Quantitative proteomic analysis reveals unique Hsp90 cycle-dependent client interactions

Abstract

Hsp90 is an abundant and essential molecular chaperone that mediates the folding and activation of client proteins in a nucleotide-dependent cycle. Hsp90 inhibition directly or indirectly impacts the function of 10–15% of all proteins due to degradation of client proteins or indirect downstream effects. Due to its role in chaperoning oncogenic proteins, Hsp90 is an important drug target. However, compounds that occupy the ATP-binding pocket and broadly inhibit function have not achieved widespread use due to negative effects. More selective inhibitors are needed; however, it is unclear how to achieve selective inhibition. We conducted a quantitative proteomic analysis of soluble proteins in yeast strains expressing wild-type Hsp90 or mutants that disrupt different steps in the client folding pathway. Out of 2,482 proteins in our sample set (approximately 38% of yeast proteins), we observed statistically significant changes in abundance of 350 (14%) of those proteins (log₂ fold change ≥ 1.5). Of these, 257/350 (∼73%) with the strongest differences in abundance were previously connected to Hsp90 function. Principal component analysis of the entire dataset revealed that the effects of the mutants could be separated into 3 primary clusters. As evidence that Hsp90 mutants affect different pools of clients, simultaneous co-expression of 2 mutants in different clusters restored wild-type growth. Our data suggest that the ability of Hsp90 to sample a wide range of conformations allows the chaperone to mediate folding of a broad array of clients and that disruption of conformational flexibility results in client defects dependent on those states.

Introduction

The cytosolic molecular chaperone Hsp90 interacts with a limited yet diverse set of clients in a dynamic, ATP-dependent cycle to ensure proper folding, transport and/or assembly into multiprotein complexes (Taipale et al. 2010; Prodromou 2012; Schopf et al. 2017). Hsp90 is an important drug target due to its role in promoting activity of proteins required at multiple stages of cancer progression (Whitesell and Lindquist 2005; Workman et al. 2016). Compounds that bind the ATP-binding pocket block Hsp90 activity and frequently result in targeting of clients to the proteasome. However, broad inhibition of Hsp90 has negative effects (Trepel et al. 2010; Butler et al. 2015), and most inhibitors induce the stress response, resulting in protective effects (Zou et al. 1998; Butler et al. 2015). Further, diverse compounds that bind the ATP-binding pocket appear to elicit similar proteomic responses (Wu et al. 2012). Selective inhibitors that target specific types of clients, such as protein kinases, and/or those that do not induce the stress response are needed, but it is unclear how to achieve that goal.

Most organisms possess 2 isoforms of cytosolic Hsp90. In Saccharomyces cerevisiae, these are HSC82 and HSP82, one of which must be present for viability (Borkovich et al. 1989). High-throughput studies that measured the impact of genetic or pharmacological inhibition of Hsp90 indicate up to 10–15% of all proteins in yeast and humans are directly or indirectly dependent on Hsp90 for function (Millson et al. 2005; McClellan et al. 2007; Zhao and Houry 2007; Franzosa et al. 2011; Wu et al. 2012; Gopinath et al. 2014). Protein kinases and transcription factors are the most studied Hsp90 clients, but there are many others that carry out essential functions. Hsp90 interacts with Bms1, a GTPase required for ribosome assembly, and reduced Hsp90 function decreased stability of 80S ribosomes (Franzosa et al. 2011). More recently, Hsp90 was shown to interact with Ded1, a protein required for translation initiation, and mutation of Hsp90 resulted in reduced translational fidelity (Kolhe et al. 2023). While a handful of reports identified specific amino acid substitutions in yeast Hsp90 that affect protein kinases differently than other clients, such scenarios are limited, and most studies utilized proteins not native to yeast (Bohen 1995; Hawle et al. 2006; Flom et al. 2012; Mishra et al. 2016). More recently, mutations throughout the Hsp90 primary sequence were shown to negatively impact yeast viability in response to diverse cellular stressors (Cote-Hammarlof et al. 2021), but the underlying reasons for these effects are unknown.

We previously identified a panel of hsc82 alleles that present similar growth defects. However, the mutant Hsp90 proteins differed in their ability to form cochaperone complexes that occur during progression through the folding cycle. They also exhibit varied sensitivity to Hsp90 inhibitors and distinct outcomes with respect to established clients (Hohrman et al. 2021; Mercier et al. 2023). Preliminary protein expressing profiling identified proteins that were affected by some mutations and not others (Hohrman et al. 2021). The current study expands the analysis using quantitative proteomic analysis to compare soluble extracts from yeast expressing wild-type Hsp90 or 9 different Hsp90 mutants. Our results demonstrate that the mutants elicit varied proteomic responses, even in cells harvested at the optimal growth temperature. We further identify cellular processes and/or likely clients that are differentially affected by Hsp90 mutation. These results provide proteomic-level insight into Hsp90 biological roles and demonstrate the feasibility of selective inhibition of Hsp90 function in living cells.

Methods

Yeast were transformed by lithium acetate methods and were grown in either YPD (1% Bacto yeast extract, 2% peptone, and 2% dextrose), YPG (1% Bacto yeast extract, 2% peptone, and 3% glycerol) or defined synthetic complete media supplemented with 2% dextrose. Strain JJ816 (hsc82::LEU2  hsp82::LEU2/YEp24-HSP82) was transformed with plasmids expressing wild-type or mutant His-tagged Hsc82 in pRS313GPD or pRS314GPD (Mumberg et al. 1995; Johnson et al. 2007). Upon initial transformation into JJ816, 5-fluorootic acid (5-FOA) (Toronto Research Chemicals) was used to cure the YEp24-HSP82 plasmid. Growth defects were conducted by spotting 10-fold serial dilutions of yeast cultures on appropriate media, followed by incubation for 2 days at 30°C or 37°C.

