Differences in the Expression of Human Class I MHC Alleles and Their Associated Peptides in the Presence of Proteasome Inhibitors¹

The recognition of antigenic peptides in association with class I MHC molecules at the cell surface provides a mechanism by which CD8⁺ T lymphocytes gain information about the proteins being made within cells. In most cells, class I-associated peptides are derived from endogenously expressed intracellular proteins and are between 8 and 11 aa long (1). The enzyme complex most often implicated in the generation of class I- associated peptides is the proteasome (2). Proteasome involvement in class I peptide production is based on several lines of evidence. First, the expression of the catalytically active proteasome β subunits LMP2 and LMP7, as well as the proteasome regulator PA28, have been shown to enhance the production of class I-associated peptides in vivo and in vitro (3, 4, 5). Second, ubiquitination, which targets proteins for proteasome-mediated degradation, is required for generation of class I- associated epitopes from some proteins (6, 7). Third, incubation of whole proteins or synthetic polypeptides with proteasomes in vitro results in the production of known epitopes (8, 9, 10). Finally, treatment of cells with various inhibitors of proteasome activity leads to a significant decrease in presentation of several peptide epitopes as well as reduced surface expression of H2-K^b and HLA-A*0201 (2). These inhibitors also diminish the ability of several human class I molecules to form stable dimers (11, 12).

Despite this breadth of evidence, other studies have suggested that proteasome involvement in class I- restricted peptide production is more limited (13). In particular, it has been suggested that ubiquitination is not involved in the production of at least some class I-associated peptides (14). More directly, cell surface expression of two murine class I alleles is only partially blocked by proteasome inhibitors (15, 16). Furthermore, the ability of two human class I molecules to form stable dimers, a property thought to depend upon peptide availability for binding, was not blocked when proteasome activity was inhibited (17). A few specific TAP-dependent epitopes have also been shown to be either insensitive to proteasome inhibition (15, 18) or destroyed by proteasome activity (19). Finally, cells grown for extended periods in proteasome inhibitors continue to express stable murine class I dimers (20). Collectively these studies strongly suggest that both proteasomes and nonproteasomal proteases can independently generate class I-associated peptides.

Neither the relative contributions of different proteolytic pathways to class I expression, nor the sequence and protein source of the peptides they produce, are well understood. Here, we have combined proteasome inhibitors, flow cytometry, and mass spectrometry to address these issues. Cells treated with acid to remove surface class I peptide complexes were allowed to re-express newly formed complexes in the presence or absence of proteasome inhibitors. We then used flow cytometry to measure total class I surface re-expression and mass spectrometry to analyze the peptides displayed under these conditions. Our results provide insight into both the relative contribution as well as the specificity of proteasome inhibitor-insensitive proteases that generate class I-associated peptides.

Lactacystin (LAC)⁴ (Calbiochem, La Jolla, CA) is a Streptomyces metabolite that irreversibly inhibits proteasomes via covalent binding to the catalytic sites of the β subunits (21, 22). N-acetyl-l-leucinyl-l-leucinal-l-norleucinal (LLnL; Calbiochem), also known as calpain Inhibitor I, reversibly inhibits proteasomes as well as several other classes of proteases (23). Carboxybenzyl-leucyl-leucyl-leucine vinyl sulfone (z-L₃VS) was a generous gift from Dr. H. Ploegh (Harvard University, Cambridge, MA) and has been shown to specifically inhibit proteasomes by binding to all of the known active sites and blocking all known in vitro measures of the proteasome function (24). Brefeldin A (BFA; Sigma, St. Louis, MO) inhibits egress of all proteins through the secretory pathway at the level of the cis-Golgi, including newly generated class I peptide complexes (25).

All tumor lines were of human origin and were maintained in RPMI 1640 medium supplemented with 5% FCS containing SerXtend (Irvine Scientific, Santa Ana, CA) and 2 mM glutamine (cell medium) in a humidified 5% CO₂ atmosphere at 37°C. The B lymphoblastoid cell line 721 and the TAP mutant line 721.174 have been described previously (26, 27). Stable transfectants of the class I-A and class I-B locus-negative cell line C1R with human HLA-A or -B molecules were maintained in cell medium containing either 300 μ g/ml G418 or 300 μg/ml Hygromycin. Various C1R transfectants were generous gifts from Dr. P. Cresswell (Yale University, New Haven, CT), Dr. A. McMichael (John Radcliffe Hospital, Oxford, U.K.), Dr. R. Colbert (Children’s Hospital Medical Center, Cincinnati, OH), or were produced by this laboratory (28).

