A Proteomic Investigation of Ligand-dependent HSP90 Complexes Reveals CHORDC1 as a Novel ADP-dependent HSP90-interacting Protein^*

Plasmids

Expression vectors for proteomics analysis were derived from the parental TAP¹ expression vector pGlue (28). The TAP tag contains a strep tag peptide sequence that binds streptavidin with high affinity, an intervening tobacco etch virus protease cleavage sequence, and a calmodulin-binding peptide. Full-length human HSP90AA1 (HSP90α) cDNA was constructed by fusing the 5′ end (GenBank^TM accession number BQ276361.1) and the 3′ end (GenBank accession number BC023006.1) using GenBank accession number X15183 as a reference sequence. N-terminal TAP fusions of the wild type HSP90α and mutants were created by site-directed mutagenesis and PCR amplification from cDNAs and verified by sequencing. Human CHORDC1 cDNA was constructed from RNA isolated from HEK293T cells and PCR-amplified. Subsequent co-expression analysis was performed using pCS2+ parental vectors for all C-terminal HSP90-TAP constructs and pSPORT6 (Invitrogen) for expression of HA-tagged CHORDC1 constructs in HEK293T cells. For E. coli expression vectors, HSP90 and HA-CHORDC1 coding sequences were PCR-amplified from the above vectors and cloned into pGWS (pGEX derivative; kindly provided by Bruce Clurman, Fred Hutchinson Cancer Research Center) and pET16b (Novagen) for N-terminal GST-tagged HSP90 and N-terminal His-tagged HA-CHORDC1 constructs, respectively. All constructs were verified by sequencing.

Cell Culture and Viability Assay

Low passage number HEK293T cells (ATCC CRL-1573) were used in all experiments. Cells were grown in high glucose Dulbecco's modified Eagle's medium containing 10% fetal bovine serum, penicillin, streptomycin, and 2 mm supplemental l-glutamine (Invitrogen). All incubations were at 37 °C with humidified 5% CO₂. HEK293T cells carrying HSP90-TAP constructs used for proteomics analysis were stably selected and maintained using puromycin (Sigma-Aldrich). The ATPlite cell viability assay (PerkinElmer Life Sciences) was used to determine relative growth rates of all clones expressing HSP90 in 96-well plates according to the manufacturer's instructions.

Nucleotide and Geldanamycin Stocks

Stocks of 250 mm ATP were made in 50 mm HEPES, 5 mm MgCl₂ (pH 8.0) and treated with 50 mm phosphocreatine and 50 μg/ml creatine kinase (Roche Applied Science) for 3 h at 20 °C. Stocks of ADP were made in 50 mm HEPES, 5 mm MgCl₂ (pH 8.0) with 100 mm glucose and 20 units/ml hexokinase (Roche Applied Science) and incubated for 3 h at 20 °C. Both stocks were then subjected to ultrafiltration at 4 °C using YM-3 (3,000-Da cutoff) filters (Millipore, Billerica, MA) to remove enzymatic activity. 20 mm geldanamycin stock (NCI, Bethesda, MD) was prepared in 100% DMSO.

Tandem Affinity Purification and Mass Spectrometry

Purifications for all experimental groups, including the vector control, were repeated independently in triplicate, and resulting protein complexes were analyzed by MS consecutively. Affinity purifications were adapted from Angers et al. (28). 7.5 × 10⁸ HEK293T cells were grown to 100% confluence on 15-cm coated plates (catalog number 168381, Nunc, Roskilde, Denmark) without puromycin. Two hours before lysis, the cells were treated with 5 μm geldanamycin in 0.1% DMSO where indicated or left untreated. The cells were gently washed twice with PBS at 4 °C, subsequently detached from the plate by aspiration with PBS, and centrifuged for 2 min at 4 °C in a clinical centrifuge. All further procedures were conducted at 4 °C unless otherwise noted. The supernatant was replaced with 7.5 ml of TAP lysis buffer containing 50 mm HEPES-KOH (pH 8.0), 10% glycerol, 0.1% Nonidet P-40, 150 mm KCL, 5 mm MgCl₂, 2 mm DTT, and an EDTA-free protease inhibitor mixture tablet (Roche Applied Science). Before lysis, buffer was supplemented with either 5 mm ATP including an ATP regeneration system (50 mm phosphocreatine and 50 μg/ml creatine kinase) (29), 5 mm ADP with an ATP depletion system (100 mm glucose and 20 units/ml hexokinase), (30), or 5 μm geldanamycin (GA) with 0.1% DMSO. ATP concentrations were monitored throughout the procedure using the Cell Titer-Glo® ATP assay (Promega, Madison, WI). Lysates were cleared by centrifugation, and supernatants were added to 200 μl of packed streptavidin beads (GE Healthcare) that were prewashed three times and equilibrated in 300 μl of TAP lysis buffer. The remaining steps were conducted essentially as reported (31) with the following exceptions: tobacco etch virus protease cleavage was eliminated as free biotin alone eluted most of the bound HSP90; ATP, ADP, and GA were added to all buffers except the elution buffer. Protein complexes were finally eluted from calmodulin beads three times in 300 μl for a total of 900 μl. Each sample was adjusted to 50 mm ammonium bicarbonate, and the disulfide bonds were reduced with 5 mm dithiothreitol for 1 h. The proteins were digested overnight with modified porcine trypsin (V5111, Promega) at 37 °C. The resulting peptide mixtures were concentrated to 100-μl volumes, desalted using C₁₈ ZipTips (Millipore), evaporated to dryness, and resuspended in distilled H₂O with 0.1% formic acid.

