A Role for a Novel Luminal Endoplasmic Reticulum Aminopeptidase in Final Trimming of 26 S Proteasome-generated Major Histocompatability Complex Class I Antigenic Peptides*

Peptides presented to cytotoxic T lymphocytes by the class I major histocompatability complex are 8–11 residues long. Although proteasomal activity generates the precise C termini of antigenic epitopes, the mechanism(s) involved in generation of the precise N termini is largely unknown. To investigate the mechanism of N-terminal peptide processing, we used a cell-free system in which two recombinant ornithine decarboxylase (ODC) constructs, one expressing the native H2-K b -re-stricted ovalbumin (ova)-derived epitope SIINFEKL (ODC-ova) and the other expressing the extended epitope LESIINFEKL (ODC-LEova), were targeted to degradation by 26 S proteasomes followed by import into microsomes. We found that the cleavage specificity of the 26 S proteasome was influenced by the N-terminal flanking amino acids leading to significantly different yields of the final epitope SIINFEKL. incubated (cid:2) Procedures” (1 and incubated (cid:2) in by of tivity after addition trichloroacetic after subtraction of

MHC 1 class I molecules associate with peptides of 8 -11 amino acids derived from the proteolytic degradation of intracellular protein antigens and present them to CD8ϩ T cells on the cell surface (1)(2)(3). Although the vast majority of class I ligands are translocated from the cytosol into the endoplasmic reticulum (ER) by the ATP-dependent transporter associated with antigen processing (TAP) (3), alternative pathways have been described including the liberation of minigene-encoded peptides from ER signal sequences (4 -6), TAP-independent processing of signal sequences (7,8), and processing of membrane-associated or soluble proteins in the secretory pathway (9 -12). Peptides that are not retained in the ER by binding to class I or glycosylation are released back into the cytosol through the Sec61 channel (13)(14)(15).
Studies using membrane-permeable inhibitors of proteasomes have indicated that the proteasome is the major proteolytic activity responsible for the generation of antigenic peptides (16 -21), although the incomplete inhibition of antigen processing by proteasome inhibitors has also demonstrated the involvement of nonproteasomal cytosolic proteases (22)(23)(24)(25)(26)(27)(28).
The proteasome is an abundant cytosolic multisubunit protease consisting of the proteolytic 20 S core particle that associates with PA700 regulatory complexes to form 26 S proteasomes or with PA28 complexes (29). The 26 S proteasome usually degrades ubiquitinated proteins (30). Ornithine decarboxylase (ODC) constitutes an exception because it is targeted to 26 proteasomes by antizyme in a ubiquitin-independent fashion (31)(32)(33). The 26 S proteasome seems to mediate the degradation of most antigenic proteins in living cells (34 -36) and, therefore, deserves particular interest. The in vitro cleavage specificities of purified 20 S proteasomes and the relevance of proteasomal processing for the generation of immunodominant epitopes have been investigated in great detail (1,37). Although 20 S proteasomes have been shown to precisely cleave a number of known class I ligands out of model polypeptide substrates, a minor part of cleavage products is longer than the canonical length of class I-binding peptides (38 -41).
Limited information is, however, so far available about protein processing by 26 S proteasomes. A direct comparison of 26 and 20 S proteasomes revealed overlapping but substantially different cleavage patterns for the protein substrate ␤-casein (41). On average, the 26 S proteasomal cleavage products were found to be slightly shorter than peptides generated by 20 S proteasomes containing greater proportions of peptides that are too short for efficient TAP-mediated translocation as well as for class I binding (40,41).
Proteasomes seem to be the dominant, if not the only, protease that generates the correct C terminus of class I ligands, whereas the N-terminal trimming of proteasomal products is insensitive to proteasome inhibitors (20,42,43). Leucine aminopeptidase, puromycin-sensitive aminopeptidase, and bleomycin hydrolase have been implicated in the cytoplasmic trimming of proteasomal products (44,45). In addition, tripeptidyl peptidase II has been suggested to play a role in the generation of class I ligands or precursors (46). Different lines of evidence suggest that precursors of class I ligands can be trimmed in the ER lumen to their final lengths. Processing of peptide imported into microsomes was directly shown by biochemical methods (14,47). The broad but not random substrate specificity of TAP precludes that 9-mer peptides containing a proline residue at position 2 or 3 are efficiently bound and transported by TAP (48 -50), which can be rescued by extending such peptides with N-and/or C-terminal flanking residues (49). Nevertheless, several MHC class I alloforms in man and mouse prefer a proline residue at position 2 or 3 in the associated 9-mer peptide (51). Also for some peptides not containing proline, the addition of flanking residues to minimal epitopes strongly improved TAP affinities (50,52,53) implicating that precursors become trimmed in the ER.
The analysis of extended minimal epitopes or tandem arrays of epitopes that were directed into the ER of TAP-deficient T2 cells by a leader sequence have clearly indicated the predominance of an aminopeptidase activity (10,47,(53)(54)(55), whereas the carboxypeptidase activity in the ER lumen seems to be very poor (5,11,(53)(54)(55). N-terminal trimming of a precursor peptide was recently suggested to depend on the presence of the correct class I restriction element (47). However, the identity and biochemical properties of the aminopeptidase(s) in charge remains elusive.
To investigate the relative contribution of the proteasome and the cytosolic and ER peptidases to the generation of the definite class I epitope, we have developed an in vitro system in which recombinant ODC containing ovalbumin (ova) peptides are targeted to degradation by 26 S proteasomes followed by import into purified microsomes. We find that the yield of a finally processed H2-K b -binding epitope was strongly influenced by N-terminal flanking amino acids in the protein sequence. We show that a slightly extended ova epitope precursor was efficiently trimmed following TAP-mediated import into microsomes. The conversion into the definite epitope could be blocked by the aminopeptidase inhibitor 1,10-phenanthroline (PNT) and was strictly dependent on the presence of MHC class I molecules known to associate with the epitope.

