Structure of Hsp90-p23-GR reveals the Hsp90 client-remodelling mechanism.

Data analysis and figure preparation

Figures were created using UCSF Chimera v.1.14 40 and UCSF ChimeraX v0.94 41 . GR ligand binding data was analyzed using Prism v.9.1.1 (GraphPad). Protein expression and purification Human Hsp90α, Hsp70 (Hsp70A1A), Hop, p23, p23 Δtail (1-133), p23 ΔhelixΔtail (1-112), and yeast Ydj1 (Hsp40) were expressed in the pET151 bacterial expression plasmid with a cleavable N-terminal, 6x-His tag. Human Bag-1 isoform 4 (116-345) was expressed in a pET28a vector with a cleavable N-terminal, 6x-His tag. Proteins were expressed and purified by the following procedure. Proteins were expressed in E. coli BL21 star (DE3) strain. Cells were grown in either LB or TB at 37°C until OD 600 reached 0.6-0.8 and then induced with 0.5 mM IPTG overnight at 16°C. Cells were harvested and lysed in 50 mM Potassium Phosphate pH 8, 500 mM KCl, 10 mM imidazole pH 8, 10% glycerol, 6 mM βME, and Roche cOmplete, mini protease inhibitor cocktail using an EmulsiFlex-C3 (Avestin). Lysate was centrifuged and the soluble fraction was affinity purified by gravity column with Ni-NTA affinity resin (QIAGEN). The protein was eluted with 30 mM Tris pH 8, 50 mM KCl, 250 mM imidazole pH 8, and 6 mM βME. For Hsp90, Hsp70, and Ydj1, an extra wash step with 0.1% Tween20 and 2 mM ATP/MgCl 2 was added to the Ni-NTA resin before eluting. The 6x-His tag was removed with TEV protease during the following overnight dialysis in 30 mM Tris pH 8, 50 mM KCl, and 6 mM βME. Cleaved protein was then loaded onto an ion exchange column, MonoQ 10/100 GL (GE Healthcare), with 30 mM Tris pH 8, 50 mM KCl, and 6 mM βME and eluted with a linear gradient of 50-500 mM KCl. Protein was further purified by size exclusion in 30 mM HEPES pH 7.5, 50 mM KCl, 10% glycerol, 1-2 mM DTT using a HiLoad 16/60 Superdex 200 (GE Healthcare) or Hi Load 16/60 Superdex 75 (GE Healthcare). For Hsp70, each peak from ion exchange was collected separately and purified by size exclusion in 30 mM HEPES pH 7.5, 100 mM KCl, 10% glycerol, 4 mM DTT, where only the monomeric peak was then collected. Protein was concentrated, flash frozen, and stored at −80° C . GR LBD expression and purification For GR, the ligand binding domain (LBD) (F602S) (520-777) was codon optimized and expressed in the pMAL-c3X derivative with an N-terminal cleavable 6x-His-MBP tag. GR LBD was expressed and purified as previously described 10 . GR-maturation complex sample preparation The GR chaperone cycle was reconstituted in vitro with purified components as previously described 10 . Buffer conditions were 30 mM HEPES pH 8, 50 mM KCl, 0.05% Tween20, and 2 mM TCEP. Proteins and reagents were added at the following concentration: 5 μM MBP-GR LBD, 2 μM Hsp40, 5 μM Hsp70, 5 μM Hop, 15 μM Hsp90, 15 μM p23, 5 mM ATP/MgCl 2 . This reaction was incubated at room temperature for 60 minutes, then 15 μM p23, 15 μM Bag-1, and 20 mM sodium molybdate (used to stabilize the closed conformation of Hsp90 11 , 42 , 43 , likely by acting as a γ-phosphate analog to stabilize the post-ATP hydrolysis transition state of Hsp90 ( Extended Data Fig. 2c )) were added, and the reaction was incubated at room temperature for another 30 minutes. Following incubation, amylose resin (New England Biolabs) was added to the reactions in a 1:1 ratio and incubated at 4°C with nutation. Resin was then washed 4 times with wash buffer (30 mM HEPES pH 8, 50 mM KCl, 5 mM ATP/MgCl 2 , 0.05% Tween20, 2 mM TCEP, 20 mM sodium molybdate) and eluted with 50 mM maltose in elution buffer (30 mM HEPES pH 8, 50 mM KCl, 2 mM TCEP, 20 mM sodium molybdate). The elution was analyzed by SDS-PAGE (4-12% acrylamide gel) ( Extended Data Fig. 1a ). The elution was concentrated and purified by size exclusion using a Shodex KW-804 on an Ettan LC (GE Healthcare) and fractions were analyzed by SDS-PAGE (4-12% acrylamide gel) ( Extended Data Fig. 1b , c ). Fractions containing the full complex were concentrated to ~2 μM. 2.5 μL of sample was applied to glow-discharged QUANTIFOIL R1.2/1.3, 400-mesh, copper holey carbon grid (Quantifoil Micro Tools GmbH) and plunge-frozen in liquid ethane using a Vitrobot Mark IV (FEI) with a blotting time of 15 seconds, blotting force 3, at 10°C, and with 100% humidity.

