Investigation of Cellular Response to the HSP90 Inhibition in Human Cells Through Thermal Proteome Profiling

Abstract

Heat shock proteins are chaperones, and they are responsible for protein folding in cells. Heat shock protein 90 (HSP90) is one of the most important chaperones in human cells, and its inhibition is promising for cancer therapy. However, despite the development of multiple HSP90 inhibitors, none of them has been approved for disease treatment due to unexpected cellular toxicity and side effects. Hence, a more comprehensive investigation of cellular response to HSP90 inhibitors can aid in a better understanding of the molecular mechanisms of the cytotoxicity and side effects of these inhibitors. The thermal stability shifts of proteins, which represent protein structure and interaction alterations, can provide valuable information complementary to the results obtained from commonly used abundance-based proteomics analysis. Here, we systematically investigated cell response to different HSP90 inhibitors through global quantification of protein thermal stability changes using thermal proteome profiling, together with the measurement of protein abundance changes. Besides the targets and potential off-targets of the drugs, proteins with significant thermal stability changes under the HSP90 inhibition are found to be involved in cell stress responses and the translation process. Moreover, proteins with thermal stability shifts under the inhibition are upstream of those with altered expression. These findings indicate that the HSP90 inhibition perturbs cell transcription and translation processes. The current study provides a different perspective for achieving a better understanding of cellular response to chaperone inhibition.

Graphical Abstract

Highlights

• •Study of cellular response to the HSP90 inhibition using thermal proteome profiling
• •Three inhibitors with different structures to minimize inhibitor-specific effects
• •Analysis of thermal stability and abundance offers complementary information
• •Translation suppression through mediating ribosome dissociation with the inhibition

In Brief

Kejun et al. systematically studied cell response to the HSP90 inhibition in human cells using thermal proteome profiling. To minimize inhibitor-specific effects and reveal intrinsic cell responses, three inhibitors (tanespimycin, onalespib, and ganetespib) with different structures and binding modes were investigated. Combined with protein expression and co-aggregation analysis, the results reveal that multiple important pathways were affected in cells with the HSP90 inhibition. This study provides valuable insights into the underlying mechanisms of cellular responses to HSP90 inhibition.

Heat shock protein 90 (HSP90) is a highly conserved chaperone protein that accounts for 1% to 2% of total copies of cellular proteins and up to 4% to 6% in stressed cells (1, 2, 3). HSP90 binds to partially folded proteins (4) using ATP as the energy source to switch between the open and closed conformations (5, 6). HSP90 has a broad spectrum of clients involved in cell growth and survival, and regulation of protein homeostasis (7, 8). In cancer cells, HSP90 is commonly overexpressed, and more than 400 proteins have been annotated to be its clients (2, 9, 10). The role of HSP90 in cancer has been associated with tumor progression, drug resistance, and metastasis (11, 12, 13). Therefore, inhibiting HSP90 is promising for anti-tumor therapy (14, 15).

Most HSP90 inhibitors target the N-terminal nucleotide-binding site, where they compete with ATP and prevent the ATPase activity of HSP90. This results in the inhibition of chaperone activity and promotes its clients to the ubiquitin-proteasome degradation pathway (16). Multiple HSP90 inhibitors entered the clinical trial, and among them, tanespimycin (17-AAG) was the most advanced one (Phase II/III) (16, 17). These HSP90 inhibitors demonstrated high potency in suppressing tumor progression, but have not been approved by the FDA due to their cellular toxicity and compromised clinical efficacy (18). Therefore, further investigation into the mechanisms of these inhibitors, including the induced cell responses and their off-targets, can facilitate the development of HSP90 inhibitors as therapeutic agents for the treatment of cancer and other diseases.

The protein abundance changes in cells with the HSP90 inhibition were previously reported (19, 20, 21, 22). For instance, Miao et al.(19) observed an upregulation of nearly all heat shock proteins (HSPs) in cells with the treatment of various HSP90 inhibitors. Besides abundance change, protein thermal stability shift represents the biophysical state alteration of a protein. It has been utilized to study protein structure and interaction alterations (23). With the recent development of thermal stability profiling, it can be used to study protein-ligand binding (24, 25), protein–protein interactions (PPI) (26), protein post-translational modifications (27), and the organization dynamics of protein complexes (28) in living cells at the proteome-wide scale. Using thermal proteome profiling (TPP), protein thermal stability changes are measured when the proteome is heated at different temperatures, providing comprehensive information about the alteration of protein biophysical states. Typically, TPP has been utilized to identify the targets of clinically relevant drugs (24, 29). Besides the targets of a drug, its off-targets may have thermal shifts as well. Furthermore, proteins with structure and interaction changes from cellular response to the drug treatment may have thermal stability changes. Therefore, TPP provides us an opportunity to study cell response.