Peptides
The peptide SIINFEKL was synthesized by Anaspec (San Jose, CA). The peptide ESIINFEKL was synthesized by Dr. M. Fridkin at the Department of Organic Chemistry, The Weizmann Institute (Rehovot, Israel). The peptides HESIINFEKL, HLESIINFEKL, QSHESIIN-FEKL, and QSHLESIINFEKL were synthesized at the peptide synthesis core facility of the German Cancer Research Center (DKFZ) (Heidelberg, Germany). All peptides were purified to ϳ95% homogeneity by HPLC, and their identity was confirmed by mass spectrometry. To accurately determine the concentration of peptides used in this study each peptide solution in double distilled water was analyzed by Edman degradation.

Chemicals
Bestatin and lactacystin were from Calbiochem. Complete EDTAfree protease inhibitors (referred to as protease inhibitors) were from Roche Molecular Biochemicals. 1,10-Phenanthroline, all standard reagents, and reagents for cell culture were from Sigma.

Cell Lines and Mice
The B3Z T cell hybridoma and the antigen-presenting cell line K b -L have been described by Karttunen et al. (56) and were granted by Dr. Chris Norbury (National Institutes of Health, Bethesda, MD). C57BL/6 (B6) and BALB/c mice were from local breeding in the animal facility of the Sackler Faculty of Medicine, Tel Aviv University (Tel Aviv, Israel). H2-D bϪ/Ϫ (B6.K b ) and H2-K bϪ/Ϫ (B6.D b ) knockout mice (57) were a kind gift of Dr. Lea Eisenbach (Weizmann Institute, Rehovot, Israel). TAP1 Ϫ/Ϫ mice (58) were courtesy of Dr. Natalio Garcia-Garbi (German Cancer Research Center (DKFZ), Heidelberg, Germany).

Preparation of 26 S Proteasome, Maltose-binding Protein-Antizyme Fusion Protein and Recombinant ODC-ova Derivatives
The 26 S proteasome was purified from B6 livers, and the recombinant proteins, maltose-binding protein-antizyme fusion protein, and ODC-ova derivatives were expressed in bacteria and purified by affinity chromatography as previously described (59). To generate the ODC-LEova expression vector, an adaptor encoding the peptide LESIIN-FEKL with BstXI-compatible ends was generated by annealing the two synthetic oligonucleotides 5Ј-ATCTGGAAAGTATAATCAACTTCGA-AAAACTGAGCC-3Ј and 5Ј-CAGTTTTTCGAAGTTGATTATACTTTC-CAGATGGCT-3Ј. The adaptor was then inserted into the ODC sequence as previously described for ODC-ova (59). The production of 35 S-labeled ODC-ova and ODC-LEova was in the methionine auxotroph strain B834(DE3) (Novagen Inc., Madison, WI), and purification of the 35 S-labeled proteins was carried out as previously described (59).