Cryo-EM data acquisition

The images were collected on a FEI Titan Krios electron microscope (Thermo Fisher Scientific) operating at 300kV using a K3 direct electron camera (Gatan) and equipped with a Bioquantum energy filter (Gatan) set to a slit width of 20 eV (example micrograph Extended Data Fig. 1d ). Images were recorded at a nominal magnification of 105,000×, corresponding to a physical pixel size of 0.835Å. A nominal defocus range of 0.8 μm −2.0 μm underfocus was used. A total exposure of 5.9 seconds was used with 0.05 second subframes (117 total frames). The total accumulated electron dose was 60 electrons/Å 2 and 0.5128 electrons/Å 2 /frame. Data was acquired using SerialEM software v.3.8-beta 44 . A small dataset on the GR-maturation complex was collected before the larger dataset described above. The smaller dataset was collected from the same GR-maturation complex sample preparation concentrated to 1.2 μM with grids prepared in a similar manner. Images were collected on a FEI Titan Krios electron microscope (Thermo Fisher Scientific) operating at 300kV using a K3 direct electron camera (Gatan). Images were recorded at a nominal magnification of 105,000×, corresponding to a physical pixel size of 0.835Å. A nominal defocus range of 0.8 μm −2.0 μm underfocus was used. A total exposure of 3.0 seconds was used with 0.0255 second subframes (118 total frames). Data was acquired using SerialEM software.

