Hsp90-mediated Assembly of the 26 S Proteasome Is Involved in Major Histocompatibility Complex Class I Antigen Processing*

Heat shock protein 90 (hsp90) and the proteasome activator PA28 stimulate major histocompatibility complex (MHC) class I antigen processing. It is unknown whether hsp90 influences the proteasome activity to produce T cell epitopes, although association of PA28 with the 20 S proteasome stimulates the enzyme activity. Here, we show that hsp90 is essential in assembly of the 26 S proteasome and as a result, is involved in epitope production. Addition of recombinant hsp90α to cell lysate enhanced chymotrypsin-like activity of the 26 S proteasome in an ATP-dependent manner as determined by an in-gel hydrolysis assay. We successfully pulled down histidine-tagged hsp90α- and PA28α-induced, newly assembled 26 S proteasomes from the cell extracts for in vitro epitope production assay, and we found these structures to be sensitive to geldanamycin, an hsp90 inhibitor. We found a cleaved epitope unique to the proteasome pulled down by both hsp90α and PA28α, whereas two different epitopes were identified in the hsp90α- and PA28α-pulldowns, respectively. Processing of these respective peptides in vivo was enhanced faithfully by the protein combinations used for the proteasome pulldowns. Inhibition of hsp90 in vivo by geldanamycin partly disrupted the 26 S proteasome structure, consistent with down-regulated MHC class I expression. Our results indicate that hsp90 facilitates MHC class I antigen processing through epitope production in a complex of the 26 S proteasome.

Heat shock protein 90 (hsp90) and the proteasome activator PA28 stimulate major histocompatibility complex (MHC) class I antigen processing. It is unknown whether hsp90 influences the proteasome activity to produce T cell epitopes, although association of PA28 with the 20 S proteasome stimulates the enzyme activity. Here, we show that hsp90 is essential in assembly of the 26 S proteasome and as a result, is involved in epitope production. Addition of recombinant hsp90␣ to cell lysate enhanced chymotrypsinlike activity of the 26 S proteasome in an ATP-dependent manner as determined by an in-gel hydrolysis assay. We successfully pulled down histidine-tagged hsp90␣-and PA28␣-induced, newly assembled 26 S proteasomes from the cell extracts for in vitro epitope production assay, and we found these structures to be sensitive to geldanamycin, an hsp90 inhibitor. We found a cleaved epitope unique to the proteasome pulled down by both hsp90␣ and PA28␣, whereas two different epitopes were identified in the hsp90␣-and PA28␣-pulldowns, respectively. Processing of these respective peptides in vivo was enhanced faithfully by the protein combinations used for the proteasome pulldowns. Inhibition of hsp90 in vivo by geldanamycin partly disrupted the 26 S proteasome structure, consistent with down-regulated MHC class I expression. Our results indicate that hsp90 facilitates MHC class I antigen processing through epitope production in a complex of the 26 S proteasome.
CD8 ϩ T cells recognize peptides in the context of major histocompatibility complex (MHC) 2 class I molecules, through which specific immunity is activated against infectious agents, tumors, and even self-antigens. The majority of these 8 -10amino acid peptides are proteasome-degraded products of cellular proteins (1,2). Cell surface expression of MHC class I molecules can be down-regulated with loss of peptides in the endoplasmic reticulum through functional disruption of transporter associated with antigen-processing molecules or because of a diminished peptide pool in the cytosol via disruption of immunoproteasomal subunits (3)(4)(5). Therefore, generation of peptides by the proteasome is the first and indispensable step in the entire MHC class I antigen-processing pathway.
The flow of polypeptides in the context of pre-and postproteasomal events is not fully understood, although polyubiquitinated proteins to be degraded by the 26 S proteasome seem to be mainly derived from defective ribosomal products (6). Recently, large proteolytic intermediates (Ͼ30 kDa) harboring MHC class I epitope were found to associate with hsp90␣ in living cells (7), although there is no direct evidence that the associated intermediates are physiologically relevant in terms of being T cell epitopes by proteasome-dependent processing. Because hsp90 binds E3 ligase CHIP (C terminus of hsc70-interacting protein) to polyubiquitinate those large fragments, we may speculate that hsp90 can serve as a bridge between the bound substrates and CHIP for degradation by the 26 S proteasome. In this case, hsp90 large intermediate complex is free from the proteasome complex, at least until the substrates are ubiquitinated. However, short peptides bound to hsp90 which are not to be ubiquitinated are efficiently processed to be T cell epitopes in a proteasome-dependent manner (8). In addition, we found that hsp90 stimulates MHC class I epitope production from C-but not N-terminally extended short synthetic peptides (13 ϳ 22-mer) in a proteasome-dependent manner in vivo (9). Because ubiquitination of those short peptides is unnecessary to be processed by the proteasome, how hsp90 is involved in proteasome-dependent epitope production has still remained elusive. There might be an additional mechanism of hsp90 involved in epitope processing. Mass spectrometry analysis revealed a direct association between hsp90 and the 26 S proteasome through 19 S cap (10), but the physiological significance of such an association remains unknown.