Cells (1–2 × 10⁶) were centrifuged and the resulting pellet was resuspended gently in 50 μl of 300 mM glycine (pH 2.5)/1% (w/v) BSA (acid wash) and incubated for 4 min at 37°C. The suspension was neutralized by dilution with 100 μl of cell medium containing 0.5 N NaOH and 0.2 M HEPES and centrifuged. Cells were resuspended into 200 μl of cell medium in the presence or absence of 10 μg/ml BFA or various concentrations of LAC, LLnL, or z-L₃VS and incubated for 5 h at 37°C to allow class I re-expression. In some experiments, cells were incubated for 2 h at 37°C in the presence or absence of inhibitors before acid treatment and further incubation with inhibitors. Cells were subjected to flow cytometry analysis as described earlier (19). Live cells were gated, 10,000 events were counted, and the mean fluorescence intensity was recorded.

Both PA2.1 and BB7.2 recognize HLA-A*0201 molecules (29, 30). 4D12 recognizes HLA-B5 molecules (31). SFR8-B6 recognizes HLA-B molecules that express the Bw6 public Ag. HLA-B8 contains Bw6 while HLA-B5 does not (32). B123.2 recognizes all HLA-B and HLA-C molecules. W6/32 recognizes all human mature peptide containing HLA-A, HLA-B, and HLA-C class I molecules (33). Biotinylated Ab to HLA-A1 was obtained from One Lambda (Canoga Park, CA). FITC-labeled goat anti-mouse IgG and streptavidin were used as secondary staining reagents and were obtained from Cappel (Durham, NC).

Cells (5 × 10⁸–1 × 10⁹) were pretreated with 250 μM LLnL for 2 h, acid treated with 50 μl of acid wash per 1 × 10⁶ as described above, and then incubated with 250 μM LLnL for another 5 h. Alternatively, 1 × 10⁸–5 × 10⁸ cells were acid treated and then incubated in cell medium for 5 h. Greater numbers of cells were used for the inhibited samples to compensate for lower class I expression levels at the cell surface at the end of the experiment. Class I surface expression was determined by flow cytometry on 1 × 10⁶ cells as described above. The majority of cells were washed twice in PBS at 4°C, snap frozen in liquid nitrogen, and stored at −80°C for later use. Frozen cell pellets were then resuspended at a concentration of 1 × 10⁸ cells/ml in 1% 3-[(3-cholamidopropyl)dimethyl-ammonio]-1-propane sulfonate, 20 mM Tris-HCl pH 8.0, 100 μM iodoacetamide, 5 μg/ml aprotinin, 10 μg/ml leupeptin, 10 μg/ml pepstatin A, 5 mM EDTA, 0.04% sodium azide, and 1 mM PMSF (lysis buffer; all from Boehringer Mannheim, Indianapolis, IN). After rocking for 1 h at 4°C, lysates were clarified by centrifugation at 100,000 × g for 1 h. Lysates were precleared for 4 h at 4°C with recombinant protein A-Sepharose beads (Pharmacia, Piscataway, NJ). Class I-specific Ab-saturated recombinant protein A-Sepharose beads were then added to lysates for 12 h at 4°C. Beads were spun down and lysate was removed to a separate tube. Beads saturated with Abs of other class I specificities could then be added sequentially to purify additional alleles. Beads were washed twice in lysis buffer, twice in 50 mM Tris-HCl/1 M NaCl (pH 8.0), twice in 50 mM Tris-HCl/250 mM NaCl (pH 8.0), and three times in 50 mM Tris-HCl (pH 8.0). Beads were then transferred in 50 mM Tris-HCl to the top insert of an Ultra-free-MC 5000 NMWL filter from Millipore (Bedford, MA) and excess buffer was spun through the filter. Filter insert was then moved into a Teflon tube (Savillex, Minnetonka, MN). Peptides were eluted from class I/β₂-microglobulin/Ab and washed through the filter with 10% acetic acid.