Cryo-EM data processing

The smaller dataset consisted of ~1500 dose-fractionated image stacks, which were motion corrected using UCSF MotionCor2 45 and analyzed with RELION v.3.0.8 18 . Motion corrected images without dose weighting were used for contrast transfer function (CTF) estimation using CTFFIND v.4.1 46 and template-based particle picking was done with Gautomatch v.0.53 ( http://www.mrc-lmb.cam.ac.uk/kzhang/ ) with the Hsp90:p23 crystal structure (PDB ID: 2CG9) as a reference to select a total of 718,080 particles. Multiple rounds of 3D classification were performed with 2CG9 as a low-pass-filtered (40 Å) initial model until a medium-resolution (~8 Å) GR:Hsp90:p23 reconstruction was obtained from 13,570 particles. This reconstruction was used as a reference for the larger dataset. The larger dataset consisted of 5,608 dose-fractionated image stacks, which were motion corrected using UCSF MotionCor2 and analyzed with RELION v.3.0.8. Motion corrected images with dose weighting were used for contrast transfer function (CTF) estimation using CTFFIND v.4.1 and reference-free particle picking was done with RELION v.3.0.8 Laplacian-of-Gaussian auto-picking to select a total of 6,062,152 particles. The processing scheme is depicted in Extended Data Fig. 2a . An initial round of three-dimensional (3D) classification was performed without symmetry using a reference model from the previously collected smaller dataset (see above). The class with clearly recognizable Hsp90 density was used for a second round of 3D classification. In this second round, a class with only Hsp90:p23 density was obtained (454,385 particles). This class was refined after per-particle CTF refinement and beam-tilt correction in RELION to a nominal resolution of 2.66 Å. Particles from two other classes, which contained GR density, were combined (~1 million particles) for a third round of 3D classification. After the third round of 3D classification, particles from classes with the best GR density were then combined (~340,000 particles) and refined. To improve the resolution of GR and the p23 tail-helix , the particles were further classified using focused classification with a mask including GR and the p23 tail-helix . The best focused classes were combined (140,217 particles) and refined to a nominal resolution of 2.56 Å. Using the 2.56 Å reconstruction, per-particle CTF and beam-tilt were refined using RELION. Although the FSC showed slightly improvement over the pre-refined reconstruction at medium resolution range (5–10 Å), the nominal resolution at 0.143 FSC remained unchanged. Nevertheless, we used the CTF/beam-tilt refined particles for the following focused refinement on GR:p23 tail-helix and for the resulting reconstructions used for model building. To further improve the resolution of GR and the p23 tail-helix for model building, these regions were refined using focused refinement with a mask including GR and the p23 tail-helix . From the third round of 3D classification, particles from 3D classes with MBP density were combined (~650,000 particles) and refined. To improve the resolution of MBP, the particles were further classified using focused classification with a mask on MBP. The best focused 3D classes were combined (31,556 particles) and refined to a nominal resolution of 3.63 Å after per-particle CTF refinement and beam-tilt correction in RELION. No ligand-free GR complexes were identified during image analysis, despite many rounds of focused classification using masks of different sizes on GR at various stages of data processing. Only classes with clear ligand density in the GR ligand binding pocket were obtained, suggesting ligand-free GR is either too dynamic or quickly released from the complex. All final reconstructions were post-processed in RELION in which the nominal resolution was determined by the gold standard Fourier shell correlation (FSC) using the 0.143 criterion ( Extended Data Fig. 2b ). Maps were sharpened and filtered automatically as determined by RELION according to an estimated overall map B-factor and filtered to their estimated resolution. RELION was used to estimate the local resolution of each map ( Extended Data Fig. 2a ). For GR:Hsp90:p23, a composite map was generated by combining the overall refinement map with the GR:p23 tail focused refinement map using vop maximum in Chimera. Note that the composite map was only used for presentation in Fig. 1a , but not used in atomic model building or refinement.

Model building and refinement

For the GR-maturation complex atomic model, the dexamethasone-bound human GR crystal structure (PDB ID: 1M2Z) and the human p23 crystal structure (PDB ID: 1EJF) were used as a starting model for model building. A homolog model of human Hsp90α was derived from human Hsp90β from the Hsp90:Cdk4:Cdc37 cryo-EM structure (PDB ID: 5FWK) with the sequence alignment (86% sequence identity) obtained from HHpred server 47 and this was also used as a starting model for model building ( Supplementary Table 1 ). Models were refined using Rosetta v.3.11 throughout. Following the split map approach 19 to prevent and monitor overfitting, the Rosetta iterative backbone rebuilding procedure was used to refine models against one of the half maps obtained from RELION, with the other half map only used for validations. The structurally uncharacterized p23 tail-helix was first de novo built into the focused map of GR:p23 tail-helix using RosettaCM 48 and then was further refined using the same Rosetta iterative backbone rebuilding procedure. With a proper density weight obtained using the half maps, the final model of the GR:Hsp90:p23 complex was refined against the full reconstruction allowing only sidechain and small-scale backbone refinement. The final refinement statistics are provided ( Supplementary Table 1 ). For the Hsp90:p23 and MBP:Hsp90:p23 cryo-EM maps ( Extended Data Fig. 8a , b and Extended Data Fig. 9a , b ), the Hsp90:p23 atomic model from the GR-maturation complex was rigid-body docked into the maps using Chimera. For MBP:Hsp90:p23, the apo MBP crystal structure 49 (PDB ID: 1OMP) was rigid-body docked into the map using Chimera. For Extended Data Fig. 9c , the maltose-bound MBP crystal structure 50 (PDB ID: 1ANF) was rigid-body docked into the map for comparison.

