The FNIP co-chaperones decelerate the Hsp90 chaperone cycle and enhance drug binding

Heat shock protein-90 (Hsp90) is an essential molecular chaperone in eukaryotes involved in maintaining the stability and activity of numerous signalling proteins, also known as clients. Hsp90 ATPase activity is essential for its chaperone function and it is regulated by co-chaperones. Here we show that the tumour suppressor FLCN is an Hsp90 client protein and its binding partners FNIP1/FNIP2 function as co-chaperones. FNIPs decelerate the chaperone cycle, facilitating FLCN interaction with Hsp90, consequently ensuring FLCN stability. FNIPs compete with the activating co-chaperone Aha1 for binding to Hsp90, thereby providing a reciprocal regulatory mechanism for chaperoning of client proteins. Lastly, downregulation of FNIPs desensitizes cancer cells to Hsp90 inhibitors, whereas FNIPs overexpression in renal tumours compared with adjacent normal tissues correlates with enhanced binding of Hsp90 to its inhibitors. Our findings suggest that FNIPs expression can potentially serve as a predictive indicator of tumour response to Hsp90 inhibitors.

T he molecular chaperone heat shock protein-90 (Hsp90) is responsible for folding, stability and activity of many proteins also known as 'client proteins', including many responsible for tumour initiation, progression and metastasis 1 . This makes the chaperone Hsp90 an attractive target for cancer therapy 2 . Hsp90 has the ability to bind and hydrolyse ATP, which is essential for its chaperone function 3 . Small molecule inhibitors bind to the ATP-binding pocket of Hsp90 and inhibit its chaperone function. Consequently, this prevents Hsp90 interaction with client proteins, leading to their degradation by the proteasome. In contrast to other anticancer drugs, Hsp90 inhibitors simultaneously inhibit multiple drivers of oncogenesis.
Hsp90 chaperone cycle is tightly regulated by another group of proteins referred to as 'co-chaperones'. Their stability does not depend on Hsp90 function but they interact with distinct Hsp90 conformational states, providing directionality to the Hsp90 cycle 4 . Furthermore, certain co-chaperones, such as HOP and Cdc37 p50 inhibit the Hsp90 chaperone cycle, assisting in delivery of distinct sets of client proteins (steroid hormone receptors and kinases, respectively) to the Hsp90 chaperone machine. In contrast, the co-chaperone Aha1 facilitates energy-intensive conformational changes necessary to establish Hsp90 ATPase competence, markedly increasing the weak endogenous ATPase activity of Hsp90. Aha1 is thus considered to be a crucial component of active Hsp90 chaperone complexes 5,6 .
Here we show that the stability of the tumour suppressor folliculin (FLCN) depends on the chaperone function of Hsp90. Germline mutations and loss of function of FLCN causes Birt-Hogg-Dubé syndrome, a rare inherited cancer syndrome that predisposes affected individuals to develop kidney tumours, pulmonary cysts and benign skin tumours (fibrofolliculomas) 7 . FLCN interacts and forms a complex with folliculin-interacting proteins 1 and 2 (FNIP1 and FNIP2, also referred to as FNIPs) [8][9][10] . The function of FNIPs, however, remains elusive. Our results indicate that FNIPs act as co-chaperones of Hsp90. They inhibit its ATPase activity, 'tailoring' Hsp90 to chaperone kinase and non-kinase clients. We have further shown that Aha1 co-chaperone can displace FNIPs and stimulate Hsp90 ATPase activity. Finally, FNIPs also enhance the binding of Hsp90 to its inhibitors such as ganetespib (GB); therefore, overexpression of FNIPs in specific tumours can be an indicator of their response to Hsp90 inhibitors.