Stress response pathway transcriptional activation assays

Cells expressing the STRE-lacZ reporter pCT31/32 (Schmitt and McEntee 1996) were grown to mid-log phase at 30°C in selective media (Trott et al. 2005). β-Galactosidase activity was measured by adding 50 µl of cell suspension and 50 µl of Beta-Glo reagent (Promega, Madison, WI) into a white 96-well plate (Lumitrac 200, Greiner) and incubating for 30 min at 30°C, followed by luminescence detection using a Synergy MX Microplate Reader (Garcia et al. 2017). The plot was prepared using Prism 8 (GraphPad Software, San Diego, CA).

Lysis and protein extract preparation

Strain JJ816 expressing wild-type or mutant Hsc82 was lysed as described (Hohrman et al. 2021). Briefly, strain JJ816 (hsc82hsp82) expressing WT or mutant His-Hsc82 were grown overnight in rich media (YPD), diluted into fresh media and grown to an OD₆₀₀ of 0.45–0.6. Cells were harvested, washed with water, and frozen at −80°C. To lyse, pellets were resuspended in lysis buffer [20 mM Tris, pH 7.5, 100 mM KCl, 5 mM MgCl₂ containing a protease inhibitor mixture (Roche Applied Science)]. Cells were disrupted in the presence of glass beads with 8 × 30 s pulses. Yeast extracts were analyzed by SDS-PAGE and Coomassie blue staining to ensure sample quality. Protein concentration was quantified by Bradford assays. Triplicate samples were shipped to the IDeA National Resource for Quantitative Proteomics at the University of Arkansas for Medical Sciences.

Chloroform/methanol extraction methods—Orbitrap Exploris DIA

Mass spectrometry analysis was performed at the IDeA National Resource for Quantitative Proteomics at the University of Arkansas for Medical Sciences. Proteins were reduced, alkylated, and purified by chloroform/methanol extraction prior to digestion with sequencing grade modified porcine trypsin (Promega). Tryptic peptides were then separated by reverse phase XSelect CSH C18 2.5 um resin (Waters) on an in-line 150 × 0.075 mm column using an UltiMate 3000 RSLCnano system (Thermo). Peptides were eluted using a 70 min gradient from 98:2 to 65:35 buffer A:B ratio (buffer A = 0.1% formic acid, 0.5% acetonitrile: buffer B = 0.1% formic acid, 99.9% acetonitrile). Eluted peptides were ionized by electrospray (2.2 kV) followed by mass spectrometric analysis on an Orbitrap Exploris 480 mass spectrometer (Thermo). To assemble a chromatogram library, 6 gas-phase fractions were acquired on the Orbitrap Exploris with 4 m/z DIA spectra (4 m/z precursor isolation windows at 30,000 resolution, normalized AGC target 100%, maximum inject time 66 ms) using a staggered window pattern from narrow mass ranges using optimized window placements. Precursor spectra were acquired after each DIA duty cycle, spanning the m/z range of the gas-phase fraction (i.e. 496–602 m/z, 60,000 resolution, normalized AGC target 100%, maximum injection time 50 ms). For wide-window acquisitions, the Orbitrap Exploris was configured to acquire a precursor scan (385–1,015 m/z, 60,000 resolution, normalized AGC target 100%, maximum injection time 50 ms) followed by 50 × 12 m/z DIA spectra (12 m/z precursor isolation windows at 15,000 resolution, normalized AGC target 100%, maximum injection time 33 ms) using a staggered window pattern with optimized window placements. Precursor spectra were acquired after each DIA duty cycle.

Following data acquisition, data were searched using an empirically corrected library against the UniProt Saccharomyces cerevisiae database (October 2019) and a quantitative analysis was performed to obtain a comprehensive proteomic profile. Proteins were identified and quantified using EncyclopeDIA (Searle et al. 2018) and visualized with Scaffold DIA using 1% false discovery thresholds at both the protein and peptide level. Protein exclusive intensity values were assessed for quality using ProteiNorm (Graw et al. 2020). The data were normalized using cyclic loess and statistical analysis was performed using linear models for microarray data (limma) with empirical Bayes (eBayes) smoothing to the standard errors (Huber et al. 2002; Chawade et al. 2014; Ritchie et al. 2015; Alhamdoosh et al. 2017; Bolstad 2023). For this study, proteins with an FDR adjusted P ≤ 0.055 and a fold change > 1.5 were considered significant.

Analysis and visualization of data

Figures were generated using Rstudio (“Mountain Hydrangea” Release; 547dcf86, 2023-07-07). Dot-chart generated using base-R functions. Bar-plots and Pie-charts generated using “ggplot2” R-package (Hadley 2016). Heatmaps were generated using “ComplexHeatmap” R-package (Gu 2022). Hierarchal clustering in heatmaps was calculated using Euclidean distance (default), but dendrograms were not shown in final figures. Visuals were improved aesthetically using “Circlize” R-package (Gu et al. 2014). The “Network D3” R-package was used to visualize GO Term analysis with a Sankey diagram.

Statistical analysis was performed using R programming language. For literature comparisons (Fig. 2c), hypergeometric distribution was used to calculate the significance of overlap between proteins with significant abundance changes identified in this study and protein hits identified in other independent studies (Supplementary Table 2: sheet 4). The protein hits from the referenced studies were pooled to calculate P values for “Total” and “Total w/ Micro”. The background genome used consists of 6,486 genes/proteins. P values less than 0.05 are considered significant.

Fig. 2.

