Two Distinct Pathways Mediated by PA28 and hsp90 in Major Histocompatibility Complex Class I Antigen Processing

Intracellular or endogenous antigens are processed by the proteasome, generating MHC class I ligands (equivalent to CTL epitopes) recognized by CD8⁺T cells (1–3). The 20S proteasome consists of two outer α-rings and two inner β-rings, stacked in the order of αββα. Both rings comprise seven α- or β-subunits. Substrate proteins are thought to go through the α-ring of the proteasome in order to reach the internal β-ring cavity where the catalytic sites are located. As the pore of the α-ring is very narrow or almost closed, substrate access is usually prevented (4), and thus the 20S proteasome by itself cannot generate peptides efficiently. To operate active proteolysis, the enzyme needs to associate with regulatory complexes to allow proteins or peptides to access the catalytic sections.

To date, two types of regulatory complexes have been identified in the cell. One is the 19S regulatory complex (also called PA700), comprising multiple subunits including six ATPases, which binds to the 20S proteasome to form the 26S proteasome capable of degrading ubiquitinated proteins in an ATP-dependent manner (5). The other is PA28 or the 11S regulator (REG) that greatly stimulates the peptidase activity of the 20S proteasome but does not assist the degradation of large proteins, even if they have already been ubiquitinated (6). PA28 is composed of two homologous subunits, PA28α and PA28β.

These two subunits assemble into a heterohexamer (α3β3) with alternating α and β subunits (7), or a heteroheptamer (α3β4 or a mixture of α3β4 and α4β3) (8) with a mass of ∼200 kD. PA28 is known to be directly associated with the 20S proteasome, forming the ‘football-shaped proteasome’ in which PA28 is attached to both ends of the central 20S proteasome or the ‘hybrid proteasome’ comprising the 20S proteasome armed by PA28 on one side and PA700 on the other (2, 9).

Evidence indicates that PA28 accelerates the in vitro processing of MHC class I ligands from their polypeptide precursors by the 20S proteasome in a coordinated dual-cleavage manner (10, 11). In vivo analysis has shown also that overexpression of PA28α subunit enhances the processing of some viral epitopes in cells (12). Moreover, it was reported that mice deficient in the PA28β gene have defective production of CTLs after viral infection, probably due to impaired assembly of the ‘immunoproteasome’ (13), a modified version of the 20S proteasome whose three constitutively expressed β-type subunits (X/MB1, Y/delta, and Z) are replaced by IFN-γ–inducible subunits (low-molecular weight polypeptide [LMP]7, LMP2, and multicatalytic endopeptidase complex-like [MECL]-1, respectively) (2, 14).

These findings indicate that PA28 plays an important role in the processing of MHC I class ligands. In contrast, we recently showed that induction of the immunoproteasome by IFN-γ is not impaired in PA28α^−/−/β^−/− cells and that specific CTLs are normally generated against OVA_257–264 epitope, and influenza virus A (PR-8) derived K^b- and D^b-restricted NS2 and NP antigen peptides in these knockout mice, whereas processing of an epitope derived from a tumor antigen of murine B16 melanoma, tyrosinase-related protein (TRP2)^*_181–188 is entirely dependent on PA28 (15). These observations suggest that PA28αβ is not a prerequisite for antigen processing in general, but plays an essential role in the processing of certain antigens. Hence, it is likely that there exists a compensatory system(s) for PA28 function in the processing of antigens, but the entity remains unknown so far.

The molecular chaperone hsp90 is one of the most abundant proteins in eukaryotic cells, comprising 1–2% of total cellular proteins even in conditions of nonstress. Hsp90 is an evolutionarily conserved protein and contributes to a wide variety of fundamentally common and species-specific processes in cells (16–18). For example, it is essential for maintenance of functional integrity of various fragile proteins, such as steroid hormone receptors and many of protein kinases (18).

It is also notable that hsp90 functions as a protein-folding machinery collaborating with other chaperone molecules, such as Hsp70 and hsp40, and cochaperones containing p23 and Hop (16, 17). Indeed, Hsp90 can bind nonnative proteins through N- and COOH-terminal domains for refolding (17). In addition, hsp90 appears to be closely linked to the protein degradation in the cell. In fact, hsp90 directly associates with the 20S proteasome, and thus can influence the enzyme activity (19, 20). Recent studies have also shown that hsp90 links misfolded proteins of aberrant structures to the ubiquitination pathway for selective elimination (21). It is noteworthy that the COOH-terminal domain is involved in binding the antigenic octapeptide of vesicular stomatitis virus G protein (22).