Preparation of Subcellular Fractions
Lactacystin-treated Cytosol-The preparation of cytosol (100,000 ϫ g supernatant) from RMA cells was carried out as previously described (60). To inhibit proteasome activity, a 2-ml aliquot of the cytosol (20 mg of protein) was incubated with 35 M lactacystin for 30 min at 37°C. The lactacystin-treated cytosol was then dialyzed for 16 h at 4°C against 500 ml of 20 mM Tris-HCl (pH 7.5), 0.5 mM DTT and stored in aliquots at Ϫ70°C. Inhibition of proteasome activity in the cytosol was greater than 90% as determined by the inhibition of fluorogenic peptide Suc-LLVY-methyl coumarin hydrolyzing activity (61).
Microsomes-Microsomes were prepared according to the protocol described by Shepherd et al. (58). Briefly, mice (8 -10 weeks old) were injected intraperitoneally with 0.2 mg of poly(inosine-cytosine)/mouse 24 h prior to sacrifice. The livers and spleens were thoroughly washed with phosphate-buffered saline and then homogenized in a motordriven Potter-Elvehjem Teflon tissue grinder. Homogenization was in 3 ml/liver buffer A (50 mM triethanolamine acetate (pH 7.5), 1 mM DTT, 5 mM magnesium acetate, 50 mM potassium acetate, 250 mM sucrose, and a 1:25 (w/v) solution of protease inhibitors). The extract was then subjected to fractional centrifugation at 1,000 ϫ g and then at 10,000 ϫ g. The supernatant from the 10,000 ϫ g spin was then placed on top of a sucrose cushion containing 1.3 M sucrose in buffer A at a ratio of 2:1 (sample:cushion) and then centrifuged at 140,000 ϫ g for 2.5 h. The resulting pellet was then resuspended in buffer B (50 mM Hepes-KOH (pH 7.5), 1 mM DTT, and 250 mM sucrose) to a concentration of 200 A 280 nm units/ml. The microsome suspension was then stored in aliquots at Ϫ70°C.

Stripping Off Microsome-associated Proteins
The microsome suspension was incubated on ice for 15 min in 10 volumes of buffer C (50 mM Hepes-KOH (pH 7.5), 1 mM DTT, 0.5 M KCl, and 15 mM EDTA). The membrane suspension was then centrifuged at 10,000 ϫ g in a microcentrifuge. The membrane pellet was then resuspended in 10 volumes of buffer D without EDTA and immediately centrifuged as described above. The final membrane pellet was resuspended in the processing reaction mixture. The protein content of the stripped microsomes was less than 40% of that of untreated membranes as determined by measurement of the absorbance at 280 nm.

Assay of TAP-mediated Peptide Transport
TAP-mediated peptide transport was determined by measuring the ATP-dependent transport of the radioiodinated peptide TNKTRIDGQY into isolated microsomes as previously described (63). Purified microsomes (1 unit) were incubated in a reaction mixture containing the following components in a final volume of 100 l: 50 mM Hepes-KOH (pH 7.5), 1 mM DTT, 5 mM MgCl 2 , 10 mM creatine phosphate, 2.5 units of creatine phosphokinase, and 125 I-labeled peptide (40 ng, 10 6 cpm). In reactions without ATP, 2-deoxyglucose (20 mM) and hexokinase (3 g) were added instead of ATP and the ATP-regenerating system. Following incubation for 10 min at 37°C the microsomes were pelleted by centrifugation at 10,000 ϫ g in a microcentrifuge. The supernatant was removed, and the membrane pellet was resuspended in 1 ml of lysis buffer (20 mM Hepes-KOH (pH 7.5), 5 mM MgCl 2 and 1% Nonidet P-40) and then sonicated for 30 s at 50% output power in a water bath sonicator (Ultrasonic Processor, Heat Systems Inc., Farmingdale, NY). The detergent extract was centrifuged for 10 min at 10,000 ϫ g in a microcentrifuge. The resulting supernatant was then mixed with 25 l of concanavalin A-Sepharose beads (Amersham Pharmacia Biotech), and the mixture was then mixed gently for 16 h at 4°C. The beads were pelleted by centrifugation at 1,000 ϫ g in a microcentrifuge and then washed twice with 1-ml portions of lysis buffer. The ATP-dependent transport of peptide was then determined by measuring the amount of radioactivity associated with the beads in the presence of ATP after subtraction of the concanavalin A-associated radioactivity obtained in a parallel reaction carried out in the absence of ATP.

