Tapasin Enhances Assembly of Transporters Associated with Antigen Processing-dependent and -independent Peptides with HLA-A2 and HLA-B27 Expressed in Insect Cells*

Assembly of HLA class I-peptide complexes is assisted by multiple proteins that associate with HLA molecules in loading complexes. These include the housekeeping chaperones calnexin and calreticulin and two essential proteins, the transporters associated with antigen processing (TAP) for peptide supply, and the protein tapasin which is thought to act as a specialized chaperone. We dissected functional effects of processing cofactors by co-expressing in insect cells various combinations of the human proteins HLA-A2, HLA-B27, β2-microglobulin, TAP, calnexin, calreticulin, and tapasin. Stability at 37 °C and surface expression of class I dimers correlated closely in baculovirus-infected Sf9 cells, suggesting that these cells retain empty dimers in the endoplasmic reticulum. Both HLA molecules form substantial quantities of stable complexes with insect cell-produced peptide pools. These pools are TAP-selected cytosolic peptides for HLA-B27 but endoplasmic reticulum-derived, i.e. TAP-independent peptides for HLA-A2. This discrepancy may be due to peptide selection by human TAP which is much better adapted to the HLA-B27 than to the HLA-A2 ligand preferences. HLA class I assembly with peptides from TAP-dependent and -independent pools was enhanced strongly by tapasin. Thus, tapasin acts as a chaperone and/or peptide editor that facilitates assembly of peptides with HLA class I molecules independently of mediating their interaction with TAP and/or retention in the endoplasmic reticulum.

HLA class I (HCI) 1 molecules present short peptides mainly derived by cytosolic degradation of cellular proteins to cytotoxic T cells. Assembly of these peptides with newly synthesized HCI molecules in the endoplasmic reticulum (ER) is assisted and controlled by a multitude of proteins (1). HCI heavy chains associate initially with the ER chaperone calnexin which, although not essential for HCI assembly (2), has been reported to facilitate folding of heavy chains and prevent their aggregation (3). Upon binding of ␤ 2 -microglobulin (␤ 2 m), HCI heavy chains dissociate from calnexin. Empty HCI/␤ 2 m dimers are found in complexes including the soluble ER chaperone calreticulin (4), the putative chaperone tapasin (4 -6), and Erp57, described previously as thiol-dependent oxidoreductase (7)(8)(9). These complexes associate then with the heterodimeric TAP1/TAP2 transporter which delivers cytosolic peptides into the ER. Once a peptide has bound to HCI molecules, these are released from assembly complexes, leave the ER, and transit to the cell surface.
Since most protein interactions in HCI loading complexes appear to be formed simultaneously, probably in a cooperative manner (10), the functions of individual proteins, and the precise nature of the formed contacts are only partially understood (11). So far only peptide supply by TAP (12) and HCI interaction with tapasin (5) have been shown to be essential for HCI assembly with peptides. Tapasin mediates association of empty HCI dimers with TAP complexes (5,13). This role of an intermediary involves binding of a tapasin moiety within the carboxyl-terminal 128 residues to TAP and interaction of the 50 amino-terminal tapasin residues with HCI molecules (10,14). Reconstitution of normal peptide assembly in the tapasin-deficient cell line .220.B8 by a soluble tapasin molecule unable to mediate TAP association suggests that tapasin may also act as a chaperone facilitating peptide assembly with HCI molecules (15).
HCI molecules display great polymorphic variation that determines their peptide ligand preferences but may also affect their associations with other proteins. Both parameters may affect the efficiency and mode of HCI assembly with peptide. Peptide preferences of individual HCI molecules may be more or less well adapted to the products of other components of the antigen processing machinery, for example peptides generated by proteasome or pumped by TAP. Variable interaction with processing cofactors in the ER may also affect HCI peptide loading.
Some evidence for HCI polymorphism-related variation in assembly of HCI molecules has been reported. The speed of assembly in complexes and progression to the cell surface has been reported to vary significantly among HCI alleles (16). Absence of TAP or tapasin affects HCI molecules to a different degree. HLA-A2, the most frequent HCI allele in caucasian populations, is affected the least by both deficiencies (17,18), presumably because of its capacity of binding signal sequencederived peptides which is shared with few other HCI alleles (1). More recently, HLA-B27 has been shown to depend less than two other HLA-B alleles on tapasin for peptide assembly (19). Moreover, HCI molecules have been found to display considerable variation with respect to the strength of TAP interaction (20) and may also vary with respect to their dependence on the proteasome for generation of antigenic peptides (21). However, the molecular mechanism of these differences has so far not been elucidated.
One candidate mechanism affecting efficiency of peptide presentation by HCI molecules is peptide supply by the TAP complex. Studies on rodent transporters have demonstrated that strongly incompatible ligand preferences of TAP and MHC class I molecules result in poor peptide supply to the latter (22,23). Although the human transporter is clearly less selective than mouse TAP and rat TAP1-TAP2 b complexes (24), we have found that ligand preferences of individual HCI molecules vary greatly with respect to their adaptation to human TAP preferences (25,26). While HLA-A2 preferences are poorly adapted to those of TAP, ligands for HLA-B27 show the highest TAP affinities among all HCI molecules studied so far (25). These observations raised the question whether peptides translocated by human TAP are more likely to assemble with HLA-B27 than with HLA-A2.