Summary of proteomic changes in the Hsc82 mutant strains. a) Out of 350 total, the number of proteins that exhibited a log₂ fold change ≥ 1.5 (P ≤ 0.055) in the presence of each mutant relative to cells expressing wild-type Hsc82 (Supplementary Table 2, sheet 1). b) The relative proportions of proteins with increased or decreased abundance in the presence of each mutant. c) Comparison of hits to proteins linked to Hsp90 in high-throughput studies (see text for details). Left side shows 204 shared hits with studies identifying genetic or physical interactors of Hsp90. These were grouped together since they are likely clients due to evidence that they physically interact with Hsp90 and/or their function is decreased when Hsp90 function is decreased. Including studies identifying impacts of altered Hsp90 function using microarray analysis (right side), a total of 257 shared hits were identified. Hits affected at the mRNA level are likely due to indirect effects, such as downstream targets of Hsp90 clients. Hypergeometric distribution was used to calculate the significance of overlap between the current and prior studies (***P < 0.001; **P < 0.01; *P < 0.05).

Principal component analysis (PCA) was performed using “FactoMineR” R-package (Lê et al. 2008). To extract as much information as possible, the entire proteomic dataset consisting of log₂FC values of 2,482 proteins was used. No method of normalization was applied to the data before PCA. Initial PCA (Fig. 4a) included all mutants used in this study. Percentage of variances explained by principal components (PCs) 1–8 in descending order: 41.3, 18.1, 13.6, 8.6, 6.9, 4.4, 3.8, 3.3. Second PCA (Fig. 4b) was performed excluding W296A mutant. Percentage of variances explained by principal components (PCs) 1–7 in descending order: 30.6, 22.7, 15.7, 11.6, 7.4, 6.4, 5.6. Only first 2 principal components are used for further analysis. Results of PCA visualized using ggplot2.

Fig. 4.

Functional clustering of the Hsc82 mutants. a) Principal component analysis (PCA) of the entire dataset of 2,482 proteins shows distinct effects. b) PCA analysis of the dataset lacking W296A separates the remainder of the mutants in to 3 groups: (1) G309S and S481Y, (2) Q380K, and (3) the remainder of the mutants. c) The top 70 loading scores explaining variability along the 1st component axis in b) were extracted. Heat map showing log₂FC of the extracted 70 proteins showing all mutants. Green represents increased abundance in the mutant and red indicated decreased abundance in the mutant. Since this analysis was performed on the entire set of 2,482 proteins, some proteins shown are not in the top 350 hits.

Results

Prior studies that examined the impact of reduced Hsp90 function on cellular functions used either an Hsp90 inhibitor (Zhao et al. 2005; McClellan et al. 2007; Franzosa et al. 2011), a temperature-sensitive allele of Hsp90 (Zhao et al. 2005), or generated cells expressing very low levels of Hsp90 (Gopinath et al. 2014). This study is the first to examine the impact of multiple temperature-sensitive alleles of HSC82 on cellular functions on a proteome-wide scale. Each mutant alters a single amino acid, exhibits similar steady state protein levels, and shows a similar growth defect at 37°C when expressed in a strain containing a chromosomal deletion of both isoforms of endogenous Hsp90 (an hsc82hsp82 strain). However, the mutants affect distinct stages of the Hsp90 folding cycle and exhibit various phenotypes with regard to cochaperone interaction, sensitivity to Hsp90 inhibitors, and activity of select clients (Hohrman et al. 2021; Mercier et al. 2023). In a model based on interaction of Hsp90 with the glucocorticoid receptor (Schopf et al. 2017), client is targeted to Hsp90 by Hsp70 and the cochaperone Sti1 (Hop in mammals). Hsp90 adopts the closed conformation characterized by interaction with the Cpr6 and Sba1 cochaperones (Cyp40 and p23 in mammals, respectively). ATP hydrolysis results in return to the open conformation and client release. Mutants in the loading group (R46G, G309S) disrupt Hsp70–Hsp90 interaction (Kravats et al. 2018; Wang et al. 2022). Mutants in the closing group (S481Y, A583T) disrupt formation of the closed conformation and limit interaction with Sba1 and Cpr6 (Johnson et al. 2007). The reopening mutants (S25P, K102E, and Q380K) are located in regions important for regulation of ATP hydrolysis and/or associated conformational changes (Meyer et al. 2003; Ali et al. 2006; Mercier et al. 2019). We also included 2 mutants do not fit into either of those groups (referred to as “Miscellaneous”; G424D, W296A) (Fig. 1) (Hohrman et al. 2021).

Fig. 1.

Model of the Hsp90 folding cycle and the steps affected by each Hsc82 mutant. Model of Hsp90-client interaction. Client is targeted to Hsp90 by Hsp70 and Hop/Sti1. Hsp90 adopts the closed conformation characterized by interaction with Cyp40/Cpr6 and p23/Sba1. ATP hydrolysis results in return to the open conformation and client release. Mutants in the loading group disrupt Hsp70–Hsp90 interaction. Mutants in the closing group disrupt formation of the closed conformation and interaction with Sba1 and Cpr6. Mutations in the reopening group disrupt ATP hydrolysis and/or associated conformational changes. The W296A and G424D mutants do not fit within those groups and thus are marked “miscellaneous”.