Antigen Processing Assays
Antigen processing reaction mixtures contained the following components in a final volume of 75 l: 50 mM Hepes-KOH (pH 7.5), 5 mM MgCl 2 , 0.5 mM DTT, 1 mM ATP, 10 mM creatine phosphate, 1.8 units of creatine phosphokinase, 10 g of recombinant ODC, 15 g of maltosebinding protein-antizyme fusion protein, and 10 units of 26 S proteasome (for the definition of proteasome units see Ref. 59). Where indicated purified microsomes (1 A 280 unit) and lactacystin-treated cytosol (60 g of protein) were also supplemented. Following incubation for 10 min at 37°C, the reaction was extracted in 400 l of ice-cold lysis solution (0.5% trifluoroacetic acid, 1% Nonidet P-40). The mixture was then sonicated for 60 s at 50% output power in an ice-cooled bath sonicator. In processing reactions containing microsomes the membranes were first pelleted by centrifugation at 4°C for 15 min at 10,000 ϫ g in a microcentrifuge and then extracted in lysis buffer as described above. The acid extract was then filtered through a 10-kDa cut-off filter (Microcon 10 concentrator, Millipore Corp., Bedford, MA), and the filtrate was collected and lyophilized. The lyophilized material was chromatographed by reverse phase HPLC, and the fractions corresponding to the elution position of synthetic SIINFEKL were pooled and lyophilized exactly as previously described (59).

Peptide Processing Assays
Reactions were carried out in a final volume of 100 l containing the following components: 50 mM Hepes-KOH (pH 7.5), 5 mM MgCl 2 , 0.5 mM DTT, 1 mM ATP, 10 mM creatine phosphate, 2.5 units of creatine phosphokinase, microsomes (1 unit), and 50 fmol of synthetic ESIINFEKL or HLESIINFEKL. Where indicated, 1,10-phenanthroline (2 mM) or bestatin (250 M) were added. In reactions without ATP, hexokinase (3 g) (Roche Molecular Biochemicals) and 2-deoxyglucose (20 mM) were added instead of ATP and the ATP-regenerating system. The reaction was preincubated for 4 min at 37°C, and then ESIIN-FEKL was added for a further incubation for 8 min at 37°C. The reaction was stopped by the addition of 450 l of lysis buffer and then processed as described above. We detected low SIINFEKL activity even in the absence of microsomes. This activity was probably because of residual SIINFEKL present in the synthetic ESIINFEKL preparation that we estimated to be ϳ5%. Therefore, the conversion of ESIINFEKL to SIINFEKL was calculated after subtraction of the value obtained in a parallel experiment without microsomes.

Quantification of SIINFEKL
Except for Fig. 1B where the amount of SIINFEKL was quantified by flow cytometry using the H2-K b -SIINFEKL-specific monoclonal antibody 25-D1.16 (62), SIINFEKL was quantified by the B3Z T cell hybridoma activation assay (56). Briefly, B3Z cells (5 ϫ 10 4 ) were co-cultured overnight in 100 l of phenol red-free RPMI 1640 medium at 37°C with K b L cells (3 ϫ 10 4 ) and in the presence of either synthetic peptides or HPLC-purified processed peptides. After activation, cells were lysed by the addition of 50 l/well stop buffer (50 mM Na 2 HPO 4 , 34 mM NaH 2 PO 4 , 10 mM KCl, 1 mM MgSO 4 , 0.125% Nonidet P-40, 150 mM ␤-mercaptoethanol) containing 0.45 mM chlorophenol red ␤-galactopyranoside (Calbiochem). The cells were then further incubated for 4 h at 37°C, and the absorbance at 595 nm in each well was subsequently determined using a 96-well enzyme-linked immunosorbent assay reader. The amount of SIINFEKL was calculated based on the activity of known amounts of synthetic peptide that were tested in parallel.

