Glucocorticoid receptor complexes form cooperatively with the Hsp90 co-chaperones Pp5 and FKBPs

The function of steroid receptors in the cell depends on the chaperone machinery of Hsp90, as Hsp90 primes steroid receptors for hormone binding and transcriptional activation. Several conserved proteins are known to additionally participate in receptor chaperone assemblies, but the regulation of the process is not understood in detail. Also, it is unknown to what extent the contribution of these cofactors is conserved in other eukaryotes. We here examine the reconstituted C. elegans and human chaperone assemblies. We find that the nematode phosphatase PPH-5 and the prolyl isomerase FKB-6 facilitate the formation of glucocorticoid receptor (GR) complexes with Hsp90. Within these complexes, Hsp90 can perform its closing reaction more efficiently. By combining chemical crosslinking and mass spectrometry, we define contact sites within these assemblies. Compared to the nematode Hsp90 system, the human system shows less cooperative client interaction and a stricter requirement for the co-chaperone p23 to complete the closing reaction of GR·Hsp90·Pp5/Fkbp51/Fkbp52 complexes. In both systems, hormone binding to GR is accelerated by Hsp90 alone and in the presence of its cofactors. Our results show that cooperative complex formation and hormone binding patterns are, in many aspects, conserved between the nematode and human systems.

HSP-90 co-chaperones differentially influence GRLBDm complexes. Having observed that PPH-5 forms a complex with HSP-90 and GRLBDm and supports the nucleotide-induced rearrangement of the chaperone, we aimed at testing whether other cofactors of the nematode HSP-90 system are influencing the assembly of the GRLBDm·HSP-90 complex. To this end, we purified the cofactors STI-1, FKB-6 and AHSA-1, all of which had been shown to interact with HSP-90 in the absence of client proteins [45][46][47] and to modulate GR activity in the vertebrate or yeast systems 10,11,48 . Addition of STI-1 strongly reduces binding of the GRLBDm to the HSP-90 protein, while AHSA-1 binds in addition, but does not affect the affinity for GRLBDm (Fig. 1B). Interestingly, the addition of FKB-6 leads to a marked increase in affinity and to formation of HSP-90·GRLBDm·FKB-6 ternary complex (Fig. 1B). Thus, the large PPIase FKB-6, like the phosphatase PPH-5, is apparently capable of strengthening the interaction between GRLBDm and the HSP-90 dimer.
We then tested whether these cofactors' interactions are also sensitive to the nucleotide bound state of the HSP-90 protein, by adding ATPγS to the cofactor containing GRLBDm·HSP-90 complexes (Fig. 1C). This leads to a further diminishing of the complex peak in the case of AHSA-1 and continued depletion in the case of STI-1. In contrast, the FKB-6 containing complex is still intensely formed and its sedimentation coefficient is increased from 6.9 to 7.3 S. This implies that, in the presence of FKB-6, as in the presence of PPH-5, nematode HSP-90 is able to perform its nucleotide-induced rearrangements.
Cep23/DAF-41 binds to closed GRLBDm·HSP-90·FKB-6/PPH-5 complexes with no influence on the affinity for the client. Given that PPH-5 and FKB-6 support GR complex formation, we aimed at testing how the cofactor Cep23/DAF-41 influences the affinity of the Hsp90 machinery for GRLBDm. The interaction of the Hsp90 machinery with p23 critically depends on nucleotide binding to Hsp90 and the closed conformation of the chaperone 34,[49][50][51][52] . This interaction reflects a late step during the maturation cycle of GR, at the point where the active hormone binding site is formed 33 .
AUC experiments were performed to evaluate the effect of DAF-41 in the presence of ATPγS. The ternary GRLBDm HSP-90·DAF-41 complex displayed a reduced affinity for the client compared to HSP-90 alone, as judged from the free GRLBDm fraction ( Fig. 2A,B). Binding of GRLBDm to HSP-90·PPH-5 decreased in the presence of ATPγS. The addition of Cep23/DAF-41 did not change this affinity, as almost the same amount of GRLBDm molecules were retained in the chaperone complex. Cep23/DAF-41 binding to the complex can Scientific RepoRtS | (2020) 10:10733 | https://doi.org/10.1038/s41598-020-67645-8 www.nature.com/scientificreports/ nevertheless be seen from the significant increase in the complex's sedimentation coefficient towards 8.3 S ( Fig. 2A). This was also tested for the PPIase FKB-6 ( Fig. 2B). Here, as well, DAF-41 bound in addition, while maintaining the affinity for GRLBDm. It therefore seems that Cep23/DAF-41 has a similar effect on both TPRcofactor complexes, not influencing the affinity for the client further than the TPR cofactor itself, while interacting with the closed HSP-90 conformation.