Results
FLCN is a new client of Hsp90. To determine the binding partners of the tumour suppressor FLCN, we transiently expressed an amino-terminally FLAG-tagged FLCN (FLAG-FLCN) in human embryonic kidney 293 (HEK293) cells and identified its intracellular binding proteins by immunoprecipitating FLAG-FLCN with anti-FLAG M2 affinity gel and mass spectrometry (MS) analysis ( Fig. 1a and Supplementary Table 1). We found molecular chaperones heat shock protein-70 (Hsp70) and Hsp90, and their regulators HOP, CHIP and Aha1, and CCT2, CCT4, CCT7 and CCT8, which are members of the chaperonin system TRiC (TCP-1 ring complex), (Fig. 1a). We validated our data by immunoprecipitating the endogenous FLCN (Fig. 1b) or the FLAG-FLCN (Fig. 1c) from HEK293 cells and showed its interaction with the molecular chaperone machineries Hsp70, Hsp90 and a subunit of the chaperonin TRiC, CCT2 (Fig. 1b,c). We also observed FLCN interaction with the Hsp70 and Hsp90 co-chaperones including HOP, CHIP, Cdc37 p50 , PP5, p23 and Aha1 (Fig. 1b,c). In general, molecular chaperones are involved in folding and stability of proteins. We first treated the HEK293 cells with the Hsp70 inhibitor JG-98 (ref. 11) and showed the degradation of FLCN after a 2 h treatment in both soluble and insoluble protein fractions (Fig. 1d). These data suggest that inhibition of Hsp70 does not lead to an increase in misfolded FLCN but instead to its degradation. The molecular chaperone Hsp90 however is more selective towards its 'client proteins' and is also involved in protecting them from degradation 12 . Therefore, we treated the HEK293 cells with different inhibitors of Hsp90 such as GB 13 (Fig. 1e), SNX2112 (ref. 14) and PU-H71 (ref. 15) (Supplementary Fig. 1a,b), to evidence the degradation of FLCN. Previous works have shown that inhibition of Hsp90 generally leads to ubiquitination and degradation of its client proteins in the proteasome 16 . We investigated this possibility by first demonstrating that inhibition of Hsp90 causes its dissociation from FLCN (Fig. 1f). We further showed that HEK293 cells treated with 50 nM proteasome inhibitor bortezomib for 2 h before addition of GB blocked FLCN degradation (Fig. 1g). We did not obtain similar results when we treated the cells with the lysosomal inhibitor Bafilomycin A1 ( Supplementary Fig. 1c), suggesting that the lysosome is not involved in degradation of FLCN. We next immunoprecipitated and salt-stripped (with 0.5 M NaCl) FLAG-FLCN from HEK293 cells treated with either 50 nM bortezomib or 1 mM GB for 4 h and showed its ubiquitination by western blotting (Fig. 1h). Taken together, our results suggest that FLCN is a client of Hsp70 and Hsp90. Inhibition of Hsp90 leads to ubiquitination and degradation of FLCN in the proteasome.
FNIP1 and FNIP2 facilitate FLCN binding to Hsp90 chaperone. FNIP1 and FNIP2 are homologous binding partners of FLCN 8,10 ; however, their molecular function remains elusive. In addition, FNIP1 was shown to interact with Hsp90, but the significance of this observation was not researched further 8 . We first confirmed these data by immunoprecipitating the endogenous FNIP1 and FNIP2 from HEK293 cells and detecting Hsp90 (Fig. 2a). We also observed FNIPs interaction with Hsp70 and co-chaperones p23, HOP and Cdc37 p50 (Fig. 2a). In addition, we co-immunoprecipitated Hsp90 clients such as glucocorticoid receptor (GR), B-Raf, Cdk4 and FLCN (Fig. 2a). We confirmed our data by transiently expressing and immunoprecipitating HA-FNIP1 and HA-FNIP2 from HEK293 cells and then probing for the chaperones, co-chaperone and client proteins ( Supplementary Fig. 2a). Previous work has shown that FNIPs are homo-and heterodimer proteins 10 . We isolated FNIP1 and FNIP2 homo-and heterodimers by transiently co-expressing cMyc-FNIP1/HA-FNIP1, cMyc-FNIP2/HA-FNIP2 and HA-FNIP1/cMyc-FNIP2 in HEK293 cells. We first immunoprecipitated cMyc-tagged proteins using anti-cMyc agarose affinity gel. We then competed the cMyc proteins from the affinity gel with cMyc peptides. The eluted proteins were dialysed and subjected to a second round of immunoprecipitation (IP) using anti-HA agarose affinity