Quantitative analysis of HSP90-client interactions reveals principles of substrate recognition

Abstract

HSP90 is a molecular chaperone that associates with numerous substrate proteins called clients. It plays many important roles in human biology and medicine, but determinants of client recognition by HSP90 have remained frustratingly elusive. We systematically and quantitatively surveyed most human kinases, transcription factors, and E3 ligases for interaction with HSP90 and its cochaperone CDC37. Unexpectedly, many more kinases than transcription factors bound HSP90. CDC37 interacted with kinases, but not with transcription factors or E3 ligases. HSP90::kinase interactions varied continuously over a 100-fold range and provided a platform to study client protein recognition. In wild-type clients, HSP90 did not bind particular sequence motifs, but rather associated with intrinsically unstable kinases. Stabilization of the kinase in either its active or inactive conformation with diverse small molecules decreased HSP90 association. Our results establish HSP90 client recognition as a combinatorial process: CDC37 provides recognition of the kinase family, whereas thermodynamic parameters determine client binding within the family.

Figure 1. LUMIER with BACON Assay Reveals the Quantitative Nature of HSP90::Client Interactions

(A) Principle of the assay. 3 × FLAG-tagged bait constructs (putative HSP90 clients) are transfected into a 293T cell line stably expressing Renilla luciferase-tagged HSP90 (prey). Cell lysates are incubated in 384-well plates coated with anti-FLAG antibody. After washing off nonspecific proteins, luminescence is measured. Interaction of HSP90 with the bait can be detected as luminescence. In the second step, the amount of bait is measured with anti-FLAG ELISA. The log2 ratio between bait and prey concentration is the interaction score. (B) Distribution of luminescence in HSP90 interaction assays with 420 kinase clones (red), 498 E3 ligase clones (green), and 1,093 transcription factor clones (blue). As a control, 176 kinase clones were tested against a cell line expressing Renilla luciferase only (black). Gaussian curve was fitted to each data set to establish a cutoff for true interactions (dashed black line). (C) Quantitative interaction score was calculated for all 193 kinases that interacted with HSP90. Scatter plot shows the interaction scores from two biological replicates. See also Figure S1 and Table S1.

Figure 2. CDC37 Is a Universal Kinase-Specific Cochaperone

(A–C) HSP90 interaction profiles were compared with CDC37 interaction profiles for kinases (A), transcription factors (B), and E3 ligases (C). Plots show mode-normalized, log₂-transformed luminescence values. (D) 293T cells were infected with a lentivirus expressing an shRNA construct targeting Cdc37 or GFP. (E) CDC37 is required for kinase::HSP90 interactions. Client proteins were transfected in duplicate into Renilla-HSP90 expressing cells infected with either Cdc37 or GFP shRNA. Forty-eight hours after transfection, interaction with each client was measured with LUMIER. Boxes indicate 25th and 75th percentiles and whiskers 10th and 90th percentiles. p value was calculated with Wilcoxon rank sum test. (F) 3 × FLAG-tagged client proteins were transfected into 293T cells infected with either Cdc37 or GFP shRNA. Forty-eight hours later, clients were immunoprecipitated with an anti-FLAG antibody. Coimmunoprecipitation with endogenous HSP90 was assayed with a monoclonal HSP90 antibody. See also Figure S1.

Figure 3. An HSP90 Inhibitor Induces Dissociation of Most HSP90::Kinase Complexes, but HSP90 Interaction Strength Correlates Poorly with Subsequent Degradation

(A) HSP90::kinase interactions were measured after treating cells with 500 nM ganetespib or vehicle for 1 hr before cell lysis and LUMIER assay. HSP90β::CDC37 interaction (blue) and HSP90β homodimerization (green) is not affected by ganetespib. Dashed black line indicates no change in HSP90 interaction upon ganetespib treatment. Kinase clones marked with filled black circles were further validated with coimmunoprecipitation with endogenous HSP90 (Figure S2). (B) Ganetespib-induced degradation of client kinases correlates poorly with HSP90 interaction score. 3 × FLAG-tagged kinases were transfected into 293T cells, which were subsequently treated with 100 nM ganetespib or vehicle for 20 hr. Protein levels were measured with anti-FLAG ELISA. Relative protein level is plotted against Hsp90 interaction score from LUMIER assay. (C) HSP90 inhibition can lead to either degradation or aggregation of strong client proteins. Left panel: 3 × FLAG-tagged kinases were transfected into 293T cells, which were treated with 100 nM ganetespib for indicated times. Protein levels were measured with an anti-FLAG antibody. Right panel: 3 × FLAG-tagged kinases were transfected into 293T cells, and after 24 hr treatment with 100 nM ganetespib or vehicle, the soluble fraction (0.1% Triton) was separated from the insoluble fraction. Protein levels were measured with an anti-FLAG antibody. (D) Degradation or aggregation of weak client proteins after HSP90 inhibition. Experiment was performed as in (C) but with weak HSP90 client proteins.