insight into ternary complex topology. To determine a representative topology of the ternary GRLBDm·HSP-90 complexes with PPH-5 and FKB-6, we searched for binding interfaces by chemical crosslinking and mass spectrometry. The protein complexes were treated with the isotopically labelled crosslinker H 12 / D 12 -BS 3 and analysed by SDS-PAGE. The complex-representing bands (Fig. 3A) were subjected to tryptic digestion, followed by high resolution mass spectrometry. Our analysis, carried out independently by the in-house software xMASS and pLink, aimed at detecting inter-protein crosslinked peptides, reporting on potential contact sites of the proteins 41,53 . Our data suggests that GRLBDm is positioned in the middle (M) domain of HSP-90, as we identified peptides of GRLBDm crosslinked to the lysines 326, 329 of HSP-90 (Tables 1, 2). Additional contacts were detected, linking GRLBDm to the C-terminal dimerization domain (CTD) at positions 551,555 and 601. These crosslinking sites were identified in both the PPH-5 (Table 1) and FKB-6 complexes ( Table 2).
PPH-5 contains a TPR domain located at its N-terminus and a C-terminal αJ subdomain 13,54 . Based on the identified crosslinking sites from the ternary GRLBDm·HSP-90·PPH-5 complex, the phosphatase appears bound to HSP-90 in a similar arrangement, as previously reported for the binary HSP-90·PPH-5 assembly 41 .   Table 1). Importantly, crosslinked products between PPH-5 and GRLBDm can also be found, hinting at an interaction on the chaperone scaffold that possibly sets the basis for the observed cooperativity during complex formation. Lysines 576 and 579 of GRLBDm are linked to lysine 319 that resides in the catalytic domain of the phosphatase. Performing a similar study with the cofactor FKB-6, crosslinking sites can be identified that help position the cofactor relative to GRLBDm and HSP-90. FKB-6 contains two peptidyl prolyl isomerase (FKBP) domains (FK1, FK2) and a C-terminal tetratricopeptide repeat (TPR) region (Fig. 3D). We identified contacts between the chaperone and FK1 and TPR domains of FKB-6, as summarized in Table 2. The identified crosslinked product that pairs lysine 11 of FKB-6 and lysine 402 of HSP-90, indicates a contact between the FK1 domain of FKB-6 and HSP-90's M domain (Fig. 3C). It has been defined that the TPR cofactors interact with the Hsp90 machinery via its C terminal MEEVD motif which, lacking lysine residues, cannot be observed as a crosslinking product 55,56 . We then aimed at confirming the relevance of the FKBP domains to complex formation based on biochemical experiments. To this end we constructed FKB-6 variants that contain the TPR and either the FK1 (FK1-FKB-6) or the FK2 domain (FK2-FKB-6). Sedimentation velocity AUC experiments with FKB-6 truncated constructs revealed that deletion of the FK1 domain maintains robust binding to GRLBDm·HSP-90 whereas, upon removal   Rates of hormone-binding to GR-LBD are modulated by the Hsp90-state. Although Hsp90 is required for hormone binding in vivo, it is known that GRLBD is capable of ligand binding in vitro in the absence of Hsp90 35,38 . It is unclear how the Hsp90 system primes GR for ligand binding, while cofactor-induced effects have been observed 57 . We therefore tested to what extent the observed complexes are competent in hormone binding. A fluorescein-labelled variant of dexamethasone (F-DEX) was used to monitor the kinetic of hormone binding to GRLBDm by fluorescence polarization. We recorded the binding kinetics first in the absence and presence of HSP-90 and then supplemented with ATP, ATPγS, ATPγS/DAF-41. In these reactions ATP presence is crucial to accelerate hormone binding, as any other state of HSP-90 retains binding close to control levels (Fig. 4A). The combination of HSP-90, Cep23/DAF-41 Table 1. Inter-protein crosslinked peptides detected for GRLBDm·HSP-90·PPH-5 complexes. Shown are the crosslinked products detected by mass spectrometry for GRLBDm·HSP-90·PPH-5 complexes, using two different search algorithms, xMASS and pLink. Amino acid positions refer to full length, non-tagged proteins as numbered in the Uniprot database. Scan number refers to the MS2 scan with the most hits for the specified peptide. Shown are also the number of hits for this scan, the per cent spectral intensity as well as the agreement between xMASS and pLink.  Table 2. Inter-protein crosslinked peptides detected for GRLBDm·HSP-90·FKB-6 complexes. Shown are the crosslinked products detected by mass spectrometry for GRLBDm·HSP-90· FKB-6 complexes, using the search algorithms, xMASS and pLink. Amino acid positions refer to full length, non-tagged proteins as numbered in the Uniprot database. Scan number refers to the MS2 scan with most hits for the specified peptide. Shown are also the number of hits for this scan, the per cent spectral intensity as well as the agreement of xMASS with the pLink algorithm.    