The aim of the present study was to determine the effects of hsp90␣ on the function and structure of the proteasome in terms of epitope production in vivo and in vitro, compared with those of PA28. Our results showed that hsp90␣, by stimulating assembly of the 26 S proteasome, is directly involved in production of epitopes recognized by CD8 ϩ T cells.
Recombinant Proteins-Plasmid expressing human hsp90␣ was kindly provided by T. K. Nemoto (Nagasaki University School of Dentistry). Protein expression was induced by 1 mM isopropyl-␤-D-thiogalactoside and purified with nickel-nitrilotriacetic acid (Ni-NTA) as recommended by the manufacturer (Qiagen).
Peptidase Assay-Peptidase activity was measured using suc-LLVY-amc (chymotrypsin-like activity). Substrate (0.1 mM) was incubated with whole-cell extracts (6 g of PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells), and peptidase activity was estimated by fluorescence (excitation at 380 nm, emission at 460 nm). In-gel hydrolysis assays were performed by irradiation of the native-PAGE with 360-nm UV light and detection by the 460-nm filter as described (9).

Preparation of Cell Lysates and Pulldown of the Proteasome-EL4 cells or PA28
Ϫ/Ϫ ␤ Ϫ/Ϫ cells were lysed with the 26 S buffer (25 mM Tris-HCl, pH 7.5, 250 mM sucrose, 1 mM dithiothreitol, and 1 mM phenylmethylsulfonyl fluoride) containing 1% Nonidet P-40 with 0.2 mM ATP. After centrifugation at 16,000 ϫ g for 10 min to remove insoluble materials, the supernatant was used as whole-cell extract. To inhibit the ATPase activity of endogenous hsp90, 12.5 M GA was added to the cell extract followed by incubation for 30 min at room temperature. To chelate Mg 2ϩ , 0.1 mM EDTA was added. Whole-cell extract was also used for proteasome pulldowns. PA28␣ (5 g) or hsp90␣ (5 g) was mixed with whole-cell lysate (250 g/100 l) derived from 1 ϫ 10 6 EL4 cells or 5 ϫ 10 5 PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells and incubated at 4°C for 3 h. Ni-NTA-agarose (100 l, 30% volume) was added for the final hour. Ni-NTA-agarose was washed twice with 26 S buffer containing 25 mM imidazole by centrifugation at 2,500 ϫ g for 5 min, and PA28␣or hsp90␣-associated proteasome was eluted with 100 l of 26 S buffer containing 250 mM imidazole.
Quantification of MHC Class I Peptide Produced by the Proteasome-C-terminally extended, MHC class I precursor peptides (4 nmol) were digested with the proteasomes (10 l derived from 25 g of lysate) pulled down by PA28␣ and hsp90␣ in 26 S buffer (total 100 l), in the presence of 0.2 mM ATP, with or without 12.5 M GA or 10 M epoxomicin at 37°C for 3 h. After the addition of 1% trifluoroacetic acid (10% volume) and centrifugation at 16,000 ϫ g, the soluble material containing the digested peptide (10 l) was used for pulsation on the target cells for CTL assay. Serum-free plain RPMI 1640 medium was used throughout the entire CTL assay to avoid serum peptidase activity that might induce peptide cleavage during the assay. MHC class I epitopes, OVA 257-264 (K b ), TRP2 180 -188 (K b ), and CSP 281-289 (K d ) (10 Ϫ9 to 10 Ϫ14 M) were used for pulsation on the target cells to determine the sensitivity of CTLs.
Loading of Peptides and Antigen Presentation Assay-Cells (2 ϫ 10 6 ) were incubated for 90 min in the presence or absence of 5 M GA or 50 M LC in serum-free RPMI. Osmotic introduction of peptides and proteins, followed by CTL assay, was performed as described (9).