Overall proteomic results

Hsp90 inhibition or mutation results in inhibition of client activity, and in many cases results in proteasomal degradation of clients (Nathan and Lindquist 1995; An et al. 2000). In other cases, Hsp90 inhibition results in decreased degradation (McClellan et al. 2005). Microarray analysis of cells treated with Hsp90 inhibitors demonstrated that Hsp90 inhibition results in induction of the stress response (Echeverria et al. 2011). Collectively, over 15% of all cellular proteins are impacted by genetic or pharmacological reduction of Hsp90 function, which includes clients and their downstream targets (Zhao et al. 2005; McClellan et al. 2007; Zhao and Houry 2007; Taipale et al. 2010; Franzosa et al. 2011; Gopinath et al. 2014). We used the data independent acquisition mass spectrometry (DIA-MS) quantitative proteomic platform to compare soluble protein levels in extracts from yeast cells expressing wild-type Hsp90 or 9 different mutants. This allowed us to capture both direct and indirect effects of Hsp90 perturbation. DIA-MS was chosen as the method of detection as improvements in sensitivity and speed in recent years have enabled proteome-wide levels of detection with increased reproducibility compared to other MS methods (Li et al. 2020; Krasny and Huang 2021). Three independent cultures of strains expressing plasmid-borne wild-type or mutant Hsc82 as the sole Hsp90 isoform in the cell were grown at the permissive temperature of 30°C and harvested during exponential growth phase (OD₆₀₀ 0.45–0.6). After trypsin digestion and analysis of comparative expression, 2,558 unique proteins were detected, but only 2,482 were quantified across all samples, representing approximately 38% of the yeast genome (2,482/6,486 annotated genes) (Supplementary Table 1). To determine if our results are biased toward more abundant proteins, we compared the distribution of the abundance of the 2,482 proteins identified in this study to a prior study that estimated the abundance of yeast proteins on a genome-wide scale (Ghaemmaghami et al. 2003). As shown in Supplementary Fig. 1, the proteins identified in this study exhibit a distribution of abundance similar to the overall yeast proteome, indicating that our results are not biased toward more abundant proteins.

The log₂ fold change (log₂FC) describes the change in abundance of proteins in pairwise comparisons with wild-type and mutant Hsc82 strains. The total number of proteins affected (log₂FC ≥ 1.5; P ≤ 0.055) was 350, but the number in each mutant varied (Fig. 2a). Changes in the abundance of additional proteins were detected, but these were excluded due to lack of statistical significance (P > 0.055) (Supplementary Table 2; sheet 1). For example, the abundance of 244 proteins changed in cells expressing the hsc82-W296A (hereafter W296A) mutant, but only 27 proteins were affected by hsc82-S25P (hereafter S25P). Consistent with prior studies, we observed proteins with both increased and decreased abundance (Gopinath et al. 2014), but the relative proportion of proteins with increased/decreased abundance varied (Fig. 2b). When we pooled the list of hits with large and statistically significant changes in abundance spread across the mutant samples, a total of 350/2,482, or 14%, of our available dataset were affected, consistent with Hsp90 inhibition impacting >15% of cellular proteins (Zhao et al. 2005). The overall pattern remained the same when a lower threshold (log₂FC ≥ 1.0; P ≤ 0.055) was examined, with 740 proteins impacted across mutant samples (Supplementary Table 2; sheet 2) (Supplementary Fig. 2, a and b). A simplified list of the impact of mutation on all proteins detected is shown (Supplementary Table 3; sheet 3).

We compared our list of the 350 top hits with high-throughput studies that identified genetic or physical interactors of Hsp90 (Millson et al. 2005; Zhao et al. 2005; McClellan et al. 2007; Franzosa et al. 2011; Truman et al. 2015; Kolhe et al. 2023). We also included a study that identified proteins with altered protein, but not mRNA, levels in a strain expressing low levels of yeast Hsp90 (Gopinath et al. 2014). This analysis showed that 204/350 proteins with significant changes in abundance were previously linked to Hsp90 (Fig. 2c, left and Supplementary Table 2; sheet 4). Two recent studies using crosslinking approaches to identify proteins that interact directly with Hsp90 had the highest overlap with our study. A crosslinking approach with the artificial amino acid p-benzoyl-L-phenylalanine shared 111 hits (Kolhe et al. 2023), while one that used formalin-crosslinking followed by tandem mass spectrometry shared 63 hits (Girstmair et al. 2019). Another 54 hits were shared with the study comparing changes in protein vs mRNA abundance in cells expressing low levels of Hsp90 (Gopinath et al. 2014). Chemical genetic screens performed with an Hsp90 inhibitor using homozygous (McClellan et al. 2007) or heterozygous (Franzosa et al. 2011) diploid deletion strains identified 15 and 18 hits, respectively. A large-scale analysis to identify putative Hsp90 clients shared 33 hits (Zhao et al. 2005), a study examining Hsp90 interactomes before and after DNA damage shared 14 hits (Truman et al. 2015), and a yeast 2-hybrid analysis had 9 shared hits (Millson et al. 2005). When we also considered studies measuring changes in mRNA transcript levels resulting from reduced Hsp90 function, the list of shared hits expands to 257/350 (73%) (Fig. 2c: right). This includes hits affected at both the mRNA and protein level (92 hits) (Gopinath et al. 2014), prior microarray analysis of cells expressing W296A (50 hits) (Flom et al. 2012), and a microarray analysis of yeast cells treated with Hsp90 inhibitor (32 hits) (Echeverria et al. 2011). We separated these hits into the broad categories to attempt to separate them into likely clients (genetic or physical interactors) vs indirect effects (due to downstream effects of kinases and transcription factors) based on methodology used to identify the hits. However, further work is needed to clearly establish the role of Hsp90 in mediating the folding of the majority of these proteins.