Evidence also suggests that hsp90 binds to tumor-associated MHC class I ligands or their precursors (23). The latter part of these observations strongly suggests that hsp90 might be involved in MHC class I antigen presentation, but the molecular basis of such role is still unclear. Using x-ray crystallographic analysis, Prodromou et al. (24) indicated that hsp90 comprises a deep pocket containing binding sites for geldanamycin (GA), an anti-cancer agent, as well as ATP/ADP at the NH₂-terminal region. Therefore, the ATPase and hence chaperone activity of hsp90 is inhibited by ansamycin drugs such as GA or herbimycin A (HA), indicating that GA and HA could serve as hsp90-specific inhibitors (25).

Here, we examined the roles of PA28 and hsp90 in the MHC class I–restricted antigen processing using GA and HA (inhibitors of hsp90) and cells derived from mice lacking PA28α and PA28β. We show that both PA28 and hsp90 accelerate the processing of C- but not NH₂-terminal flanking regions of MHC class I ligands.

Importantly, the results showed that hsp90 is essential for compensating the loss of PA28 function in LPS blasts of PA28α^−/−/β^−/− mice, resulting in nearly normal antigen presentation of OVA_257–264. In addition, we show that IFN-γ–induced up-regulation of MHC class I is defective in PA28α^−/−/β^−/− macrophages and dendritic cells. Our data provide novel insights into the roles of PA28 and hsp90 in MHC class I antigen processing.

Hsp90, hsc70, and gp96 were purified from ascitic fluid from mice injected with RL ♂ 1 cells as described previously (23). cDNA of murine PA28α was amplified by reverse transcription (RT)-PCR from the mRNA of EL4 cells and cloned into 5′ BamHI and 3′ KpnI sites of the pQE31 expression vector (QIAGEN). The protein expression was induced in Escherichia coli strain M15 and purified as described previously (26). The OVA_257–269 (SIINFEKLTEWTS), OVA_248–264 (EVSGLEQLESIINFEKL), and OVA_248–269 (EVSGLEQLESIINFEKLTEWTS) peptides were synthesized by standard solid phase methods using F-mock chemistry in a peptide synthesizer (AMS422; Applied Biosystems) and purified by reversed-phase HPLC on a C8 column. TRP2_181–193 (VYDFFVWLHYYSV) was purchased from SAWADY Technology Co. GA was purchased from Sigma-Aldrich. HA was provided by Dr. Y. Uehara, the National Institute of Infectious Diseases, Tokyo, Japan.

E.G7 is an OVA cDNA transfected EL4 line (27). OVA_257–264 and TRP2_181–188 specific CTLs were induced from spleen cells of mice immunized with these peptides fused to hsc70 as described previously (26) and maintained by weekly stimulation with E.G7 and B16 melanoma cells, respectively, in the presence of syngeneic feeder cells and IL-2. LPS blasts and murine embryonic fibroblasts (MEFs) were prepared from PA28α^+/+/β^+/+ and PA28α^−/−/β^−/− mice, as described previously (15).

Mini-genes encoding OVA_257–269, OVA_248–264, and OVA_248–269 were cloned into pMSCVhyg (CLONTECH Laboratories, Inc.). PT67 packaging cells were transfected with 10 μg of these constructs by DOTAP Liposomal Transfection Reagent (Boehringer). The cells were selected by 300 μg/ml hygromycin for one week to obtain stable virus-producing cell lines. The virus titers were 1–3 × 10⁷ CFU/ml for OVA mini-genes, as evaluated by culture with NIH3T3 cells. These supernatants were used for transfection analysis. cDNA of human hsp90 was obtained by RT-PCR from mRNA of peripheral blood mononuclear cells. For transfection of genes of PA28α or hsp90, cDNAs were cloned into 5′ HpaI and 3′ EcoRI sites for PA28α and XhoI site for hsp90 of pMSCVpuro. (CLONTECH Laboratories, Inc.) and virus producing PT67 cells were selected by 2 μg/ml puromycin. E.G7 cells were transfected by these retrovirus vectors and stable cell lines expressing the molecules were established as E.G7 PA28α and E.G7 hsp90 and E.G7 mock cells, then maintained in the presence of 5 μg/ml puromycin.