HPLC Mass Spectrometry
Synthetic and proteasomal digest-derived peptide mixtures were analyzed by a reverse phase HPLC system (ABI 140D, Applied Biosystems) coupled to a hybrid quadrupole orthogonal acceleration time of flight tandem mass spectrometer (Q-TOF, Micromass, Manchester, England) equipped with an electrospray ionization source. Solvent A was 4 mM ammonium acetate adjusted to pH 3.0 with formic acid. Solvent B was 2 mM ammonium acetate in 70% acetonitrile/water adjusted to pH 3.0 with formic acid. Loading and desalting of typical sample volumes of 100 l was achieved by preconcentration on a 300 m ϫ 5 mm C18 -Precolumn (LC Packings, San Francisco, CA). A syringe pump (PHD 2000, Harvard Apparatus, Inc., Holliston, MA) equipped with a gastight 100-l syringe (1710 RNR, Hamilton, Bonaduz, Switzerland) was used to deliver solvent and sample at a flow rate of 2 l/min. For peptide separation, the preconcentration column was switched in line with a 75 m ϫ 250 mm C18 column (LC Packings). A binary gradient of 25-60% solvent B within 70 min was performed, applying a flow rate of 27 l/min reduced to ϳ300 nl/min with a precolumn split using a TEE-piece (ZT1C, Valco, Schenkon, Switzerland) and a 300 m ϫ 150 mm C18 column as a backpressure device. A gold-coated glass capillary (PicoTip, New Objective, Cambridge, MA) was used as the needle in the electrospray ionization source. A blank run was performed prior to any subsequent HPLC mass spectrometry run to ensure that the system was free of any residual peptide.
For on-line nanocapillary HPLC tandem mass spectrometry experiments, fragmentation of the parent ion was achieved at the given retention time by collision with argon atoms. Q1 was set to the mass of interest Ϯ 0.5 Da, and collision energy was applied for 1-2 min.
Quantitative analysis of different compounds in HPLC mass spectrometry is possible by calibrating with synthetic substances of known amounts. For quantifying relative amounts of different SIINFEKLcontaining peptides in one digest, relative signal intensities have been evaluated by calibration with different amounts of the corresponding synthetic peptides. For comparison of the amounts of SIINFEKL and ESIINFEKL in the digests of the two different substrates, the efficiencies of the digests were normalized using three proteasomal cleavage products from various parts of ODC, which for this reason should be generated in equal amounts in all digests.