Fig. S2) seems to stimulate hormone binding to the same extent as HSP-90 and ATP alone, leaving uncertain whether an interaction with Cep23/DAF-41 has taken place, as this interaction is known to be better stabilized by ATPγS and not so much by ATP 52 . By employing F-DEX in analytical ultracentrifugation measurements, only the F-DEX-bound complexes can be observed, allowing us to confirm the formation of protein complexes from an independent approach. F-DEX alone does not sediment in these experiments, but F-DEX bound to GRLBDm is readily observable at 2.7 S. Complex formation with chaperones then leads to the detection of larger species based on the bound fluorescent hormone. Judging from the reduction in the amount of monomeric F-DEX-bound receptor (Fig. 4B), the HSP-90 complex with hormone-bound GRLBDm is formed most efficiently, if ATP is also present (Fig. 4B). This implies that the rate increase may correlate with the formation of this complex.
Testing the influence of the TPR-cofactors PPH-5 and FKB-6 on the hormone binding to GRLBDm, we find a further modulation of the binding rates compared to GRLBDm·HSP-90·ATP (Fig. 4C,E). The rate of hormone binding is slightly reduced but remains accelerated compared to control reactions. To get clarity on the complex formation, theses assemblies were further investigated by AUC experiments with F-DEX. C. elegans ternary complexes with the cofactors were formed as efficiently as with HSP-90/ATP in the case of FKB-6 and slightly more efficiently with PPH-5, as judged from the reduction of monomeric F-DEX-bound receptor at 2.7 S (Fig. 4D). Based on this data, ATP-binding to HSP-90 seems to accelerate the binding of hormone to GRLBDm and slighter modifications to this rate are observable, when cofactors enter the complex (Fig. 4E). Given that residual hormone might be present in the binding pocket, it cannot be excluded that the observed rates represent exchange kinetics and not binding rates. Nevertheless, such acceleration shows that nematode HSP-90 has the ability to influence the hormone binding properties of GRLBDm.
Hsp90β's conformation is more restricted in the human Hsp90 system. We then aimed at understanding, to what extent the main principles are also conserved in the human system. The complexes assembled from the nematode proteins in this study correspond to those identified for the human system in the 1990s, but only limited in vitro data on the GR-complexes, even on the human ones, are available to date. To compare the two systems, we purified the corresponding human proteins and repeated the experiments described for the nematode system. The human system is known to behave differently from the nematode HSP-90 regarding its ATP-turnover and conformational flexibility 58,59 . We could form GRLBDm containing protein complexes with the protein phosphatase Pp5 and Hsp90β (Fig. 5A). In contrast to the nematode system, no ATPγS induced changes in the human GRLBDm·Hsp90β·Pp5 complex are observable and also no changes in s 20,w are observed for the binary GRLBDm·Hsp90β complex after ATPγS addition. Only upon binding of p23 is a strong increase in s 20,w observed and the closed form is obviously stabilized (Fig. 5A). Interestingly, we find a cooperative action of ATPγS, p23 and Pp5, which leads to a strong increase in affinity and binding of almost all GRLBDm to the complex. It here seems that a closed Hsp90β complex with the phosphatase Pp5 is very favorable, but only achievable if p23 stabilizes the closed conformation (Fig. 5A).
We then tested both human PPIase homologs, Fkbp51 (Fig. 5B) and Fkbp52 (Fig. 5C). Both PPIases support GR-binding to the chaperone complex, albeit the cooperative effect is reduced compared to the nematode system. In contrast to the nematode FKB-6 complexes, the response of GRLBDm·Hsp90β·Fkbp51/52 complexes to ATPγS is not observable. The presence of ATPγS, however, slightly decreases the affinity for the client. As in the complexes with Pp5, binding of p23 eventually leads to the compaction of the PPIase-containing protein complex (Fig. 5B, C), indicating that ATPγS may not be sufficient to initiate the closing reaction in the human system. This data shows that the nematode and human chaperone systems share a conserved interaction pattern with the TPR-cofactors involved in the processing of steroid receptors. The human Hsp90β however, shows a reduced ability to form the closed state under the examined conditions and seems to strictly require the cofactor p23 to perform this conformational change.