Analysis of proteins affected in hsc82-W296A

W296A is notable for 2 reasons: it had the biggest impact on protein abundance (244/350 total hits), and the largest percentage of proteins with increased abundance (Fig. 2, a and b). We previously reported that cells expressing W296A exhibited elevated levels of Hbt1 protein and mRNA levels, and we conducted a microarray analysis to compare the transcriptome of cells expressing Hsc82 wild-type or W296A. In that study, the levels of 5 mRNAs decreased and 132 increased (log₂FC ≥ 2.0) (Flom et al. 2012). Over 80% of the affected genes contained at least 1 stress response element (STRE), the binding site for the Msn2/Msn4 transcription factors (Boy-Marcotte et al. 1999). A comparison of W296A proteomic data and microarray data shows 43 hits that exhibited a similar increase at both the transcript and protein level, validating that increased protein levels are due to increased mRNA levels (Fig. 3a, bottom 2 rows). An additional 139 proteins had elevated protein abundance in W296A. Of those, either there was no significant change in mRNA level in the prior study (90 hits) (Flom et al. 2012), or the corresponding microarray data were unavailable. Most of these hits (103/139) contain STREs or have been linked to Msn2/Msn4 (Teixeira et al. 2006). Consistent with our prior study, Hbt1 protein levels were elevated in cells expressing W296A but not G309S, S481Y, or A583T (Flom et al. 2012) (Fig. 3a: red). However, Hbt1 levels were elevated in some mutant strains, the highest being R46G (log₂FC ∼2.5), compared to W296A (log₂FC ∼4.5). R46G shows a similar proportion of proteins with increased abundance when compared to W296A (Fig. 2b), and 16 proteins had statistically significant increases in abundance in both R46G and W296A (Fig. 3a and Supplementary Table 2). Since we identified R46G, S25P, K102E, and G424D in a screen for temperature-sensitive mutants after the prior study (Hohrman et al. 2021), we directly tested the effect of these mutants on activation of a STRE-lacZ reporter construct. Consistent with the changes we observed at the protein level (Fig. 3b), cells expressing R46G result in a more than 2-fold increase in basal activity, whereas no increase was observed in cells expressing G309S, K102E, or G424D. Overall, these results indicate that W296A causes a dramatic induction of genes regulated by Msn2/Msn4, and that a weaker induction was observed in R46G.

Fig. 3.

Analysis of proteomic changes in W296A. a) Heat map showing log₂FC of 43 proteins with increased abundance (green) in W296A and a corresponding increase in mRNA levels using published microarray data [W296A (m)] (Flom et al. 2012). Hbt1 protein is highlighted in red.b) β-Galactosidase activity of cells expressing STRE-lacZ reporter plasmid in indicated Hsc82 mutant. P = 0.05*, 0.005**, 0.0005***. c) Heat map showing log₂FC of 62 proteins with significantly decreased abundance in W296A. In heat maps, green represents increased abundance in the mutant and red indicated decreased abundance in the mutant. Potential clients shown in red and/or bold in c). Proteins for which there was no significant change in mRNA levels in prior study are marked in red (Flom et al. 2012). Proteins that crosslinked to Hsp90 shown in bold (Girstmair et al. 2019; Kolhe et al. 2023).

We next focused on the 62 proteins with reduced abundance in W296A. For 25 of those proteins, we observed decreased protein levels without the corresponding decrease in mRNA level in the prior microarray analysis (Fig. 3c: red), suggesting that they are client proteins dependent on Hsp90. Of the 62, 42 exhibited log₂FC ≥ 1.5 only in that mutant, indicating selective effects. As further evidence that some of these are likely client proteins, 24/62 were shown to crosslink to Hsp90 (Fig. 3c; bold) (Girstmair et al. 2019; Kolhe et al. 2023). One set of proteins with decreased abundance in W296A is Vtc2, Vtc3, and Vtc4. Along with Vtc1, these are components of the vacuole transporter chaperone (VTC) complex, and all 4 were shown to crosslink to Hsp90 (Kolhe et al. 2023). The VTC complex has important roles in vacuolar membrane fusion, autophagic processes, and other transport related processes (Uttenweiler et al. 2007; Armstrong 2010). Vtc4 is involved in production of inorganic polyphosphate and cells starved for inorganic phosphate exhibit altered expression of genes involved in the phosphate signal transduction (PHO) pathway. Prior studies showed that yeast lacking VTC4 have reduced expression of PHO5 and PHO84 (Tomashevsky et al. 2021). As shown in Fig. 3c, levels of Pho5 and Pho84 proteins were decreased in the presence of W296A, suggesting that decreased levels of Vtc2/3/4 proteins result in functional defects.

Analysis of the similarities and differences between Hsc82 mutants

We noticed that some proteins with increased abundance in W296A showed decreased abundance in cells expressing other mutants, mainly G309S and S481Y (Fig. 3a). To better characterize mutant-specific effects, we conducted PCA using the entire dataset of 2,482 proteins. Unsurprisingly, initial PCA clearly separated W296A from the other mutants along the 1st principal component (PC1) axis (Fig. 4a), which explained 41.3% of the variability between the mutants. Upon removing W296A, the amount of variability explained by PC1 and PC2 is 30.6 and 22.7%, respectively, for a total of 53.3% (Fig. 4b). This results in separation into 3 groups: Q380K; G309S, and S481Y; and then the remaining mutants. G309S and S481Y are in the “loading” and “closing” groups (Fig. 1), respectively, and we previously showed they have similar effects in vivo (Hohrman et al. 2021; Mercier et al. 2023).

Loading scores in PCA denote the weights assigned to variables (in this case, log₂FC values for each protein), which reflects their contribution to the respective principal components. The proteins with the top 70 loading scores from the 1st principal component in Fig. 4b were extracted to identify hits that best explain the separation of effects of G309S and S481Y from the other mutants (Fig. 4c). From left to right, there is a gradient of patterns: (1) overall decrease in G309S, S481Y, and W296A; (2) overall decrease in W296A and mutants other than G309S and S481Y; (3) proteins decreased mainly only in G309S and S481Y, (4) proteins decreased G309S and S481Y but increased in W296A, and finally (5) proteins increased in W296A, G309S, and S481Y. A prior study showed that Hsp82 mutation results in altered activity of the protein kinase client Gcn2, resulting in constitutive expression of a GCN4-lacZ reporter (Donze and Picard 1999). Gcn2 stimulates translation of the yeast transcription factor Gcn4 upon amino acid starvation. Consistent with signs of amino acid starvation, many of the proteins shown to have increased abundance in G309S and S481Y are required for amino acid synthesis (such as Arg7, Arg4, His5, and Lys1) (Fig. 4c) (Teixeira et al. 2006; Rawal et al. 2018). A mutant used in the prior study, Hsp82-G313N, alters the residue in Hsp82 that is homologous to Hsc82-G309, and these results suggest that Gcn2/Gcn4 activity is selectively affected by Hsp90 mutation.