Three Distinct Patterns Mediated by Hsp90 and PA28 in the
Processing of Peptides in Vivo-For in vivo antigen presentation assay, we used osmotic shock introduction of antigenic peptides, as shown (9). To compare the function of hsp90 with PA28, we prepared six different T cell epitopes spanning the cognate C-terminal flanking regions. We osmotically loaded these synthetic peptides, OVA 257-269 (K b ) and TRP2 180 -193 , CSP 279 -296 (K d ), and influenza A NP 147-160 (K d ) into P815 (H-2 d ) cells together with hsp90 or recombinant PA28␣. The efficiency of the processing of each epitope was assessed by cytolysis of the loaded cells with specific CTLs. The processing of all peptides was completely inhibited by pretreatment of cells with a proteasome inhibitor, LC ( Fig. 1 and data not shown for JAK1 355-368 , pRL1b-C5 and influenza A NP 147-160 ). We observed the following. First, the processing was enhanced by both PA28␣ and hsp90. This pattern could be applied to OVA 257-264 (and JAK1 355-363 , pRL1a, an exact epitope of pRL1b-C5, data not shown) (Fig. 1, A and D). GA treatment of EL4 partially suppressed the processing (Fig. 1D). Second, the processing of TRP2 180 -188 depended solely on PA28␣ (Fig. 1, B and E), and neither hsp90 nor its specific inhibitor, GA, affected the processing (Fig. 1E). Third, like CSP 281-289 (and influenza A NP 147-155 , data not shown), the processing was enhanced by hsp90, but not by PA28␣, and it was profoundly inhibited in P815 cells treated with GA ( Fig. 1, C and F). Thus, processing of antigen peptides can be categorized into three patterns based on PA28 and hsp90 dependence, either on one or the other or on both. There are apparently distinct molecular mechanisms underlying the roles of hsp90 and PA28. Thus, hsp90 is a molecule independent of PA28 and actively participates in the processing, together with the proteasome.
Next, interaction of hsp90 and PA28␣ with the proteasome was investigated. We used recombinant histidine-tagged hsp90␣ and PA28␣ throughout the following experiments. Hsp90␣ or PA28␣ was incubated for 30 min with extracts of PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells in which the effect of endogenous PA28 could be ignored, and the mixtures were applied to native-PAGE gel, followed by in-gel hydrolysis assay. The addition of hsp90␣ stimulated the hydrolysis of suc-LLVY-amc at singly capped 26 S proteasome regulatory particle-core particle (RC) and/or the doubly capped 26 S proteasome, regulatory particle-core particle-regulatory particle (RCR) level, and both EDTA and GA abrogated the stimulatory effect (Fig. 2B, lanes 3). PA28␣ stimulated the formation of the homo-PA28 as well as hybrid proteasome, which was not suppressed by EDTA and GA (Fig. 2B, lanes 2). The kinetics of the proteasome activity was also shown (Fig. 2C). The results indicate that the molecular mechanism through which hsp90␣ stimulates the proteasome might be quite different from that for PA28␣.
To visualize the migration of recombinant PA28␣ and hsp90␣, mixtures of those proteins and PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cell extracts were applied to native-PAGE followed by in-gel hydrolysis assay and Western blotting with the appropriate antibodies. Because of apparently distinct migratory positions,  RC was clearly distinguished from the hybrid proteasome resulting from just association of PA28␣ to RC (Fig. 2D, upper left panel). PA28␣ migrated to the level of the hybrid and homo-PA28 -20S proteasomes, recognized by histidine immunoreactivity (Fig. 2D, upper panels). Hsp90␣ migrating to the RC was visible also by histidine reactivity (Fig. 2D, lower panels, lanes 2  and 3), although the RCR level was not clear, probably because of the sensitivity limits of the Western blotting assay being reached. Western blotting with antibodies against the 20S␣2 or Rpt6 subunit of the 19S cap showed increased levels of RC and RCR after the addition of hsp90␣, and the increment was suppressed in the presence of EDTA and GA (Fig. 2D, lower panels). EDTA did not affect the binding of hsp90␣ to RC, but it inhibited the quantitative increase of RC and RCR as revealed by Western blotting (Fig. 2D, lower panels, lanes 3). These results correlated well with the increased hydrolytic activity of the hsp90␣-stimulated proteasome (Fig. 2D, lower left panel), indicating that hsp90␣ stimulates the assembly of the RC (and RCR) in an ATP hydrolysis-dependent manner. This is consistent with a previous study showing in yeast that reassembly of the heat-disrupted proteasome depended, at least in part, on hsp90 (19). Notably, GA inhibited the association of hsp90␣ and RC and forced it to migrate to the position of the 20 S proteasome or the lower (Fig. 2D, lower panels, lanes 4). Intriguingly, a significant portion of Rpt6 was apparently separated from proteasome by GA treatment (Fig. 2D, lower panels,  lanes 4), suggesting association of this molecule with hsp90.