Similar patterns between the two systems can also be observed for the interaction with the hormone. As in the nematode system, only Hsp90β/ATP can accelerate the binding rate beyond control levels (Fig. 5D). AUC data show that F-DEX-bound protein complexes are formed more efficiently with ATP and also the more compacted ATPγS/p23-bound states of Hsp90β (Fig. 5E). Regarding the influence of human TPR-cofactors, only the hormone-bound quaternary complexes with ATPγS/p23 and either Fkbp51 or Fkbp52 form as efficiently or stronger than the binary GRLBDm·Hsp90β/ATP complex, judging from the reduced F-DEX-bound GRLBDm at 2.7S (Fig. 5G). While the Pp5-containing ternary complex with F-DEX-bound GR-LBDm is weakly associated (Fig. 5G), it is nevertheless able to strongly stimulate the binding rate (Fig. 4E). Thus, like in the nematode system, ATP binding to Hsp90β appears to accelerate the exchange of hormone in GRLBDm and further modifications to the exchange kinetics are observable, when TPR-cofactors enter the complex (Figs. 5F, 4E).

Discussion
GR·Hsp90 complexes have been studied extensively since the 1980s in cellular lysates of reticulocytes and wheat germ, leading to the identification of Hsp90 as a chaperone bound to GR and also to the description of assemblies involving several Hsp90 co-chaperones 40,[60][61][62][63] . Much of our current structural understanding of GR complexes in vitro originates from yeast Hsp90 and its cofactors and only recently, structural data on higher eukaryotic species became available 5,35,38,50,64 . The extent and timing of movements performed by Hsp90 and the biochemical contribution of the co-chaperones are still rather unclear, but the recent purification of client proteins makes in vitro studies possible [35][36][37] .
We here focus on the nematode and human Hsp90 chaperone systems to clarify which conserved principles are observable between the different eukaryotic species. The nematode system contains only one homolog of all major cofactors and a highly conserved Hsp90 protein, which has cellular contact to client proteins of all known www.nature.com/scientificreports/ Hsp90-client classes. There are at least 200 nuclear receptors in nematodes, but which of them are clients of the Hsp90 machinery is currently unknown. In this study, we used a stabilized GRLBDm mutant of the ligand binding domain of the human GR to model the wild-type protein, which is unstable 65 . The LBD domain of the human GR shows homology to the C. elegans nuclear hormone receptors NHR-25, NHR-47 and, FAX-1 that bind so far unknown ligands and to some extent, to the receptor for dafachronic acid, DAF-12. Based on our results, C. elegans HSP-90 appears to perform its closing reaction with an efficiency in-between yeast and human. Even under conditions where the nucleotide ATPγS is present, formation of the closed, twisted state does not proceed for the isolated HSP-90 or Hsp90β (data not shown). This contrasts with the homologous yeast Hsp82 protein. Instead, the nematode or human Hsp90s remain in an open-like conformation and also the presence of client protein does not induce the closing reaction. The TPR cofactor STI-1 and the ATPase activator AHSA-1 could hardly be detected in GRLBDm·HSP-90 complexes. This is in agreement with experiments performed in yeast 35 . The human homologue of STI-1, Hop, is known to deliver the GRLBD bound to the chaperone Hsp70 to Hsp90 and ATP hydrolysis by Hsp90 is thought to induce the release of Hsp70 and Hop 38 . Thus, STI-1 alone may not be detected as a component of GRLBDm·HSP-90 complexes in the absence of the HSP-70 chaperone. Further, we see that the binary GRLBDm·HSP-90 complex is disrupted once excess STI-1 is added, hinting at a competition for a binding site on HSP-90. Technically, the STI-1·HSP-90 complex cannot be observed by fluorescence AUC with labeled GRLBDm, it is, however, noticeable, since GRLBDm is exclusively observed in the unbound fraction. STI-1 interacts primarily with the C-terminal MEEVD-peptide of Hsp90, but also the M domain, which is likely the binding site that STI-1 and GRLBDm compete for 59,66,67 . For AHSA-1, we also observe competitive binding with GRLBDm, which could be due to the different conformational requirements or overlapping binding sites on the Hsp90 chaperone. In particular, the binding site in the middle domain of the chaperone may be used by both proteins and therefore such ternary complexes may not form efficiently 64,68 . Only in trimeric complexes with the cofactors PPH-5 or FKB-6 does HSP-90 perform nucleotide-induced conformational rearrangements and reach a compacted state. These cofactors also strongly facilitate the binding of GRLBDm to the chaperone. Thus, the nematode HSP-90 cycle seems to be based on cooperative events between TPR-proteins, nucleotide and client protein (Fig. 6).