We have used the insect cell/baculovirus system to study functional interactions of proteins involved in class I antigen processing. This system has two advantages. First, since insects do not possess a specific immune system and the associated specialized antigen-processing proteins, fly cells can be used to freely combine sets of processing proteins, thereby creating cells lacking more than one component of the processing machinery. Second, overexpression of proteins in this system may in some cases allow for detecting protein interactions that are difficult to observe at physiological expression levels. By using this system, we addressed the following issues: (i) what are the effects of tapasin, calnexin, and calreticulin on assembly of TAP-dependent and independent peptides with HCI molecules; (ii) are there allelic variations in these effects; and (iii) which interactions between HCI molecules, TAP, calnexin, and calreticulin can be observed in insect cells?

EXPERIMENTAL PROCEDURES
Viruses-Baculoviruses expressing wild type human TAP1.0101 and TAP2.0101 proteins have been described previously (27). For expression of wild type TAP1-TAP2 complexes, a dual promoter virus containing human TAP1 and TAP2 cDNAs (a gift from Dr. R. Tampé, MPI Martinsried) was used. Mutant TAP1 and TAP2 proteins were generated by replacement of the Walker A sequence motifs by a peptide linker; single and double mutant TAP subunits form dimeric complexes that do not translocate peptides. 2 Correct sequences of all cDNAs to be expressed in baculovirus were verified by PCR-based complete sequencing of plasmid inserts using an Applied Biosystems Inc. automated sequencer. cDNAs were sequenced after cloning of PCR products or restriction fragments into pCRII (Invitrogen, Carlsbad, CA), pTAG (R&D Systems, Abingdon, UK), or pBluescript SKϩ (Stratagene, La Jolla, CA) vectors. Human ␤ 2 m cDNA was PCR-amplified in 22 cycles from plasmid HS4 provided by Dr. S. Kvist, Stockholm, Sweden, using primers with internal BamHI sites. Then the cDNA was cloned into pCRII and finally subcloned into the BamHI sites of pVL1393 (Invitrogen) and of the dual promoter vector pAcUW51 (PharMingen, San Diego, CA) already containing HLA-A2 or B27 inserts.
An HLA-A*0201 cDNA cloned into an M13 phage vector was obtained from Drs. P. Parham and J. Gumperz (Stanford). Insert was amplified from phage-derived double-stranded DNA in a 22-cycle PCR using primers with an internal BglII (5Ј) or EcoRI (3Ј) site, respectively, and cloned into pCRII. A2 cDNA was then cloned as Bgl/Eco fragment into pVL1392 (Invitrogen) and pAcUW51. An HLA-B27*05 cDNA cloned into pUHD was obtained from Dr. K. Frü h, San Diego. Insert was amplified in an 18-cycle PCR using primers with a BamHI (5Ј) or EcoRI (3Ј) site, respectively, cloned into pMOS Blue (Amersham Pharmacia Biotech) and then into pBluescript SKϩ for sequencing. Sequencing revealed that the amplified as well as the original pUHD-cloned cDNA differed from the published consensus sequence by a single nucleotide replacement resulting in an aspartic acid for asparagine substitution at codon 151. To correct the sequence, the B27 insert was transferred as Bam/Eco fragment into BglII/EcoRI-digested pAcUW51, and site-directed mutagenesis using the QuikChange kit (Stratagene) was performed to obtain the correct B*2705 sequence.
A human calnexin cDNA cloned in pMCFR-PAC was obtained from Drs. P. Cresswell and T. Novak, Yale University, and subcloned into pBluescript SKϩ containing six histidine codons followed by a stop codon. Calnexin stop codon and 3Ј-untranslated sequence were then removed by loop out mutagenesis so that the last calnexin codon was joined in frame to the first histidine codon. To shorten and modify the 5Ј-untranslated calnexin sequence, a 425-base pair KpnI/HindIII 5Ј fragment was replaced by an equivalent PCR-amplified fragment with a 5Ј BamHI site. Finally, the complete calnexin cDNA was subcloned into pVL1393. A human calreticulin cDNA cloned into pTZ18U was obtained from Dr. J. D. Capra, Dallas. Insert was PCR-amplified using primers with an internal BglII (5Ј) or EcoRI (3Ј) site, respectively, and cloned into pTAG before subcloning as Bgl/Eco fragment into pVL1392 and pAcUW51.
A human tapasin cDNA with a 3Ј polyhistidine extension was generated by PCR amplification of reverse-transcribed (Copy Kit, Invitrogen) cDNA from the cell line U937 using a high fidelity mixture of thermostable DNA polymerases (Advantage HF kit, CLONTECH). The tapasin cDNA was sequenced in pCRII before subcloning into pVL1393. Two TAP-independent peptides were expressed in the baculovirus system as minigenes transferred from a vaccinia virus expression plasmid. Briefly, the vaccinia vector pSC11s was first modified by insertion of complementary oligonucleotides encoding the signal peptide of the adenovirus E3/19K protein. Then additional oligonucleotides coding for an HLA-A2-restricted epitope (FLPSDFFPSV (28)), or an HLA-B27-binding peptide (RRYQKSTEL (29)), respectively, were inserted 3Ј of the signal peptide encoding sequence, and the complete sequences were transferred into pVL1393. Both epitopes have high binding affinity for their restricting HLA molecules, and vaccinia-expressed sig/core 18-27 gives rise to high levels of TAP-independent target cell lysis by A2-restricted CTLs. 3 All recombinant viruses were produced by co-transfection of Spodoptera frugiperda (Sf9) cells with 100 ng of baculovirus DNA (Baculo-Gold, PharMingen) and 3 g of pVL1392, pVL1393, or pAcUW51cloned cDNAs as described (30). Control baculoviruses used in this study express human 65-kDa glutamic acid decarboxylase (GAD65) or the intracellular portion of the tyrosine phosphatase IA-2 (IA-2ic), two autoantigens targeted in type 1 diabetes (31).