Here, we systematically analyzed protein thermal stability shifts induced by three different HSP90 inhibitors in MCF7 cells using TPP, together with the quantification of protein abundance changes in each case. The three inhibitors, that is, 17-AAG, onalespib (AT13387), and ganetespib (STA-9090), have different structures and binding modes with HSP90, providing a unique opportunity to study the cellular response to the HSP90 inhibition with minimal effect from the individual inhibitor. A list of proteins exclusively stabilized by 17-AAG was identified as its putative off-targets. Comparative analysis of protein thermal stability shifts and expression changes uncovered the cell response to the HSP90 inhibition, demonstrating that protein thermal stability changes provide complementary information to protein abundance changes. The PPI changes were also investigated by the thermal proximity coaggregation analysis, showing altered composition within the HSP90 complex during the inhibition. Overall, we comprehensively studied protein thermal stability and expression changes in cells with the inhibition of a very important chaperone (HSP90), facilitating our understanding of the cellular response to its inhibition that can be further considered in the chaperone inhibition-related therapies.

Experimental Procedures

Cell Culture

MCF7 cells (American type culture collection (ATCC)) were grown at 37 °C in 5.0% CO₂ in Dulbecco's Modified Eagle's Medium (Gibco) supplemented with 10% fetal bovine serum (Thermo) and 100 units/ml penicillin-streptomycin in a humidified incubator with 5.0% CO₂ at 37 °C. When cells reached ∼80% confluency, they were treated with 10 μM of either 17-AAG, STA-9090, AT13387 (MedChemExpress), or dimethyl sulfoxide (DMSO) (as the vehicle), respectively. Cells were further cultured for 12 h before being harvested.

Thermal Proteome Profiling

TPP was performed as previously described with slight modifications (30). Briefly, cells were harvested through trypsinization, pelleted at 300g for 5 min at 4 °C, and washed with ice-cold PBS. The supernatant was removed, and cells were resuspended in 1.1 ml of ice-cold PBS. Then, 100-μl aliquots of the cell suspension were transferred to a PCR tube and centrifuged at 300g for 5 min at 4 °C. Next, 80 μl of the supernatant was removed, and each sample was heated at a certain temperature within a defined temperature gradient (37.0–66.1 °C) for 3 min followed by incubation at room temperature for 3 min. Heated cell pellets were resuspended in 130 μl ice-cold lysis buffer (0.5% NP-40, 1 mM MgCl₂, complete mini EDTA free protease inhibitors (Roches), 250 U/ml benzonase (Thermo) in PBS) through pipetting, and cells were lysed at 4 °C for 1 h. The lysates were transferred to a 0.45-μm 96-well filter plate (Millipore, prewetted with the lysis buffer), and the filter plate was transferred to an extraction plate vacuum manifold (Promega). Then, the sample was filtered to remove protein aggregates. Next, 100 μl of each sample was transferred to a new 96-well plate and kept at −80 °C until further prepared for MS analysis.

Protein Digestion and Peptide Labeling With the TMT Reagents

Proteins in the lysates were reduced with 5 mM dithiothreitol at 56 °C for 30 min, then alkylated using 14 mM iodoacetamide at room temperature in the dark for 30 min. Proteins were then precipitated using the methanol-chloroform method, followed by digestion with trypsin (protein: trypsin = 100: 1, w/w) in a digestion buffer (5% ACN, 50 mM HEPES pH = 8.5, and 1.6 M urea) at 37 °C for 16 h. After digestion, the solution was acidified with trifluoroacetic acid to a final pH value of ∼2. Peptides were desalted using the tC18 Sep-Pak cartridges (Waters) and then lyophilized. The purified peptides were labeled with the tandem mass tag (TMT) reagents following the manufacturer's instructions with minor modifications. Briefly, peptides were resuspended in 100 mM HEPES pH = 8.5, and their concentrations were measured using the BCA assay. Peptides in each sample were labeled with each channel of the 10-plex TMT reagents (Thermo) in 30% ACN/HEPES solution for 1 h at room temperature. The reaction was quenched using 5% hydroxylamine for 15 min. Ten labeled peptide samples were combined, desalted, and lyophilized. The mixed samples were then fractionated using high-pH reversed-phase HPLC (pH = 10). The sample was separated into 24 fractions using a 4.6 × 250 mm reversed-phase column packed with 5 μm particles (Waters) using a 40 min gradient of 5 to 50% ACN with 10 mM ammonium acetate. Every fraction was further purified with StageTip before liquid chromatography (LC)-MS analysis.