Osmotic introduction of peptides or proteins into EL4 cells was performed as described previously (26). Briefly, 2 × 10⁶ pelleted cells were suspended in 200 μl warm (37°C) hypertonic buffer (0.5 M sucrose in 10% wt/vol polyethylene glycol 1,000 in RPMI) with or without synthetic peptides and 100 μg of hsp90, hsc70, gp96, or recombinant PA28α, and incubated for 10 min. Then, 15 ml of warm hypotonic buffer (RPMI 1640/dH₂O: 60%) was added immediately followed by a 2-min incubation. After centrifugation, the cells were washed twice and further incubated in the presence or absence of GA (5 μM), HA (5 μM), or lactacystin (LC; 50 μM) in serum-free RPMI for 3 h at 37°C, under 5% CO₂. For retroviral expression of OVA peptides, 5 × 10⁵ EL4 cells were transfected for 3 h by retrovirus vector encompassing mini-genes encoding OVA_257–269, OVA_248–264, and OVA_248–269 at the doses indicated in the figure legends, and then the virus was washed off. These cells were labeled by 3.7 MBq Na₂⁵¹CrO₄ (NEN Life Science Products) and used for standard ⁵¹Cr-release assay. The CTL assay was performed in the presence of brefeldin A (BFA) to block the egress of newly assembled MHC class I molecules from the endoplasmic reticulum to the cell surface. In all cases, control cultures were incubated in DMSO at a concentration equivalent to those in the inhibitor preparations.

After disruption of the cells by sonication and centrifugation at 100,000 g for 1 h, the resulting supernatant and precipitate were used as the cytosol and membranous fractions, respectively. The resultant total materials were dissolved with sample buffer for SDS-PAGE and run on a SDS-PAGE gel, blotted onto a PVDF membrane, and probed with anti-hsp90 mAb (rat IgG, SPA-835, clone 16F1; StressGen Biotechnologies), anti-β₂ microglobulin (M-20: affinity purified goat polyclonal antibody; Santa Cruz Biotechnology, Inc.), anti-hsp70 mAb (rat IgG, SPA-815, clone 1B5; StressGen Biotechnology) and anti-grp94 (gp96) mAb (rat IgG, SPA850; StressGen Biotechnology), and anti-PA28α, β rabbit polyclonal antibodies (15). Biotin-conjugated goat anti–rat Ig(Fc) or rabbit IgG (H+L) was used as secondary Ab (Jackson ImmunoResearch Laboratories) and peroxidase-conjugated avidin was used for visualization. The quantity of each protein was assessed by densitometric analysis of ECL-exposed films. Polyclonal antibodies against PA28α and PA28β and transporter associated with antigen processing (TAP)2 were produced as serum from rabbits immunized with synthetic peptides, human PA28α_22–37 (CTKTENLLGSYFDKKI) and PA28β_3–18 (KPCGVRLSGEARKQVE) and human TAP2_688–703 (CCQEGQDLYSRLVQQRLMD), respectively, and purified with peptide bound affinity columns.

RT-PCR was used for quantification of transcripts of K^b and its control glyceraldehyde-3-phosphate dehydrogenase (G3PDH). For the amplification of K^b transcript, forward primer CCGAATTCATGGTACCGTGCACGCTGC and reverse primer CCTCTAGATCACGCTAGAGAATGAGG were used. For the amplification of G3PDH, ACCACAGTCCATGCCATCAC and TCCACCACCCTGTTGCTGTA for forward and reverse primers were used.

Cells were incubated for 30 min with ascites obtained from 25.D1–16 hybridoma (murine IgG1, specific for K^b-OVA_257–264 [28], a gift from Dr. R. Germain, National Institutes of Health, Bethesda, MD) at 4°C, then washed twice followed by staining with FITC-conjugated rabbit anti–mouse IgG (H+L; Jackson ImmunoResearch Laboratories) for 30 min at 4°C. For staining of K^b molecules, biotin-conjugated anti–mouse K^b mAb derived from CTKb (Cedarlane Laboratories) and FITC-conjugated streptavidin (Jackson ImmunoResearch Laboratories) were used. For staining of LFA-1, M17/4.2 (rat IgG2a) and FITC-conjugated mouse anti–rat Ig(H+L) were used. The acid wash recovery assay was performed, as described previously (29). Briefly, cells were treated for 1.5 min with acid (1:1 mixture of 0.263 M citric acid and 0.132 M NaH₂PO₄ at pH 3.0), neutralized to pH 7.5 by adding 1 ml of 0.15 M NaH₂PO₄ and washed twice with PBS. The treated cells were further incubated for 3 h with GA or OVA_257–264 in the presence or absence of IFN-γ before FACS^® analysis. Flow cytometric analysis was performed with FACScan™ (Becton Dickinson), and the data were analyzed by CELLQuest™ (Becton Dickinson). The experiments with FACScan™ were performed several times and results of one representative experiment are shown. The relative fluorescence was calculated after subtraction of the fluorescence of control (staining with only second antibody or FITC-conjugated streptavidin).