Figure 4. HSP90 Associates with the Majority of the Human Kinome

Quantitative HSP90 interaction scores are plotted on the human kinome tree (courtesy of Cell Signaling Technologies and Science Magazine). Interactions were divided into eight bins and color coded. Client proteins were further divided into weak clients (interaction score 1–3) or strong clients (interaction score 4–7). Inset left, distribution of clients within kinase families. Tyrosine kinases are the only family that contains more clients than expected by chance (Bonferroni-corrected p value 0.027, Wilcoxon rank sum test). Kinase tree from Manning et al. (2002). Reprinted with permission from the American Association for the Advancement of Science (AAAS). See also Figure S3.

Figure 5. HSP90 Interaction Is Determined by a Distributed Set of Residues and Is Affected by Alternative Splicing

(A) Multiple residues regulate ARAF interaction with HSP90. 3 × FLAG-tagged kinase constructs were transfected into Renilla-HSP90 cells and their interaction with HSP90 was measured with LU-MIER. Wild-type BRAF interacts weakly with HSP90, whereas RAF1 (CRAF) and ARAF are among the strongest HSP90 clients (gray). Constructs containing the kinase domain of ARAF but regulatory domains of BRAF interact strongly with HSP90 (blue), but single point mutations in BRAF do not confer robust interaction with HSP90 (green). Clones with chimeric kinase domains display intermediate phenotypes (orange). Error bars indicate SDs (n = 4). (B) Location and the sequence of the alternatively spliced αD-αE loop in AMPKα1 (PRKAA1) are shown in the structure of the AMPKα2 isoform (PDB 2H6D). Kinase N-lobe is colored blue, C-lobe in green, αD helix in red, and αE helix in salmon pink. (C) Alternative splicing of AMPKα1 regulates HSP90 association. 3 × FLAG-tagged kinase constructs were transfected into 293T cells and interaction with endogenous HSP90 was assayed by coimmunoprecipitation. AMPKα1 loop was inserted into AMPKα2 or CDK2 that do not have loops between the two helices. As a control, the DNA sequence encoding the loop sequence was also inserted inverted (inv) into the same location. See also Figure S4.

Figure 6. HSP90 Associates with Thermodynamically Unstable Kinases

(A) Distribution of kinase domain crystal structures deposited to PDB by client class and expression system extracted from the PDB coordinate file. p value was calculated with Fisher’s exact test. (B) Strong HSP90 clients are less soluble than nonclients or weak clients. Kinase domains that have been expressed in a soluble manner by Structural Genomics Consortium are divided into client classes. p value was calculated with Fisher’s exact test. (C) ATP-competitive inhibitors decrease HSP90:: BCR-ABL interactions. 3 × FLAG-tagged kinases were transfected into Renilla-HSP90 cells, and HSP90 interaction was measured by LUMIER after 1 hr treatment with 5 µM inhibitor or DMSO. Average relative interaction with HSP90 (DMSO = 1) from four independent replicates is shown with SD. Statistical significance is shown as asterisks (*p < 0.05; ***p < 0.001). (D) Stabilization of the kinase fold with allosteric modulators decreases HSP90::BCR-ABL interactions. Experiment was done as in Figure 5C, using 5 µM GNF-2 or DPH. (E) BCR-ABL point mutations have opposing effects on HSP90 interaction. Interaction of wild-type BCR-ABL (black), M472I mutant (blue) and M244I mutant (red) with HSP90 was measured with LUMIER. Average relative interaction (wild-type = 1) from four independent replicates is shown with standard deviation. (F) Thermal stability of ABL mutants correlates with HSP90 association. Thermal denaturation curves of wild-type ABL and M472I and M244I mutant proteins were determined by circular dichroism. See also Figure S5.

Figure 7. Model for HSP90::Kinase Interactions

Kinase domains are in equilibrium between the fully folded and HSP90-binding competent conformations. CDC37 cochaperone recognizes the kinase fold and recruits kinases to HSP90. HSP90 binds the alternative kinase conformation and assists the kinase in adopting the fully folded conformation. Client kinases thus undergo repeated rounds of chaperoning, whereas nonclient kinases are stable when fully folded. Binding of an inhibitor to its target kinase increases the stability of the kinase fold and thus decreases HSP90 interaction. HSP90 inhibitors block the loading of the client to the chaperone, which leads to aggregation of the partially unfolded kinase or degradation by the ubiquitin-proteasome system.