Each digested protein mixture was analyzed by automated microcapillary LC-MS/MS. A 75-μm-inner diameter PicoFrit silica capillary nanospray emitter (New Objective, Woburn, MA) was slurry-packed with 20 cm of 5-μm C18AQ (100-Å) packing material (Michrom Bioresources, Auburn, CA) using a pressure bomb. The column was placed in line with a guard column created from a 100-μm-inner diameter IntegraFrit column (New Objective) packed with 1.5 cm of 5-μm C18AQ (200-Å) packing material. The column setup was connected in line to an Eksigent 2D nanoLC system (Eksigent Technologies, Dublin, CA) and interfaced to an LTQ-FT ion trap mass spectrometer (ThermoFisher Scientific, San Jose, CA).

The mobile phase for the HPLC separation was 0.1% formic acid in water (Buffer A) and 0.1% formic acid in acetonitrile (Buffer B). The LC-MS/MS experiments consisted of a 100-min HPLC run where the HPLC gradient started at 2% Buffer B for 5 min, ramped to 10% Buffer B over 3 min, and then ramped to 40% Buffer B over 60 min followed by washing between samples at 80% Buffer B for 12 min and re-equilibration at 2% Buffer B for 20 min.

The LTQ-FT mass spectrometer was operated in the data-dependent acquisition mode using the Xcalibur 2.0SR2 software. The experiment consisted of a single full-scan mass spectrum in the FT (400–1800 m/z, 100,000 resolution at m/z 400) followed by five data- dependent MS/MS scans in the ion trap at 35% normalized collision energy. The mass spectrometer was set to only perform MS/MS on ions with identifiable charge states of 2+ or 3+. The dynamic exclusion parameters were as follows: repeat count, 1; repeat duration, 30 s; exclusion list, 100 m/z values; and exclusion time, 45 s.

Raw Data Analysis and Database Searching

The MS/MS scans from each LC-MS/MS run were converted from the .RAW file format to .mzXML files using the program ReAdW.exe (v. 1.1) with the default parameters. The spectra were matched to peptides using X!Tandem (v. 2007.07.01.2) (32) configured with a pluggable scoring algorithm (33) compatible with the PeptideProphet^TM inference tools (34) bundled in the Computational Proteomics Analysis System (CPAS) data analysis system (v. 8.2) (35). The following parameters were used in the database search: full trypsin enzyme specificity; missed cleavages allowed, 2; peptide mass tolerance, 2.0 Da; fragment ion tolerance, 0.5 Da; monoisotopic molecular weight for both peptide and fragment ion masses; b/y ion search; and variable modification at Met (+15.995), Gln (−17.027) only if on the peptide N terminus, and Glu (−18.011) only if on the N terminus.

The MS/MS spectra were searched against a custom database created by combining the IPI human database (v. 3.52, released November 2008) with a pseudo-reversed version of each protein sequence for the purpose of estimating the false discovery rates (FDRs). The pseudo-reversed sequences were created by identifying all the tryptic protein cleavage sites and holding all terminal lysine and arginine residues static while reversing all other amino acids present within the internal peptide sequences of the protein. The .mzXML peak lists were filtered by X!Tandem during the database search with the following parameters: maximum charge state, 4+; minimum peptide mass, 600 Da; maximum peptide mass, 4000 Da; and minimum number of allowed fragment peaks, 6. The MS/MS database search results were then analyzed using PeptideProphet and ProteinProphet (v. 3.0, April 1, 2004 (TPP v. 2.9 GALE revision 9, Build 200703221424)) both under default settings for peptide and protein identification scoring, respectively. The final results were then organized in CPAS.

Comparison, Normalization, Protein List Filtering, and Statistical Analysis

Peptide and protein scores were generated using PeptideProphet and ProteinProphet software (30, 32). Spectral counting was used to quantify protein abundances, and the following filtering protocol was used to eliminate proteins from additional analysis. Peptide spectra counts for each protein were calculated by CPAS as the total number of individual spectra (unique or non-unique sequences) matched to a protein using ProteinProphet in which the PeptideProphet scores for individual spectra exceeded a 5% FDR. Therefore, the cutoff scores for accepting individual spectra were different for each of the 33 individual MS runs. The FDR was estimated using a target-decoy strategy developed by Gygi and co-worker (36) by searching all spectra against a combined pseudo-reversed and forward IPI database (for details see above). Proteins having less than a 5-fold average spectral abundance over the vector control were considered to be contaminants and were eliminated. Only proteins with four or more spectra total and at least one peptide in each replicate (prenormalization) were allowed. A protein was considered to be identified if it passed these criteria for at least one experimental group. The total spectra count for all proteins from each run was then determined individually and used to normalize all runs. Then all N-TAP runs and C-TAP runs were normalized separately to the spectra counts for HSP90α (HSP90AA1). The spectral abundance factor (SAF) was then determined by dividing the normalized spectra count for an individual protein by the length of the protein in amino acids (37, 38). This calculation accounts for the fact that longer proteins contain more tryptic sites than shorter proteins and forms a basis for determining relative protein abundance. Finally, the list was then searched for common contaminants such as keratin and cytoskeletal components, and these proteins were eliminated.