To monitor MHC class I antigen processing pathway, we used a well-characterized K^b-restricted OVA_257–264 dominant antigen. Graded doses of synthetic peptides extending COOH-terminally (OVA_257–269), NH₂-terminally (OVA_248–264), or both sides of the epitope (OVA_248–269) were osmotically introduced into EL4 cells. Alternatively, these peptides were ectopically expressed by infection of a retrovirus encoding each of these peptides. Then, antigen presentation was measured by a standard ⁵¹Cr-release CTL assay using CTLs specific for OVA_257–264. As shown in Fig. 1 a, cytolysis was dependent on the dose of peptides and on the virus titers. Cytolysis was observed in RMA cells, but not in TAP-2-mutated RMA-S cells (Fig. 1 b), indicating that the presentation of OVA_248–269 is TAP-dependent. In addition, a proteasome inhibitor LC completely abolished the killing of EL4 cells expressing OVA_257–269 and OVA_248–269, but not OVA_248–264 (Fig. 1 b), indicating that the proteasome is involved in the processing of COOH- but not NH₂-terminally extended peptides, as reported previously (30).

Next, we examined the roles of hsp90 and PA28 on OVA antigen presentation. Loading of hsp90, rPA28α, gp96, and hsc70 by osmotic shock resulted in ∼2-3-fold increase of each protein in both the cytosol and membranous fractions of the EL4 cells compared with nonloaded cells (Fig. 2 a). OVA_248–269 was introduced with PA28α or hsp90 into EL4 cells by osmotic shock, and the presentation was assessed by CTL assay. Cytolysis was enhanced by hsp90 or PA28α but not by gp96, hsc70, or BSA (Fig. 2 b). We then examined the effects of GA or HA. Pretreatment of hsp90 with GA or HA in vitro abrogated the effect of hsp90, indicating the ATPase activity of hsp90 is required for hsp90-mediated antigen presentation. Although gp96, which is an endoplasmic reticulum (ER) resident member of Hsp90 family, is known to have a peptide chaperoning property similar to hsp90, it did not show any effect, suggesting that hsp90-assisted presentation is not simply due to its activity as a molecular chaperone that may protect the peptides from nonspecific degradation. It was previously reported that hsp90-peptide complex is more active in antigen presentation than free peptides when loaded into cells, presumably due to its chaperone activity (31). To minimize the effect of hsp90 as a chaperone, we sequentially loaded peptide and hsp90 (first peptide, second hsp90), and cytolysis by CTLs was examined. As shown in Fig. 2 c, sequential loading obviously enhanced the presentation of the peptide compared with loading peptide alone and was comparable to that of admixing of Hsp90 and peptides. This result also suggests that hsp90 enhances antigen presentation in a novel fashion other than chaperoning peptides.

To further analyze the effects of PA28α and hsp90, we introduced the three OVA peptides with or without PA28α and hsp90 and measured antigen presentation by quantifying the cell-surface K^b-OVA_257–264 complex using mAb 25D.1–16 (28) as a function of time (at 0, 4, and 8 h). As shown in Fig. 3, a and b, when COOH-terminally extended peptides (OVA_257–269, OVA_248–269) were introduced without PA28α or hsp90, the cell surface expression of K^b-OVA_257–264 complex reached the maximum level 8 h after introduction, whereas cointroducing PA28α or hsp90 caused rapid presentation of the complex, resulting in maximum expression level after 4 h, although their intensities were nearly equal to each other. In contrast, NH₂-terminally extended peptide, OVA_248–264, was processed and presented regardless of the presence or absence of PA28α or hsp90. 12 h after peptide loading, the expression of K^b-OVA_257–264 complex in each case decreased (data not shown). Subsequently, we also performed CTL assay. Three peptides were introduced into EL4 cells either by osmotic shock or retrovirus infection together with rPA28α or hsp90. rPA28α (Fig. 3 c) or hsp90 (Fig. 3 d) enhanced the cytolysis of cells expressing COOH- but not NH₂-terminally extended peptides. These enhancements by both proteins were completely abolished when EL4 cells were pretreated with LC, suggesting that hsp90 as well as rPA28α are functionally linked to the activity of the 20S proteasome. As shown in Fig. 3 e, PA28α and hsp90 appreciably augmented the processing of OVA_257–269, at any concentration of the peptides, while they had no effect on the processing of OVA_248–264 even at a limiting concentration. These data indicate that PA28 and hsp90 accelerate the processing of the COOH-terminally extended OVA peptides by the proteasome without changing its substrate specificities. It is of note that hsp90 did not affect the presentation of NH₂-terminally extended peptide, precluding again the effect of hsp90 as peptide chaperone and indicating that hsp90 acts through the function of the proteasome as PA28.