Mass spectrometric data were acquired on a home-built Fourier transform ion cyclotron resonance mass spectrometer (FTMS) (34) equipped with a nano-HPLC microelectrospray ionization source. Nano-HPLC columns were constructed of 50-μm inside diameter fused silica and packed with ∼8 cm of 5-μm diameter reverse-phase beads. An integrated micro-ESI emitter tip (∼1-μm diameter) was located a few millimeters from the column bed. Typically, 0.3% of a given class I peptide extract (∼3 × 10⁶ cell equivalents) was loaded onto a column and eluted directly into the mass spectrometer with a linear 30-min gradient of 0–70% acetonitrile in 0.1% acetic acid. Full scan mass spectra, over a mass-to-charge (m/z) range 300 ≤ m/z ≤ 2500, were acquired at a rate of 1 scan/s. After acquisition, data were saved in the NetCDF format and imported into Matlab v5.3 technical computing software (The Mathworks, Natick, MA) using the NetCDF Toolbox (publicly available from Dr. Charles R. Denham, U.S. Geological Survey, Woods Hole, MA 02543). m/z and scanline coordinate data were superimposed on a rectangular grid. Line/peak intensity data artifacts were displayed and removed by setting the intensity values for the corresponding grid coordinates to zero. The filtered data were then saved in the CDF format using the Matlab NetCDF Toolbox and imported into Finnigan Xcalibur (Finnigan, San Jose, CA) for further analysis and display.

MS/MS data were acquired on a Finnigan LCQ quadrupole ion trap mass spectrometer (Finnigan) equipped with a nano-HPLC micro-ESI source as described above. Typically, 4% of a given class I peptide extract (∼4 × 10⁷ cell equivalents) was loaded onto a column and eluted directly into the mass spectrometer with a linear 120-min gradient of 0–70% acetonitrile in 0.1% acetic acid. Data-dependent spectral acquisition was performed as follows. A full scan mass spectrum was acquired over 280 ≤ m/z ≤ 2000. The instrument control computer then selected the top five most abundant ion species, which were subjected to MS/MS analysis over the next five scans. After acquiring MS/MS data on a particular ion species, its corresponding m/z value was ignored for a period equal to the observed chromatographic peak width (∼1.5 min for the data shown herein). This data acquisition procedure minimized redundancy and allowed MS/MS analysis on peptide species spanning a wide abundance dynamic range. A typical chromatographic run contained ∼2000 total MS/MS scans, of which 35–50% contained features characteristic of peptide dissociation spectra. After acquisition, data were searched using SEQUEST; an algorithm that matches uninterpreted MS/MS spectra to theoretical peptides generated from user-specified databases (35). All data herein were searched against both human-only and nonredundant protein databases compiled at the National Center Biotechnology Information (National Institutes of Health, Bethesda, MD), and all reported peptide sequences were verified by manual interpretation.

The 20S proteasomes were purified from 721 cells as previously described (36). Proteasomes were incubated for 2 h at 24°C in 50 mM Tris-HCl and 5 mM MgCl₂ in the presence and absence of 100 μM LAC, 250 μM LLnL, or 100 μM z-L₃VS. The fluorosubstrates z-Leu-Leu-Glu-AMC, Suc-Leu-Leu-Val-Tyr-AMC, or Boc-Leu-Arg-Arg-AMC (Calbiochem) were then added to give a final concentration of 100 μM along with additional inhibitor necessary to maintain original inhibitor concentration. Fluorescence was measured every 20 min for 8–12 h using a Cytofluor II Multiwell Plate Reader (Bio-Rad, Hercules, CA) with an excitation wavelength of 380 nm and an emission wavelength of 460 nm. For each substrate, control samples from which 20S proteasomes were omitted were used to establish the background. End points chosen for all experiments were within the linear portion of the activity profile for samples incubated in the absence of inhibitors.

To examine the involvement of different proteolytic activities in the generation of peptides presented by class I MHC molecules, cells were treated with acid to remove surface class I peptide complexes, and then allowed to re-express newly synthesized complexes in the presence or absence of proteasome inhibitors for 5 h (19). Immediately after treatment of the B-LCL 721 with acid, the expression of HLA-A*0201 was nearly undetectable as quantitated by flow cytometry (Fig. 1 A). In the absence of inhibitors, cells re-expressed 30% of the pretreatment level of HLA-A*0201 in 5 h. BFA, which blocks egress of newly generated class I peptide complexes at the level of the cis-Golgi (25), inhibited re-expression almost entirely. The other class I alleles on 721 (HLA-A1, HLA-B8, and HLA-B51) behaved similarly (data not shown). Treatment of cells with acid therefore makes it possible to selectively measure the surface expression of newly generated class I peptide complexes.

