hsp90αS52A), and 153 and 162 μM (Hsp90β. Binding constants were determined by fitting the data to a one-site model using Origin 7 software. Values reported represent the average of the two independent titrations, and errors are the standard error of the mean.

K_i calculation

K_i values for competition binding experiments were calculated using the method of Nikolovska-Coleska et al.³⁵ according to the following formula:

$$K_i={[I]}_{50}/(\frac{{[L]}_{50}}{K_d}+\frac{{[P]}_0}{K_d}+1)$$

Where [L]₅₀ is the concentration of the free tracer ligand at 50% inhibition, [P]₀ is the concentration of the free protein in the absence of inhibitor, and K_d is the dissociation constant of the protein-tracer complex. The value of [I]₅₀, the concentration of the free inhibitor at 50% inhibition, is given by the formula:

$${[I]}_{50}=I{C}_{50}-{[P]}_0+{[P]}_{L50}(1+\frac{K_d}{L_{50}})$$

and

$${[P]}_0=-((K_d+L-P)+{[{(K_d+L-P)}^2+4P{K}_d]}^{1/2})/2$$

where L and P are the total tracer and protein concentrations, respectively, and

$$[P{]}_{L0}=P-[P{]}_0,[P{]}_{\text{L}50}=[P{]}_{\text{L}0}∕2,L_0=L-[P{]}_{\mathrm{L}0},\text{and}{L}_{50}=L-[P{]}_{\mathrm{L}50}$$

Results and Discussion

Direct Binding Measurements of Hsp90 α / β and PU-11-trans

To probe the direct binding properties of PU-11-trans (Figure 1B), we measured its affinity for Hsp90α and Hsp90β NTDs using isothermal titration calorimetry (ITC). As shown in Figure 1C,D, Hsp90α and Hsp90β bind PU-11-trans with a K_D of 2.0 ± 0.14 μM and 4.2 ± 0.74 μM, respectively, representing a 2.1 ± 0.4-fold factor of selectivity in favor of Hsp90α (Table 2). A previous analysis using a fluorescence polarization (FP) competition-binding assay yielded IC₅₀ values of 18.6 ± 0.3 μM and 89.8 ± 12.6 μM for the α and β paralogs, respectively, representing a 4.8 ± 0.7-fold factor of α-selectivity²¹. The same report determined IC₅₀ values of 111.4 ± 9.7 and 172.9 ± 7.3 for Grp94 and Trap-1, respectively, showing that the compound is selective for the two cytoplasmic paralogs.

Table 2.

Thermodynamic parameters of PU-11-trans binding

Protein	Temp (K)	N	KD (μM)	ΔG (kcal/mol)	ΔH (kcal/mol)	TΔS (kcal/mol)	C-values
Hsp90α	298	1.03 ± 0.18	2.00 ± 0.14	−7.77 ± 0.06	−10.91 ± 0.18	−3.15 ± 0.23	16-24
Hsp90β	298	0.79 ± 0.01	4.24 ± 0.74	−7.34 ± 0.11	−9.75 ± 0.07	−2.42 ± 0.18	20-26
Hsp90α S52A	298	1.16 ± 0.11	4.15 ± 0.03	−7.35 ± 0.38	−11.50 ± 0.19	−4.16 ± 0.19	8-10

Values are the average of 2 independent measurements; errors are standard error of the mean

Discrepancies between FP competition and ITC data have been observed previously for hsp90 inhibitors^22,36. These differences may be reconciled, in part, by converting FP competition IC₅₀ values to inhibition constants (K_i)³⁵. This conversion helps account for the paralog-specific differences in intrinsic affinity for the fluorescent tracer probe used in the displacement FP assays. Converting the earlier reported IC₅₀ values to K_i’s using the reported K_D values of the tracer, cy3B-Geldanamycin (cy3B-GM), of 0.7 nM for Hsp90α and 2.2 nM for Hsp90β^21,37 yielded K_i’s of 1.5 ± 0.02 μM and 16.9 ± 2.2 μM for Hsp90α and Hsp90β (Table 3), which is in better agreement with the ITC data. The calculated K_i values for Grp94 and Trap-1 were 22.8 ± 2.0 μM and 28.6 ± 1.2 μM, respectively. Interestingly, the fold-selectivity for Hsp90α increases from ~5-fold to ~11-fold when comparing K_i’s, suggesting that the cy3B-GM FP tracer underestimates Hsp90β affinity. Taken together, however, both competition and direct-binding experiments indicate that PU-11-trans has a preference for Hsp90α over Hsp90β.