We next examined the epistasis of PA28- and hsp90-mediated pathways. We previously reported that the generation of OVA_257–264 epitope is not affected in PA28α^−/−/β^−/− LPS blasts (15) and expression of total K^b of LPS blasts derived from PA28α^+/+/β^+/+ and PA28α^−/−/β^−/− mice is almost comparable (data not shown), suggesting that there exists an alternative pathway for antigen processing, which compensates for the loss of PA28 in PA28α^−/−/β^−/− LPS blasts. To test whether hsp90 acts in place of PA28 in these cells, LPS blasts from PA28α^+/+/β^+/+ and PA28α^−/−/β^−/− mice were treated for 1.5 h with GA before osmotic loading of the three OVA peptides, and antigen presentation was monitored by CTL assay. Consistent with our previous report (15), the generation of OVA_257–264 epitope was not influenced by the absence of PA28. Furthermore, GA had no effect on PA28α^+/+/β^+/+ cells. Surprisingly, the processing of COOH-terminally extended peptides was profoundly inhibited by GA in PA28α^−/−/β^−/− blasts (Fig. 4 a, top and middle panels). GA had no effect on the presentation of NH₂-terminally extended peptide (OVA_248–264) irrespective of the presence of PA28αβ. To examine whether the role of hsp90 is restricted in PA28α^−/−/β^−/− LPS blasts or generally observed in other cells, we assessed the OVA epitope presentation in EL4 cells pretreated with GA or HA. As shown in Fig. 4 a (bottom panel), partial inhibition by GA or HA was observed in EL4 cells loaded with OVA_257–269 or OVA_248–269, but not in cells loaded with OVA_248–264.

To verify that the loaded peptide OVA_248–264 were processed and presented internally in cells, we treated LPS blasts and EL4 cells with BFA during antigen processing and performed CTL assay. BFA treatment completely inhibited the presentation of the peptide, and pulsing of the epitope peptide to BFA treated cells restored the cytolysis (Fig. 4 b). The processing of COOH-terminally extended peptides OVA_257–269 or OVA_248–269 was also inhibited by BFA (data not shown) like LC (Fig. 3, c and d), indicating that all peptides used were processed and presented internally in cells.

These results indicate that the hsp90-dependent processing pathway is common, but the contribution of hsp90 in antigen processing seems to depend on the cell type. As reported previously (15), PA28α^−/−/β^−/− LPS blasts almost completely lost the ability to process a melanoma antigen TRP-2 peptide (TRP2_181–188), indicating that PA28 is definitely required for this antigen presentation. Considering the partial suppression by GA or HA for OVA_257–264 presentation in EL4 cells, it is interesting to test how PA28 or hsp90 affects the TRP2_181–188 presentation in these cells. We introduced TRP2_181–193 (COOH-terminally extended) or OVA_248–269, together with rPA28α or hsp90, into EL4 cells pretreated with or without GA. As shown in Fig. 4 c, neither hsp90 nor GA-treatment influenced the presentation of TRP2_181–188, whereas rPA28α considerably increased the presentation of this epitope, suggesting that presentation of TRP2_181–188 is absolutely dependent on PA28 and that hsp90 does not have any role for this epitope. On the other hand, experiments using OVA_248–269 again showed that both PA28α and hsp90 contributed to the presentation of the epitope in EL4 cells (Fig. 4 c), indicating that these two pathways operate in these cells. Taken together, the PA28- and the hsp90-dependent pathways for antigen processing can work both redundantly and specifically, which depends on the type of antigen.

It is clear that both the PA28- and hsp90-dependent pathways for OVA epitope presentation operate redundantly in wild-type LPS blasts and EL4 cells, but the PA28-dependent pathway seems to be dominant in LPS blasts compared with EL4 cells, based on GA sensitivity. The difference may be due to the balance of PA28 and hsp90 expression. Therefore, we next examined the effect of IFN-γ on the two pathways of antigen processing in E.G7 cells, which express the full-length OVA protein. We transfected retrovirus vectors encoding cDNAs of either PA28α or hsp90 into E.G7 cells, and measured the cell-surface quantity of K^b-OVA_257–264, with or without GA and/or IFN-γ treatment. The total cell extracts from these transfected cells resulted in 2–3-fold increase in hsp90 or PA28α, as determined by densitometric analysis (Fig. 5 a, top panel). Whereas both PA28α and hsp90 enhanced the expression of K^b-OVA_257–264 complex, the stimulatory effect of hsp90, was completely prevented by GA treatment, unlike PA28α (Fig. 5 b). Interestingly, preincubation with IFN-γ for 24 h markedly induced both PA28α and PA28β (Fig. 5 a, bottom panel), and at the same time almost completely abrogated the inhibitory effect of GA (Fig. 5 b).