LC-MS Analysis

The peptide samples were dissolved in a solvent containing 5% ACN and 4% FA, and 2 μl of the solution was loaded onto a microcapillary column packed with C18 beads (ReproSil-Pur 120 C18-AQ, 1.9 μm, 120 Å, 50 μm × 16 cm, Dr Maisch Gmbh) using a Dionex WPS-3000TPL RS autosampler (UltiMate 3000 thermostatted Rapid Separation Pulled Loop Wellplate Sampler). Peptides were separated by a nanoflow reversed-phase-HPLC (Ultimate 3000 RSLC, Dionex) with a 60-min gradient of 1 to 20% ACN (with 0.125% FA). Peptides were detected with a data-dependent Cycle2s method in an Orbitrap mass spectrometer (Oribitrap Exploris 480, Thermo Scientific). For each cycle, one full MS scan (resolution: 120,000) in the Orbitrap cell at the automatic gain control (AGC) target of 5 × 10⁵ was followed by MS/MS recorded in the Orbitrap cell with a resolution of 45,000 for the most intense ions in 2 s. The isolation window is 0.7 m/z, and the isolation specificity was set at 90% to reduce co-isolation. The selected ions were excluded from further analysis for 60 s. Ions with single or unassigned charges were not selected for fragmentation. For MS/MS scans, high-energy collision dissociation with 35% normalized collision energy was used to fragment ions, and fragments were detected in the Orbitrap cell.

Database Search and Peptide and Protein Filtering

The resulting raw files were converted to mzXML files and then searched against the human (Homo sapiens) protein database (downloaded from Uniprot, 09/12/2021; common lab contaminates included; 20,500 entries) using the SEQUEST algorithm (version 28) (31). The following parameters were used during the search: 10 ppm precursor mass tolerance; 0.025 Da product ion mass tolerance; Trypsin digestion with three missed cleavages; variable modifications: oxidation of methionine (+15.9949 Da); static modifications: TMT (+229.1629) on the lysine residue and the peptide N-terminus; carboxyamidomethylation (+57.0215) on the cysteine residue. Peptides with shorter than seven amino acid residues in length were discarded. XCorr score was controlled at >1.0 for each peptide. The false discovery rates (FDR) of peptide and protein identifications were evaluated and controlled by the target-decoy method (32). Each protein sequence was listed in both forward and reversed orders. Linear discriminant analysis was employed to control the quality of peptide identifications using multiple parameters, including XCorr, mass accuracy (ppm), peptide length, and charge state (33). Furthermore, FDRs were filtered to <1% at both the peptide and protein levels.

Data Analysis for Protein Quantification and Protein Thermal Stability Changes

For protein quantification, peptides with summed signal-to-noise level (S/N) <50 were excluded, representing an average S/N >5 for all retained peptides. For thermal stability analysis, the abundance of protein was calculated by summing the intensities of all peptides from this protein. The melting point of each protein was determined as previously described with modifications (30). In the first step, the relative abundance ratios of the TMT reporter ions compared with the lowest temperature point (37 °C) were calculated. The ratios were then normalized using the TPP R script, and the melting curve was fitted according to the chemical denaturation theory (34):

$$f\left(T\right)=\frac{1-plateau}{1+e^{-\left(\frac{a}{T}-b\right)}}+plateau$$

In this equation, T is the temperature, and a, b, and plateau are constants. The melting point of a protein is determined as the temperature where half of the protein has denatured: f(T) = 0.5. The generated melting curves were inspected for a change in melting behavior. All melting curves shown were generated in OriginLab 2023. The melting curves were checked for the significant difference by using the nonparametric analysis of response curves (NPARC) method developed by Childs et al. (35), where a curve is determined to be different if adj. p ≤ 0.05. Additionally, we applied more criteria: (i) the melting point differences of a protein in the duplicated treated samples versus the control ones must have the same sign; (ii) the difference between the melting points of the drug-treated groups and the control groups must be larger than the difference between the melting points of both control groups, (iii) plateau value < 0.5, and (iv) R² ≥ 0.8.

Protein abundances were calculated by the previously reported top3 method (36) with slight modification (30). In brief, the TMT intensities from the sample treated at the lowest temperature (37 °C) in each experiment were extracted for all proteins. The intensity of each protein was calculated by summing the intensities of the three most abundant peptides belonging to this protein. Proteins that were not quantified in all eight experiments were discarded. The total protein abundance of each experiment was normalized to the median of the eight experiments, and internal reference scaling (IRS) was then used to control the batch effect (37). For the replicated experiments, only proteins with a coefficient of variance (CV) <0.5 were further analyzed. PCA analysis was performed based on IRS-transformed data using Originlab 2023.

PPI alterations were determined by the Thermal proximity coaggregation (TPCA) approach (38). Briefly, soluble protein ratios at each temperature were averaged between two replicated experiments to generate the denaturation curve for each protein. Then the coaggregation behavior differences of each high-confidence PPI (downloaded from String database (Homo sapien), 03/10/2023, combined score≥950) between the vehicle and treated samples were determined by the Rtpca package (39). PPI changes with p-value < 0.05 were selected as significantly changed interactions. Protein complex coaggregation analysis was performed using the ProSAP algorithm (40). Only proteins with curves fitting R² ≥ 0.8 in all eight experiments were used. Protein complexes with a number of subunits >3 were used in the analysis (downloaded from the CORUM database, 03/12/2023), with a p-value <0.05 as the significance threshold.