Table 3.

Comparison of PU-11-trans Binding Measurements

	KD (ITC)	IC50 (FP)†	Ki (FP)
Hsp90α	2.0 ± 0.1	18.6 ± 0.3	1.5 ± 0.02
Hsp90β	4.2 ± 0.7	89.8 ± 12.6	16.9 ± 2.2
Grp94	N/D‡	111.4 ± 9.7	22.8 ± 2.0
Trap-1	N/D	172.9 ± 7.3	28.6 ± 1.2
α/β selectivity	2.1 ± 0.3	4.8 ± 0.7	11.3 ± 1.5

^†From Patel et al. Nat. Chem Biol. 2013. FP=fluorescence polarization competition assay.

^‡N/D, not determined.

To compare the binding of PU-11-trans with that of ATP and ADP, the natural ligands of Hsp90, we carried out ITC measurements between the NTDs of Hsp90α and Hsp90β and these nucleotides. As seen in Table 4, Hsp90α and Hsp90β bind to ATP with K_Ds of 209 ± 22 μM and 281 ± 40 μM, representing a smaller 1.3 ± 0.2 fold preference. For ADP, Hsp90α and Hsp90β bind with K_Ds of 8.6 ± 0.3 μM and 12.4 ± 1.8 μM, representing a 1.4 ± 0.2 fold preference for Hsp90α. These values are in agreement with previously published measurements³⁸ and indicate that fold selectivity of PU-11-trans for Hsp90α over Hsp90β is greater than the fold selectivity for the naturally occurring nucleotide ligands.

Table 4.

Thermodynamic parameters of ATP and ADP binding

Protein	Ligand	Temp (K)	N	KD(μM)	ΔG (kcal/mol)	ΔH (kcal/mol)	TΔS (kcal/mol)	C-values
Hsp90α	ATP	298	0.691 ± 0.066	209 ± 21.9	−5.05 ± 0.07	−17.1 ± 1.77	−12.0 ± 1.84	1.13 – 1.44
Hsp90β	ATP	298	0.707 ± 0.112	281 ± 39.6	−4.90 ± 0.14	−17.2 ± 2.40	−12.3 ± 2.55	1.00 – 1.19
Hsp90α S52A	ATP	298	0.592 ± 0.026	239 ± 0.71	−4.95 ± 0.07	−17.7 ± 0.78	−12.7 ± 0.85	1.26 – 1.29
Hsp90α	ADP	298	0.752 ± 0.014	8.59 ± 0.26	−6.90 ± 0.00	−19.9 ± 0.28	−13.0 ± 0.28	8.40 – 11.1
Hsp90β	ADP	298	0.813 ± 0.042	12.4 ± 1.77	−6.75 ± 0.07	−19.7 ± 0.92	−12.9 ± 0.99	6.74 – 14.1
Hsp90α S52A	ADP	298	0.756 ± 0.100	9.06 ± 0.93	−6.85 ± 0.07	−19.8 ± 0.14	−13.0 ± 0.21	16.7 – 18.2

Values are the average of 2 independent measurements; errors are standard error of the mean

Structures of Hsp90 α / β bound to PU-11-trans

To determine the structural basis for the differences in affinity for PU-11-trans, we solved their co-crystal structures with the Hsp90α and Hsp90β NTDs at 1.50 Å and 1.55 Å resolution, respectively (Table 1). A comparison of the two apo- and two ligand-bound complexes shows that the binding of PU-11-trans requires the remodeling of helices 3 and 4 of both apo forms of the two NTDs in order to accommodate the 8-aryl moiety into Site 1 (Figure 2A). In the apo form of Hsp90α, Site 1 is blocked by Leu107, while in Hsp90β, in a 3.1 Å resolution structure determined here, it is blocked by the equivalent residue, Leu102, that remodels upon ligand binding.