This observation was further confirmed by an acid-wash recovery assay. E.G7 cells preincubated with or without IFN-γ were briefly exposed to acid medium (pH 3.0) in order to destroy surface MHC-peptide complexes of cells, and then monitored the recovery of K^b-OVA_257–264 complex expression. As shown in Fig. 5 c, GA significantly suppressed the recovery of K^b-OVA_257–264 complex, but this suppression was cancelled by treatment with IFN-γ. These results were also confirmed by CTL assays, in which OVA_248–269 was introduced into EL4 cells pretreated with or without GA. As shown in Fig. 5 d, GA considerably inhibited OVA presentation in EL4 cells that were not treated with IFN-γ, whereas no significant suppression of cytolysis by GA was observed when EL4 cells were cultured in the presence of IFN-γ for 24 h before subjected to the CTL assay.

We also tested the expression of total K^b molecules on cell surface of E.G7 cells transfected with mock, PA28α or hsp90 cDNA. To exclude the possible nonspecific influence of GA on the stability of cell-surface molecules, we also examined the surface expression of LFA-1. As shown in Fig. 6 a, K^b expression was significantly suppressed by GA, especially in E.G7 hsp90, whereas that of LFA-1 was not affected. The GA-dependent reduction of K^b was completely restored to the level of E.G7 mock cells by incubation with the K^b ligand OVA_257–264 (Fig. 6 b), indicating that the reduction of K^b was due to a disturbed supply of K^b ligands. Incubation of E.G7 cells with IFN-γ greatly enhanced the expression of surface K^b molecules, which could not be inhibited by GA (Fig. 6 b). These results were further confirmed by an acid-wash recovery assay as mentioned above. As shown in Fig. 6 c, GA significantly inhibited the recovery of K^b, which was restored by incubation with the epitope peptides, but IFN-γ abrogated the inhibition. These results suggest that IFN-γ treatment reduces the contribution of the hsp90-dependent processing pathway.

Finally, we examined whether IFN-γ can induce the cell surface expression of K^b molecules in cells derived from PA28α^−/−/β^−/− mice. To this end, peritoneal macrophages (Mϕ) were prepared from PA28α^−/−/β^−/− mice and tested for K^b expression on their cell surfaces. IFN-γ markedly enhanced cell surface K^b expression in wild-type Mϕ, however, no induction of K^b was observed in PA28α^−/−/β^−/− Mϕ (Fig. 7, a and b) and pulsing of OVA_257–264 peptides to IFN-γ–treated PA28α^−/−/β^−/− Mϕ markedly increased the expression of K^b to a level comparable with that of wild-type cells (Fig. 7 b). These results indicate that PA28-deficient cells induce K^b molecules as well as wild-type cells upon IFN-γ stimulation, but that K^b ligands were not supplied in sufficient amount in PA28-deficient cells. In other words, PA28-dependent pathway is the main route responsible for antigen processing when cells were stimulated with IFN-γ. Another important feature is the profound inhibition of K^b expression by GA in IFN-γ-treated PA28α^−/−/β^−/− Mϕ, which was in contrast to that in wild-type Mϕ (Fig. 7 b). Similar results were obtained using both bone marrow–derived dendritic cells and MEFs (data not shown). These intriguing observations were confirmed by the acid-wash recovery assay. Indeed, in the absence of IFN-γ, GA completely inhibited the recovery of K^b expression on Mϕ of PA28α^−/−/β^−/− and partially of PA28α^+/+/β^+/+, which was restored by incubation with pulsing of OVA_257–264. Treatment with IFN-γ significantly enhanced the level of K^b expression of wild-type Mϕ and the enhanced K^b expression was not inhibited by GA. On the other hand, IFN-γ treatment of PA28α^−/−/β^−/− Mϕ did not enhance K^b expression, although OVA_257–264 pulsing enhanced the K^b expression to a level similar to that of the wild-type cells treated with IFN-γ (Fig. 7 c). The expression of heavy chain of K^b and β₂ microglobulin was normally induced by IFN-γ in MEFs derived from PA28α^−/−/β^−/− mice, as assessed by RT-PCR and Western blotting, respectively (Fig. 7 d), which is consistent with the enhancement of K^b expression when OVA_257–264 was pulsed. Furthermore, GA treatment completely inhibited K^b expression on PA28α^−/−/β^−/− Mϕ even when they were treated by IFN-γ. In addition, there was no difference in TAP2 induction by IFN-γ between wild-type and PA28-deficient cells (data not shown), showing IFN-γ signaling and MHC class I assembly components were apparently normal in PA28-deficient mice. Taken together, our results indicate that in IFN-γ treated cells PA28-dependent pathway is dominant for epitope supply to MHC class I molecules, whereas hsp90-dependent pathway becomes negligible, even for epitopes processed by both pathways under normal circumstances such as OVA.