gel. Immunoprecipitated proteins have shown that FNIP1 homodimer has a stronger affinity for binding to Hsp90 and FLCN than FNIP2 (Fig. 2b). We next asked whether FNIP1 or FNIP2 are clients or co-chaperones of Hsp90. Treating HEK293 cells with the Hsp90 inhibitor GB did not affect FNIP protein stability (Fig. 2c). These data suggest that FNIP1 and FNIP2 are co-chaperones of Hsp90. We obtained further evidence by demonstrating a direct interaction between FNIP1 and Hsp90. We were able to bacterially express and purify a small amount of FLAG-FNIP1. Our purified FLAG-FNIP1 can directly interact with bacterially expressed and purified Hsp90a-His 6 ( Fig. 2d) and FLCN-His 6 ( Fig. 2e). Surprisingly, Hsp90a and FLCN did not directly interact with each other in vitro; however, pre-incubation of Hsp90a with FNIP1 facilitated the Hsp90-FNIP1-FLCN complex formation (Fig. 2f). It is noteworthy that we were unable to express and purify an adequate amount of FNIP2 protein from bacteria, yeast or baculovirus expression systems for biophysical analysis.   affinity gel, we observed FNIP1 interaction with the Hsp90a M-and C-domains, whereas FNIP2 interacted with only the M-domain of Hsp90a (Fig. 3a). Association of FLCN with the Mand C-domains of Hsp90a was also observed (Fig. 3a). We next determined the region in FNIP1 that interacts with Hsp90. FNIP1 does not have a known functional domain; however, based on previous studies 8, 10 , we constructed and transiently expressed four regions of HA-FNIP1 designated A-D in HEK293 cells (Fig. 3b). Consistent with the previous work 8, 10 , truncation of FNIP1 at the C-terminus (deletion of fragment D) abrogated its interaction with FLCN (Fig. 3b). However, the concomitant deletion of 300 amino acids (fragment A) from the N-terminus of FNIP1 restored its binding to FLCN (Fig. 3b). This result is not in agreement with the previous study 8 and underlies the complexity in FNIP1 binding to FLCN. Our data also indicated that the C-domain of FNIP1 (amino acid 929-1166 or fragment D) preferentially interacts with Hsp90 (Fig. 3b). We confirmed these data by carrying out a reciprocal experiment and co-expressing FNIP1-D-HA and Hsp90a-FLAG full-length (wild type) and its different domains in HEK293 cells. We immunoprecipitated FNIP1-D-HA and observed co-IP of Hsp90a full length and its M-domain (Fig. 3c). We next bacterially expressed and purified fragment D (FNIP1-D-His 6 ), ( Supplementary Fig. 2b,c) and examined its affinity for binding to Hsp90a in vitro by fluorescently labelling FNIP1-D-His 6 with Texas Red Maleimide, and measured the K d by fluorescence anisotropy (Fig. 2j). Our bacterially expressed and purified HSP90a had ATPase activity. The titration fit to a single-site binding equation with a K d of 1.3±0.7 mM (Fig. 3d). This K d was unaffected by the presence, absence or identity of adenosine nucleotide bound to the protein ( Supplementary Fig. 2d). Taken together, our findings suggest that FNIPs behave as cochaperones of Hsp90 and they are involved in loading of FLCN to Hsp90.
FNIPs regulate Hsp90 activity and chaperoning of the clients.
To gain further insight into FNIP1 and FNIP2 function as co-chaperones of Hsp90, we used small interfering RNA (siRNA) to silence either FNIP1 or FNIP2 (Supplementary Table 2), or both in HEK293 cells, and monitored the stability and activity of the Hsp90 kinase clients such as B-Raf, Ulk1 and Cdk4, and nonkinase clients, for example, GR, ER and FLCN ( Fig. 4a and Supplementary Fig. 3a). Silencing either FNIP1 or FNIP2 caused a modest decrease in client protein levels. This data suggests that the absence of either FNIP1 or FNIP2 allows for compensation by the other FNIP isoform. Silencing of both FNIP1 and FNIP2 significantly decreased the stability of the selected kinase and non-kinase client proteins (Fig. 4a). We next transiently overexpressed cMyc-FNIP1 or cMyc-FNIP2 in HEK293 cells and examined the stability and the activity of the Hsp90 clients. Overexpression of FNIPs caused an increase in GR, Ulk1 and FLCN protein levels and hyperphosphorylation of B-Raf and pY416-c-Src ( Fig. 4b and Supplementary Fig. 3b).