Redundancy was then analyzed using ProteinProphet where a peptide is matched either to a single protein or to a group of protein sequences sharing that peptide sequence. A probability score was then assigned by the software to reflect the likelihood that a set of peptides originated from a particular protein in the group. ProteinProphet scores for all protein identifications are shown in supplemental Table S3. All proteins reported in this study exceeded a ProteinProphet score of >0.9 in at least one run. One protein group containing HSP70 isoforms is reported where no distinction between these proteins could be made. These proteins were treated as a single group. In cases of bioinformatic redundancy where duplicate or truncated versions of the same protein were included in a protein group along with the full-length protein, the duplicated sequences were eliminated, and spectra counts were reported for only one full-length protein for purposes of calculating the SAF. Where several splicing variants were identified, results were simplified by reporting the spectra counts and calculating the SAF for the isoform with the longest amino acid sequence. The SAFs for the resulting list of interacting proteins were then subjected to ANOVA to determine whether variances were due to any particular experimental condition, and the threshold was set at p = 0.01. A post hoc pairwise t test determined p values for all experimental conditions among N-TAP and C-TAP constructs separately. All computations were performed using R statistical analysis software.

Transfections, Affinity Precipitations, and Western Blotting

1.3 × 10⁶ HEK293T cells were plated in 60-cm plates 24 h before transfection by calcium phosphate precipitation. The medium was changed 18 h post-transfection, and cells were allowed to recover 24 h before lysis. Drug treatments were as described above. Lysates were prepared in 500 μl of TAP lysis buffer, cleared by centrifugation, incubated with 5 μl of streptavidin beads for 5 h in a 1.5-ml deep well 96-well plate, and transferred to an Orochem OF1100 fritted 96-well plate for washing. The bead solution was centrifuged briefly to remove supernatant, washed four times by centrifugation with 250 μl of TAP lysis buffer containing the appropriate nucleotide or drug, and eluted with 20 μl of hot 1.2× SDS-PAGE loading buffer. Western blotting was performed using anti-HA (catalog number MMS-101R, Covance, Emeryville, CA), horseradish peroxidase-conjugated streptavidin (GE Healthcare), or anti-HSP90α (catalog number V10051, Biomedia). ImageJ was used to collect all densitometry data.

Recombinant Protein Expression, Purification, and in Vitro Binding Assays

C-terminal GST-tagged HSP90 and HA-CHORDC1 were expressed in BL21 codon plus E. coli (Stratagene, La Jolla, CA) by induction with 0.5 mm isopropyl 1-thio-β-d-galactopyranoside for 4 h at 37 °C. Cells were lysed in TAP lysis buffer (see “Tandem Affinity Purification and Mass Spectrometry” above) using pulsed sonication. HSP90-GST supernatants were incubated with GST beads for 4 h, washed extensively, and stored at −20 °C. His-tagged HA-CHORDC1 supernatants were applied to nickel-nitrilotriacetic acid resin, purified using standard techniques, and stored at −20 °C. All proteins were purified to >80% purity as determined by Coomassie staining. 1 μg of HSP90-GST was incubated with 100 ng of His-tagged HA-CHORDC1 for 5 h in the presence of 5 mm ATP or ADP or TAP lysis buffer alone at 4 °C. Washes and blotting were as described above for affinity precipitations.

Real Time RT-PCR

Total RNA was isolated from HEK293T cells expressing N-terminal TAP-tagged HSP90 either treated for 2 h with 5 μm geldanamycin or DMSO alone. 900 ng of total RNA was subjected to RT using an Affinity Script cDNA synthesis kit (Stratagene) according to the manufacturer's instructions using oligo(dT) primers. An amount equivalent to 20 ng of total RNA was used in real time PCRs using Brilliant II Fast SYBR® Green (Stratagene). All conditions were according to the manufacturer's instructions. PCRs commenced in an ABI 7900HT thermocycler. Primer sequences are available in supplemental Table S9. All reactions were analyzed for multiple products using melting curve analysis. Data were analyzed by REST software (Corbett Lifescience) using β-actin as a reference gene.

Identification of Novel HSP90 Interactions and Those Influenced by Nucleotides and Pharmacological Inhibition

N- and C-terminal TAP-tagged human HSP90α cDNAs were stably expressed in the human kidney epithelial cell line HEK293T, and TAP was performed on HSP90 complexes in the presence of 5 mm ATP, 5 mm ADP, or 5 μm GA. In addition to the wild type HSP90α, we chose to use the E47A mutant that is unable to hydrolyze ATP yet able to bind the nucleotide with affinities comparable to the wild type protein (15, 17, 39). This mutant has been shown to strengthen interactions between HSP90 and substrate peptides in vitro and is predicted to function as a kinetic trap to detect interactions with client proteins and co-chaperones in vivo. Indeed, the analogous mutation in yeast was previously shown to strengthen the interaction between the yeast p23 homolog Sba1p and the yeast HSP90, an interaction known to be positively influenced by ATP (40). Therefore, HSP90 E47A is predicted to serve as a suitable probe in this system to identify HSP90 interactions that are stabilized by excess ATP. TAPs were performed in three independent replicate experiments containing the following four experimental groups for both N- and C-terminal constructs: 1) wild type HSP90 with ATP (NW ATP and CW ATP), 2) ADP (NW ADP and CW ADP), 3) geldanamycin (NW GA and CW GA), and 4) HSP90 E47A mutant with ATP (NE ATP and CE ATP). An empty vector control with ATP (VO ATP) was included to control for nonspecific binding. Relative quantities of the TAP-tagged proteins were determined by Western blot, and N- and C-terminal constructs were shown to have comparable expression levels (supplemental Fig. S1). All stably expressed constructs were tested for altered growth rates using an ATP assay, and no differences between clones were observed, suggesting that the essential functions of endogenous HSP90 complexes are largely undisturbed by exogenous expression of TAP-tagged HSP90 mutants (supplemental Fig. S2). An ATP regeneration or depletion system was used for all purifications containing ATP or ADP, respectively (see “Experimental Procedures”). For the WT GA group, cells were pretreated with drug for 2 h before harvesting. Purified complexes were then consecutively subjected to LC-MS/MS, and the resulting raw spectra were analyzed using spectral counting and filtered. Spectra counts for all proteins were converted to SAFs to gain a relative measure of the protein abundance among interacting proteins (37, 41). The SAFs were compared across experimental groups, and the average SAFs for identified proteins are presented in Fig. 1. All N-terminal constructs and treatments were compared independently from C-terminal constructs to account for the fact that N-terminal constructs captured about 5-fold more HSP90 than C-terminal constructs. A complete record of raw spectra counts, error analysis, protein and peptide certainty values, and other statistics for each identified protein are found in supplemental Table S3. Spectra and PeptideProphet (certainty) scores for all single peptide identifications are shown in supplemental Fig. S4 and supplemental Table S5. For presentation purposes, all IPI accession numbers for the final table of interacting proteins were converted to the corresponding NCBI protein name.