Antibodies-Monoclonal antibody (mAb) AF8 specific for human calnexin (32) was provided by Dr. M. Brenner (Harvard Medical School, Boston), and mAb BBM.1 specific for human ␤ 2 m was kindly provided by Dr. G. Moldenhauer, Heidelberg, Germany. Hybridomas producing mAb W6/32 (recognizing HCI/␤ 2 m dimers), BB7.2 (specific for HLA-A2), and B27M1 (specific for HLA-B27) were obtained from American Type Culture Collection (Manassas, VA). Hybridoma HC10 with specificity for free HCI heavy chains (33) was obtained from Dr. H. Ploegh (Harvard). mAb 148.3 recognizing the carboxyl terminus of human TAP1 and mAb 429.3 (used for Western blots) and 435.3 (used for immunoprecipitations) with specificity for the carboxyl-terminal domain of human TAP2 have been described previously (27). mAb were purified from ascites obtained from hybridoma-inoculated Balb/c mice by affinity purification on protein G-Sepharose columns (Gamma-Bind Plus, Amersham Pharmacia Biotech). Rabbit serum R425 recognizing denatured human HCI heavy chains was provided by Drs. S. Kvist and P. Wang, Stockholm, Sweden. Rabbit sera specific for human calnexin and calreticulin were purchased from StressGen Biotechnologies Corp. (Victoria, Canada), and a rabbit serum with specificity for human ␤ 2 m was obtained from Dako (Glostrup, Denmark).
Metabolic Labeling and Immunoprecipitation-5 ϫ 10 6 Sf9 cells adhering to 6-cm inner diameter tissue culture dishes were infected with 2 ϫ 10 7 pfu of HCI/␤ 2 m double insert virus supernatants together with 4 ϫ 10 7 pfu of each of two other viral supernatants for 1 h. Infectious supernatants were then replaced by 5 ml of complete TMN-FH (Roche Molecular Biochemicals or Sigma). To assess steady state levels of HCI molecules, cells were labeled after 24 h infection by incubation for 40 min in 0.6 ml of Grace's medium without methionine supplemented with 0.2 mCi of [ 35 S]methionine (EasyTag, NEN Life Science Products). Then 0.9 ml of methionine-deficient Grace's medium with 10% fetal calf serum dialyzed against PBS was added, and labeling was continued overnight. Labeled cells were recovered by rinsing in cold PBS, washed once in PBS with 1 mM PMSF, and lysed by incubating 1 h in 1 ml of a buffer containing 150 mM NaCl, 40 mM Tris, pH 7.4, 1% Nonidet P-40 (Pierce), and a mixture of protease inhibitors: 0.2 mM 4-(2-aminoethyl)-benzenesulfonyl fluoride-HCL (ICN, Costa Mesa, CA), 1 g/ml aprotinin (ICN), 1 mM EDTA, 1 mM benzamidine (Calbiochem), 5 M leupeptin (Alexis, Laufelfingen, Switzerland), and 10 M pepstatin (Alexis). Lysates were clarified by centrifugation at 20,000 ϫ g for 10 min and precleared by incubation for 2 h at 4°C with 40 l of protein G-Sepharose beads (Gamma-Bind Plus, Amersham Pharmacia Biotech) or Sepharose 4B beads coupled to a control mAb. After removal of beads by centrifugation for 1 min at 1,000 ϫ g, cleared lysates were split in equal parts (for HLA-A2) or at a ratio of 1:2 (for HLA-B27) for immunoprecipitation by mAb HC10 and W6/32 (10 g each), respectively. 2.5 h later, 12.5 l of protein G-Sepharose was added for a further 30 min to recover immunoprecipitated material. Then beads were washed with a series of Tris/NaCl buffers containing 0.1% Nonidet P-40 with increasing pH (2ϫ pH 7.4, 1ϫ pH 8.0, 1ϫ pH 9.0, 1ϫ pH 9.0 with 250 mM NaCl added), washed once in 50 mM Tris, pH 6.8, with 0.01% Nonidet P-40, and finally boiled in 20 l reducing SDS-PAGE sample buffer.
In experiments on HCI stability at 37°C, precleared lysates were split in two equal parts that were incubated for 1 h at 4 or 37°C, respectively. Then HCI molecules were immunoprecipitated by incubation for 45 min with 10 g of mAb HC10 or W6/32 pre-absorbed onto 8 l of protein G-Sepharose. Beads were washed as described above. Precipitated proteins were separated in 10% SDS-PAGE gels. Gels were fixed in 50% trichloroacetic acid, destained in 50% methanol, 10% acetic acid, enhanced in 1 M sodium salicylate, pH 6.0, and finally autoradiographed for 24 h to 1 week. Autoradiographs were scanned and analyzed by densitometry using NIH Image 1.62 software.
Co-precipitation Experiments-To analyze protein interactions, 5 ϫ 10 7 plastic-adherent Sf9 cells were infected for 1 h with 1.5 ϫ 10 8 pfu in a total volume of 5 ml of TMN-FH. Infectious supernatant was then replaced by 20 ml of fresh medium, and cells were cultured for 2.5 to 3 days. Cells were harvested, washed once in cold PBS with 1 mM PMSF, and unless used immediately, snap-frozen in aliquots of 5 ϫ 10 6 in liquid nitrogen and stored at Ϫ80°C until used. 5 ϫ 10 6 cells were lysed for 15 min at 4°C in 0.65 to 1.0 ml of precipitation buffer (25 mM Tris, pH 7.4, 150 mM NaCl) containing 1% digitonin (Sigma) and protease inhibitors (see above). After clarification and preclearing as described above, a lysate volume corresponding to 1-2 ϫ 10 6 cells was added to 10 g of mAb and incubated for a period ranging from 1 to 16 h. Immune complexes were recovered by incubation with 10 l of protein G-Sepharose during 30 min, followed by three washes in 500 l of cold precipitation buffer with 0.5% digitonin. Finally, complexes were solubilized by boiling for 10 min in 35 l of reducing SDS-PAGE sample buffer. In the experiment shown in Fig. 1, protein concentrations of Nonidet P-40 lysates of Jesthom and Sf9 cells were quantified by the BCA assay (Pierce). Precipitated proteins were separated by SDS-PAGE in minigels with the appropriate percentage of acrylamide (7.5% for TAP and calnexin, 10% for HCI and calreticulin, and 12.5% for ␤2m). Separated proteins were then blotted onto polyvinylidene difluoride membranes (PVDF, Amersham Pharmacia Biotech) at 80 V for 1 h in 48 mM Tris, 390 mM glycine, 0.1% SDS, 20% methanol. Blotted proteins were then visualized with a standard ECL protocol, using primary antibody dilutions of 1:10,000 to 1:50,000.