Extraction of the top loading scores from the 2nd principal component in Fig. 4b identified many mitochondrial proteins (Supplementary Fig. 3a). Using Gene Ontology Term finder (Ashburner et al. 2000; Boyle et al. 2004), we found that Q380K affected many mitochondrial proteins (33/58 hits). When the specific culture used for the proteomic analysis was tested, Q380K cells were unable to grow using glycerol as the carbon source. However, a fresh version of that strain was able to grow on glycerol (Supplementary Fig. 3b). Thus, it is likely that some of the proteomic changes in Q380K are secondary mutations that resulted in loss of mitochondrial function and are not due to the Q380K alteration in Hsc82. To examine mutant-specific changes further, we identified proteins with the biggest changes in abundance using volcano plots, focusing on hits shared among mutants. Three proteins, Mck1, Sam4, and Bul1 showed similar changes in W296A, G309S, and S481Y (Supplementary Fig. 4; blue). Three proteins (Rad3, Dal7, and Hsp26) showed changes only G309S and S481Y (Supplementary Fig. 4; black). Another 4 proteins (Pof1, Anb1, Ubc4, and Rpc19) showed similar changes in the majority of mutants (Supplementary Fig. 4; red). This suggests that some proteins, such as Hbt1, are signatures of defects in particular aspects of Hsp90 function, while others, such as Anb1, may be a more general signal of reduced function. We performed 1 additional analysis to determine how data in our proteomic dataset compare to other studies of Hsp90 inhibition. Surprisingly, the levels of 32 proteins that crosslinked to Hsp90 did not change significantly in the presence of an Hsp90 inhibitor (Kolhe et al. 2023). Consistent with their results, we did not detect large decreases in the abundance of 29/32 proteins that were present in our dataset (Supplementary Fig. 5), although levels of 3 proteins were elevated. This includes Anb1, which we identified above as a hit with increased abundance in many mutants. Anb1 is normally expressed only under anerobic conditions (Lowry and Zitomer 1984). Two other proteins, Nsr1 and Mak5, showed selectively decreased abundance. The overall conclusion from this test is that results from our proteomic analysis of genetic mutants largely conform with direct tests of the impact of pharmacological Hsp90 inhibition on cellular proteins.

Analysis of the ability of Hsc82 mutants to complement each other

A comparison of the proteins affected in W296A and G424D shows that the levels of only 6 proteins were significantly decreased in both mutants (Supplementary Table 2). Since the effects of the mutants were so distinct, we tested the effect of co-expression of the 2 mutants, reasoning that client defects caused by 1 mutant may be complemented by a mutant that affects other clients. We used sets of plasmids expressing wild-type or mutant Hsc82 that differ only by the selective marker. As shown in Fig. 5, a and b, cells expressing either G424D or W296A exhibit growth defects at 37°C, and the growth defect of either mutant is relieved by co-expression of wild-type. Co-expression of the same mutant does not improve growth, and in the case of G424D, may even further impair viability. Strikingly, co-expression of G424D and W296A restored wild-type growth. This experiment provides further evidence that W296A and G424D affect nonoverlapping sets of Hsp90 clients. We further tested rescue of the growth defects of G309S (Fig. 5c). The growth defect was rescued by co-expression of wild-type, but the presence of additional G309S or S481Y, which are in the same PCA cluster, did not result in significantly improved growth. Co-expression of K102E or G424D partially rescued the growth defects, while W296A resulted in near wild-type growth. This is consistent with the results in Supplementary Fig. 4 that some mutants have a mixture of shared and distinct effects. To identify shared features of likely client proteins that were affected by some mutants and not others, we analyzed proteins with reduced abundance using GO Term analysis (Supplementary Fig. 6). This identified functional classes of proteins similarly affected by S25P, K102E, and G424D, consistent with the PCA groupings. This also showed that protein kinases were specifically affected by R46G, G309S, and S481Y, which is consistent with our prior results that mutants in the loading and closing groups had the largest effect on activity of the v-src kinase (Hohrman et al. 2021). A list of proteins within in each category is provided in Supplementary Table 3.

Fig. 5.

Co-expression of Hsc82 mutants with distinct effects restores growth. An hsc82hsp82 strain expressing either G424D a) or W296A b) was transformed with vector, a plasmid expressing wild-type HSC82, or plasmids expressing either W296A or G424D. Cells were grown overnight in selective media, serially diluted 10-fold, then grown for 2 days at 37°C. c) An hsc82hsp82 strain expressing G309S was transformed with vector, a plasmid expressing wild-type HSC82, or plasmids expressing the indicated mutant. Cells were grown overnight in selective media, serially diluted 10-fold, then grown for 2 days at 37°C.