Three different inhibitors were used, either alone or in combination, to block proteasome activity in 721 cells. LLnL, also known as calpain inhibitor I, reversibly inhibits proteasomes as well as several other classes of proteases (23). LAC and z-L₃VS specifically inhibit proteasomes by covalent binding to all known catalytic sites (21, 22, 24). Each of these inhibitors strongly blocked the re-expression of two class I alleles on 721 cells. HLA-A1 re-expression after 5 h was only 5–15% of that observed in the absence of inhibitor, whereas that of HLA-B51 was 15–25% (Fig. 1,B). In contrast, re-expression of HLA-A*0201 under the same conditions was 50–60% of that observed in the absence of inhibitor, whereas that of HLA-B8 was 40–50%. The concentrations of each inhibitor used in the above experiments had been determined to be saturating for inhibition of class I re-expression (data not shown). In addition, we measured the effect of inhibitors on the cleavage of model fluorosubstrates by proteasomes purified from 721 cells and which classically measure the chymotrypsin-like (ChT-L), trypsin-like (T-L), and peptidyl glutamyl peptide-hydrolyzing (PGPH) cleavage activity of proteasomes. Using the same high concentrations and preincubation conditions (see below) as were used in vivo, each inhibitor blocked 89–99% of all three cleavage activities of proteasomes, and each activity was inhibited by a minimum of 95% by at least one of the inhibitors (Table I). In particular, LAC gave essentially complete inhibition of both the ChT-L and T-L activities. Also, T-L was the major activity that persisted in the presence of z-L₃VS (10% residual activity) while PGPH was the major activity that persisted in the presence of either LLnL or LAC (10 and 11%, respectively). It is noteworthy that these differences in residual activities were not reflected in significantly different levels of re-expression in vivo (Fig. 1). In addition, using combinations of the three inhibitors at these saturating concentrations did not inhibit re-expression further. Collectively, these results demonstrate that a significant fraction of the surface expression of these class I molecules was dependent on proteolytic activities that were insensitive to inhibitors of the proteasome.

Re-expression of class I molecules in the presence of proteasome inhibitors could be explained by a preexisting pool of class I-binding peptides or peptide-MHC complexes. This pool would have been generated by proteasome activity before inhibition but not yet expressed on the cell surface at the time of acid treatment. To examine this possibility, we pretreated cells with proteasome inhibitors for 2 h before treatment with acid. Pretreatment with either LAC or LLnL decreased the re-expression observed by an additional 10% (Fig. 1 C). Longer pretreatment times did not further decrease re-expression (data not shown). This is consistent with the hypothesis that there was a small internal pool of peptides that were either preexistent or made shortly after inhibitor addition and had not yet reached the surface at the time of acid treatment. On the other hand, after complete exhaustion of this pool and in the presence of saturating levels of proteasome inhibitors, nearly half of the HLA-A*0201 and HLA-B8 molecules were still re-expressed on this cell line. Therefore, the high level re-expression of HLA-A*0201 and HLA-B8 in the presence of proteasome inhibitors cannot be accounted for by a preexisting pool of class I-binding peptides or peptide-MHC complexes.

The cell surface expression of most class I MHC molecules, including HLA-B8 (37), is almost completely dependent on the function of TAP, which transports peptides generated in the cytosol into the endoplasmic reticulum (ER) lumen. HLA-A*0201 binds to peptides produced by this pathway, but also those produced in a TAP-independent manner. We and others have suggested that this pathway consists of peptides generated entirely within the ER lumen by as yet poorly described proteases (38, 39). Thus, the re-expression of HLA-A*0201 on 721 cells treated with proteasome inhibitors might be due to this alternate pathway. We therefore examined HLA-A*0201 re-expression in the B-LCL 721.174, a mutant cell line derived from 721 that lacks a functional TAP transporter (26, 40). The constitutive surface expression of HLA-A*0201 on 721.174 is ∼17% of that on 721 (Table II). Both cells re-expressed ∼30% of their pretreatment levels of HLA-A*0201 after they were acid treated and incubated for 5 h in the absence of inhibitors. Significantly, the level of HLA-A*0201 that was re-expressed on 721 cells in the presence of LAC (405 fluorescent units) was substantially greater than the level expressed on 721.174, even in the absence of LAC (104 fluorescent units). The lower level of expressed HLA-A*0201 on 721.174 cells is not likely to reflect differences in the translation of class I or β₂-microglobulin molecules because restoration of TAP normalizes expression of HLA-A*0201 in 721.174 cells (27). Therefore ∼75% of the HLA-A*0201 that is re-expressed on 721 cells in the presence of proteasome inhibitors is dependent on TAP function. Collectively, these results suggest that the proteasomeinhibitor-insensitive proteolytic activity responsible for HLA-B8 and HLA-A*0201 re-expression resides in the cytosol.