Next, we pulled down the proteasome complex with recombinant proteins from cell extracts. From EL4 cell extracts, PA28␣ pulled down homo-PA28-20S and hybrid proteasomecontaining endogenous hsp90 (Fig. 3A). Hsp90␣ isolated mainly the hybrid proteasome (and/or RC) containing endogenous PA28 (Fig. 3B). From extracts of PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells, hsp90␣ pulled down the RC (Fig. 3D), whereas PA28␣ pulled down homo-PA28-20S and hybrid proteasomes containing endogenous hsp90 (Fig. 3C). Because all of the hybrid proteasomes (and/or RC) contained hsp90 derived from either endogenous hsp90 (Fig. 3 A and C) or recombinant protein per se (Fig. 3 B  and D), hydrolysis of suc-LLVY-amc by those proteasomes was examined in the presence or absence of GA. Intriguingly, GA down-regulated all activities of the hybrid proteasome (and/or RC, RCR) but not those of the homo-PA28-20S proteasome (Fig. 3,  right panels). The GA-induced undermined activity was attributed to impaired structures of the proteasomes as determined by Western blotting with an anti-20S␣2 subunit antibody (Fig. S1).
Generation of a Unique Epitope by Proteasome Pulled Down by Hsp90␣-To answer a question about whether the proteasome complexes produce a suitable T cell epitope for sensitization of specific CTL, we performed peptide digestion assay using synthetic peptides used in Fig. 1. The peptides were incubated with the proteasome obtained from extracts of the PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells in Fig. 3 C and D, with or without GA and epoxomicin, a specific inhibitor of the proteasome. The proteasome pulled down by PA28␣ or hsp90␣ contained both active PA28 and hsp90 that could locate on the opposite sides of the 20 S proteasome; PA28␣ directly on one ␣-ring of the 20 S and hsp90␣ on the other, possibly in a complex with the 19 S cap regulatory particle. The digested peptides were diluted and then pulsed onto EL4 cells or P815 cells. The quantities of the produced epitopes were evaluated by cytolysis of target cells. The sensitivity of CTLs used in this study is also shown (Fig. 4 G-I). The proteasome pulled down by PA28␣ produced epitopes that could sensitize CTLs from all three peptides ( Fig.  4 A, C, and E). GA completely inhibited the epitope production from CSP 279 -296 ( Fig. 4E) but had no inhibitory effect on TRP2 180 -193 (Fig. 4C). The proteasome obtained by hsp90␣ produced epitopes from OVA 257-269 (Fig. 4B) and CSP 279 -296 (Fig. 4F) but not from TRP2 180 -193 (Fig. 4D). GA again induced complete inhibition on CSP 279 -296 (Fig. 4F) and partial suppression on OVA 257-269 (Fig. 4B). The results indicate that hsp90␣ associated with the proteasome contributes to the production of some but not all T cell epitopes from C-terminally longer precursors and that this is blocked by GA.
GA Treatment of Cells Transiently Disrupted 26 S Proteasome Structures-To examine the role of hsp90 in vivo, we treated EL4 and PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells with GA for the indicated hours as in Fig. 5. Toxicity of GA was monitored by counting cell number and viability in each chase time (Fig. S2). Both RC and RCR expressions were significantly decreased, as indicated by in-gel hydrolysis assay and/or Western blot analysis (Fig. 5 A  and B). The results indicated that inhibition of hsp90 transiently but significantly disrupted at least some, if not all, RC and RCR structure. The cell surface expressions of K d , D d , and L d (PA28␣ Ϫ/Ϫ ␤ Ϫ/Ϫ cells in BALB/c background) after a 10-h GA treatment were then examined by FACS analysis. GA down-regulated all MHC class I molecules, and pulsation of CSP 281-289 restored the K d expression (Fig. 5C).