Figure 2. PU-11-trans bound to Hsp90α and Hsp90β.

a) Comparison of the apo forms of Hsp90α (“open” (orange, PDB ID 1YES) and “closed” (cyan, PDB ID 1YER)) and Hsp90β (purple) shows that binding of PU-11-trans requires the remodeling of helices 3 and 4 (H3/H4) of the NTDs in order to accommodate the 8-aryl moiety into Site 1. In the apo forms of Hsp90α, Site 1 is blocked by Leu107, while in Hsp90β it is blocked by the equivalent residue, Leu102. b) Overlay of PU-11-trans bound structures of Hsp90α (cyan) and Hsp90β (purple) showing residues within the ATP-binding pocket that surround the inhibitor. Conserved residues between paralogs are labeled according to Hsp90α numbering (black). c) A small sub-pocket of the ATP-binding cavity captures 4 bound water molecules. One of the waters (4) forms a hydrogen bonding network with Ser52 of Hsp90α and the 6-amino group of PU-11-trans. The equivalent residue in Hsp90β, Ala47, cannot form the same interactions with the conserved water molecule.

Although the apo forms of Hsp90α and Hsp90β differ slightly from each other in the region of helices 3 and 4 (Figure 2A), the backbones of the two paralogs in complex with PU-11-trans are essentially identical to one another, with an RMSD between α-carbons of 0.16 Å. The conformations of PU-11-trans bound to Hsp90α and Hsp90β are also essentially identical. PU-11-trans binds in a similar fashion to all previously solved structures of Hsp90α NTDs bound to PU compounds^21,39,40, with the purine ring inserted into the central ATP binding cavity and the tri-methoxy phenyl moiety of PU-11-trans occupying the conserved and largely hydrophobic “Site 1” that consists of Met98, Leu103, Leu107, Ala111, Phe138, Tyr139, Val150, and Trp162 (Hsp90α numbering). The N9-tails of the ligands point outward towards the mouth of the ATP-binding pocket (Figure 2B).

For each paralog, PU-11-trans interacts with 19 of the 21 residues that line the ATP-binding pocket (interaction distance cutoff = 5 Å)(Figure 2B). In addition to the Site 1 residues listed above, these pocket residues include Phe22, Leu48, Asn51, Ser52, Ala55, Ile91, Asp93, Ile96, Gly97, Gly135, Val136, Thr184, and Val186 (Hsp90α numbering). All but two of these residues are conserved between the two paralogs (Figure 1A). The differing residues, Ser52 and Ile91, are unique to Hsp90α, as the equivalent residues in Hsp90β are Ala47 and Leu86, respectively. Both Ser52 and Ile91 are located in a small sub-pocket of the ATP-binding cavity of Hsp90α that captures a local network of 4 bound water molecules. Neither Ser52 nor Ile91 interact directly with the bound ligand. Ile91 is 6.6 Å away from the PU-11-trans at its closet approach. Similarly, the hydroxyl group of Ser52 provides little to no stabilization to the bound purine ring on its own, as it is too distant (4.1 Å) to form a direct hydrogen bond with the exocyclic amine at the 6 position. Inspection of the 4 surrounding waters, however, revealed a key difference between Hsp90α and Hsp90β when bound to PU-11-trans. Three of the 4 water molecules are superimposable with each other. For Hsp90α, however, the fourth bound water molecule forms stabilizing hydrogen bonds with Ser52 (3.3 Å) and the 6-amino group (3.6 Å), bridging the hydroxyl of Ser52 with PU-11-trans (Figure 2C). In contrast, the corresponding water molecule in Hsp90β cannot establish the same hydrogen-bonding network with the nonpolar Ala47, loses one hydrogen bond, and is shifted 0.9 Å from its position in Hsp90α.