Identification of Known and Novel HSP90 Interactions

We identified 37 known HSP90-interacting proteins and 24 putative novel interactions (Table I). Of the known interactors, we identified 20 co-chaperones, including seven containing the TPR domain, an established HSP90 interaction element. We also identified a group containing eight HSP70 family members. We were unable to clearly differentiate between HSP70 family members using MS, so these proteins were treated as a single group, and HSPA8 was chosen to represent this group because of its high spectral abundance. All HSP70 family members identified bound HSP90 in a similar manner with respect to ligands (see Table I and supplemental Fig. S6 for data on all HSP70s). Furthermore, we identified seven CHORD-Sgt1 (CS) domain-containing proteins believed to function as co-chaperones, including SUGT1, PTEGS3 (p23), NUDC, CACYBP (SIP), CHORDC1, NUDCD3, and USP19 (23, 42–44). USP19 and NUDCD3 are predicted to form interactions with HSP90 based upon homology to p23 and SUGT1 (45); however, to our knowledge these two interactions have not been previously reported. These findings provide further support for the CS domain as a bone fide HSP90 interaction domain. The C-terminal TAP-tagged HSP90 constructs identified considerably fewer proteins with lower spectra counts than the N-terminal TAP constructs, and none of the identifications were unique to the C-terminal constructs. This is most likely because of a relatively lower recovery of C-terminal tagged HSP90 in our purifications as both the N- and C-terminal constructs are expressed at comparable levels (see supplemental Fig. S1). Overall, the identification of many known HSP90-interacting proteins in this data set supports the possibility that we have also identified novel HSP90 interactions.

Our data set also contained five known client proteins. We identified two kinases forming novel HSP90 interactions (CSNK1A1 and MYLK2). The presence of the known HSP90 client kinases CHUK and CDC2L2 (CDK11) taken together with the identification of the kinase-specific co-chaperone CDC37 in our purifications further strengthens the possibility that these kinases may be HSP90 client proteins. However, it must be noted that several known kinase client proteins are absent from our data. The detection of particular client proteins may be dependent upon their expression levels, and some HSP90 clients may be conditionally expressed, cell cycle-dependent, or not expressed in the renal epithelium from which the HEK293T cells were derived.

Identification of Ligand-dependent HSP90 Interactions

It is well established that some interactions with co-chaperones and client proteins are dependent upon the nucleotide- or drug-bound state of HSP90. We sought to identify new ligand-dependent interactions and to observe effects of ligands on the dynamics of the HSP90 interactome. We searched our data set to identify proteins with spectral abundances that were significantly altered in purifications containing one HSP90 ligand over the others. We subjected the SAFs for each protein across experimental groups to ANOVA and identified 52 proteins (including HSP70 homologs) that were below the ANOVA p value cutoff of 0.01 (Table I). These proteins represent 84% of the total identifications in this data set, suggesting that HSP90 complexes are sensitive to the ligand present in the purification reactions. We also sought to identify interactions that are specifically formed in the presence of a particular ligand. To this end, we used a post hoc pairwise t test to identify 36 proteins that were significantly overrepresented in one group over the others. Proteins were considered to be increased in abundance by a specific ligand if p values in a post hoc t test were less than 0.05 for all pairwise tests for that ligand (Table I and supplemental Table S3).

ATP-dependent Interactions

We hypothesized that the use of the ATP hydrolysis-deficient E47A mutant would increase the strength of interaction between HSP90 and ATP-dependent interacting proteins as has been observed previously with the homologous yeast HSP90 (HSC82p) E33A mutant (40). Under this assumption, we expected that proteins identified in the E47A ATP group would also be represented in the WT ATP group at a lower abundance because of the ability of wild type HSP90 to hydrolyze ATP. Consistent with this concept, we found that 26 proteins identified in the E47A ATP complexes were also present in the WT ATP complexes, eight of which were identified at lower abundances in the WT ATP complexes (Fig. 1; FKBP5, SUGT1, AHSA1, NUDCD3, PTGES3 (p23), CSNK1A1, UNC45A, and FAM83H; t test p value <0.02). These results are consistent with previous findings showing p23 and AHSA1 to have an increased binding affinity for HSP90 in its ATP-bound state (13, 40) and suggest that other interactions in this category may also be positively regulated by ATP.