Cystic fibrosis transmembrane conductance regulator (CFTR) is also an Hsp90 client 17,18 and relies on a 'slow' Hsp90 chaperone cycle for proper folding. To examine the effect of FNIPs on Hsp90 chaperone function, we assessed their impact on steady-state expression of CFTR protein in mammalian cells. HEK293 cells were transiently co-transfected with CFTR, and HA-FNIP1 and HA-FNIP2. Empty plasmid pcDNA3 (empty vector, EV) was used as a negative control. Western blot analysis of these samples using anti-CFTR antibody detected a doublet (Fig. 4c), with the upper band representing the mature Golgi-processed glycoform of CFTR found at the cell surface and the lower band an immature core-glycosylated protein (Fig. 4c). Overexpression of the FNIPs produced a significant increase in CFTR protein (Fig. 4c). These data suggested a reduction in Hsp90 chaperone activity and, therefore, an increase in CFTR expression.
Hsp90 chaperone function is coupled to its ATPase activity 3 . We therefore examined the impact of FNIPs on Hsp90 ATPase activity. We transiently expressed Hsp90a-HA in the prostate cancer PC3 cell line and HA-FNIP1, HA-FNIP1-D (amino acids 929-1166) and HA-FNIP2 in HEK293 cells. Proteins were immunoprecipitated, salt stripped and competed off the HA affinity beads with the relevant peptides ( Supplementary Fig. 4a). The quality of the isolated proteins was examined by Coomassie staining of the SDS-polyacrylamide gel electrophoresis (SDS-PAGE). These purified proteins were quantified and used in the FNIP2 on Hsp90 clients. Stability and activity of the indicated clients were assessed by immunoblotting. Densitometry of the western blotting for FNIP1 and FNIP2 is represented as mean ± s.d. A Student's t-test was performed to assess statistical significance (**Po0.005 and ***Po0.0005). (b) Transient overexpression of cMyc-tagged FNIP1 or FNIP2 in HEK293 cells and their impact on levels of Hsp90 clients was assessed by immunoblotting. Empty vector (EV) was used as a control. (c) HEK293 cells were co-transfected with CFTR and indicated HA-FNIP1 and HA-FNIP2 constructs. After 24 h, CFTR and FNIPs-HA were detected by immunoblotting; GAPDH was used as loading control. Empty vector (EV) was used as a control. (d) In vitro ATPase activity of Hsp90a-FLAG isolated from PC3 cells. Inhibitory effects of purified HA-FNIP1, FNIP1-D-HA or HA-FNIP2 on ATPase activity of Hsp90a-FLAG. All the data represent mean±s.d. A Student's t-test was performed to assess statistical significance (***Po0.0005 and ****Po0.0001). (e) ATPase activity of Hsp90a-FLAG from d was inhibited by addition of 10 mM GB. All the data represent mean ± s.d. A Student's t-test was performed to assess statistical significance (****Po0.0001). molar ratio indicated in Supplementary Fig. 4b of Hsp90a:FNIP in the PiPer Phosphate Assay Kit (Thermo Fisher Scientific), in the presence of ATP as substrate as previously described (see Methods) 19 . We measured the ATPase activity of isolated Hsp90a in vitro as previously described 19 (Fig. 4d and Supplementary  Fig. 4c). GB (10 mM) inhibited ATPase activity (Fig. 4e). Addition of FNIP1, FNIP1-D and FNIP2 also significantly inhibited the ATPase activity of Hsp90a ( Fig. 4d and Supplementary Fig. 4c). Percentage ATPase activity was based on mmol P i per mol min À 1 for Hsp90a alone and in the presence of FNIPs titrated until inhibition was achieved (Hsp90a ¼ 291.5 nM, FNIP1 ¼ 189.8 nM, FNIP1-D ¼ 899.3 nM and FNIP2 ¼ 1.6 mM) ( Supplementary  Fig. 4d). These data suggest that FNIPs are potent inhibitors of the Hsp90 chaperone cycle.