Experimental Design and Statistical Rational

MCF7 cell line is used in this work because it is a common model in anticancer therapy research (41, 42). All experiments were run with two biological replicates, including each of the three HSP90 inhibitors and the DMSO vehicle. The concentration of each inhibitor is high enough to ensure the inhibition of HSP90 in MCF7 cells. The NPARC output was first filtered by the algorithm generated Benjamin-Hochberg adj. p-value; proteins with adj. p-value below a threshold of 0.05 were retained, and further filtered by the quality of all denaturation curves (see above for detailed criteria). In the analysis using Rtcpa and PROSAP algorithms (38, 39, 40), a p-value < 0.05 was considered statistically significant.

Results

Profiling Protein Thermal Stability Changes Under the 17-AAG Treatment in MCF7 Cells

We performed thermal proteome profiling in human breast cancer cells, that is, MCF7 cells, a classical model in anticancer therapy research (41, 42). MCF7 cells were treated with 10 μM 17-AAG in duplicates for 12 h to fully inhibit HSP90 (43, 44), and then heated at the temperature ranging from 37.0 to 66.1 °C, respectively (Fig. 1A). Aggregated proteins were removed by a membrane filter, as in the previous report (30), and the soluble fractions were analyzed using TMT-based multiplexed proteomics. Each of the three inhibitors was used to treat cells, and their structures are illustrated in Figure 1B.

Fig. 1.

Global Analysis of protein thermal stability changes in MCF7 cells treated with each of three HSP90 inhibitors using TPP.A, Workflow of the TPP analysis. B, Structures of three HSP90 inhibitors used in this work. HSP90, Heat shock protein 90; TPP, thermal proteome profiling.

In the 17-AAG treated samples, more than 7000 proteins were quantified (supplemental Table S1). Based on the criteria described in the experimental section, the melting points of 5437 proteins were found in all replicates with good reproducibility (Fig. 2A and supplemental Fig. S1A; supplemental Table S2). The NPARC method (35) was then applied to determine differential melting behavior induced by 17-AAG with a melting point difference cutoff of 0.5 °C and adjusted p-value < 0.05 (see Supplemental Data for details). In total, 1233 proteins had their thermal stability significantly changed under the 17-AAG treatment (Fig. 2B). We studied the thermal stability changes of the nuclear and cytoplasmic HSP90 isoforms, HSP90AA1 and HSP90AB1, which are the designated targets of the three inhibitors. As expected, we observed positive melting point shifts for both HSP90AA1 and HSP90AB1 under the treatment (ΔTm = 4.0 °C and 3.0 °C, adj. p-value = 1.57E-11 and 3.81E-9, respectively; Fig. 2, C and D), demonstrating their strong interactions with 17-AAG, consistent with the previous reports (16, 17).

Fig. 2.

Proteome thermal stability changes induced by HSP90 inhibition.A, The number of quantified proteins in each experiment. B, Volcano plot showing the quantification of protein thermal stability shifts with the 17-AAG treatment. Proteins with significant thermal stability changes are indicated in colors (detailed criteria in the text, adj. p value <1E-13 are assigned as 1E-13). C and D, Thermal denaturation curves of HSP90AA1 (C) and HSP90AB1 (D). The curves generated from different experiments are indicated in gray (DMSO), red (17-AAG), dark green (STA-9090), or light green (AT13387). Squares and circles indicate different replicates. DMSO, dimethyl sulfoxide; HSP90, Heat shock protein 90.

Besides HSP90AA1 and HSP90AB1, the thermal stability of the two other HSP90 isoforms was detected as well. It was reported that 17-AAG has similar inhibition efficiency on the endoplasmic reticulum (ER) isoform of HSP90, i.e., GRP94 (45). Indeed, GRP94 was also identified as a hit in this work but with a smaller melting point shift (ΔTm = 2.0 °C, adj. p-value = 2.12E-3, Fig. 2B and supplemental Fig. S2A). The relatively less thermal shift is possibly due to the fact that GRP94 is intrinsically more soluble than other HSP90 isoforms because of its heavy N-glycosylation (6 sites according to the Uniprot), making its thermal stability less affected. This is further supported by the result that >30% of GRP94 remains soluble after heating at 66.1 °C (supplemental Fig. S2A). The mitochondrial HSP90 isoform TRAP1 exhibited insignificant thermal stability change (ΔTm = 0.2 °C, adj. p-value = 0.826, supplemental Fig. S2B), which is consistent with its previously reported low binding affinity with 17-AAG (45). Together, these results agree with the known targets of 17-AAG and validate the quality of the current results.