Geldanamycin-dependent Interactions

Most evidence supports the hypothesis that pharmacological inhibition of HSP90 disrupts its interactions with client proteins and co-chaperones such as p23. Indeed, in several cases, we show that known and novel HSP90 interacting proteins identified in reactions containing ATP are either entirely excluded from HSP90 complexes treated with geldanamycin or are severely reduced in abundance (Fig. 1). However, we also identified 15 HSP90 interactions that are specifically induced after treatment with geldanamycin (t test p value <0.05). Interestingly, these complexes are enriched for proteins known to function in basic processes of RNA metabolism including POLR2E (RNA polymerase II subunit), C19orf2 (a transcriptional co-repressor known to associate with POLR2E (46, 47)), ASCC3L1 (RNA helicase), RPAP3 and PIH1D1 (ribonucleoprotein assembly factors) (48), and YTHDC2 (putative RNA helicase/splicing factor).

It is well established that HSP70 expression is dramatically up-regulated after treatment with HSP90 inhibitors. As several proteins indentified in geldanamycin-dependent complexes are known to be involved in stress responses, we considered the possibility that these interactors may be likewise up-regulated. We were interested in whether the observed overabundance in complexes treated with geldanamycin may be a result of increased total cellular protein rather than a direct effect of geldanamycin acting on HSP90. We tested this possibility by determining whether a subset of geldanamycin-dependent interactors were up-regulated using real time RT-PCR. Of the eight proteins tested, only HSP70 homologs HSPA1A and HSPA8 were up-regulated after geldanamycin treatment (supplemental Fig. S7). It is therefore unlikely that the observed increase of these proteins in HSP90 complexes is due to increased cellular abundance. Furthermore, if the binding interaction between HSP90 and HSP70 was only influenced by cellular abundance of these proteins, then we would expect that any increase in HSP70 mRNA expression after GA treatment would be directly reflected in the peptide spectra count. We found that the mRNA abundances for the HSP70 homologs HSPA8 and HSPA1A are 2- and 18-fold up-regulated, respectively, after GA treatment, whereas the spectral abundances for the corresponding proteins were not found to be statistically different (see supplemental Fig. S8; t test p = 0.31).

As a control, we went further to test the inactive HSP90 ATPase mutant D93N to determine whether the presence of geldanamycin-dependent interactors in HSP90 complexes depended upon an active HSP90 ATPase or the ability of HSP90 to bind nucleotides or geldanamycin (49). The D93N mutation prevents nucleotide binding in the HSP90 ATPase active site and is also predicted to perturb geldanamycin binding based upon the observation that D93N makes contacts with both GA and adenine nucleotides through a similar mechanism (50). We also tested a mutant lacking the C-terminal MEEVD motif (ΔMEEVD), which is known to bind HSP90 interactors containing TPR motifs (51–53). We reasoned that interactions with geldanamycin-dependent complex members that contained TPR motifs, such as RPAP3, would be significantly impaired. We stably expressed N-terminally TAP-tagged HSP90 mutants in HEK293T cells, repeated the purifications in the presence of geldanamycin, and compared the results with our original data set. We found that the abundance of the geldanamycin-dependent HSP90 interactors identified in the WT GA group was severely decreased or absent in all cases. Fig. 2 shows the average SAF for all proteins identified in geldanamycin-dependent complexes that were also identified with the D93N or ΔMEEVD mutants. One notable exception was the CS domain-containing protein USP19, which was increased slightly in the ΔMEEVD purifications (t test p = 0.0335). However, USP19 did not bind the D93N mutant, which is in accordance with the prediction that this CS domain-containing co-chaperone is predicted to bind the N-terminal ATPase domain (45). These results suggest that the effects of geldanamycin on HSP90 complexes are intrinsic to HSP90 and are dependent upon a full ATPase activity and/or the C-terminal MEEVD motif.

CHORDC1 Forms an ADP-dependent Interaction with HSP90

Using the data from the N-terminal TAP-tagged HSP90, we identified c10orf137 (EDRF1) and the CS domain-containing protein CHORDC1 as ADP-regulated HSP90 interactors (t test p < 0.0116 and 5.0 × 10⁻⁸, respectively). We were intrigued that ADP stimulated the interaction between HSP90 and CHORDC1 when other CS domain-containing family members in our data set appear to bind the ATP-bound HSP90 (PTGES3 (p23), SUGT1, CACYBP, NUDC, and NUDCD3). It is well established that the binding of ATP and ADP to the N-terminal ATPase domain of HSP90 induces conformational changes that influence the client protein and co-chaperone constituents of HSP90 complexes (7, 18–20). The CHORD domain in CHORDC1, like other CS domain-containing proteins, has been shown to interact with the N-terminal ATPase domain of HSP90; however, the role of nucleotide binding in the interaction between HSP90 and CHORDC1 has not been clearly demonstrated (42, 45, 55). It is important to note that slightly more CHORDC1 peptides were identified in the MS/MS analysis of wild type N-TAP complexes versus the E74A mutant (Fig. 1a, compare NW ATP with NE ATP); however, this apparent increase was not found to be significant (t test p = 0.055). Furthermore, we were unable to identify CHORDC1 as an ADP-dependent interactor in our MS/MS experiments using the C-terminal TAP-tagged HSP90. Only a single peptide was identified in each of the CE ATP replicates, and no CHORDC1 peptides were observed in the CW ADP group. We attribute this to the generalized capture inefficiency observed for the C-terminal TAP-tagged constructs. Despite this discrepancy, we were interested to verify and further investigate the basis for this observed ADP-dependent interaction.