The cofactor p23 has been thought to stabilize Hsp90-client interactions, which is supported in our experiments with the human chaperone system 36,69 . A recent study, however, shows that p23 stimulates GRLBD dissociation from the chaperone with ATP but not with the non-hydrolysable nucleotides and that it thus, can function as a substrate release factor for Hsp90 70 . For our AUC analysis, we utilize the non-hydrolysable nucleotide ATPγS in an attempt to stabilize the closed states of the client·Hsp90 and client·Hsp90·TPR cofactor complexes in the presence and absence of p23. In this set-up the affinity for Cep23/DAF-41 is rather low. In complexes consisting of the nematode proteins, 0.22 ± 0.05 of the GRLBDm fraction is bound to the chaperone in the presence of ATPγS while 0.28 ± 0.03 is bound when Cep23/DAF-41 is added. In contrast, 0.49 ± 0.05 of GRLBDm is bound to the chaperone in the absence of ATPγS. We therefore, can say with conviction that ATPγS reduces the affinity of HSP-90 to GR, but the further integration of Cep23/DAF-41 may not change the affinity significantly (p = 0.1493). Given that the nematode protein responds much stronger to the binding of ATPγS, it can be assumed that the observed differences between nematode HSP-90 and human Hsp90β relate to the different flexibility and response of the protein to the nucleotide.
Based on mass spectrometry data, we investigated whether the observed cooperativity in complex formation may originate from contacts between the cofactor and client proteins on the chaperone scaffold. Our structural interpretation for GRLBDm·HSP-90 is in good agreement with the structural model proposed previously for yeast Hsp90 in complex with GRLBDm 35 . Electron microscopy structural studies also report GRLBD being bound to the MD and CTD domains of human Hsp90, as part of a complex with Hsp70, Hop and Hsp90 38 . This binding site is also in line with the client-binding region of E. coli Hsp90 described by Genest et al. 71 . Our results on the GRLBDm·HSP-90·FKB-6 topology are consistent with recent findings regarding the human Fkbp51 72 . This study also postulates a stepwise interaction with Hsp90, with affinities decreasing in the order TPR > FK2 > FK1. In the case of PPH-5, the arrangement of the cofactor in GRLBDm·HSP-90·PPH-5 complexes is similar to our previous study on the binary HSP-90·PPH-5 complex 41 . To illustrate these topologies, we used the identified crosslinked peptides in docking calculations with HADDOCK and obtained structural models for the dimeric GRLBDm·HSP-90 complex (Fig. 7A) as well as the trimeric complexes with the two TPR cofactors (Fig. 7B,  C) 73,74 . These calculations bring both cofactors' catalytic domain in close proximity to the client (Fig. 7B, C), which may be setting the basis for the observed cooperativity. The client may then be accessible to undergo transformations towards dephosphorylation and receptor maturation.
The cooperative complex formation and hormone binding patterns seem similar in the nematode and human Hsp90 systems, albeit the persistence of the "open" state is stronger for the human Hsp90β protein. Here, Fkbpcontaining and Pp5-containing complexes appear unable to perform the closing reaction induced by ATPγS and only with the support of p23 are these rearrangements possible and lead to the progression of the Hsp90 cycle. This data highlights that, despite the conserved features of the cofactors, the different degree of flexibility within the Hsp90 protein from the two organisms influences the principles during chaperone·cofactor·client complex formation.  Experiments contained 600 nM of labelled GRLBDm (*GRLBD) and 3 μM of the unlabelled chaperones and cofactors of interest. Nucleotides were added at a concentration of 2 mM. GRLBD alone showed a weak tendency to aggregate, which was absent once Hsp90 and cofactors were present. This tendency is visible in the dc/dt plots as an extension towards higher s-values and a deviation from the symmetric peak shape. AUC-experiments monitoring F-DEX (Thermo Fischer Scientific, Bremen, Germany) utilized 400 nM F-DEX and 3 μM of the unlabelled proteins of interest. These measurements were carried out in 20 mM HEPES, 20 mM KCl, 5 mM MgCl 2 , pH 7.5. Centrifugation was performed at 42,000 rpm at 20 °C. Data analysis was performed by calculating differences between scans from a selected time range and averaging over several of these differentials. The dF/dt data was then normalized against the initial fluorescence intensity. To ensure comparable sample handling, plots were generated from samples measured in the same experiment with automated data processing in the inhouse software diffUZ 41 . s 20,w values were derived from a bi-Gaussian fitting of the dF/dt plots and the error is based on the standard deviation of this approach plus an additional contribution from performing the meniscus picking procedure. In cases were trends were not always conserved between replicate analyses, all replicates were used to calculate significance levels for potential effects. Significance was achieved, if p-values from a Student's t-test were below 0.01. In general repeatability of AUC results decreased if three or more weakly binding components were added at concentrations below saturation levels.