Classification of Proteins With Different Thermal Stability Changes and Identification of HSP90 Clients

Protein thermal stability is directly related to its structure and interactions. When an inhibitor is added to cells, it is expected that both its intended targets and unintended off-targets are stabilized. For chaperones like HSP90, direct interactions with their client proteins are required to aid their folding. Both chaperones and client proteins are stabilized when they interact, and thus the perturbation of such interactions often results in protein thermal stability changes (46). Besides HSP90 clients, the cell response induced by the inhibition also causes the thermal stability changes of relevant proteins. Hence, a significant melting point shift of a protein may be attributed to several sources, including drug targets, interactors of the drug targets, off-targets of the drug, and proteins with structure and interaction changes due to the cell response to the drug treatment.

In this work, we used three HSP90 inhibitors with different structures (Fig. 1B and supplemental Fig. S1, A–E, supplementals Table S1 and S2), and all three inhibitors are bound to the HSP90 N-terminal nucleotide-binding site (16). Despite this commonality, AT13387 and STA-9090 have relatively similar backbone structures that greatly differ from that of 17-AAG (47, 48). As each inhibitor exhibits distinct off-targets and side effects, it is necessary to filter out inhibitor-specific effects for confident identification of true cellular responses to the HSP90 inhibition. Here, the structural differences of these inhibitors resulted in different thermal stability shifts of the HSP90 isoforms. Specifically, the AT13387 and STA-9090 treatments caused very similar thermal stability changes in all HSP90 isoforms (Fig. 2, C and D and supplemental Fig. S2). Moreover, both AT13387 and STA-9090 resulted in more pronounced melting point shifts in HSP90AA1 and HSP90AB1 compared with 17-AAG (Fig. 2, C and D), which is consistent with their higher binding affinities. Overall, these results further validate the accuracy and reliability of the current results.

By comparing the protein melting point shifts among the treatments by three inhibitors, we were able to identify both HSP90 clients and putative off-targets of inhibitors. Using the melting point shift >0.5 °C for thermal stability changes as the threshold, proteins that altered their thermal stability under the HSP90 inhibition were grouped accordingly in Figure 3A. Due to the folding assistance activity of HSP90, it is expected that most HSP90 clients would have decreased thermal stability under the HSP90 inhibition because they cannot fold well. We identified 175 proteins (Group B in Fig. 3A) that were thermally destabilized in the cells with all three HSP90 inhibitors. Among them, 65 proteins had previously been reported to be HSP90 clients (Fig. 3B and supplemental Table S3A) (8, 49, 50). Although they could be due to the cell response to the HSP90 inhibition, it is highly likely that this group of proteins contain more clients of HSP90 besides 65 previously reported ones. Notably, the majority of chaperones were stabilized by all three inhibitors, suggesting that the binding with their clients was strengthened under the HSP90 inhibition (Fig. 3C and supplemental Table S3B, the chaperone list from Shemesh et al. (51)). Typically, the members of the HSP40 and HSP70 families were stabilized. Importantly, these 175 proteins were not thermally stabilized, indicating that they did not bind to other chaperones because such binding tends to increase the melting points of client proteins. Consequently, they may be the exclusive substrates of HSP90.

Fig. 3.

The results of protein thermal stability changes in cells treated by each of the three inhibitors using TPP.A, Heatmap of significantly changed thermal stability of proteins by each inhibitor. B, A list of proteins in Group B categorized in Fig. 2A. C, Melting point shifts of quantified chaperones from different families by each inhibitor. The name of each chaperone family is indicated at the bottom of each subgroup. TPP, thermal proteome profiling.

Potentially, off-targets of the HSP90 inhibitors may be identified by comparing melting point shifts among all three treatments. For instance, 89 proteins (group C in Fig. 2A) were found to be stabilized only under the 17-AAG treatments, among which 69 proteins were not destabilized by the other two inhibitors. To more confidently identify the off-targets of 17-AAG, we applied more stringent criteria among these proteins: (1) higher thermal shifts by more than 1 °C with the 17-AAG treatment compared with the STA-9090 and AT-13387 treatments; (2) adj. p-values < 0.05 in all treatments. Using these criteria, two proteins were found to be potential off-targets of 17-AAG, ARF GTPase-activating protein GIT2 (GIT2), and Tripartite motif-containing protein 25 (supplemental Fig. S3, A and B). To the best of our knowledge, neither of them was reported to interact with 17-AAG previously. Notably, GIT2 has a significantly higher melting point shift (ΔTm = 1.8 °C, adj. p-value = 6.0E-6) under the 17-AAG treatment while STA-9090 and AT13387 cause no significant thermal stability change (ΔTm = −0.3 °C and 0.1 °C, adj. p-value = 4.81E-3 and 1.84E-3, respectively). These results strongly suggest that GIT2 is a probable off-target of 17-AAG. Meanwhile, for ARF GTPase-activating protein GIT1 (GIT1), no stability change was found under all the treatments (supplemental Fig. S3C), which shares 65% of sequences with GIT2 (52). However, we cannot exclude the possibility that these proteins are not the direct binding partners of the inhibitor, and further investigation is needed.