In this study, we provide evidence to show the existence of two distinct pathways responsible for generation of MHC class I ligands, one is hsp90 dependent and the other PA28 dependent. These two pathways were clearly characterized by using hsp90-specific inhibitors GA and HA and cells derived from PA28α^−/−/β^−/− mice. Our results showed that antigen presentation in LPS blasts of PA28α^−/−/β^−/− mice was profoundly inhibited by GA, whereas GA never and only partially inhibited the generation of OVA_257–264 epitope in PA28-positive LPS blasts and EL4 cells, respectively, indicating that these two pathways cooperate in these cells (Fig. 4 a). In contrast, in the absence of GA treatment, the generation of OVA epitope was not impaired in PA28α^−/−/β^−/− blasts and rather comparable to that of PA28α^+/+/β^+/+ blasts (Fig. 4 a), consistent with our previous findings (15). In addition, the presentation of OVA_257–264 epitope from native OVA was clearly enhanced by PA28α and hsp90 (Fig. 5 b). The enhancement by hsp90 but not by PA28α was almost completely inhibited by GA, indicating that the ATPase activity of hsp90 is required for antigen presentation although GA inhibits not only hsp90 but also gp96 in the ER. Taken together, our results indicate that PA28 and hsp90 redundantly contribute to the OVA_257–264 epitope-processing pathway. It is clear that hsp90 plays an essential role in the presentation of OVA_257–264 epitope in the absence of PA28, which was clearly demonstrated using GA-treated PA28α^−/−/β^−/− LPS blasts, whereas PA28 is dominant in PA28α^+/+/β^+/+ LPS blasts.

On the other hand, presentation of TRP2_181–188 epitope was solely dependent on PA28, even in the presence of hsp90 in PA28α^−/−/β^−/− LPS blasts. Consistent with this observation, rPA28α had a stimulatory effect on the TRP2_181–188 processing in EL4 cells, whereas introduction of hsp90 and GA treatment had no effect (Fig. 4 c, left panel). Taken together, it is concluded that the two pathways mediated by hsp90 and PA28 redundantly contribute to OVA processing, but the TRP2_181–188 epitope processing is specifically mediated by the PA28-dependent pathway. Thus, it appears that utilization of these two pathways is dependent on antigen species.

In wild-type LPS blasts and EL4 cells treated with IFN-γ, the PA28-dependent pathway is dominant, as GA had no obvious inhibitory effects on OVA_257–264 epitope generation (Figs. 4 a and 5, b–d). It is possible that the PA28-dependent pathway suppresses the hsp90-dependent pathway for processing of the same antigens. To understand this phenomenon, it is essential to determine the molecular mechanism of hsp90 involved in the MHC class I ligand production. How can hsp90 activate antigen processing? We first investigated whether hsp90 enhanced peptide hydrolyzing activity of the proteasome like PA28 in vitro, but the results showed it did not (data not shown), consistent to the previous reports (19, 20). In this regard, Goasduff and Cederbaum (32) have recently reported that hsp90 inhibited the cleavage of the fluorogenic peptide by purified proteasome in vitro, whereas the use of hsp90 and cytosolic fraction enhanced the hydrolysis, suggesting that hsp90 together with other unidentified cytosolic factors could activate the hydrolysis by the proteasome. Furthermore, several studies have shown the binding of hsp90 to the 20S proteasome (19, 20, and unpublished data). In considering the redundant roles between PA28 and hsp90 pathways, hsp90 (with or without other proteins) might be directly involved in proteasome-dependent processing of antigens by forming a complex with 20S proteasome, as PA28 does. This model may explain in part the dominance of the PA28 over hsp90, although the role of the hsp90-proteasome complex remains to be determined.