DISCUSSION
In the present study, we observed that epoxomicin, a highly specific proteasome inhibitor, abrogated the production of CTL epitopes from C-terminally extended peptides by the proteasome affinity-isolated by hsp90␣, indicating no association of hsp90␣ with cytosolic peptidases such as tripeptidyl protease, which acts downstream of the proteasome to process certain peptides independently (18). Recently, both C-and N-terminal extended large peptides harboring T cell epitope (to be processed by the proteasome) were found to be associated with hsp90 (7). In addition, N-terminal extended versions of T cell epitopes (peptides once processed by the proteasome) were shown to be associated with hsp90 in vivo (19,20). The peptide carrier effect of hsp90 at both pre-and postproteasomal events may be important in peptide traffic in the cytosol. Our results provide an additional function of hsp90 in MHC class I antigen processing; thus, hsp90␣ is linked with the assembly of the 26 S proteasome.
We presented here evidence for the direct involvement of hsp90␣ in epitope production by the 26 S proteasome, whose structures were sensitive to GA. In vivo processing of CSP 281-289 depends on hsp90 but not on PA28, whereas that of TRP2 180 -188 depends on PA28 but not hsp90. Furthermore, processing of OVA 257-264 depends on both PA28 and hsp90 (Fig. 1). In vitro production of those epitopes exactly depends on the same protein combinations used for the proteasome pulldowns (Fig. 4). We do not know the reason why the processing pattern is different between hsp90 and PA28. Hsp90 binds short peptides, probably with some preference for partic-  Fig. 3, C and D. The digestion mixtures were then serially diluted and pulsed onto EL4 cells (for OVA 257-269 and TRP2 180 -193 ) or P815 cells (for CSP 279 -296 ). The pulsed cells were used as targets for the CTL assay (mean Ϯ S.E.; n ϭ 3). Right panels (G-I), the sensitivities of CTLs used were determined by sensitization assay using the serially diluted exact epitopes whose concentrations are indicated as 10 ϪX M (mean Ϯ S.E.; n ϭ 3). The all-CTL assays were performed in plain RPMI 1640 medium to avoid serum peptidase activity. X indicates the cytolysis of target cells pulsed with 0.4 nmol of each peptide and incubated with CTLs, confirming no significant peptidase activity involved in this assay system. The results shown are representative of three separate experiments. ular amino acids, which might determine peptides to be processed. With this point of view, because hsp90 itself could bind OVA 257-269 , TRP2 180 -193 , and CSP 279 -296 ( Fig. 2A), the binding of peptides to hsp90 per se would not account for the independence of TRP2 180 -193 on hsp90. Alternatively, hsp90␣-(in a complex with 19 S cap) or PA28␣-driven proteasomal activity might be different and involved in their unique epitope production, which is our most preferred hypothesis. Indeed, peptide repertoires produced from Insulin-like growth factor by the 26 S and 26 S plus PA28 is definitely different, even though the 20 S proteasome is same (21).
Recent observations suggest that the 26 S proteasome is a dynamic machine that disassembles into the 20 S and 19 S cap subcomplexes during the catalytic cycle rather than remaining as a stable complex (22). Based on this scenario, the 26 S proteasome must reassemble immediately after substrate degradation. The observation gave us a view that the steady-state 26 S proteasome is net-balanced upon disassembly and reassembly in catalytic cycle in living cells. GA may impair the balance by suppressing the reassembly of the 26 S proteasome after proteolysis, which down-regulates the level of the proteasome. The down-regulated expressions of RC and RCR by GA treatment (Fig. 5) supported an idea that hsp90 is involved in maintaining the steady-state level of the 26 S proteasome. We must note the following possibility that the proteasome pulled down by hsp90␣ is an intermediate form assembling closely into the fully matured 26 S proteasome; therefore, its structure is relatively fragile and sensitive to GA.
Hsp90 is clearly not the sole molecule that facilitates assembly of the 26 S proteasome because GA-induced down-regulation of the 26 S proteasome was transient up to 3-6 h after starting the treatment and thereafter observed a vigorous rescue of the activity, as determined by native-PAGE (Fig. 5A). Therefore, additional molecules must play critical roles in reassembly of the 26 S proteasome, which need to be clarified in the future experiments.