Proteins With Thermal Stability Changes Caused by the HSP90 Inhibition are Related to Cell Stress Response and Translation Process

By eliminating proteins that exhibited distinct thermal stability changes under the treatment of different inhibitors, protein thermal stability changes caused by the inhibition of HSP90 can be confidently profiled. Surprisingly, almost half of the proteins with significant thermal shifts in the same direction are thermally stabilized in all the treatments, while reduced thermal stability was found on a relatively smaller portion of proteins. The possible reason is that the unfolded protein response led to an escalated activity of other chaperones, which interacted with and stabilized their substrates. This was supported by the fact that many identified chaperones exhibited significantly increased melting points under the HSP90 inhibition (Fig. 3C). Additionally, the treatment time of 12 h is sufficient for cells to degrade some unfolded proteins that were not rescued by other chaperones.

We then conducted Gene Ontology analysis to investigate proteins stabilized with the inhibition of HSP90 (supplemental Fig. S4A and supplemental Table S4). Those associated with protein stabilization and unfolded protein binding were highly enriched, demonstrating that the protein quality control machinery was activated. After the accumulation of unfolded proteins in the cells, the protein quality control machinery was essential to cope with them. Besides, transcription-regulating proteins, like DNA methylation and chromatin remodeling proteins, were also overrepresented. Interestingly, many proteins that participated in the translation process exhibited increased thermal stability. It was reported that inhibiting HSP90 activity could decrease the global protein synthesis rates, and upregulate the expression of some proteins related to stress response (53). Increased stability of proteins related to the translation after the HSP90 inhibition may be because upregulated proteins for stress response need to be synthesized by the ribosome, which requires ribosomal proteins to bind with mRNA and other macromolecules. For proteins destabilized by all HSP90 inhibitors, their functions are related to a wide range of processes, as shown in supplemental Fig. S4B and supplemental Table S4B.

Comparative Analysis of Thermal Stability and Expression Changes of Proteins Under 17-AAG Treatment

Proteome thermal stability changes can provide unique and valuable information that is complementary to the abundance analysis results. By combining them together, a more comprehensive view of cellular response to the HSP90 inhibition can be generated. To extract protein abundance changes, we used the intensities of the reporter ions of peptides quantified from cells heated at the lowest temperature (37.0 °C), which is close to physiological temperature and has minimal extra aggregates produced during the heat treatment. It should be noted that the protein abundances represent only the soluble fraction of the proteome, which is different from normally referred protein abundances that include insoluble proteins extracted by NP-40, such as some insoluble aggregates (54). However, it is believed that the activity of a protein predominantly comes from its soluble form rather than the aggregated one (55). Hence, the current abundance analysis can still provide valuable information about cell response to HSP90 inhibition. To control the data quality, only proteins with their intensity coefficient of variance <0.5 among the duplicate experiments were included for further analysis (supplemental Fig. S5 and supplemental Table S5). The principal component analysis of the protein expression level in each experiment showed that the replicated samples were grouped together while different treatments resulted in distinct clustering (Fig. 4A), indicating the high reproducibility of the current experiments. All HSP90 inhibitor-treated samples possess great differences compared with the control ones in the first dimension. This result further demonstrates that proteins showing different expressions under different drug treatments should be excluded to unravel the real cellular response to the HSP90 inhibition.

Fig. 4.

Analysis of protein abundance changes in cells under the HSP90 inhibition.A, Principal component analysis of the samples treated with each of the three HSP90 inhibitors and the control ones. B, Scatter plots show protein abundance changes under the treatment of different HSP90 inhibitors. Proteins upregulated in 17-AAG are in yellow, and downregulated proteins are in blue. The distributions of protein abundance changes are presented at the top (AT13387 in dark green and STA9090 in light green) and right side (17-AAG in red). C, GO analysis of upregulated (yellow) and downregulated (blue) proteins in the cells with the HSP90 inhibition. HSP90, Heat shock protein 90.

Using the fold change >1.5 for expression changes as the threshold, proteins were grouped into upregulated and downregulated subgroups. Among 355 proteins that exhibited the same trends of abundance changes in all the experiments (Figs. 4B), 259 proteins were downregulated, which is consistent with the previous report that most proteins were downregulated with the HSP90 inhibition (22). Next, we attempted to discover the relevance between protein expression changes and thermal stability shifts. The majority of proteins had either the abundance change or thermal stability shift, showing that thermal stability and expression changes occurred in different groups of proteins (supplemental Fig. S6). It is well known that besides the abundance, protein activity changes may be from its structure and interaction changes. For example, many enzymes alter their activities without their abundance changes. GO enrichment analysis revealed that the functions of proteins with altered thermal stability or expression were relevant (Fig. 4C). For upregulated proteins, those related to response to unfolded protein stress were highly enriched. This represented an activated unfolding stress response, which is consistent with the stabilization of chaperone proteins discussed above. In the downregulated group, proteins associated with translation, typically ribosomal proteins, were overrepresented, which contributed to the reduced global protein synthesis under the HSP90 inhibition. Notably, those downregulated ribosomal proteins are downstream of the translation initiation proteins. This demonstrates that investigation of protein thermal stability and expression changes can provide us complementary information about cell response to external stimulus.