Discussion

The essential and abundant molecular chaperone Hsp90 is required for the folding of a significant portion of the eukaryotic proteome. Since many Hsp90 clients are critical for progression of cancerous growth, a driving force in the field has been the development of Hsp90 inhibitors (Trepel et al. 2010). Inhibitors that bind the ATP-binding pocket have shown promising results in treating a variety of cancers, but only recently have pan-inhibitors found limited clinical success (Kurokawa et al. 2022). Efforts to develop isoform-specific inhibitors, or inhibitors that target Hsp90-cochaperone interaction are ongoing, but the extent of the selective effects of these inhibitors is unknown (Smith et al. 2015; Khandelwal et al. 2018; Mishra et al. 2021; Serwetnyk and Blagg 2021). We previously described a series of mutations in yeast Hsp90 that result in similar temperature-sensitive growth defects and showed that these mutants had differing effects in a limited proteomic analysis using 2D-DIGE (Hohrman et al. 2021; Mercier et al. 2023). Here we identified 350/2,482 proteins (∼14% of the yeast proteome) that exhibited at least a 1.5 log₂FC in abundance in the presence of at least 1 Hsp90 mutant. Over 73% of the affected proteins (257/350) were previously linked to Hsp90 (Millson et al. 2005; McClellan et al. 2007; Zhao and Houry 2007; Franzosa et al. 2011; Gopinath et al. 2014; Truman et al. 2015; Girstmair et al. 2019; Kolhe et al. 2023). Every mutant tested had differing effects, with one altering the level of 244/350 proteins and another altering the levels of only 13. PCA analysis supports clustering of the mutants into 3 broad groups based on impacts on protein abundance. This clearly demonstrates that the in vivo consequences of yeast Hsp90 mutants differ and suggests that it will be possible to selectively alter critical functions of mammalian Hsp90 pharmacologically. Co-expression of cells expressing separate temperature-sensitive mutants from different PCA groups results in near wild-type growth, indicating that client defects due to 1 mutant may be rescued by another mutant that has distinct client defects. Further studies, such as those described elsewhere (Wayne et al. 2010), are needed to determine whether heterodimerization of the mutants contributes to the ability to rescue growth defects.

Hsp90 undergoes a series of dramatic conformational changes and may exist in a dynamic equilibrium of conformations (Krukenberg et al. 2011). Mutations that disrupt the timing of progression through open and closed states disrupt the essential function of Hsp90 (Zierer et al. 2016). Most of the mutants in our study were grouped according to their impact on formation of Hsp90-cochaperone complexes associated with progression through the folding cycle (Hohrman et al. 2021; Mercier et al. 2023). The loading mutants R46G and G309S disrupt physical interaction between Hsp70 and Hsp90 and are predicted to disrupt the efficient transfer of clients from Hsp70 to Hsp90 (Kravats et al. 2018; Wang et al. 2022). The closing mutants S481Y and A583T disrupt the ability of Hsp90 to adopt the closed conformation in the presence of AMP-PNP (Ali et al. 2006; Johnson et al. 2007). The reopening mutants S25P, K102E, and Q380K alter residues near the nucleotide binding site or are in a flexible loop involved in ATP hydrolysis (Meyer et al. 2003; Ali et al. 2006). The residue altered in the W296A mutant has been described as a critical switch point in the Hsp90 folding cycle. Mutations in the homologous residue of yeast Hsp82 (W300) disrupted communication between client interaction and conformational changes required to progress through the folding cycle (Rutz et al. 2018). Another study with human Hsp90 alpha showed that the W320A alteration resulted in a more open conformation (Peng et al. 2023). The residue altered in G424D has been predicted to be within a hinge region important for allosteric communication within Hsp90, and thus it is also likely to alter Hsp90 conformational changes (Blacklock and Verkhivker 2014; Hohrman et al. 2021).

Our model (Fig. 6) postulates that Hsc82 mutants impact different aspects of the conformational cycle, resulting in client-specific effects. W296A is predicted to be in the most open conformation (Peng et al. 2023). G309S and S481Y have defects prior to formation of the closed conformation, while S25P and K102E appear to affect steps after formation of the closed conformation. The conformational defect of G424D is unknown. Mutants that likely result in a more open conformation had the largest overall effects. Our prediction is that conformational defects disrupt either cochaperone or client interaction. Detailed structural information about Hsp90-client interaction is available for only a few clients (Karagoz et al. 2014; Verba et al. 2016; Wang et al. 2022), but presumed Hsp90 clients crosslinked to multiple sites within Hsp90 (Kolhe et al. 2023). Cochaperones are known to interact with each domain of Hsp90 and many bind in a conformation-specific manner (Karagoz and Rudiger 2015; Rehn et al. 2016; Mader et al. 2020; Lopez et al. 2021). Our prior classifications of mutants based on their effect on Hsp90 complexes and structural information is mostly consistent with the PCA grouping. The main exceptions are R46G and A583T, which are also members of the loading and closing group, respectively. We predicted that their effects would also cluster with G309S and S481Y. The reason for this discrepancy is unknown, but 2 possibilities are that the overall defects of R46G and A583T are relatively mild under these conditions, or that the mutants affect client activity without affecting the steady state levels of clients. Further studies are needed to clarify these alternative scenarios.

Fig. 6.

Model of effect of Hsp90 conformation on client specificity. Our prior grouping of Hsc82 mutants identified 4 groups: loading, closing, reopening, and other. Based on these studies, a refined model is shown. The W296A mutant is proposed to be in an extended conformation (Peng et al. 2023), G309S and S481Y are proposed to favor open conformations, and S25P and G424D are proposed to be in various closed conformations (Mercier et al. 2023). These conformation groups fit well with the clusters identified by principal component analysis (PCA) and our prior groupings. According to this model, Hsc82 mutants that affect steps prior to adoption of the closed conformation impact the most proteins.