Our results with 721 cells suggested that class I alleles differ in their ability to bind peptides generated by proteasome inhibitor-insensitive activities. Most of the positions at which residues occur that are relevant for class I binding are not at either the amino or carboxyl termini and are therefore unlikely to be affected by protease specificity. However, previous studies have suggested that there is limited trimming of the carboxyl termini of class I- associated peptides, and they therefore are the result of initial endoprotease cleavage (41, 42, 43, 44, 45, 46). Since the carboxyl-terminal residue is an important component of every known class I- binding motif, we hypothesized that the carboxyl-terminal binding motifs of class I molecules would correlate with their surface expression in the presence of proteasome inhibitors. We therefore examined the cell surface expression of a set of human class I alleles with different carboxyl-terminal binding motifs, which were expressed after transfection in the B-LCL C1R.

C1R expresses only a low level of endogenous HLA-Cw04 and does not express endogenous HLA-A or -B class I products (47). We used the pan class I- specific mAb w6/32 to detect HLA-Cw04 in untransfected C1R cells as well as all transfected products. As measured with this Ab, surface expression levels of all transfected class I MHC molecules were similar and at least 10- to 20-fold higher than that of endogenous HLA-Cw04 (data not shown). LAC and LLnL both inhibited the re-expression of the HLA-Cw04 by 70% (Fig. 2). This percentage inhibition was as great as that observed with any other C1R transfectant, and combined with a much lower level of initial expression, indicates that the endogenous C locus accounts for <5% of the w6/32 staining observed with any of the other C1R transfectants. In keeping with our earlier results, the cell surface expression of all 13 HLA-A, -B, and -C alleles was at least partially insensitive to inhibitors of proteasome activity, but the extent of this insensitivity varied among alleles (Fig. 2). As observed with 721, preincubation of the transfectants with inhibitors for longer than 2 h did not decrease re-expression further (data not shown). Similar recovery patterns were also observed when either C1R transfectants or melanoma cell lines were stained with mAbs specific for HLA-B*2702, HLA-B*2704, HLA-B*2705, HLA-B8, and HLA-A*0201, indicating that recovery was not cell type or Ab dependent (data not shown).

To test the hypothesis that re-expression was correlated with the carboxyl-terminal binding residue, we used the binding motif information available on the SYFPEITHI web site (48). Interestingly, all alleles that bind peptides with basic carboxyl termini (HLA-B*2705, HLA-A68, and HLA-A3.1) demonstrated relatively high levels of re-expression in the presence of proteasome inhibitors. This is consistent with the idea that peptides with basic carboxyl termini are preferentially generated by proteasome inhibitor-insensitive proteolytic activity. Further support for this idea was evident when the re-expression of structurally related alleles was compared. For example, HLA-A69 consists of the α1 domain of HLA-A68 linked to the α2 domain of HLA-A*0201. Consequently, its binding specificity at the P2 anchor is that of HLA-A68, but it prefers peptides with hydrophobic carboxyl termini like HLA-A*0201 (49, 50). Although the re-expression of HLA-A68 was ∼65% in the presence of proteasome inhibitors, that of HLA-A69 was only 30%. In addition, HLA-B*2705 binds peptides with basic, aromatic, or aliphatic carboxyl termini, whereas HLA-B*2704 prefers peptides with hydrophobic C termini (51, 52). Again, re-expression of HLA-B*2704 was only about half that of HLA-B*2705 in the presence of proteasome inhibitors. These data imply that class I-associated peptides with basic carboxyl termini represent a significant component of the peptides generated by proteasome inhibitor-insensitive proteolytic activities.

Despite the above evidence, HLA-B8 and HLA-B*2702 show relatively high re-expression in the presence of proteasome inhibitors although their published motifs do not include basic carboxyl-terminal residues. Interestingly, within the group of alleles that do not bind to peptides with basic carboxyl termini, there is a suggestion that re-expression in the presence of proteasome inhibitors correlates with the ability to bind to a larger range of carboxyl-terminal residues. This is most evident when the motifs of HLA-B*2702 and HLA-B*2704 are compared, as well as those of HLA-B51 and HLA-B35. Collectively, these data suggest that some proteases active in the presence of proteasome inhibitors generate class I- associated peptides with carboxyl termini that do not include basic residues.