Fluorescence polarization assays

Fluorescence polarization of fluorescent dexamethasone (F-dex) (Thermo Fisher) was measured on a SpectraMax M5 plate reader (Molecular Devices) with excitation/emission wavelengths of 485/538 nm, temperature control set at 25°C. Buffer conditions were 50 mM HEPES pH 8, 100 mM KCl, 2 mM DTT. For equilibrium ligand binding in Fig. 2d and Extended Data Fig. 7b , proteins were pre-equilibrated together at room temperature for 60 minutes prior to F-dex addition. Proteins and reagents were added at the following concentration: 20 nM F-dex, 250 nM GR, 2 μM Hsp40, 15 μM Hsp70, 15 μM Hsp90, 15 μM Hop, 15 μM p23 or p23 tail mutants, and 5 mM ATP/MgCl 2 . Note that the dissociation constant (K D ) between GR and F-dex is ~150 nM 10 . Ligand binding was initiated with 20 nM F-dex and association was measured until reaching equilibrium. For the GR control sample, 3 experiments were done with a 1-hour room temperature preincubation of GR in the reaction buffer and 3 experiments were done without a preincubation of GR to account for small effects on equilibrium ligand binding from the preincubation. The plotted equilibrium values in Fig. 2d and Extended Data Fig. 7b represent the mean of 3 biologically independent samples (except the GR control reaction, which represents the mean of 6 biologically independent samples), with error bars representing the standard deviation. Statistical significance was evaluated by an ordinary one-way ANOVA with post-hoc Dunnett’s multiple comparisons test using Prism v.9.1.1 (GraphPad). For equilibrium ligand binding in Extended Data Fig. 7c , proteins were pre-equilibrated together at room temperature for 60 minutes prior to F-dex addition. Proteins and reagents were added at the following concentration: 20 nM F-dex, 250 nM GR and 15 μM p23 or p23 tail mutants. Ligand binding was initiated with 20 nM F-dex and association was measured until reaching equilibrium. The plotted data points for each reaction represent 7 biologically independent samples (6 biologically independent samples for the GR + p23Δtail reaction) and polarization values were baseline subtracted in accordance with the measured F-dex only baseline polarization value. Statistical significance was evaluated by an ordinary one-way ANOVA using Prism v.9.1.1 (GraphPad). GR ligand binding behavior was affected by buffer conditions; therefore, reactions were always normalized such that each reaction had equivalent amounts of buffer reagents. Sequence alignments and p23 tail-helix motif search For the p23 sequence alignments in Extended Data Fig. 6f , sequences were obtained from Uniprot 51 , aligned in Clustal Omega 52 ( https://www.ebi.ac.uk/Tools/msa/clustalo/ ), and visualized in JalView 2.11.1.0 53 . Sequences in the alignment are: H. sapiens p23, M. musculus p23, R. norvegicus p23, G. gallus p23, X. tropicalis p23, D. melanogaster p23, A. thaliana p23, and S. cerevisiae p23 (Uniprot accession codes: Q15185 , Q9R0Q7 , P83868 , Q90955 , Q5U4Z0 , Q7SZQ8 , A0A0B4K6D2, Q8L7U4 , P28707 , respectively). Secondary structure prediction for the S. cerevisiae p23 protein sequence was performed using Psipred v.4.0 54 ( http://bioinf.cs.ucl.ac.uk/psipred/ ). For Fig. 2b , the ConSurf server 55 , 56 ( https://consurf.tau.ac.il/ ) was used to select and align 87 GR protein sequences as follows: the human GR crystal structure 57 (PDB ID: 4P6X) was used to select sequences from UNIREF90 with maximal percent ID at 95% and minimal percent ID at 65%. Conservation scores were calculated and provided by the server. The conservation scores calculated by the ConSurf server were mapped onto GR from the maturation complex atomic model using Chimera. For Extended Data Fig. 6c , the sequences were obtained from Uniprot 51 , aligned in Clustal Omega 52 ( https://www.ebi.ac.uk/Tools/msa/clustalo/ ), and mapped onto GR from the maturation complex using Chimera. Sequences in the alignment are the human steroid hormone receptors: glucocorticoid receptor, mineralocorticoid receptor, androgen receptor, progesterone receptor, estrogen receptor α and β (Uniprot accession codes: P04150 , P08235 , P10275 , P06401 , E3WH19, Q92731 , respectively). Conservation was calculated using AL2CO 58 parameters (unweighted frequency estimation and entropy-based conservation measurement). Relating to Extended Data Fig. 6f , the p23 tail-helix motif search was performed using the ScanProsite server 28 ( https://prosite.expasy.org/scanprosite/ ). The motif “FXXMMN” was used to search the UniProtKB sequence database with taxonomy restricted to Homo Sapiens . There were 10 total hits on the motif, which included p23 and NCoA3/SRC-3. Relating to Extended Data Fig. 6d , the human p23 helix predictions were performed using state-of-the-art secondary structure prediction algorithms: Porter 4.0 59 ( http://distillf.ucd.ie/porterpaleale/ ), RaptorX 60 ( http://raptorx.uchicago.edu/StructurePrediction/predict/ ), and Psipred v.4.0 54 ( http://bioinf.cs.ucl.ac.uk/psipred/ ). The human p23 structure prediction is available from AlphaFold v2.0 25 with the accession code P83868 Yeast survival assays Relating to Extended Data Fig. 7d , yeast survival assays were performed using S. cerevisiae strain hsc82hsp82Δ (JJ816) or sba1hsc82hsp82Δ (JJ94) expressing either wild-type Hsc82 or a mutant in pRS313GPDHis-Hsc82 ( hsc82-I588A, M589A ) 61 . The sba1hsc82hsp82Δ strain expressing hsc82-I588A, M589A was transformed with either empty vector (pR416GPD), human p23, or p23 tail mutants constitutively expressed from the p416GPD plasmid. Cells were grown overnight at 30°, then serially diluted 10-fold and grown on selective media at the indicated temperature for 2 days.