Analysis of Cell Surface Expression of HCI Molecules-For fluorescence-activated cell sorter stainings, 5 ϫ 10 6 Sf9 cells were infected as for metabolic labelings with a combination of three viral supernatants. 36 h after infection, cells were recovered by rinsing, washed once in cold PBS/PMSF, and resuspended at 2 ϫ 10 7 cells/ml in staining buffer (PBS with 0.05% NaN 3 , 1% fetal calf serum). Aliquots of 1 ϫ 10 6 cells were incubated for 30 min at 4°C with mAb at 10 g/ml, washed twice, incubated 30 min at 4°C with fluorescein isothiocyanate-labeled goat Ab to mouse Ig (SBA, Birmingham, AL), diluted 1:100, and again washed three times. Immediately before analysis on a FACSCAN cytometer (Becton Dickinson, San Jose, CA), propidium iodide (0.8 g/ml) was added to stained cells. Only live cells excluding propidium iodide were evaluated.

Expression of HLA-A2 and B27 and Human Tapasin, Calnexin, and Calreticulin in Baculovirus-infected Insect Cells-
We expressed full-length cDNA clones coding for human calnexin, calreticulin, tapasin, ␤ 2 m, HLA-A*0201, and HLA-B*2705, and two leader sequence-coupled peptides with high affinity for HLA-A2 or B27, respectively, in recombinant baculoviruses under the control of the polyhedrin promoter. Cal-nexin and tapasin were expressed as fusion proteins joined to six carboxyl-terminal histidine codons. For expression of HLA class I heavy chain/␤ 2 m dimers and of TAP1-TAP2 complexes, we also produced viruses expressing two proteins under the control of the polyhedrin and the p10 promoter, respectively.
To verify expression of proteins with correct molecular weights, we blotted Nonidet P-40 lysates of insect cells infected for the 3 days with one or several viruses onto PVDF membranes and quantified expressed proteins using specific mAb or sera and an ECL protocol (Fig. 1). As controls, lysates from the human B cell line Jesthom expressing HLA-A2 and HLA-B27 were analyzed. As shown in Fig. 1, all recombinant proteins had molecular weights that were similar or identical to their physiological counterparts expressed in the human B cell. This includes HLA-A2 and B27 heavy chains, suggesting that the insect cell-expressed HCI molecules are glycosylated to a similar extent as in human cells. Recombinant calnexin migrated slightly more slowly than its physiological equivalent due to the carboxyl-terminal polyhistidine extension. Recombinant tapasin could be purified as a 49-kDa protein based on the interaction of its polyhistidine extension with Ni 2ϩ -nitrilotriacetic acid resins (not shown).
Recombinant proteins represented a higher percentage of total Nonidet P-40-solubilized cellular protein than their physiological counterparts in the human B cell line. The two TAP subunits and the chaperones calnexin and calreticulin showed the highest relative levels of expression; equal protein amounts of B cells contained on average 10-fold (between 5-and 30-fold in several experiments) less of these proteins than insect cells infected with the relevant viruses. HCI heavy chains and ␤ 2 m were expressed at relatively more moderate levels; in several independent experiments, the ratio of expression in insect cells to B cells (normalized for equal protein amounts) was between 1 and 3. Tapasin expression levels were not compared with those in human B cells.
Effect of Chaperones and TAP on Assembly in Dimers of HLA-A2 and HLA-B27-To study the effect of antigen processing cofactors on steady state levels of HCI dimers, insect cells were infected with various combinations of three viruses, driving expression of up to five human proteins. 24 h after infection, cells were labeled metabolically for 12 h, followed by immunoprecipitation of unfolded free heavy chains or folded HCI dimers with an excess of mAb HC10 or W6/32, respectively. Precipitated HCI molecules were separated by SDS-PAGE and quantified by densitometry. For each condition,   2 shows three parameters as follows: (i) the percentage of HCI molecules recovered by dimer-specific mAb W6/32 relative to the sum of HCI molecules precipitated by HC10 and W6/32 (given as numbers); (ii) the amounts of recovered free heavy chains and (iii) of dimers relative to control infections, in which cells expressed HCI/␤2m together with two control proteins (expressed as a bar graph). Fig. 2 is representative of three independent experiments. Similar results were obtained in four experiments in which the amounts of dimers and free heavy chains were quantified by Western blot staining with a serum (R425) recognizing denatured heavy chains (not shown).
In control experiments on Jesthom B cells, 10 -20 times less HCI molecules were recovered by HC10 than by W6/32 (not shown). Only HC10-reactive heavy chains could be recovered from insect cells lacking ␤2m expression (not shown).