Material and methods
Fluorescence polarization. Hormone binding kinetics were monitored by fluorescence polarization on a Jasco FP-8500 fluorescence spectrometer (Jasco, Groß-Umstadt, Germany) equipped with polarizers. 1 µM apo-GRLBDm, after extensive dialysis to remove dexamethasone as described, was added to various chaperone mixtures with a chaperone and cofactor concentration of 3 µM 35,38 . Binding kinetics to 50 nM fluorescently labelled dexamethasone (F-DEX, Thermo Fischer Scientific, Bremen, Germany) were recorded at 20 °C  NanoLC-ESI-MS/MS. 200 ng tryptically digested samples were measured by nanoLC-ESI-MS/MS as described previously 79 . An UliMate 3,000 RSLC nano LC system (Thermo Fischer Scientific, Bremen, Germany) was utilized for nano HPLC analysis using the following solvent system: (A) 0.1% FA; (B) 84% ACN, 0.1% FA. Samples were initially loaded on a trap column (Thermo Fischer Scientific, 100 μm × 2 cm, particle size 5 μm, pore size 100 Å, C18) with a flow rate of 30 μl/min with 0.1% TFA. After sample concentration and washing, the trap column was serially connected with an analytical C18 column (Thermo Fischer Scientific, 75 μ m × 50 cm, particle size 2 μm, pore size 100 Å), and the peptides were separated with a flow rate of 400 nl/min using a solvent gradient of 4% to 40% B for 95 min at 60 °C. After each sample measurement, 1 h of column washing was performed for equilibration. The HPLC system was on-line connected to the nano-electrospray ionization source of a Q Exactive HF mass spectrometer (Thermo Fischer Scientific, Bremen, Germany). The mass spectrometer was operated in a data-dependent mode with the spray voltage set to 1,600 V in positive mode and a capillary temperature of 275 °C. Full scan MS spectra (mass range 350-2000 m/z) were acquired in the Orbitrap analyzer at a mass resolution of 60,000. The twenty most intensive ions per spectrum were subsequently fragmented using collision-induced dissociation (35% normalized collision energy) and scanned in the linear ion trap. The m/z values triggering MS/MS were set on a dynamic exclusion list for 30 s.
Identification of crosslinked peptides. Initial data analysis was performed with MaxQuant 1.5 to obtain lists of all peptides and peaks from the raw data files 80 . These tables were then searched with xMASS as described 41 , yielding crosslinks in four categories with either crosslinker attached at one (Type 1) or both ends (Type 2) on one peptide, crosslinker bridging peptides from the same protein (Type 3) or crosslinker bridging peptides from different proteins (Type 4). These crosslinked structures could be filtered for the same intensity of the peaks separated by 12.07 Da, co-elution from the column, fragmentation spectrum in MS2 and other potential solutions with a similar score. In parallel, the software pLink was used to search the same datasets with the default parameter settings 53 .

Molecular modelling and Docking calculations. A homology model for the open HSP-90 conforma-
tion was generated with ADP-bound HtpG from E. coli (PDB 2IOP) as a template, using the Chimera interface to MODELLER 50,81-83 . FKB-6 and PPH-5 were modelled based on human FKBP51 (PDB 5NJX) and rat PP5 (PDB 4JA9) respectively 84,85 . For GR-LBD we used the structure solved by Hemmerling et al. (PDB 5NFP) 86 . The validity of each crosslinked pair was first confirmed using the algorithm DisVis 87,88 . Docking was performed in the expert interface of HADDOCK 73 . Lysine residues identified in crosslinking products were defined as active residues, enforcing a distance restraint of 30 Å between their Cb-atoms, to direct the docking calculations. Structures from the best binary solutions were used to proceed with docking the third protein and assemble the trimeric complexes.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.