Expression of human p23 tail mutants

Human p23 and Sba1 protein expression levels in wild-type strain (JJ762 (PJ51-3a), a derivate of W303) 62 were assessed by SDS-PAGE (12.5% acrylamide gel) followed by immunoblot analysis with polyclonal antisera raised against human p23 (Novus NBP1-85485)(1:200 dilution) or Sba1 61 (1:500 dilution)( Extended Data Fig. 7f ). In vivo GR activity assays Relating to Extended Data Fig. 7e , GR transactivation was measured in the wild-type S. cerevisiae strain (JJ762) expressing GR on a single copy plasmid (p414GPD-GR) and a single copy LacZ reporter plasmid (pRS317-GRE-lacZ) that were constructed using multicopy versions of each plasmid 62 . Human p23, p23 tail mutants, or Sba1 were constitutively expressed from the p416-GPD plasmid. Cells were grown at 30°C with shaking overnight in selective media, diluted 10-fold and grown to OD 600 0.4-0.5. Cultures were split in two and one set was induced with ligand (10 μM DOC, deoxycorticosterone) (Sigma) for 1 hour. After 1 hour, cultures were put on ice to stop growth. The β-galactosidase (β-gal) activity of paired samples in the presence and absence of hormone was measured as described using the yeast β-galactosidase assay kit from Thermo Scientific (Catalog number #75768). GR activity was determined by the fold difference in β-gal activity between the hormone treated duplicate relative to the untreated duplicate. Relative GR activation was calculated by normalizing the GR activity of each experimental sample to the average GR activity of strain JJ762 expressing p416GPD (empty vector [e.v.]). GR activity was measured with 18 biologically independent samples (10 biologically independent samples for the + Sba1 condition). Statistical significance was evaluated by an ordinary one-way ANOVA with post-hoc Šídák’s multiple comparisons test using Prism v.9.1.1 (GraphPad).

Quantification and statistical analysis

All data were tested for statistical significance with Prism v.9.1.1 (GraphPad). Statistical significance was determined by ordinary one-way ANOVA (with post-hoc Dunnett’s or Šídák’s multiple comparisons test). All experiments were performed at least three times. Statistical details (including sample sizes ( n ), F-statistics, p-values, and degrees of freedom) are included in the figure legends for each experiment. Relating to Fig. 2d and Extended Data Fig. 7b , the one-way ANOVA with post-hoc Dunnett’s test p-values are as follows: p(p23 vs. p23 Δtail ) = 0.1512, p(p23 vs. p23 Δhelix-tail ) = 0.0002, p(p23 vs. no p23) =