Both the percentage assembling in dimers in the absence of processing cofactors and the effect of cofactors differed substantially between HLA-A2 and B27. In the case of HLA-A2 ( Fig.  2A), 35% of heavy chains formed dimers with ␤ 2 m in the absence of any human cofactor. Surprisingly, this percentage was not affected by co-expression of chaperones or peptide sources including a TAP-independent peptide with high A2 binding affinity. However, although not affecting the percentage of A2 molecules forming dimers, calnexin and calreticulin co-expression resulted in modest (30 -50%) simultaneous increases in the amount of free heavy chains and dimers relative to control infections. Dimer formation of HLA-A2 was limited to less than 40% of HLA-A2 molecules under all conditions.
Less than 10% of B27 molecules formed dimers in the absence of cofactors (Fig. 2B). Co-expression of peptide sources was sufficient to increase this percentage and the amount of HLA-B27 dimers relative to control infections without affecting the amount of free heavy chains. A TAP-independent peptide increased the percentage of dimers much more efficiently (factor 3.9) than TAP alone (factor 1.6). Co-expression of chaperones did not enhance the effect of the TAP-independent peptide with high B27 affinity. However, tapasin, which had only a small effect when co-expressed alone or with calreticulin, and a modest effect together with calnexin, enhanced the TAP effect 2-fold so that a similar percentage of dimers as in the presence of the TAP-independent peptide was formed. Thus, formation of B27 dimers required co-expression of peptide sources and, in the case of TAP, was enhanced by co-expression of tapasin.
Note that the percentage of dimers did not exceed one-third of B27 molecules in all settings.
It has been proposed that TAP may also act as a chaperone for HCI molecules, for example by retaining them in the ER or stabilizing them until peptides bind. To determine whether this was the case in the insect cell system, we tested the effect on dimer formation of TAP proteins with mutated Walker A sequences (not shown). These mutant TAP proteins assemble normally in complexes but cannot transport peptide. 2 Co-expression of mutant TAP dimers had no effect on HLA-B27 dimer formation or cell surface expression (see below). This demonstrates that active peptide transport by TAP is required for its effect on HLA-B27 dimer formation and argues against an important role of TAP complexes as chaperones for HLA-B27 dimers.
Effect of Chaperones and TAP on Cell Surface Expression of HLA-A2 and B27-In vertebrate cells, newly formed MHC dimers are retained in the ER until they acquire peptide ligands with sufficient affinity (34). Therefore, the amount of MHC dimers acquiring mature N-linked glycans in the Golgi compartments and then expressed on the cell surface reflects the functional performance of the machinery for generation, delivery, and assembly of peptides in the cytosol and ER. However, murine MHC class I molecules can reach the surface of vertebrate cells when these are incubated at 26°C (35,36), whereas most human MHC class I molecules cannot (37,38). Empty murine and human MHC class I molecules have also been reported to be expressed on the cell surface of Drosophila cells lacking antigen processing cofactors, presumably because insect cells are cultured at 27°C (39). We therefore analyzed the level of HLA-A2 and B27 dimers on the surface of Sf9 cells in the absence or presence of processing cofactors. HCI surface expression was analyzed under the conditions also applied in analysis of intracellular dimer formation, i.e. 36 h after triple infections by 2-fold higher infectious doses of cofactor viruses relative to HCI viruses.
Insect cells expressing HLA-A2/␤ 2 m plus two irrelevant proteins expressed significant amounts of W6/32-reactive dimers on the surface, whereas only small amounts of surface B27 dimers were expressed under these conditions (mean fluores- had negligible effects on expression of A2 dimers but increased B27 surface expression dramatically (A and C). Additional expression of tapasin increased most substantially surface expression of HLA-A2 in the absence of a peptide source and of HLA-B27 in cells co-expressing TAP (B and D). Equivalent results were obtained with additional conformation-specific mAb recognizing A2 (BB7.2) or B27 (B27M1) dimers; the latter mAb has been reported to recognize a subset of peptide-filled B27 molecules (40) whose assembly with B27 may be highly tapasin-dependent (19).
A quantitative view of the effects of processing cofactors on HCI surface expression is provided in Fig. 4. HLA-A2 surface expression was modestly increased by calnexin and the TAPindependent peptide, more by calreticulin, and most significantly by tapasin. TAP co-expression alone or in combination with other factors had no effect. Expression of the TAP-independent peptide was synergistic with all three chaperones, i.e. tapasin, calreticulin, and calnexin. Thus, different from the percentage of intracellular dimers, cell surface expression of HLA-A2/␤ 2 m was affected by processing cofactors, especially availability of chaperones. Moderate effects of calnexin and calreticulin may be due to increased total cellular amounts of HLA-A2 in their presence ( Fig. 2A). In contrast, since tapasin did not affect the total amount of cellular HLA-A2, its stronger effect was likely to be related to more efficient peptide assembly and thereby stabilization and export of A2 dimers in its presence. Importantly, in these experiments, tapasin exerted its effect on a pool of TAP-independent peptides. Fig. 4 also illustrates the striking effect of peptide sources on surface expression of HLA-B27 dimers (7-fold increase with TAP and 10-fold with the TAP-independent peptide). Chaperones enhanced these effects but in a distinct fashion according to the peptide source. Formation of cell surface-expressed dimers in the presence of the TAP-independent peptide was most strongly (but still modestly) enhanced by calreticulin, whereas formation of exported dimers with TAP-supplied pep-tides was strongly increased by tapasin (factor 1.7), little by calreticulin, and not at all by calnexin.