The results above demonstrate that class I- associated peptides can be generated by at least one cytosolic protease activity that is resistant to proteasome inhibitors. However, they provide only suggestive information about the nature of the peptides presented. We therefore extracted the peptides associated with class I molecules re-expressed on acid-treated cells that were incubated with or without 250 μM LLnL. These peptide mixtures were analyzed by liquid chromatography interfaced directly to a LC/FTMS (34). In addition to temporal separation of peptide species, this method provides a 1000-fold improvement in mass-resolving power and a 10- to 100-fold improvement in mass accuracy compared with previously used triple quadrupole mass analyzers (9, 38, 53). This enables us to unambiguously distinguish peptides differing by as little as 0.01–0.1 Da compared with ∼1.5 Da afforded by triple quadrupole instrumentation. When combined with software allowing two-dimensional displays of the data, this method allowed us to visualize 2500 different peptide species in a single display.

We first analyzed peptides associated with HLA-A1, the class I MHC allele that was the most sensitive to proteasome inhibitors of all those tested above. By comparing m/z vs retention time plots of peptide extracts from equivalent numbers of inhibitor-treated and untreated cells, we observed that the vast majority of individual peptides normally presented by HLA-A1 were decreased in abundance by at least 90% in the sample from cells treated with LLnL (Fig. 3, A and B). This demonstrates that proteasomes are responsible for the generation of most of the peptides associated with HLA-A1. In addition, the remaining species that were visible in the sample from proteasome inhibitor-treated cells were frequently distinct from those in control cells. Also, when the intensity values of the sample from inhibitor-treated cells were normalized for the level of HLA-A1 as determined from flow cytometry, the peptide profile was clearly different from that of untreated cells (Fig. 3 C). The persistence of a small but distinct subset of peptides at elevated levels in the presence of proteasome inhibitors indicates that proteolytic activities with specificities that are distinct from the uninhibited proteasome are responsible for their generation.

We next looked at the impact of proteasome inhibitors on the distribution of peptides associated with HLA-B*2705 and HLA-Cw04. As shown above, HLA-Cw04 is expressed constitutively at only 5% of the level of HLA-B*2705 and is re-expressed poorly in the presence of LLnL, while HLA-B*2705 re-expression is largely unaffected by proteasome inhibitors. Therefore, the vast majority of peptides in these extracts should have been associated with HLA-B*2705. The m/z vs retention time plots of peptides expressed in the presence of LLnL was similar in complexity to that of peptides expressed on untreated cells (Fig. 4). Thus, the re-expression of HLA-B*2705 molecules in cells treated with LLnL was not due to the presentation of a small number of highly abundant peptides. Proteasome inhibition resulted in an observable decrease in only a subset of early eluting peptides (boxes in Fig. 4). Unlike what was seen with HLA-A1, the majority of the peptides that eluted later in the gradient seemed unaffected (two areas are circled as examples in Fig. 4).

To determine more accurately the relative abundances of individual peptide species in the two samples, we analyzed all of the masses visible in a representative 1-s interval of the two m/z vs retention time plots (indicated by the arrows in Fig. 4). In this time slice, 20 of 21 observed peptides changed in abundance by no more than a factor of 5 (data not shown). In contrast to what was observed with HLA-A1, these were equally distributed between groups that increased or decreased in abundance. The abundance of the remaining peptide increased 9-fold upon proteasome inhibition. These results demonstrate that activities sensitive to proteasome inhibitors are required for the generation of at most a small subset of peptides associated with HLA-B*2705, and that proteasome inhibitor-resistant activities can generate the majority of peptide species that are associated with HLA-B*2705 under normal conditions.

To gain information about the proteasome inhibitor-resistant pathways of class I epitope generation in the cytosol, we characterized a subset of peptides re-expressed in the presence of proteasome inhibitors using a combination of online microcapillary HPLC and collision-activated dissociation on a quadrupole ion trap mass spectrometer. Forty-six peptides were sequenced from the mixture associated with the HLA-B and -C alleles on 721 cells in the presence of LLnL (Table III), while 36 were sequenced from the mixture associated with HLA-B*2705 and HLA-Cw04 on C1R/B*2705 cells in the presence of LLnL (Table IV).

All of the sequenced peptides were derived from endogenously expressed proteins. Forty-four of the 82 peptides were from known proteins with nuclear, cytosolic, mitochondrial, or peroxisomal locations and are therefore translated in the cytosol on free polyribosomes. Based on the TAP dependence of the presenting class I molecules and the cytosolic localization of these proteins, generation of these peptides most likely occurs in the cytosol. Of the 22 peptides derived from known membrane or secreted proteins, only 6 were from signal sequences, while 16 were derived from luminal or transmembrane regions. Both TAP-independent (38) and -dependent pathways (53, 54, 55) for the presentation of these peptides have been described. Regardless of the exact mechanism, these results demonstrate that proteasome inhibitor-resistant proteases responsible for class I epitope generation degrade a variety of proteins localized throughout the cell as well as proteins synthesized in either the cytosol or the ER.