Hsp90 is an evolutionarily conserved chaperone and based on structural homology detection methods Lst4 is a potential orthologue of FNIPs in Saccharomyces cerevisiae 20 . However, we were unable to detect Lst4-GST interaction with yeast Hsp90 (yHsp90, also known as Hsp82) (Fig. 5a). In addition, unlike FNIPs, Lst4-GST was unable to inhibit ATPase activity of yHsp90 in vitro (Fig. 5b and Supplementary Fig. 4e,f). We next decided to express FLAG-FNIP1 and FLAG-FNIP2 galactose-inducible promoter of GAL1 in yeast containing either human or yeast Hsp90 as the sole functional Hsp90. Overexpression of FNIPs in yeast containing human Hsp90a caused lethality (Fig. 5c). However, overexpression of FNIPs in yeast with only yHsp90 did not cause any growth defects. We examined the expression of the FNIPs in these cells by western blot analysis and were only able to detect FLAG-FNIP1 and FLAG-FNIP2 in yeast expressing Hsp90a (Fig. 5d). These data suggest that expression of FNIPs in yeast cause lethality, because they inhibit the essential chaperone function of Hsp90a. Taken together, our results suggest that (i) Lst4 is not an orthologue of FNIPs in yeast and (ii) FNIPs probably arose in higher eukaryotes as co-chaperones to inhibit or slow the complex chaperone cycle of Hsp90 in cellular milieu.
FNIPs compete with the Aha1 co-chaperone for binding to Hsp90.
To determine the effects of FNIPs on Hsp90 interaction with co-chaperones, we used FNIP1-and FNIP2-specific siRNA to silence expression of these genes in HEK293 cells transiently expressing Hsp90a-FLAG. We next immunoprecipitated Hsp90a-FLAG from these cells and examined its interaction with selected co-chaperones. Our data show that siRNA-mediated silencing of both FNIP1 and FNIP2 increased Hsp90 interaction with the co-chaperones Aha1 and PP5 (Fig. 6a). We next transiently overexpressed HA-FNIP1 and HA-FNIP2 in HEK293 cells that also express Hsp90a-FLAG. The overexpression of FNIPs compared with the endogenous FNIP1 and FNIP2 were examined by immunoblotting ( Supplementary Fig. 5a). Following IP of Hsp90a-FLAG, we observed a marked reduction in its interaction with the co-chaperones Aha1 and PP5 (Fig. 6b). Aha1 is the activator of the Hsp90 ATPase activity 5,6 . We next queried whether Aha1 has the ability to stimulate the ATPase activity of Hsp90 that has already bound to FNIP1, FNIP1-D or FNIP2. In agreement with our earlier data, FNIPs inhibited the ATPase activity of Hsp90a and, conversely, addition of 1.3 mM Aha1 to these reactions stimulated Hsp90 ATPase activity (Fig. 6c and Supplementary Fig. 5b-f). Taken together, these data suggest that Aha1 has the ability to displace FNIPs from Hsp90. We obtained additional evidence for this possibility by using the bacterially expressed and purified Hsp90a, Aha1-FLAG and FNIP1-D fragment. First, we bound FNIP1-D-His 6 to Ni-NTA agarose and incubated with 100 ng recombinant Hsp90a. After washing the agarose with buffer, we added different amounts of Aha1-FLAG. Our data show that addition of 10 ng recombinant Aha1-FLAG completely dissociated Hsp90a from FNIP1-D-His 6 (Fig. 6d). We next carried out a similar experiment by first immobilizing Aha1-FLAG onto anti-FLAG M2 affinity gel. We then added 100 ng recombinant Hsp90a, to form an Aha1-Hsp90a complex. After washing the agarose beads with buffer, we added different amounts of FNIP1-D-His 6 . We observed that addition of 100 ng FNIP1-D-His 6 disrupted the Aha1-Hsp90a complex (Fig. 6e). We repeated the experiments in Fig. 6d,e, but using bovine serum albumin as a negative control ( Supplementary Fig. 5g-i). Our data confirm that Aha1 and FNIP1 compete for binding to Hsp90, either accelerating or decelerating its chaperone cycle. In addition, Aha1 appears to have higher affinity than FNIP1 towards Hsp90, as 10 ng of Aha1 was sufficient to displace FNIP1-D from Hsp90 (Fig. 6d).
FNIPs overexpression enhances Hsp90 binding to ATP and drugs. ATP binding and hydrolysis by the N-terminus of Hsp90 is essential for its chaperone function (Fig. 7a) 3 . We examined the impact of FNIPs on Hsp90 binding to ATP or drugs by first overexpressing HA-FNIP1 and HA-FNIP2 in HEK293 cells. Overexpression of FNIPs caused an increase in Hsp90 binding to ATP agarose (Fig. 7b). Furthermore, overexpression of cMyc-FNIP1 and cMyc-FNIP2 in HEK293 cells significantly increased the Hsp90 binding to 0.01 mM biotinylated-GB (Fig. 7c,d) and 0.01 mM biotinylated-SNX2112 (ref. 14 and Fig. 7e,f) compared with control (EV). We next examined the effects of FNIP1 and FNIP2 knockdown on Hsp90 binding to ATP and drug. siRNA knockdown of FNIP1, FNIP2 or both in HEK293 cells did not have an impact on Hsp90 binding to ATP (Fig. 7g) but significantly decreased Hsp90 binding to GB (Fig. 7h).