MHC class I complexes expressed on the surface of Drosophila (39) and, as recently reported (41), also Aedes insect cells can be devoid of peptide. This can be revealed by incubation of the cells at 37°C for 1 h (41) which leads to disappearance of unstable empty molecules. To determine whether HLA-A2 and B27 molecules expressed on Sf9 cells are peptide-filled, we incubated cells 36 h after infection for 60 min at 37°C followed by staining of cell surface HCI dimers by mAb W6/32 (not shown). These incubations resulted in a dramatic reduction in the number of viable insect cells, presumably due to the cumulated cell damage from viral infection and heat shock. However, cells surviving after 37°C incubations expressed about 70 -80% of surface HCI dimers of cells incubated at 27°C, regardless of the combination of human proteins expressed. This suggested that under all conditions the vast majority of HCI dimers reaching the surface of infected Sf9 cells were peptide-filled.

FIG. 3. Expression of HCI dimers on the surface of insect cells.
Sf9 cells were infected for 36 h with a combination of three viruses, harvested, and expression of HCI⅐␤ 2 m complexes on the surface of live cells was determined by staining with mAb W6/32. Mean fluorescence is plotted on the logarithmic x axis. Virus 1 was A2/␤ 2 m in A and B, and B27/␤ 2 m in C and D. Virus 2 was GAD65 control in A and C, and tapasin in B and D. Virus 3 is indicated between the panels, with Control corresponding to IA-2ic, TAP to TAP1/2, and PEP to the TAPindependent high affinity peptide restricted by the co-expressed HCI molecule.

Effect of Chaperones and TAP on Stability of HLA-A2 and
HLA-B27 Dimers-Similar to empty cell surface dimers, empty detergent-solubilized HCI/␤ 2 m dimers dissociate at 37°C, a property that can be used experimentally to distinguish empty and peptide-filled dimers (42). We used this method to analyze peptide filling of HLA-A2 and B27 dimers that were metabolically labeled for 12 h at the end of a 36-h infection period (Fig.  5). Equal aliquots of Nonidet P-40 lysates were incubated for 1 h at 4 or 37°C before HCI dimers or free heavy chains were recovered in a rapid immunoprecipitation to avoid stabilization of empty dimers by mAb during the precipitation (36). Dimers precipitated from human B cells were completely stable under these conditions (not shown). For each condition, Fig. 5 indicates the percentage of stable dimers (numbers), and the amount of dimers recovered at 4 or 37°C relative to the amounts recovered in cells expressing HCI/␤ 2 m dimers together with two control proteins (bar graph).
Less than 20% of cellular HLA-A2 dimers were stable at 37°C in cells devoid of processing cofactors (Fig. 5A). This proportion was increased slightly by calreticulin and the TAPindependent peptide and doubled by tapasin. TAP had no effect on A2 dimer stability (not shown). Simultaneous expression of calreticulin and the TAP-independent peptide was synergistic and also doubled the proportion of stable dimers. Thus, stability at 37°C of A2 dimers paralleled closely their expression at the cell surface; chaperones, especially tapasin, enhanced A2 assembly with peptides derived from an endogenous TAP-independent pool, and assembly of the defined TAP-independent peptide with high A2 affinity appeared to be facilitated mainly by calreticulin.
Similar to cell surface expression, stability of B27 dimers was strikingly enhanced by co-expression of peptide sources (Fig. 5B). The defined TAP-independent peptide alone increased the percentage of stable molecules almost 10-fold, resulting in a more than 50-fold increase in the amount of stable dimers relative to control cells. Expression of additional cofactors did not enhance stable dimer formation with the TAPindependent peptide. Expression of TAP alone also increased the proportion of stable dimers more than 5-fold and thereby their total amount 10-fold relative to control cells. Tapasin, which alone or in combination with calreticulin had little effect on stability, increased the percentage and total amount of dimers formed in the presence of TAP substantially. Thus, also in the case of HLA-B27, results for dimer stability were closely related to those for cell surface expression.
We also studied stability of metabolically labeled free heavy chains recognized by mAb HC10 (Fig. 5, A and B, right lanes). In the absence of processing cofactors, the majority of heavy chains was lost during 37°C incubations, probably due to aggregation. Co-expression of housekeeping chaperones calnexin (for A2 and B27) and calreticulin (B27 only), but also of the TAP-independent peptide alone (B27), increased the amount of heavy chains stable at 37°C (Fig. 5, and not shown). Small but significant amounts of HC10-reactive A2 and B27 heavy chains could also be detected on the surface of live insect cells (not shown). Surface expression of free heavy chains was also increased in the presence of calnexin, calreticulin (A2 and B27), and the TAP-independent peptide (B27).
Interactions of TAP and Chaperones with HCI Molecules-Taking advantage of high recombinant protein levels and the possibility of freely combining proteins for expression in the insect cell system, we also analyzed interactions between HCI molecules and processing cofactors (with the exception of tapasin). Before performing experiments on lysates from co-infected cells, we asked whether the various proteins could associate after cell lysis. We lysed insect cells infected with single viruses and expressing high levels of HCI heavy and light chains, TAP1-TAP2 complexes, calnexin, or calreticulin in digitonin, mixed lysates containing HCI molecules with a lysate containing another protein of interest, and incubated 16 h before recovering free HCI heavy chains with mAb HC10. As shown in Fig. 6A, prolonged incubation of large amounts of digitoninsolubilized calreticulin or TAP with A2 or B27 heavy chains did not result in detectable formation of complexes between any two proteins; only when calnexin and B27 heavy chains were mixed, a small quantity of the chaperone associated with the free heavy chain.