RESULTS
The Efficiency of Generation of the H2-K b Ova-derived Epitope Is Dependent upon Its N-terminal Flanking Residues-We have generated two recombinant ODC derivatives that express the optimal H2-K b epitope SIINFEKL (ODC-ova) and the naturally extended ovalbumin-derived peptide LESIINFEKL (ODC-LEova). We have previously shown that ODC-ova is processed very efficiently to generate almost exclusively the optimal epitope SIINFEKL (59). When we compared the proteolytic processing of ODC-ova and ODC-LEova by purified 26 S proteasome under linear kinetic conditions we noticed that the amount of SIINFEKL produced from ODC-ova was consistently 3-4-fold higher than that produced from ODC-LEova (Fig. 1A). The difference was not due to decreased proteolysis of ODC-LEova because the rate of degradation of the two derivatives was similar (Fig. 1B). Consequently mass spectrometry analysis indicated that the distribution of the SIINFEKL-containing peptides produced by the two derivatives was markedly different in respect to the proportion of the optimal epitope relative to the other SIINFEKL-containing peptides. Indeed the proportion of SIINFEKL relative to the overall SIINFEKL-containing peptides from ODC-ova and ODC-LEova was 90 and 25%, respectively. Instead the predominant peptide produced from ODC-LEova was ESIINFEKL (62%). Consistent with previous observations that the proteasome is the sole proteolytic activity that generates the correct C terminus of MHC class I-restricted epitopes, no SIINFEKLcontaining peptides with C-terminal extensions could be detected (Table I). The finding that ESIINFEKL became the dominant peptide generated from ODC-LEova is probably because of the introduction of the leucine residue, a preferential proteasomal cleavage site (39), upstream of the epitope in ODC-LEova.
Processing of ODC-ova and ODC-LEova in the Presence of Cytosol and Isolated Microsomes-It has been previously reported that antigenic peptide precursors with N-terminal extensions are further processed either in the cytoplasm or in the ER (see the Introduction). To investigate the role of cytoplasmic and ER peptidases in the generation of the final antigenic epitope, we subjected the recombinant antigens to proteolytic processing by the 26 S proteasome in the presence of a cytosolic fraction and isolated microsomes. Following incubation of the antigens in the cell-free system, the microsomes were pelleted, and the yield of SIINFEKL was measured by the ability to activate the K b /SIINFEKL-specific T cell hybridoma B3Z.
When ODC-ova was incubated in the presence of 26 S proteasome and B6.K b microsomes, 54% of the SIINFEKL initially produced by the proteasome alone was recovered (6 of 11.2 fmol) ( Fig. 2A). However, when ODC-LEova was processed under the same conditions, the yield of SIINFEKL was 175% (6.3 compared with 3.6 fmol) (Fig. 2B). These results suggested that the N-terminally extended SIINFEKL precursors generated from ODC-LEova and specifically ESIINFEKL were further trimmed by the microsomes to produce the optimal epitope.
To exclude the possibility that the augmentation in SIINFEKL yield was because of exceedingly higher affinity of ESIINFEKL to TAP resulting in preferential transport followed by further trimming in the ER, we tested the affinities of the extended peptides to TAP. Using a radioiodinated and glycosylatable reporter peptide (63), we determined the IC 50 values of SIINFEKL, ESIINFEKL, HLESIINFEKL, and QSHLESIINFEKL to be 4.5 M, 3.0 M, 2.3 M, and Ͼ1 mM, respectively. Thus, we found that the affinity of all the Nterminally extended ova peptides generated from ODC-LEova was similar to that of SIINFEKL (except for the 13mer) suggesting that the increment in the recovery of SIINFEKL was mainly because of processing of the predominant proteasomal product, ESIINFEKL. The increment in SIINFEKL activity when ODC-LEova was processed could also not be explained by recognition of K b -bound ESIINFEKL because the B3Z T cell hybridoma recognized this peptide at least 10-fold less efficiently than SIINFEKL (Fig. 2D).
The addition of cytosol to the processing reaction caused a considerable reduction in the yield of SIINFEKL from both substrates (compare ϩ and Ϫ cytosol in Fig. 2, A and B). These results indicated that the proteasome-processed peptides were subjected to nonspecific degradation in the cytosol. Nevertheless, the significant recovery of SIINFEKL in microsomes indicated that the transport of peptides into the ER lumen rescues a significant amount of peptides from complete cytosolic degradation.