We next confirmed that overexpression of either FNIP1-HA or FNIP2-HA (Fig. 8a,b) sensitized the HEK293 cells to GB, as evidenced by induction of the pro-apoptotic markers cleaved caspase-3 and cleaved poly-ADP-ribose polymerase (PARP) (Fig. 8c). Conversely, siRNA knockdown of FNIP1 or FNIP2 in HEK293 cells significantly reduced apoptosis with 1 mM GB and knockdown of both FNIPs completely abolished the induction of apoptotic markers (Fig. 8d). Taken together, these data suggest that expression of FNIPs correlates to Hsp90 binding to drugs and also sensitivity of cells to Hsp90 inhibitors.

FNIPs overexpression sensitizes cancer cells to Hsp90 drugs.
Tumours are generally sensitive to Hsp90 inhibitors 19,21,22 . We therefore asked whether FNIPs are overexpressed in cancer cells and also whether they are contributing to cancer cell sensitivity towards Hsp90 inhibitors. We used prostate (LNCaP), bladder (T24), breast (MCF7), lung (H1299) and colorectal adenocarcinoma (HT29) cancer cell lines. Previous work by Synta Pharmaceuticals Corp. (data available at www.syntapharma.com/) has established these cell lines' sensitivity to Hsp90 inhibitor GB and these are: LNCaP (IC50 ¼ 8 nM), T24 (IC50 ¼ 7 nM), MCF7 (IC50 ¼ 25 nM), H1299 (IC50 ¼ 6 nM) and HT29 (IC50 ¼ 50 nM). We first showed that FNIPs were highly expressed in these cancer cells compared with HEK293 cells (Fig. 9a). We next immunoprecipitated the endogenous Hsp90 from the above cancer cells and detected its interaction with FNIPs (Fig. 9a). To link the high levels of FNIPs with sensitivity of cancer cells to Hsp90 inhibitors, we used siRNA to knock down FNIP1 and FNIP2 in LNCaP, T24, MCF7, H1299 and HT29 cells. The apoptotic markers cleaved caspase-3 and cleaved PARP were abundant in cancer cells treated with 0.1 mM GB; however, this effect was abrogated in the siRNA FNIPs knockdown samples (Fig. 9b-f). These data suggest that overexpression of FNIPs is a contributing factor towards cancer cell sensitivity to Hsp90 inhibitor GB. It is noteworthy that cleaved caspase-7 was used to evaluate apoptosis in MCF7 cells, as they lack caspase-3 (Fig. 9d).
Elevated FNIPs sensitize renal tumours to Hsp90 inhibitor. To gain further insight into FNIPs expression and sensitivity of tumours to Hsp90 inhibitors, we examined tumours and adjacent normal tissues from patients with renal cell carcinoma (RCC), the most common type of kidney cancer 23 . Histopathologically, RCCs are classified into subtypes. We used tumours from patients with clear cell RCC, papillary type I and type II RCC, and oncocytoma ( Fig. 10a-d). Within 10 min of removal of tumours, by radical or partial nephrectomy, RCC tumours and adjacent normal tissues were dissected into 3 mm 3 pieces followed by protein extraction and drug binding assay. Our data showed that both FNIP1 and FNIP2 were overexpressed in RCC tumours compared with adjacent normal tissues (Fig. 10a-d). In addition, we discovered that the Hsp90 from RCC tumours had a higher affinity for binding to biotinylated GB compared with the adjacent normal tissues (Fig. 10a-d). We next showed greater association between Hsp90 and FNIPs compared with normal tissues (Fig. 10e). This is not surprising, because FNIPs are highly expressed in tumours compared with adjacent normal tissues. However, we found that Hsp90 bound stronger to Aha1 in normal tissues compared with the tumours, even though both tissues expressed equal levels of Aha1 (Fig. 10e). Our earlier data showed that FNIPs compete with the activating co-chaperone Aha1 for binding to Hsp90 (Fig. 6e). We therefore asked whether addition of purified FNIP1-D-His 6 to the protein lysates from normal tissues could disrupt the Hsp90-Aha1 complex. Addition of 100 ng FNIP1-D-His 6 to protein lysates from normal tissues displaced Aha1 interaction with Hsp90 and also increased Hsp90 binding to 0.1 mM biotinylated GB (Fig. 10f). Taken together, our data suggest that FNIPs make renal tumours sensitive to Hsp90 inhibitors. Their expression level can potentially serve as a predictive indicator of tumour response to Hsp90 inhibitors.