The results shown in Fig. 3, c and d, also suggested the linkage of hsp90 and the proteasome. COOH- and NH₂-terminally extended OVA_248–269 was presented more efficiently than only NH₂-terminally extended OVA_248–264 by PA28 and hsp90, while lactacystin blocked only the presentation of COOH-terminally extended peptides. This suggests a mechanism that facilitates delivery of peptides processed by the proteasome to the MHC class I assembly system on the ER. In this regard, it is interesting that proteasomes, PA28, and hsp90 are associated with the ER membranes, probably the cytosolic face (33, 34), imaging a speculative model that the proteasome processing machinery involving PA28 and hsp90 may be directly linked to the TAP antigen transport system, as proposed previously by Rechsteiner et al. (6). One may attribute the effect of hsp90 to its function as a peptide carrier. Although we showed in this study that enhanced processing of OVA peptides by hsp90 was not attributed to its role as a peptide chaperone because hsp90 did not affect the presentation of NH₂-terminally extended peptide and that the effect of hsp90 was exactly linked to the proteasome, it is still possible that hsp90 plays a role as post-proteasomal peptide carrier. That is, hsp90 might bind peptides generated by the proteasome and safely carry them to TAP transporter, preventing further degradation by cytosolic peptidases as suggested by Binder et al. (31). Indeed, when hsp90 and PA28 were osmotically introduced, their levels were increased 2–3-fold in the cytosol as well as the membranous fractions (Fig. 2 a). Therefore, considerable amounts of transfected proteins are thought to be associated with the membranes, probably the ER, as considerable amounts of hsp90 and PA28 were recovered in the membrane fractions in cells, even without osmotic loading of these proteins (Fig. 2 a).

Another important aspect of the two distinct pathways is their physiological roles. In this study, we provided evidence that the PA28 pathway is up-regulated by IFN-γ. In this context, we propose the existence of a constitutive pathway and adaptive pathway regulated by IFN-γ in MHC class I antigen processing. The process requiring hsp90 can be regarded as a constitutive pathway. On the other hand, the PA28-mediated pathway can be regarded as an adaptive pathway, because PA28 is markedly induced by a major cytokine IFN-γ. In fact, we showed that GA had no significant effect on OVA_257–264 epitope generation in PA28α^+/+/β^+/+ LPS blasts (Fig. 4 a). This is probably because the PA28α^+/+/β^+/+ LPS blasts express a large amount of PA28, and hence the hybrid proteasome. Consistent with this model, GA had no effect on OVA_257–264 epitope generation in IFN-γ–treated EL4 cells, nevertheless it considerably inhibited the presentation in E.G7 cells without treatment with IFN-γ (Fig. 5, b and c). In addition, GA treatment of E.G7 cells down-regulated the expression of total K^b molecules (Fig. 6) as well as K^b-OVA_257–264 complex on the cell surface (Fig. 5, b and c). Conversely, transfection of hsp90 cDNA by retrovirus enhanced cell surface K^b molecules (Fig. 6 b) and other tumor cell lines (data not shown). Regarding the expression of K^b-OVA_257–264 complex and total K^b, the inhibitory effect by GA was abolished by incubation of cells with IFN-γ (Fig. 5, b and c, and Fig. 6, b and c). Thus, IFN-γ shifts the functional balance from hsp90 toward PA28. The defect of IFN-γ–induced up-regulation of K^b molecules on cell surface of macrophages (Fig. 7), dendritic cells, MEFs, and LPS blasts (data not shown) derived from PA28α^−/−/β^−/− mice also supports this notion. The reason for this shift could attribute to the fact that IFN-γ can induce PA28, resulting in an increase in the hybrid-proteasome (9). As hsp90 was not induced by IFN-γ (data not shown), the substantial role of hsp90 in antigen presentation would be replaced by abundant PA28. Indeed, down-regulation of total K^b molecules by GA was markedly enhanced even in IFN-γ–treated macrophages of PA28α^−/−/β^−/− mice (Fig. 7 b), which was in contrast to that of wild-type mice.

In summary, we have reported a novel function for hsp90, and demonstrated the crucial role of PA28 on IFN-γ–induced up-regulation of MHC class I molecules. Our results provide fundamental insight into the function of PA28 and hsp90 in MHC class I antigen processing/presentation.

We thank Ms. Hiroiwa for the excellent technical support, especially for preparation of recombinant proteins.

This work was supported by a Grant-in-Aid for Scientific Research on Priority Areas from the Ministry of Education, Science, Sports and Culture, Japan.