When processing was performed with B6.D b microsomes no SIINFEKL could be recovered (Fig. 2, A and B). Furthermore, SIINFEKL could not be detected when ODC-LEova was processed in the presence of BALB/c microsomes (Fig. 2C). These results clearly indicate that the correct restriction element was essential for precursor peptide processing and retention of the definite epitope. These results are also consistent with previous reports that unless the peptide is retained in the ER by binding to the appropriate MHC class I molecule it is degraded either in the microsomal lumen or after release to the cytosol (14). No SIINFEKL could be recovered when the processing of ODC-LEova was carried out in the presence of TAP1-deficient microsomes instead of B6.K b microsomes, indicating that entry of the peptides into the microsomal lumen was necessary for further processing (Fig. 2C).
Processing of Synthetic ESIINFEKL by Isolated Microsomes-To investigate the peptidase activity responsible for the N-terminal trimming, we studied the conversion of synthetic ESIINFEKL to SIINFEKL by isolated microsomes. As observed in the antigen processing experiments, incubation of synthetic ESIINFEKL with isolated B6.K b microsomes resulted in a remarkable enhancement of SIINFEKL recovery (Fig. 3). In agreement with the results of the ODC-LEova processing experiment, TAP1 Ϫ/Ϫ , BALB/c as well as B6.D b microsomes could not generate SIINFEKL (Fig. 3 and data not  shown).
To further characterize the specific trimming activity associated with the microsomes, we tested the sensitivity of the putative peptidase to the metalloaminopeptidase inhibitor bestatin (64) as well as to the ion chelator PNT that was previously shown to inhibit various metallopeptidases (65). We found that when both inhibitors were present, the ESIINFEKL to SIINFEKL conversion was completely blocked (Fig. 4). PNT alone was sufficient to sustain the block. Surprisingly however, in the presence of bestatin alone, there was a significant augmentation in the yield of SIINFEKL. These results suggest that at least two types of peptidases influence the fate of ESIINFEKL: a bestatin-sensitive peptidase that had a deleterious effect and a PNT-sensitive "trimmase" that converted ESIINFEKL to SIINFEKL. The PNT-mediated inhibition was not due to inhibition of peptide transport into the ER because  the compound even slightly stimulated TAP-mediated peptide transport (data not shown). Although the processing of ESIINFEKL was dependent on TAP transport and was class I-specific it was still possible that peptidases operating at the cytoplasmic leaflet of the microsomes initially converted ESIINFEKL to SIINFEKL and that only subsequent transport into the ER lumen and binding to the correct class I molecule prevented further degradation of the peptide epitope. Although the ultimate proof for luminal localization of the trimming aminopeptidase would have been resistance to proteolysis, this approach could not have been applied because treatment of microsomes with trypsin, for example, would have also digested TAP and prevented peptide transport. Thus, to establish the localization of the ESIINFEKL to SIINFEKL-converting enzyme, we attempted to remove the external peptidases. To this end the microsomes were washed with 15 mM EDTA and 500 mM KCl. This washing procedure removed over 60% of the proteins associated with the microsomes (data not shown) and should have removed almost entirely any peptidase activity associated with the outer face of the microsomes. As shown in Fig. 5A, when synthetic SIINFEKL was incubated with the washed microsomes in the absence of ATP (i.e. when TAP transport was blocked) there was only a small reduction in the amount of SIINFEKL relative to the input level obtained in the control incubation with buffer alone (Fig. 5A, first two columns on the left). Low and intermediate PNT concentrations (50 -200 M) were sufficient to restore the input level of SIINFEKL indicating that the remaining peptidase activity on the cytosolic face was completely inhibited (Fig. 5, A and C). However, when ESIINFEKL was incubated with the washed microsomes in the presence of ATP, the conversion of ESIINFEKL to SIINFEKL was significantly inhibited only at higher PNT concentrations (Fig. 5, B and C). These results demonstrate that under conditions in which the activity of the peptidase on the external surface of the microsomes was completely inhibited, there was an additional PNTsensitive activity that became apparent only in the presence of ATP (see also Fig. 4). We conclude that an ER-luminal metallopeptidase contributes to the conversion of ESIINFEKL to SIINFEKL.