To verify any effect of the position of the TAP tag and more clearly understand the effects of nucleotides on this interaction, we used a C-terminal TAP-tagged HSP90 and an N-terminal HA-tagged CHORDC1 (HA-CHORDC1) for co-affinity precipitation experiments in the presence of varying ratios of nucleotides. We tested whether the ADP:ATP ratio, and not solely the lack of ATP, was governing the CHORDC1-HSP90 interaction. Co-expressing HEK293T cells were harvested directly into TAP lysis buffer containing either increasing amounts of ADP or decreasing ratios of ADP:ATP by holding the concentration of ADP constant. TAP-tagged HSP90 was then affinity-precipitated with streptavidin beads and washed with lysis buffer containing the appropriate nucleotide ratios, and the bound proteins were eluted and Western blotted. Empty vector constructs (VO) were used to control for nonspecific binding. Fig. 3a clearly shows that an ADP-dependent interaction with CHORDC1 persists when the C-terminally TAP-tagged HSP90 is used, suggesting that the position of the tag has no effect on the interaction. The 37-kDa HA-CHORDC1 band became visible when ADP alone was added at a concentration as low as 1 μm and reached a maximal intensity at 1 mM (Fig. 3a, upper panel, lanes 3–7). Importantly, no interaction was detected where ADP was not added, suggesting that this interaction was not occurring because the lysates were lacking ATP (Fig. 3a, lanes 2 and 13). ATP was then titrated into the lysis buffer while holding the ADP concentration constant at the subsaturating concentration of 100 μm. The addition of ATP began to disrupt the interaction at 1 μm. At 100 μm ATP (1:1 ADP:ATP ratio), the interaction was completely abolished, demonstrating an inhibitory effect of ATP on the stability of this complex (Fig. 3a, lanes 8–13). Densitometry was then performed to quantify the Western blot signal (Fig. 3a, bottom). To further verify that the position of the TAP tag had no bearing on the interaction, affinity precipitations of HA-CHORDC1 were performed with N-terminal HSP90 TAP-tagged constructs in the presence of 5 mm ATP or 5 mm ADP. In all cases, the 37-kDa HA-CHORDC1 band was clearly visible only in reactions containing ADP (data not shown). Despite the fact that we did not identify CHORDC1 as an ADP-dependent interactor in the MS/MS analysis of affinity precipitations from cells stably expressing the C-terminal TAP-tagged HSP90 (Fig. 1), we clearly demonstrated this interaction in these experiments using transient transfections and Western blotting.

Role of Nucleotide Binding in the ADP-dependent HSP90-CHORDC1 Interaction

Because our results show that high ADP:ATP ratios are stimulating the HSP90-CHORDC1 interaction, we wanted to determine which structural elements of HSP90 are specifically responsible for this activity. We tested 1) whether the HSP90 N-terminus where the ATPase domain resides is sufficient to form the ADP-dependent CHORDC1-HSP90 interaction, 2) whether the nucleotide binding activity of HSP90 is necessary for this interaction, and 3) whether the ATPase activity of HSP90 is essential for the stimulatory activity of ADP. To this end, we co-expressed and affinity-precipitated wild type HA-CHORDC1 with a panel of C-terminal TAP-tagged HSP90 mutants consisting of the isolated N-terminal ATPase domain (HSP90-N), a mutant lacking both nucleotide binding and hydrolysis activities (HSP90 D93N) (49), or a point mutant that lacks ATP hydrolysis activity but retains wild type nucleotide binding activity (HSP90 E47A) (15, 17, 39) (Fig. 3b). An attempt was made to construct an HSP90 mutant with the ATPase domain deleted; however, this construct was not tested because of an undetectable level of expression. Affinity precipitation reactions were performed in lysis buffer containing either 1 mm ATP or 1 mm ADP.

The HSP90 E47A mutant precipitated more CHORDC1 than the full-length wild type protein in the reactions containing either ADP or ATP (Fig. 3c, compare lanes 1 and 2 with lanes 3 and 4); however, the interaction continued to be specifically stimulated by ADP as was shown for the wild type HSP90. Notably, the HSP90 D93N mutant failed to interact with CHORDC1 in either the ATP or ADP reaction despite robust protein expression. This suggests that CHORDC1 is unable to bind the apoenzyme form of HSP90 and that nucleotide binding is necessary for this interaction. The isolated ATPase domain of HSP90 (HSP90-N) interacted with CHORDC1 in both ATP and ADP reactions, yet the interaction remained positively influenced by ADP. Compared with full-length wild type HSP90, however, the ATPase domain alone appeared to interact with less specificity for ADP as evidenced by the CHORDC1 band appearing in the ATP-containing reaction (Fig. 3c, compare lanes 1 and 2 with lanes 7 and 8). This suggests that elements outside the ATPase domain may reinforce the selectivity for this ADP-dependent interaction.

The CHORDC1 protein is composed of two tandem zinc-binding CHORD domains separated by a 90-amino acid linker region and a C-terminal CS domain homologous to the HSP90 binding domain of p23 and SUGT1 (54). Previous studies have shown subsections of the CHORD domain to be essential for CHORDC1 binding to HSP90 in yeast two-hybrid assays (42, 55). Specifically, the N-terminal CHORD domain (CHORDI) was shown to be dispensable for HSP90 binding; however, the linker region and the C-terminal CHORD domain (CHORDII) were shown to be necessary for the full binding interaction. We wanted to determine which elements of CHORDC1 were involved in producing this ADP-dependent interaction with HSP90 and whether deletion of a particular domain would destroy selectivity for ADP. We created N-terminal HA-tagged truncation mutants of CHORDC1 either by deleting the first CHORD domain (ΔCHORDI), deleting CHORDI along with the intervening linker region (ΔCHORDI-Linker), or deleting the C-terminal CS domain (ΔCS) (Fig. 3b). We also attempted to test the binding of a CHORDC1 mutant consisting of only the C-terminal CS domain, but this construct failed to be expressed (data not shown). As before, we co-expressed these constructs with C-terminally TAP-tagged wild type full-length HSP90 in HEK293T cells and performed affinity precipitations and Western blots on lysates containing either 1 mm ATP or 1 mm ADP (Fig. 3d).