Discussion
Germline mutation in the tumour suppressor FLCN is responsible for Birt-Hogg-Dubé syndrome, an inherited kidney cancer syndrome 24 . Previous work has shown that FNIP1 and FNIP2 are critical components of the FLCN complex and are essential for its tumour suppressive function 9 . However, the exact molecular function of FNIP1 and FNIP2 remains elusive. In this study, we demonstrate by MS analysis that FLCN interacts with the molecular chaperones Hsp70 and Hsp90, as well as members of the chaperonin system TRiC. Our data have shown that FLCN is a bona fide Hsp70 and Hsp90 client, as treating the cells with the Hsp70 inhibitor JG-98 destabilizes FLCN, and furthermore Hsp90 inhibitor treatment GB, SNX2112 or PU-H71 leads to dissociation of FLCN from Hsp90 and consequently its ubiquitination and proteasomal degradation. Based on our current knowledge in the function of TRiC-Hsp70-Hsp90 proteins and also our findings here, we would like to suggest that TRiC-Hsp70 are involved in 'early' stages of FLCN folding and Hsp90 plays a key role in protecting FLCN from degradation. We further demonstrated that FLCN binding partners FNIP1 and FNIP2 function as co-chaperones of Hsp90. FNIP1 interacts with the M-and C-domains of Hsp90, whereas FNIP2 interacts only with its M-domain. Previous work reported that FNIP1 and FNIP2 exist as a homodimer and a heterodimer, respectively 10 . We confirmed these data by demonstrating the existence of a FNIP1:FNIP2 heterodimer and its interaction with FLCN. We have also demonstrated that three populations of FNIP1 and FNIP2 interact with Hsp90; however, the FNIP1 homodimer preferentially binds to Hsp90. Our in vitro data provide additional evidence that recombinant FNIP1 directly interacts with Hsp90; however, FLCN requires the presence of FNIP1 for its binding to Hsp90. These data suggest that FNIP1 and FNIP2, similar to the co-chaperone Cdc37, are involved in 'loading' of FLCN and perhaps other client proteins to Hsp90. Approximately 93% of all pathogenic FLCN mutations prematurely truncate the encoded FLCN protein 25 . When experimentally modelled in vitro, most mutations analysed appeared to be pathogenic by disrupting stability of FLCN 25 . In addition, FLCN binding to both FNIP1 and FNIP2 is mediated specifically through the C-terminal region of FLCN 8,26 . Truncating mutations result in loss of the C-terminus of FLCN, abolishing its interaction with FNIP1 and FNIP2, and consequently Hsp90. We speculate that loss of Hsp90 interaction with FLCN is the reason for the instability of the pathogenic FLCN mutants.
Hsp90 chaperone activity is coupled to its ATPase activity, which is tightly regulated by co-chaperones and posttranslational modification 27,28 . Our results suggest that FNIP1 is a potent inhibitor of Hsp90 ATPase activity, as 200 nM of FNIP1 inhibits Hsp90 ATPase activity by 50-fold. FNIP2 also has shown  We also found that FNIPs compete with Aha1 cochaperone binding to Hsp90. Our data have demonstrated the ability of FNIP1 and Aha1 to compete for binding to Hsp90, fine tuning its chaperone cycle. The competition between Aha1 and FNIPs binding to Hsp90 was also observed in renal tumours and their adjacent normal tissues. We found that Aha1 bound stronger to Hsp90 in normal tissues; however, overexpression of FNIPs appears to displace Aha1 from Hsp90 and increase Hsp90 binding to its inhibitors. Our data are in agreement with previously published work by Workman's laboratory 29 , suggesting that Aha1 dissociation from Hsp90 plays a role in drug sensitivity. What determines the binding of these proteins to Hsp90? We have already shown that posttranslational modifications of Hsp90 determine its binding to cochaperones 28 . In addition, our recent work has shown that c-Abl-mediated phosphorylation of Y223-Aha1 promotes its interaction with Hsp90 (ref. 19). We therefore suspect similar mechanisms exist between these three proteins, where cross-talk between posttranslational modifications is a determining factor for differential binding of Aha1 and FNIPs to Hsp90.