DISCUSSION
In this work we have reconstituted in vitro the entire antigen processing pathway, from proteolysis of the antigen by the 26 S proteasome to the binding of the final epitope to the specific MHC class I molecule in the ER. This cell-free system that is comprised of the 26 S proteasome, a protein antigen, cytosol, and isolated microsomes enables the analysis of the relative contribution of the proteasome and cytosolic and microsomal peptidases to the generation of the definite epitope. The major impediment for the analysis of the mechanism of the 26 S proteasome in antigen processing lies in the difficulty to produce large enough ubiquitinated protein antigen for in vitro processing experiments to allow mass spectrometric analysis of the cleavage products. We have overcome this crucial problem by targeting the protein antigen to the 26 S proteasome through the ubiquitin-independent mechanism used by ODC (31,59). By varying the amino acid residues immediately N-terminal to SIINFEKL we were able to generate an ODCova derivative, ODC-LEova, that was degraded at a similar rate by the 26 S proteasome; however, the proteolytic processing resulted in a significant reduction in the abundance of SIINFEKL (see Fig. 1 and Table I). The fact that ESIINFEKL, the dominant extended SIINFEKL, was a poor activator of B3Z T cells allowed us to follow for the first time postproteasomal processing activity in a physiologically relevant fully reconstituted cell-free system. Consistent with our previous results (59), we found by mass spectrometric quantification that the 26 S proteasome efficiently liberates the SIINFEKL peptide from the ODC sequence context ( . . . WQLMKQIQSH-SIINFEKL-SHGFPPEVEE . . . ), whereas the N-terminal extended peptides HSIINFEKL and QSHSIINFEKL were only minor products. When introducing the extended epitope LESIINFEKL into the same ODC sequence context, the predominant cleavage now occurred between the N-terminal Leu and Glu residues, which is in full agreement with the cleavage specificities of both the 20 S and the 26 S human proteasome assayed on the model proteins enolase and ␤-casein, respectively (39,41).
The 26 S-processed SIINFEKL is efficiently delivered into the ER lumen despite losses that are due to the activity of peptidases co-purifying with ER membranes. The cytosolic origin of the SIINFEKL-degrading peptidases is suggested by the finding that the addition of cytosol led to further degradation (see Fig. 2A). Nevertheless, the significant proportion of SIIN-FEKL that was retained in the microsomes suggested that the TAP-mediated translocation of peptide occurred at a higher rate.
Three cytosolic peptidases have thus far been implicated in the trimming of N-terminally extended peptide epitope precursors to their final size. Rock and co-workers (43,44) have shown that a bestatin-sensitive leucine aminopeptidase can cleave SIINFEKL precursors to generate the final 8-mer. In a recent study, puromycin-sensitive aminopeptidase and bleomycin hydrolase have been implicated in the N-terminal trimming of a vesicular stomatitis virus nucleoprotein-derived epitope pre-cursor (45). These observations indicate that individual peptidases contribute to limited N-terminal trimming of epitope precursors. Nevertheless, the cumulative effect of cytosolic peptidases on SIINFEKL and its precursors seems to be deleterious. This can be concluded from our finding that the addition of cytosol to the antigen processing reaction caused a major reduction in the amount of SIINFEKL retained in the ER (see Fig. 2). In other experiments not presented in this study we found that cytosol completely destroyed 26 S proteasome-processed peptides in the absence of microsomes and that this degradation could not be inhibited by bestatin. The stimulatory effect of bestatin on the yield of SIINFEKL in the absence of cytosol (see Fig. 4) likely resulted from inhibition of a particular membrane-associated peptidase(s). This also suggested that the bestatin-sensitive peptidase acted prior to TAP transport and thus limited the availability of peptide substrates.
López and co-workers (28) have recently provided evidence for a PNT-sensitive cytosolic aminopeptidase activity. When the metallopeptidase inhibitor PNT was administered to cells infected with recombinant vaccinia virus expressing a human immunodeficiency virus Env epitope either in the context of the natural human immunodeficiency virus envelope protein or of a recombinant hepatitis B virus core protein, the authors (28) noticed efficient blockade of antigen presentation. However, the processing of another recombinant hepatitis B virus construct or a long epitope precursor targeted to the ER in a TAP-independent manner was not affected by PNT suggesting that a PNT-sensitive peptidase was operative prior to TAP transport. We have also identified a PNT-sensitive peptidase associated with the external leaflet of the microsomal membranes. However, in addition we show that a distinct PNTsensitive peptidase is operative in the lumen of the ER and that this luminal peptidase in involved in the final processing of 26 S proteasome-processed peptides subsequent to introduction into the ER by TAP. This follows from our observation that the final epitope SIINFEKL was generated by EDTA/salt-stripped microsomes from ESIINFEKL when the activity of the external peptidase was inhibited and when peptide translocation was facilitated (see Fig. 5).
The presence of H2-K b was strictly required to convert ESIINFEKL to SIINFEKL, whereas only background levels could be recovered from H2-D b -or H2-K d /D d -expressing microsomes (cf. Figs. 2B, 2C, and 3). This would invoke a model FIG. 5. Inhibition of peptidase activity by PNT. A, synthetic SIINFEKL was incubated in the absence of ATP. B, synthetic ESIINFEKL was incubated in the presence of ATP with B6.K b microsomes and with the indicated concentrations of PNT. Microsomes were then extracted and the amount of SIINFEKL was determined by the B3Z activation assay. C, quantification of the inhibition of peptidase activity. The percentage of inhibition of the external peptidase activity (squares) was calculated according to the following formula: (((SIINFEKL without microsomes) Ϫ (SIINFEKL with microsomes plus PNT)) ϫ 100)/((SIINFEKL without microsomes) Ϫ (SIINFEKL with microsomes)). The percentage of inhibition of ESIINFEKL processing (triangles) was calculated as the ratio of the activity observed in the absence of PNT relative to that observed in the presence of the inhibitor after subtraction of the activity measured in the absence of microsomes in each case. The results were calculated as the mean of the values obtained in two independent experiments. according to which N-terminal trimming is initiated by binding of the precursor peptide to the appropriate class I receptor followed by recruitment of the aminopeptidase to the class I-peptide complex as originally proposed by Rammensee and co-workers (51). In this model, the class I molecules would prevent N-terminal trimming beyond the optimal peptide length.
Earlier work has provided evidence that selected peptides could only be extracted from tissues expressing class I molecules able to associate with these peptides (66). Furthermore, it has been shown that antigenic peptides that are not retained in the ER undergo retrotranslocation to the cytosol for degradation (13)(14)(15)58). The herein presented findings that SIINFEKL processed from ODC-ova was undetectable in reactions containing H2-D b -but not H2-K b -expressing B6 microsomes ( Fig.  2A) or H2-K d -expressing BALB/c microsomes 2 are fully consistent with the idea that class I molecules retain and protect TAP-translocated peptides.
Recently, Shastri and colleagues (47) demonstrated the efficient conversion of an N-terminally extended SIINFEKL derivative in H2-K b -but not H2-K d -expressing cells. We have shown that in a cell-free system the presence of H2-K b is essential for the recovery of SIINFEKL following processing of ESIINFEKL by a microsomal metalloaminopeptidase. However, we cannot conclusively determine whether binding of the precursor peptide to the correct restriction element is strictly required for the initiation of the processing or whether class I molecules merely capture partially digested peptides released from slowly acting luminal aminopeptidases in a random process and protect them from further degradation. Additional work is required to distinguish between these two potential mechanisms. Furthermore, the general role of this aminopeptidase will have to be dissected using additional protein antigens.