Thirty-eight of 46 peptides sequenced from the mixture associated with the HLA -B and -C alleles on 721 cells conformed to the established binding motifs for at least one of HLA-B51, HLA-B8, or HLA-Cw01 molecules (56) (Table III). The remaining eight peptides fit none of the published motifs well and may either be derived from an unidentified C locus product or bind to one of the known B or C alleles despite the lack of a canonical binding motif. Regardless, all of the peptides sequenced contained either an aromatic or large hydrophobic residue at the carboxyl terminus. In addition, there was no obvious constraint on the occurrence of particular aliphatic residues at this position in the presence of LLnL (Table III). Twenty-eight of 36 peptides sequenced from the mixture associated with HLA-B*2705 and HLA-Cw04 contained the Arg element of the HLA-B*2705 motif at P2 while the remaining 8 contained a canonical HLA-Cw04 motif residue at this position (Table IV) (56). Unlike all other alleles from which sequences were obtained, HLA-B*2705 binds peptides containing aromatic, hydrophobic, or basic C-terminal amino acids (51, 57). Of 28 HLA-B*2705-associated peptide sequences, 8 contained a lysine or arginine at the carboxyl terminus, while 9 others ended in aromatic residues and 9 ended in nonaromatic hydrophobic residues (Table IV). Thus, the peptides that persisted upon proteasome inhibition demonstrated the complete range of C termini known to bind to HLA-B*2705. These results demonstrate that the proteases that generate class I- associated peptides in proteasome inhibitor-treated cells collectively have a broad spectrum of specificities.

Several studies using inhibitors of proteasome activity have either supported or questioned the general involvement of proteasomes in the generation of class I MHC-associated peptides. In this study, we used 3 different inhibitors to examine the contribution of proteasomes to the generation of peptides associated with 13 different human class I MHC alleles. In the presence of these proteasome inhibitors, 9 of 13 human class I alleles continued to be expressed at >30% of control levels. This re-expression could not be accounted for by the use of peptides generated directly in the ER in a TAP-independent manner. In addition, the allelic variation did not correlate with differences in their affinities for TAP (presumably via tapasin) (58). We used a combination of Fourier transform mass spectrometry and two-dimensional data displays to examine the complexity of peptides associated with two alleles. In the presence of proteasome inhibitors, expression of HLA-A1 was substantially reduced, as was the expression of the vast majority of its associated peptides. In contrast, expression of HLA-B*2705 was largely insensitive to proteasome inhibition and a diverse array of peptides continued to be presented. Thus, the peptides expressed in the presence of proteasome inhibitors represents a significant fraction of the peptides associated with some class I MHC alleles under normal conditions.

One possible explanation for our results was that proteasome inhibitor-resistant class I peptide expression resulted from large internal pools of peptides. Although preincubation of cells with proteasome inhibitors demonstrated the existence of an internal store of peptides that could associate with class I MHC molecules, class I re-expression occurred even after depletion of these stores. The minimal size of this pool, combined with the overall inhibition of class I MHC re-expression, demonstrates that there is not a large reservoir of peptides in the cells that are destined for class I MHC binding. This is an important consideration given the role of class I peptide presentation in controlling intracellular pathogens. Upon initial infection, peptides generated from newly translated proteins should be immediately available for class I binding and surface expression.