Hsp90 alteration resulted in significant downstream effects on transcription. One of the most dramatic effects in this study were increases in abundance of the products of genes regulated by Msn2/Msn4, which mediate the general stress response in yeast (Boy-Marcotte et al. 1999). Many of the changes linked to Msn2/Msn4 were specific to W296A (Flom et al. 2012). However, R46G resulted in increased basal activity of a STRE-lacZ reporter construct and shared elevated abundance of 17 proteins with W296A. The reason for the shared effects with R46G is unknown, but 1 possibility is altered interaction with the cochaperone Sgt1, since R46 is in the Sgt1 interacting domain (Zhang et al. 2008). A mutation in Sgt1 resulted in similar transcriptional effects as W296A (Flom et al. 2012). We also observed increased abundance of proteins that are downstream targets of Gcn2, an Hsp90-dependent protein kinase (Donze and Picard 1999), and increased levels of proteins in the phosphate availability (PHO) pathway, which are likely due to decreased abundance of components of the vacuolar transporter chaperone complex (Vtc2, Vtc3, and Vtc4) (Tomashevsky et al. 2021). Further studies are needed to demonstrate that the W296A mutation affects vacuolar functions and/or the availability of vacuolar inorganic phosphate. In our prior analysis using 2D-DIGE, we showed that Hsp90 mutation had differential effects on levels of targets of the transcription factor Hsf1 (Hohrman et al. 2021). We did not observe changes in the abundance of Hsf1, but we observed slight increases in levels of 6 Hsf1 targets (Mdj1, Aha1, Ssa1, Hsp42, Hsp78, and Hsp104 showed log₂FC ≥ 1.0) (Supplementary Table 2; sheet 5) (Solis et al. 2016). However, the pattern of changes at the protein level was not consistent and it is unclear whether this is due to differing effects on transcription. The levels of 2 additional transcription factors, Dal81 and Tod6, were decreased (>1.5 log₂FC) in at least 1 Hsc82 mutant. Dal81 is a general positive regulator of genes involved in nitrogen utilization (Palavecino et al. 2015). Tod6 is a repressor of genes required for ribosomal biogenesis (Lippman and Broach 2009), and altered Tod6 activity could explain change in abundance of ribosomal subunits observed in our prior analysis (Hohrman et al. 2021).

Recent studies that identified proteins that crosslink to Hsp90 are key in trying to identify direct effects on client proteins (Girstmair et al. 2019; Kolhe et al. 2023). We identified 62 proteins that exhibited decreased abundance in cells expressing W296A, suggesting that they are clients with reduced stability in those cells. Using Yeast-GO slim mapping (Ashburner et al. 2000), these proteins were enriched in categories of rRNA processing or translation (Bud27, Hca4, Nop19, Nsr1, Rpl8A, Sen1, Utp18, Rps29B) and DNA repair (Dpb2, Msh1, Sen1). Proteins in each of these functional categories have previously been linked to Hsp90 (Flom et al. 2005; Tenge et al. 2014; Truman et al. 2015; Kolhe et al. 2023). Notably, 4 proteins are RNA helicases (Dhr2, Hca4, Mak5, and Sen1). RNA helicases have multiple roles regulating gene expression (Bourgeois et al. 2016), and this may represent an understudied class of Hsp90 clients. Two other proteins that showed decreased abundance are Mht1 and Sam4. These proteins regulate the ratio of methionine/S-adenosylmethionine (Thomas et al. 2000). Additional studies are needed to determine whether Hsp90 mutation selectively alters Mht1 and Sam4 function.

Hsp90, together with the cochaperone Cdc37, is known to interact with protein kinases (Caplan et al. 2007; Verba and Agard 2017). We previously showed that the activity of the v-src kinase was most strongly affected by mutants in the loading and closing group (Hohrman et al. 2021). We identified 12 protein kinases with decreased abundance in at least 1 mutant (>1.0 log₂FC) (Mck1, Tpk3, Dun1, Kic1, Kin28, Pkc1, Hrr25, Sgv1, Cdc28, Tpk1, Slt2, and Smk1). Many of these protein kinases have already been linked to Hsp90 or Cdc37 (Piper et al. 2006; Mandal et al. 2007). The largest changes in abundance were observed in cells expressing the loading and closing mutants, although levels of Mck1 were also strongly affected by W296A (Fig. 4c). Mck1 has roles in chromosome segregation and regulating entry into meiosis (Quan et al. 2015; Rathi et al. 2022). Mutations in Hsp90 have previously been shown to have sporulation defects (Tapia and Morano 2010), and further studies are needed to determine if those defects are due to reduced Mck1 function.

One of the practical aspects of this study is that we identified proteins highly sensitive to Hsp90 perturbation. Some, like Anb1, showed large (log₂FC > 3) increased in abundance in most mutants, while other proteins, such as Rad3, showed more selective effects. Since the Lindquist lab first developed assays of v-src and glucocorticoid receptor activity in yeast (Nathan and Lindquist 1995), those have been the predominant clients used to study Hsp90 and cochaperone function. In future studies, we will exploit our findings to develop sensitive reporter assays that will allow us to monitor changes in the level of these proteins. For example, changes in Anb1 protein level could be a sensitive assay of overall Hsp90 function, while changes in levels of others could be indicators of more selective inhibitors. Additional studies also are required to determine how existing Hsp90 inhibitors or cochaperone deletion affect some of these proteins. We will also determine how a temperature shift changes the proteome in different mutants. A critical question is whether the mutants will have more similar effects upon heat shock, suggesting that main determinant may be overall level of Hsp90 activity, or whether the effects will become more divergent, uncovering more selective effects. However, the ability of Hsp90 mutants in different clusters to complement the growth defects of each other suggests that the defects will remain distinct at elevated temperature. Future analysis of the effects of Hsp90 mutation on the proteins affected in this study will help clarify the underlying mechanism of the selective effects and may provide new clues to how to selectively modulate Hsp90 function by targeting specific conformations.

Data availability

Strains and plasmids are available upon request. R-code required to generate figures is provided in the Supplementary File 1. The authors affirm that all data necessary for confirming the conclusions of the article are present within the article, figures, and tables. The raw data for proteomic dataset are available at the MassIVE database with the identifier MSV000094197 (doi:10.25345/C5Z60CC93).

Supplemental material available at GENETICS online.

Funding

This research was primarily supported by the National Institute of General Medical Sciences under award numbers R01GM127675 (JLJ), RO1GM127675-03S1 (JLJ), and R01GM127287 and R35GM149196 (KAM). The DIA-MS analysis was conducted by the IDeA National Resource for Quantitative Proteomics and NIH/NIGMS grant R24GM137786. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.