After having demonstrated that at least TAP and calreticulin FIG. 5. Stability of insect cell-expressed HCI dimers at 37°C. Sf9 cells were infected with a combination of three viruses, labeled metabolically after 24 h, and lysed in a Nonidet P-40 buffer. Lysates were divided in equal halves which were incubated for 1 h at 4 or 37°C, respectively. HCI⅐␤ 2 m complexes or free HCI heavy chains were then precipitated with mAb W6/32 or HC10 as indicated above the panels, separated by SDS-PAGE, and quantified by densitometry. Virus 1 was A2/␤ 2 m in A, and B27/␤ 2 m in B. Viruses 2 and 3 are indicated above the panels, with abbreviations as in Fig. 2. Scanned autoradiographs were re-assembled pair-wise (4°C incubation left lane and 37°C incubation right lane) and derive from a single experiment and exposure time. Histograms indicate the amount of dimers recovered after 4 or 37°C incubation relative to that recovered in the absence of processing cofactors (infection by HCI/␤ 2 m, GAD, and IA-2ic) which was set at 100. Percent stable dimers was calculated as amount dimers at 37°C divided by amount dimers at 4°C multiplied by 100.
interaction with HCI proteins required protein co-expression in the same cell, we studied TAP interaction with HCI molecules (Fig. 6B). We found a highly significant association of HC10-reactive free A2 and B27 heavy chains with TAP complexes (left lane); in cells expressing HCI heavy chains only together with individual TAP subunits, large amounts of TAP1 can be coprecipitated with HCI heavy chains, whereas a much smaller amount of TAP2 associates with heavy chains. This association can be detected in cells expressing HCI heavy chains only or heavy chain/␤ 2 m together and is also observed when calnexin and/or calreticulin are co-expressed (not shown); the most substantial association is observed in cells expressing the TAP1 subunit and an HCI heavy chain only.
In several experiments, we also searched for a potential association between TAP and ␤ 2 m or HCI dimers. However, we have not found any association of W6/32-reactive A2 or B27 dimers with TAP complexes or individual subunits (not shown). We tried to detect association of ␤ 2 m with TAP complexes or subunits, using antibodies to TAP, to ␤ 2 m, and to free heavy chains (43) and cells infected with various virus combinations; in no case could we detect interaction of ␤ 2 m or ␤ 2 m-assembled HCI heavy chains with TAP (not shown). Reasoning that HCI dimer association may be difficult to detect due to rapid binding of TAP-delivered peptides to HCI molecules followed by HCI dissociation from TAP, we tried to co-precipitate ␤ 2 m or HCI dimers with mutant TAP complexes unable to transport peptides (see above); again, ␤ 2 m could not be co-precipitated with TAP (not shown). We conclude that in the absence of tapasin, insect cell-expressed free heavy chains associate efficiently with TAP complexes and especially with the TAP1 subunit, but HCI dimers cannot interact with TAP. There was no difference between HLA-A2 and B27 with regard to TAP association.
Next we studied calnexin association with HCI molecules. Calnexin could be co-precipitated with A2 heavy chains in cells expressing A2 heavy chains alone or in combination with ␤ 2 m; co-expression of calreticulin or TAP did not affect significantly this association (Fig. 6C). In cells expressing ␤ 2 m with HLA-A2, a small amount of the chaperone was associated with W6/32-reactive A2 dimers (Fig. 6C, right panel). As already suggested by the experiment shown in Fig. 2, expression of calnexin resulted in increased recovery of free A2 heavy chains regardless of the presence of ␤ 2 m. In analogous experiments on cells expressing HLA-B27 and calnexin, identical results were obtained (not shown).
Finally, interactions of calreticulin with insect cell-expressed HCI molecules were analyzed (Fig. 6D). Different from calnexin, calreticulin could be precipitated with HLA-A2 and HLA-B27 dimers recognized by mAb W6/32 as well as allelespecific mAbs BB7.2 and B27M1. However, larger amounts of calreticulin were recovered with free heavy chains immunoprecipitated by mAb HC10. Free heavy chains associated with calreticulin were not derived from unstable empty HCI dimers dissociating during immunoprecipitation since calreticulin could also be co-precipitated very efficiently with A2 and B27 heavy chains expressed in the absence of ␤ 2 m (not shown).

DISCUSSION
This is the first study of the functional interactions between human MHC class I molecules and TAP, tapasin, and other ER chaperones in non-vertebrate cells. Novel results obtained in this study bear on three issues as follows: (i) the effect of tapasin on the assembly of TAP-dependent and -independent peptides with HCI molecules; (ii) distinct utilization of peptide sources by the two HCI molecules; and (iii) interactions between TAP and HCI molecules.