Another possible explanation for the continued expression of some class I MHC molecules in the presence of proteasome inhibitors is that these compounds do not completely block all activities of the proteasome. Several previous studies have demonstrated that the ChT-L activity of mammalian proteasomes is rapidly inactivated by relatively low concentrations of LLnL (23), LAC (21, 22, 59), or z-L₃VS (59) and that inhibition by the latter two compounds is irreversible (21, 59). Although the T-L and PGPH activities of the proteasome are less sensitive than the ChT-L activity, they can nonetheless be blocked by higher concentrations of any of these three inhibitors (21, 22, 23, 59). Importantly, preincubation with inhibitors greatly increases their effectiveness (59, 60), and we adopted this strategy in the present work. Of the three activities, PGPH is the most insensitive to LAC (21), while T-L is the most insensitive to LLnL (23). In the present study, we found that all of these activities of purified proteasomes in vitro were blocked at least 95% by at least one of the inhibitors at the same concentrations and conditions used in vivo. If residual proteasome activity accounted for the high level re-expression of some class I alleles, it would be expected that this would vary depending on the inhibitor used because of the differences in effectiveness of the inhibitors against different proteasome activities. This was not observed. Alternatively, if re-expression were due simply to a fraction of proteasomes that were not inhibited at all, then one would not expect to see allele-specific differences in the extent of re-expression. Furthermore, saturating concentrations of LAC, LLnL, or z-L₃VS did not inhibit class I re-expression to a greater extent in combination than individually, despite the fact that they differ in chemical structure, mode of inhibition, and profile of in vitro activity that persisted in the face of inhibition. Finally, the peptide sequences presented by class I MHC molecules re-expressed in the presence of inhibitors were not enriched for those with C-terminal residues that would be generated by residual PGPH or T-L activity. Collectively, these results suggest that proteolytic activities other than the proteasome may be responsible for the high re-expression of some human class I MHC alleles in the presence of proteasome inhibitors. However, it is also possible that residual proteasome activities are responsible for our observations.

Our data suggest that the specificity for the anchor residue at the peptide carboxyl terminus is an important determinant of the expression of class I alleles in the presence of proteasome inhibitors. We observed that class I MHC molecules with a preference for basic carboxyl-terminal anchors were generally re-expressed at higher levels. In addition, by comparing pairs of highly related class I MHC alleles with differences in carboxyl-terminal binding preference, we found that those with a preference for basic residues were expressed at higher levels. Consistent with our results, Benham et al. (17) also observed that two alleles with basic carboxyl-terminal binding motifs, HLA-A3.1 and HLA-A11, continued to form peptide-dependent, SDS-stable dimers in the presence of proteasome inhibitors. However, we found that HLA-A68, which binds to basic carboxyl-terminal anchors, was also insensitive to proteasome inhibitors, whereas Benham et al. (17) found the opposite result. One possible explanation for this discrepancy is that peptide-associated HLA-A68 is more sensitive to SDS than other class I molecules. Interestingly, an endoprotease that predominately cleaves after Lys or Arg has recently been identified in the cytosol of mammalian cells (20, 61). We suggest that this protease produces at least some of the peptides with basic carboxyl termini that are associated with class I MHC molecules in the presence of proteasome inhibitors.

Despite the foregoing, a tryptic-like endoprotease activity cannot account for all of the class I-associated peptides displayed in the presence of proteasome inhibitors. Under these conditions, the majority of HLA-B*2705-associated peptides sequenced did not contain basic carboxyl termini. In addition, sequenced peptides extracted from HLA-B8, HLA-B51, and HLA-Cw01 all contained large hydrophobic residues at their carboxyl termini. Therefore, nonproteasomal proteases that generate class I-associated peptides must also include those capable of generating hydrophobic and aromatic carboxyl termini. In keeping with this, the carboxyl-terminal peptide-binding motifs of HLA-B8 and HLA-B*2702 do not include basic residues, yet these alleles are re-expressed at relatively high levels in the presence of proteasome inhibitors. Our data suggest that aside from a preference for basic carboxyl-terminal residues, another determinant of the re-expression of class I MHC alleles in the presence of proteasome inhibitors is their ability to bind to a larger range of carboxyl-terminal residues.

Although the overall complexity of HLA-B*2705-associated peptides was similar in the presence and absence of proteasome inhibitors, individual peptides in the mixture either decreased by up to 5-fold or increased by up to 9-fold. Because of the complexity of the mixture, these changes seem unlikely to be due to peptide competition. Instead, these peptides may be either generated or destroyed by different proteasome activities in addition to possibly being generated by nonproteasomal proteases. We have previously shown that the M1_58–66 epitope from influenza A virus is produced by a proteasome inhibitor-resistant protease and destroyed by proteasomes (19). The data in the present study suggest that the final level of presentation of many epitopes is the result of interplay between different proteolytic activities.

In summary, we have demonstrated that proteasome inhibitor-resistant pathways produce a significant fraction of peptide species associated with many class I alleles. These peptides are derived from a wide range of cellular proteins and display heterogeneous C termini. This suggests that they are generated either by a multifunctional protease or by multiple proteases with different cleavage specificities. Further characterization of these proteolytic activities will provide insight into mechanisms for class I epitope generation.

We thank Janet Gorman for her expert technical assistance in Ab production, Dr. Jacques Retief for his assistance in database management at the University of Virginia, and C. A. Mosse, V. L. Crotzer, and Dr. T. J. Bullock for helpful discussions.