Deletion of the CHORDI domain had no effect on either HSP90 binding or nucleotide selectivity. The amount of ΔCHORDI precipitating with HSP90 was equivalent to the full-length protein, suggesting that the CHORDI domain is dispensable for both HSP90 binding and ADP selectivity. These findings are in agreement with previous two-hybrid studies showing CHORDI to be dispensable for this interaction (42). Interestingly, deletion of the linker region in addition to the CHORDI domain caused the loss of selectivity for ADP. Despite lower protein expression, both ATP and ADP reactions precipitated equivalent amounts of CHORDC1 (Fig. 3d, lanes 5 and 6). This raises the possibility that the linker region may play a role in conferring nucleotide selectivity to the CHORDC1-HSP90 interaction. Deletion of the CS domain, however, completely abolished the interaction regardless of nucleotide composition of the reaction (Fig. 3d, lanes 7 and 8). This result is also in agreement with previous yeast two-hybrid data showing the CS domain to be necessary for HSP90 interactions (42).

HSP90 and CHORDC1 Interact in an ADP-dependent Manner in Vitro

We went further to confirm whether this ADP-dependent interaction was direct and did not require the presence or activity of additional proteins by performing in vitro binding assays using purified recombinant HSP90 and CHORDC1. cDNAs for HA-CHORDC1 and HSP90 (WT, D93N, and E47A) were N-terminally tagged with polyhistidine and GST, respectively; expressed in E. coli; and purified. HSP90-GST remained immobilized on glutathione beads throughout the purification procedure, and the beads were then used to affinity precipitate HSP90-CHORDC1 complexes in the presence of ATP and ADP (Fig. 4). As in the co-expression experiments presented above, recombinant CHORDC1 binding to WT and E47A HSP90 was greatly increased only in the presence of 1 mm ADP, whereas no binding was detected with HSP90 D93N regardless of the nucleotide present in the reaction.

HSP90 Interactome Is Dynamic with Respect to Nucleotides and Geldanamycin

Several recent studies presented profiles of the human HSP90 interactome using immunopurifications and affinity capture with immobilized HSP90 (23, 27). These studies effectively identified several new and novel HSP90 interactors; however, these experiments were not designed to identify interactions that are sensitive to HSP90 ligands. In addition to presenting evidence for 24 novel HSP90-interacting proteins, we have gone further to identify 52 proteins (84% of total identifications) that interacted with HSP90 in a ligand-dependent manner (ANOVA p < 0.01), 43 of which (HSP70 homologs included) were significantly overrepresented in one ligand group over the others, presenting evidence that these interactions hold specificity for a particular ligand-bound form of HSP90 (Table I). These findings suggest that the HSP90 interactome is strongly influenced by nucleotide and drug ligands. Of these regulated interactions, approximately half were overrepresented in complexes containing the HSP90α E47A mutant. Because this mutant binds but does not hydrolyze ATP, we predict that proteins identified with this mutant are either completely ATP-dependent or at least in some cases positively influenced by ATP binding. To date, our TAP method utilizing the E47A mutant has yielded the largest number of identifications of known HSP90 interacting co-chaperones. Most notably, we have identified seven members of the CS domain-containing co-chaperone family, including two novel interactions (USP19 and NUDCD3). This is the most complete representation of this family of HSP90 interactors presented thus far and demonstrates the ability of this technique to identify known and novel HSP90 interactions. Despite the presence of 5 mm ATP and an ATP regeneration system, the WT ATP purifications appeared to produce significantly fewer interactions than the E47A mutant. This was not entirely unexpected as it has been previously shown that interactions with the wild type HSP90 are relatively transient (17). Interestingly, we identified PLCE1 (phospholipase C type E1) as a protein that binds abundantly to WT HSP90 regardless of the ligand present in the purifications but that was not detectible in reactions containing the E47A mutant. We cannot eliminate the possibility that clonal effects may have some influence or that structural effects of the E47A mutation somehow exclude this interaction; however, the significant number of known HSP90 interactors identified with the E47A mutant suggests that clonal effects are minimal.

Geldanamycin-dependent HSP90 Complexes Are Enriched for Core Transcription Machinery

The majority of interaction studies focused on HSP90 client proteins show that pharmacological inhibition of ATPase activity leads to disruption of HSP90 complexes. Indeed, we show that most interactions that were shown to be ATP-dependent are in fact disrupted by geldanamycin treatment. Interestingly, we identified several HSP90 interactions that appear to be reinforced by geldanamycin treatment. Several proteins in these geldanamycin-dependent complexes, including HSP70, RPAP3, c19orf2, PIH1D1, and EFTUD2, were also reported in a proteomics study as interacting with core transcription machinery (57). The concordance between these two studies suggests the interesting possibility that HSP90 is involved in regulating transcription at several levels and that the activity of geldanamycin may alter fundamental transcriptional processes.