To better understand the role of Ser52 in PU-11-trans binding, and to test whether Ser52 is the residue of Hsp90α that confers a binding advantage to PU ligands, we mutated Ser52 to alanine. This mutation mimics Ala47, the natural residue found in Hsp90β, that is not capable of stabilizing the network of bound water molecules. We tested the binding of PU-11-trans to this mutant using ITC. As seen in Figure 3A, mutating Ser52 to alanine decreases the affinity of Hsp90α for PU-11-trans and precisely recapitulates the affinity observed for Hsp90β (Hsp90α S52A = 4.2 ± 0.03 μM, Hsp90β = 4.2 ± 0.72 μM). Together, these data suggest that Ser52 is responsible for creating the favorable water interactions with the 6-amino group of the PU ligand to promote higher affinity.

Figure 3. Mutation of Hsp90α Ser52 to alanine disrupts a stabilizing hydrogen bonding network between Hsp90α, PU-11-trans, and bound water molecules.

a) ITC analysis of PU-11-trans binding to Hsp90α S52A. b) Overlay of PU-11-trans bound structures of Hsp90α S52A (cyan) and Hsp90β (purple) shows that the 4 water molecules surrounding the 6-amino group of PU-11-trans are now superimposable between paralogs.

To confirm the effect of Ser52 on the bound waters in the vicinity of the 6-amino group of the ligand, we determined the structure of the Hsp90α S52A:PU-11-trans complex. As seen in Figure 3B, the waters near the 6-amino group of the ligand in the Hsp90α S52A mutant now superimpose with those of the Hsp90β:PU-11-trans complex, confirming the effect of S52 in improving the affinity for the ligand in Hsp90α. Together, these data point to the non-conserved Ser52/Ala47 pair and the surrounding pocket as a promising region to target for the design of Hsp90α selective inhibitors.

The naturally occurring ligands for Hsp90s are ATP and ADP, both of which contain purine moieties that bind in the ATP binding pocket of Hsp90s in the same position as the purine ring of the PU inhibitors. To test whether Ser52, in imparting favorable binding to Hsp90α, was specific for PU-11-trans or applied to all purine-based ligands, we tested the Hsp90α S52A mutant for binding to ATP and ADP. As seen in Table 4, the dissociation constant for ATP increases only slightly, from 209 ± 22 μM in wild type Hsp90α to 239 ± 0.7 μM in the Hsp90α S52A mutant, representing a 1.14 ± 0.12 fold change in affinity. Similarly, for ADP, the binding constant increases to 9.06 ± 0.93 μM in Hsp90α S52A from 8.59 ± 0.26 μM in wild type, representing a 1.05 ± 0.11 fold change in affinity. The negligible effect of the S52A mutation on ATP and ADP binding compared to PU-11-trans binding likely reflects the greater influence of the purine ring in driving affinity in PU-11-trans binding, compared to the naturally occurring ribonucleotides. The α/β selectivity conferred by Ser52 for PU-11-trans does, however, appear to be a general phenomenon for the PU-class of synthetic inhibitors. Where a direct comparison between Hsp90α and Hsp90β has been made, the PU ligand binds with 2-5 fold better affinity to Hsp90α than to Hsp90β²¹.

In summary, the results presented here identify Ser52 and the constellation of bound water molecules that surround it as the protein element that enhances the affinity of PU-class ligands for Hsp90α compared to Hsp90β. Previous studies ²⁴ have exploited the smaller size of Ala47 at the equivalent position in Hsp90β to develop Hsp90β-selective inhibitors. While the mechanism of selectivity for these inhibitors involves steric restriction by the larger Ser52 in Hsp90α, we have shown here that Hsp90α selectivity can be mediated by the ability of Ser52 and moieties from the ligand to enhance the stability of bound water molecules. Further development of Hsp90α selective compounds would benefit from designs that establish polar contacts between the position occupied by the exocyclic amino group of the purine ring and the hydroxyl of Ser52, perhaps displacing the bound water molecules in the process to improve the entropic component of the binding energetics.