Methods
Plasmids and yeast strains. FLCN complementary DNA was synthesized in FLAG-tagged pcDNA3.1 vector backbone (Thermo Fisher Scientific) by GeneWiz, Inc. FNIP1-His 6 and FNIP2-His 6 messenger RNA was also synthesized in pcDNA3.1 by GeneWiz, Inc. and subcloned into pCMV-Myc and pCMV-HA vectors (ClonTech). The EVs containing these tags were used and also referred to as controls. For bacterial expression of FNIP1, FNIP2, FNIP1-D and Hsp90a, pRSET-A expression vector containing a 6 Â His tag (ThermoFisher Scientific) was used. The HA-FNIP1 domain-specific plasmids were previously reported 8 . Hsp90a domain-specific plasmids were amplified using primers listed in Supplementary Table 3. FNIP1-FLAG and FNIP2-FLAG were cloned in to yeast expression plasmid pYES2 using primers FNIP1-KpnI-FLAG-F/FNIP1-XhoI-R and FNIP2-KpnI-FLAG-F/FNIP2-XhoI-R.
Mammalian cell culture. Cultured cell lines human embryonic kidney (HEK293) grown in DMEM (Sigma-Aldrich), human bladder (T24) and human colorectal adenocarcinoma (HT29) grown in McCoy's 5a medium (Sigma-Aldrich), human prostate cancer (LNCaP) and human non-small cell lung cancer (H1299) grown in RPMI 1640 medium (Sigma-Aldrich) and human breast adenocarcinoma (MCF7) grown in Eagle's minimum essential medium (Sigma-Aldrich). Cultured cell lines were acquired from (American Type Culture Collection) and their media was supplemented with 10% fetal bovine serum (Sigma-Aldrich) and grown in a CellQ incubator (Panasonic Healthcare) at 37°C in an atmosphere containing 5% CO 2 .
Ex vivo culture and analysis of human RCC tumours. Tumour and adjacent normal tissues of the patients with conventional RCC were obtained with written informed consent from the Department of Urology at SUNY Upstate Medical University. At the time of radical or partial nephrectomy, which was done with o10 min of renal ischaemia, RCC tumours were dissected into B3 mm 3 pieces and protein was extracted and quantified as previously described in detail 22 .
Bacterial expression and purification of proteins. All proteins were expressed in Escherichia coli strain BL21 (DE3) and included an N-terminal 6 Â His tag. Purification buffers included 20-50 mM Tris or phosphate pH 8.0 and 10 mM b-mercaptoethanol. Chromatography resins were purchased from GE Healthcare Bio-Sciences (Marlborough, MA), except for Ni-NTA agarose, which was purchased from Qiagen (Valencia, CA). Transformed cells were grown at 37°C in lysogeny broth (LB) with 50 mg l À 1 ampicillin until OD 600 ¼ 0.6. For Hsp90a, cultures were then cooled to 20°C and induced with 20 mg l À 1 isopropyl-b-D-thiogalactoside overnight. Cells were harvested by centrifugation and lysed enzymatically. Hsp90a expressed in the supernatant and was isolated by sequential Ni-NTA metal affinity (10-250 mM imidazole step gradient), Q-Sepharose anion exchange (0-1 M NaCl gradient) and Superdex-75 size-exclusion chromatography. Purified Hsp90a was nucleotide free evidenced by an A 280/260 ratio of 1.83. For FNIP1-D, cells were induced with 20 mgl À 1 isopropyl-b-D-thiogalactoside at 37°C for 3 h, harvested by centrifugation and lysed by sonication. FNIP1-D expressed in inclusion bodies and was purified by washing the pellet 3 Â in buffer, dissolving the pellet in 8 M urea, subjecting the dissolved pellet to Ni-NTA metal affinity chromatography (10-250 mM imidazole step gradient) and refolding by rapid dilution. Proteins were 490% pure. Concentrations were determined using calculated extinction coefficients as previously described 19 . Proteins were flash frozen on dry ice and stored at À 80°C until use.