Despite the dramatic differences in protein expression and stability changes, we found that there is a small portion of proteins exhibited both expression and thermal stability changes under the HSP90 inhibition (supplemental Fig. S6). Those overlapped proteins are likely to be the most influenced and possess important information about cell response to the HSP90 inhibition. Functional annotation clustering results show that these proteins are highly associated with RNA binding (32 out of 64, p = 1.5E-11), agreeing with the previous result that thermal stability-altered proteins are highly related to RNA synthesis and processing (56).

Function Changes of HSP90 Under the Inhibition Revealed by PPI Analysis

The HSP90 complex includes various co-chaperones and regulatory proteins to control its ATPase activity and interact with its clients. Besides HSP90 modulators, increasing evidence demonstrates that HSP90 also directly participates in multiple signaling pathways through its interactions with other proteins. Hence, profiling PPI alterations under the inhibitor treatments at the proteome level provides valuable information for a better understanding of HSP90 functions. Thermal proximity coaggregation (TPCA) was utilized to detect PPI changes based on the nature that interacted proteins present similar precipitation profiles during a gradient of increasing heat exposure (38). Using TPCA, we identified nearly 2000 significantly altered PPIs (p-value < 0.05) under the treatment by each of the three inhibitors (supplemental Fig. S7 and supplemental Table S6).

We first extracted the significant PPI changes among direct interactions between HSP90AA1 or HSP90AB1 and other proteins (Fig. 5A and supplemental Fig. S8A and B). In total, 16 proteins displayed diminished coaggregation behavior with HSP90AA1 and HSP90AB1 in all the treatments, suggesting that their function alterations as their interactions with HSP90s were reduced. The composition differences of the HSP90 complex between the treated and control samples were monitored by detecting the altered coaggregation profiles of HSP90 with its important co-chaperones. Primarily, we studied the activator of HSP90 ATPase activity 1 (AHSA1), an important co-chaperone that increases the chaperone activity of HSP90 (57). Its interactions with HSP90AA1 and HSP90AB1 were reported to be directly interrupted by 17-AAG (58). From the thermal denaturation curves generated from our dataset, AHSA1 was coaggregated with HSP90AA1 and HSP90AB1 in the control samples, but precipitated at a significantly lower temperature after the 17-AAG treatment (supplemental Fig. S8A), agreeing with its reported binding with 17-AAG. Other key HSP90 co-chaperones were also detected. For example, stress-induced-phosphoprotein 1 (STIP1) and HSP90 co-chaperone CDC37 were reported to assist HSP90 in recruiting clients (59). In the current work, they are both less coaggregated with HSP90AA1 under the STA-9090 or AT13387 treatment (supplemental Fig. S8, B and C). 17-AAG enhanced the HSP90AA1-CDC37 interaction, while decreasing the HSP90AA1-STIP1 interaction.

Fig. 5.

Protein PPI alteration after HSP90 inhibition.A, A network diagram visualizing the coaggregation of HSP90 and their interactions. The color of the linkage indicates the distance changes between the treated and control samples. The width of the linkage indicates the significance of the PPI change, as a wider linkage stands for a lower p-value. Only significantly changed interactions are included. B, Venn diagram shows the overlap between quantified significantly changed PPIs among three treatments compared with the control samples. C, Gene ontology cluster analysis on proteins with significantly changed PPIs in all three treatments. A cluster enrichment score of 1.0 is used as the threshold. D, Volcano plot shows the protein complexes with significantly changed compositions under the 17-AGG treatment. p-value < 0.05 is the threshold. E, Heatmap representing 61 complexes significantly affected by three inhibitors. HSP90, Heat shock protein 90; PPI, protein–protein interactions.

Besides HSP90 regulators, many known HSP90 clients were found to have reduced interactions with HSP90 under inhibitions. For example, ribosomal protein S6 kinase B1 (RPS6KB1) was reported to be a client of HSP90, and the degradation of RPS6KB1 was enhanced under the HSP90 inhibition (60). In our dataset, the interaction between RPS6KB1 and HSP90AA1 was significantly interrupted under the 17-AAG treatment, indicating that the folding of RPS6KB1 was negatively affected, which could contribute to the degradation of RPS6KB1. Similarly, the interaction between HSP90 and DNA methyltransferase I (DNMT1) was weakened by all three inhibitors.