The HSP90 interactor RPAP3 makes five interactions with components of core transcription machinery identified in geldanamycin-dependent complexes (c19orf2, POLR2E, EFTUD2, HSP70, and PIH1D1), suggesting that this protein may serve as a bridge for an HSP90 subcomplex. RPAP3 contains two TPR domains that have been shown in other proteins to bind the C-terminal MEEVD motif on HSP90 and HSP70. It is possible that this interaction is also mediated by the MEEVD motif on the C terminus of HSP90 because RPAP3 was absent in our MS/MS analysis of HSP90 ΔMEEVD mutant complexes (supplemental Table S3). In support of this notion, RPAP3 (hSpagh) was first identified as an HSP90-interacting protein using the isolated C-terminal domain, suggesting that this protein binds the HSP90 C terminus (23), and was recently shown to bind HSP90 directly and function in a small nucleolar ribonucleoprotein assembly complex containing HSP90, PIH1D1, RUVBL1, and RUVBL2 (48). Notably, RUVBL1 and RUVBL2 were also identified as components of geldanamycin-dependent complexes in our experiments; however, these proteins were present in our empty vector control (VO ATP) purifications and were consequently filtered (data not shown). In this case, it is unlikely that RPAP3 and PIH1D1 are contaminants being carried nonspecifically because both of these proteins are entirely absent in all groups except for the vector control and those treated with geldanamycin, and the quantity of the RUVBL1/2 proteins does not increase concordantly with PIH1D1 and RPAP3 in the geldanamycin-dependent complexes.

HSP70 (HSPA1A) is one of the most abundant proteins among the geldanamycin-dependent complexes and is known to interact with four other proteins identified in these complexes including STIP1 (HOP). It is well established that HSP70 homologs interact with HSP90 through binding to STIP1, and this suggests that STIP1 may act as an adaptor protein in our experiments (58, 59). However, this interaction was not stimulated by geldanamycin treatment in our experiments, suggesting that the interaction between HSP70 and HSP90 may either be independent of HOP or that geldanamycin treatment may fortify the interaction between an existing HSP90-HOP complex and HSP70.

CHORDC1 Interaction Is Stimulated by ADP

We have identified and characterized CHORDC1 as an ADP-dependent HSP90-interacting protein. We further show that high ADP:ATP ratios stimulate binding and that this selectivity can be traced to elements within the linker region residing between the two tandem CHORD domains. Certain lines of evidence suggest that this interaction is direct. First, we were unable to detect other proteins with increases in SAF values proportional to CHORDC1 for purifications containing ADP in our MS experiments that would indicate a cofactor or adaptor protein (Fig. 1 and supplemental Table S3). Second, the heterologously expressed purified recombinant proteins interact in vitro in an ADP-dependent manner (Fig. 4). These results taken together with the fact that the CS domain has been shown in several proteins to be a direct HSP90 binding element (3, 23, 42–44) lead us to propose the possibility that this interaction is direct.

Our results are generally in agreement with those presented by Wu et al. (42) showing that the interaction between HSP90 and CHORDC1 (Chp1) is independent of ATP. Indeed, we did observe a low affinity interaction in the presence of ATP (Fig. 3d, lane 1). We believe this to be the low affinity, nucleotide-independent interaction reported previously. Importantly, this group did not report experiments testing the effects of ADP on this complex, so to our knowledge, our results are the first to demonstrate the effects of ADP on this interaction.

Melusin (ITGB1BP2), a homolog of CHORDC1, is the only other CHORD domain-containing protein in the human proteome and has been shown to interact with HSP90 (56). We predicted that this protein, like CHORDC1, would bind HSP90 in an ADP-dependent manner. In co-expression experiments and in vitro with purified proteins, we were indeed able to detect an ADP-dependent HSP90 interaction with melusin; however, this was only observed with the HSP90 E47A mutant, and no significant increase in binding was observed with the wild type HSP90 and ADP (data not shown). Melusin and CHORDC1 show homology within the CHORD domains; however, they are only 42% identical in the region containing the 90-amino acid linker and the CHORDII domain (BlastP, NCBI) that we have shown to modulate the ADP-dependent interaction between HSP90 and CHORDC1. It is therefore possible that this sequence diversity could lead to altered HSP90 binding activity. Furthermore, the relative binding affinities of ADP for wild type HSP90 and the E47A mutant have not been determined. It is therefore possible that this mutant binds ADP more tightly than the wild type protein, thus contributing to the increased effect of ADP in the melusin-HSP90 interaction. In support of this notion, we have also observed increased CHORDC1 binding to the E47A mutant relative to WT HSP90 in the presence of ADP in co-expression experiments (Fig. 3c, compare lanes 1 and 2 with lanes 3 and 4), which suggests that the E47A mutation positively affects the ADP-dependent interaction with CHORDC1 and that this phenomenon may therefore be extended to melusin. However, the E47A mutant bound similar amounts of HA-CHORDC1 in the presence of ADP in the in vitro binding assays (Fig. 4). This may be due to a slightly higher expression of the E47A mutant in the co-expression experiments that was not detected in our Western blots (see Fig. 3).

The observation that CHORDC1 is stimulated to bind HSP90 when the ATP:ADP ratio falls below 1 raises the possibility that the HSP90-CHORDC1 interaction may be involved in the response to altered energy balance in vivo. The ADP:ATP ratio has been shown to exceed 1 under conditions of oxidative stress combined with a glycolytic blockade (60). Furthermore, melusin has been shown to be required for recovery from dilated cardiomyopathy, which is marked by acute depletion of ATP and an increase in the ADP:ATP ratio (56, 61–63). This suggests a possible role for the CHORD domain in response to metabolic stress. It remains to be tested whether the HSP90-CHORDC1 interaction can be observed in living cells under conditions that permit ATP depletion. These experiments may further develop our understanding of the function of the CHORD domain in stress responses.