Global PPI Changes Induced by the HSP90 Inhibition

We then expanded the PPI analysis to the whole proteome. In total, 322 PPIs were significantly changed in all three treatments, correlated with 355 proteins (Fig. 5B). Based on these interactions, we can extract the coaggregation behavior alterations of some protein complexes (Fig. 5C), which are highly correlated to their function changes. For instance, all quantified interactions between ANAPC2 and its interactors were enhanced by three inhibitors, including multiple other subunits from the anaphase-promoting complex, a ubiquitin E3-ligase complex. As the inhibition of HSP90 activity causes increased unfolded protein loads, the elevated assembly level of the anaphase-promoting complex is a strong indication of facilitating protein degradation to cope with unfolded proteins. Another example is that the interaction between phosphatidylinositol 4,5-bisphosphate 3-kinase catalytic subunit beta isoform (PIK3CB) and all its detected interactors were strengthened, including KRAS and AKT2.

A list of large and essential macromolecular machines like the ribosome, the mitochondrial respiratory complex, and the proteasome were also studied (Fig. 5C). However, it is challenging to summarize the function change of large complexes from each PPI change between their subunits due to the complexity. To gain a deeper insight into the PPI changes caused by the HSP90 inhibition, all protein precipitation curves were grouped based on the CORUM complex annotation, and their interaction changes were scored by the ProSAP algorithm (40). The algorithm evaluates the precipitation curve distances between all subunits in a complex with and without treatment and compares them to randomized protein groups. Then, an FDR-controlled averaged distance was obtained, which is representative of the coaggregation state of a complex (Figs. 5D and supplemental Fig. S10, A and B). For each inhibitor treatment, the coaggregation behaviors of their subunits in protein complexes were investigated and more than 100 significantly changed complexes were identified (supplemental Table S7). In total, 61 complexes were shown to be significantly altered in all the treatments (Fig. 5E). Notably, all of them exhibited the same trend (more assembled or separated) by all the inhibitions. Various ubiquitin E3 ligase complexes were more assembled, consistent with the aforementioned observation of the anaphase-promoting complex. Many other important large complexes, like the nuclear pore complex and the spliceosome, were more dissociated with the inhibition. While the exact function alteration of those complexes is difficult to be extracted by the differences in coaggregation behaviors, our data highlighted a list of key protein complexes that respond to the HSP90 inhibition by changing their compositions rather than abundances in the cells.

Discussion

The protein chaperone HSP90 has been an attractive target for cancer therapy (2, 9). Although multiple HSP90 inhibitors have been developed, none has yet been approved by the FDA because of their toxicity and side effects (18). Systematic investigation of cell response to the HSP90 inhibition can help us better understand the inhibitors, their cellular toxicities, and HSP90 functions. In this work, three different HSP90 inhibitors, that is, 17-AAG, STA-9090, and AT13387, were systematically investigated in MCF7 cells by profiling protein thermal stability and quantifying protein abundance changes at the proteome level.

Previously, 17-AAG was studied by TPP with the aim of identifying HSP90 clients (49). However, the interferences of off-targets are hard to be excluded. To differentiate the potential off-targets of each inhibitor and the cellular response of the HSP90 inhibition, we used three different HSP90 inhibitors with different structures. Using multiple functionally relevant but structurally different drugs in the TPP experiments can effectively distinguish protein thermal stability changes caused by cell response and drug-specific effects.

Thermal stability changes can provide valuable information that cannot be obtained through the quantification of protein abundances. Previously, TPP was normally used to identify drug targets. Here, we employed TPP to study cell response to the drug treatment. Using three different inhibitors to exclude drug-specific effects, we systematically investigated cell response to the inhibition of HSP90. Furthermore, a comparative analysis of protein thermal stability shifts and abundance changes demonstrates that many proteins related to unfolded protein response are greatly stabilized while their abundances have no or relatively small changes. The upregulation of proteins related to protein quality control was found in the protein expression analysis, and the thermal stability shifts were also detected on these proteins, suggesting their activity alterations. This work demonstrates that the abundance changes can only partially reflect the cellular response to the HSP90 inhibition. Comprehensive analysis of protein thermal stability changes provides valuable and complementary information to aid in a better understanding of cell response to chaperone inhibition.

In this work, we systematically studied protein thermal stability and expression changes under the HSP90 inhibition using MS-based proteomics. Using TPP, the melting behaviors of more than 5000 proteins were quantified, and alterations in their thermal stabilities under the treatment of each of three different HSP90 inhibitors were comprehensively investigated. To distinguish inhibitor-specific effects from the real cell response to the inhibition of HSP90, three inhibitors with different structures and distinct binding modes were studied. This work reveals that the HSP90 inhibition significantly diminished cell transcription and translation processes. Combining protein thermal stability and expression analysis, we demonstrated that quantification of thermal stability changes of proteins provides insights into the cellular response to the HSP90 inhibition. Furthermore, TPCA analysis provides a general map of how PPIs were modulated under the HSP90 inhibition. This study provides a different perspective to further our understanding of cells' response to external stimulus.

Data Availability

The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD037575. (Username: reviewer_pxd037575@ebi.ac.uk; Password: tnb4rODy)

Supplemental data

This article contains supplemental data.

Conflict of interest

The authors declare no competing interests.