Insect cells have previously been used to express MHC class I molecules. Baculovirus-encoded H-2K d and HLA-B27 have been reported to be expressed at relatively high levels and to assemble poorly (Ͻ5 or Ͻ10%, respectively) in dimers (44,45). Jackson and co-workers (39) expressed three murine and four FIG. 6. Interactions of insect cell-expressed proteins. A, 5 ϫ 10 6 Sf9 cells expressing calnexin, calreticulin, or TAP complexes (as indicated) were lysed in 0.5 ml of a buffer containing 1% digitonin and mixed with an equal volume of a digitonin lysate of cells expressing HCI heavy and light chains. After overnight incubation of the mixture, HCI heavy chains were immunoprecipitated with mAb HC10, and immunoprecipitates were blotted onto PVDF membranes together with 10-l aliquots of calnexin, calreticulin, or TAP-containing lysates used for mixing. Blots were stained with antibodies specific for calnexin, calreticulin, or TAP1, respectively. B, Sf9 cells were co-infected with viruses coding for TAP1 (T1), TAP2 (T2), or a double promoter TAP virus (T1/2), together with viruses coding for HCI heavy chains alone or dimers, as indicated. Cells were lysed in digitonin buffer, HCI heavy chains were immunoprecipitated with mAb HC10, immunoprecipitates were separated, blotted, and stained with mAb 148.3 specific for TAP1 (top panel) or mAb 429.3 specific for TAP2 (bottom panel). C, cells were infected with viruses coding for A2 heavy chains alone (left panel) or A2/␤ 2 m dimers (right panel) together with viruses coding for calnexin, calreticulin, and TAP complexes as indicated. Free A2 heavy chains or A2/␤ 2 m dimers, respectively, were then precipitated from digitonin lysates with mAb HC10 and W6/32. Blots were stained simultaneously with HCI heavy chain-specific serum R425 and a rabbit serum against calnexin; the lower band corresponds to HCI heavy chains and the upper band to calnexin. D, cells expressing HLA-A2 or HLA-B27 heavy and light chains together with calreticulin were lysed in digitonin buffer, free heavy chains and dimers were precipitated with indicated mAb, immunoprecipitates were blotted, and HCI-associated calreticulin was detected using a specific rabbit serum. human MHC class I molecules in stably transfected Drosophila cells, whereas a very recent study described expression of murine MHC class I molecules by Aedes insect cells infected by recombinant vaccinia viruses (41). These studies provided convincing evidence that Drosophila as well as Aedes cells synthesize and assemble MHC class I dimers (including HLA-B27) at a similar rate as vertebrate cells and express them on the surface as empty dimers that disappear almost completely upon incubation at 37°C (39,41). In striking contrast to these reports, our results suggest that baculovirus-infected Sf9 cells do not express empty HCI molecules on their surface. This conclusion is based on three findings as follows: (i) almost undetectable surface expression of HLA-B27 in the absence of peptide sources; (ii) small and invariant changes in HCI surface expression upon cell incubation at 37°C; (iii) close quantitative correlation between stability at 37°C of metabolically labeled HCI dimers and HCI dimer expression on the surface of cells incubated at 27°C (Figs. 4 and 5). Thus, baculovirusinfected Sf9 cells surprisingly appear to possess a retention mechanism for empty HCI molecules; we do not know whether this mechanism is related to insect cell or viral proteins. Degradation of empty dimers is unlikely to account for this phenomenon since, at least in the case of HLA-A2, tapasin expression doubled surface expression without affecting the total cellular amount of HCI dimers (Figs. 2 and 4). Digestion of Sf9-expressed HCI dimers by endoglycosidase H suggested that empty HCI dimers carried immature glycan moieties and were therefore retained in the ER (not shown).
Notably, even in the presence of a suitable peptide source and tapasin, no more than 30 -40% of HCI molecules formed dimers. This may be due to incomplete insect cell processing of overexpressed heavy chains that migrate in SDS-PAGE as at least two molecular forms (not shown); only one of these may be able to assemble correctly. Alternatively, other factors, e.g. ERp57 also found in class I loading complexes (7), may be required for complete restoration of class I antigen processing in Sf9 cells.
Our results confirm the crucial role of tapasin for peptide assembly with human as well as murine MHC class I molecules (10, 15, 19, 46 -49). By coincidence, for this study we had selected the two HCI molecules that have been suggested to be less dependent on tapasin than other alleles (18,19,50). The significant effects observed underline that tapasin is likely to act on all HCI molecules. Our study reports several novel findings with respect to tapasin. First, since the mutant human line .220 expresses low amounts of a truncated tapasin molecule (14), this is the first study of the effect of tapasin on human MHC class I molecules in an entirely tapasin-deficient cell. Our observation of enhanced peptide binding to HLA-A2 in the presence of tapasin but absence of TAP demonstrates conclusively that tapasin acts indeed independently of its role as an intermediary between TAP and HCI dimers, as suggested by Lehner and associates (15). Moreover, that same observation is the first evidence reported so far for a role of tapasin in HCI assembly with a TAP-independent peptide pool. Very recently, Schoenhals and associates (46) reported that assembly of a peptide epitope (expressed in the cytosol or in a TAP-independent form) with K b was promoted by tapasin. Since Drosophila transfectants used in that study do not retain empty MHC class I dimers in the ER, this effect may solely have been due to retention of empty K b molecules in the ER by tapasin. As infected Sf9 cells appear to possess an unidentified tapasinunrelated mechanism of preventing surface export of empty HCI dimers, our study demonstrates that ER retention of such dimers by tapasin is at least not the sole mechanism by which it enhances peptide binding to insect cell-expressed MHC class I molecules.
Experimental reagents allowed us to assess the tapasin effect on HCI assembly of peptides from three sources as follows: a heterogeneous TAP-dependent peptide pool of cytosolic origin, a heterogeneous TAP-independent peptide pool with a probable ER-luminal origin, and two defined epitopes with high HCI binding affinity presumably generated in the ER by signal peptidase. Interestingly, tapasin enhanced HCI assembly with peptides from both heterogeneous peptide pools but had a much smaller effect on assembly of the high affinity TAP-independent epitopes. This discrepancy is compatible with the hypothesis of peptide "editing" by tapasin which has been proposed by several authors (15,19,46,51). The latter result is in accordance with a recent study in Aedes cells in which human tapasin failed to promote assembly of a TAP-independent epitope with high affinity for H-2K b (41); however, this result might also be due to species specificity in tapasin interaction with MHC class I molecules, a hypothesis supported by some (19,37,38) but not all (48) evidence.
Our study provides some insight into the surprisingly distinct peptides suitable for assembly with HLA-A2 and B27. Quantity and quality of peptides available in the ER of infected Sf9 cells in the absence of TAP appear to be sufficient for attaining maximum levels of stable A2 dimers in the system, provided tapasin is co-expressed to facilitate their assembly. In contrast, HLA-B27 depends almost entirely on TAP-supplied peptides; in the presence of these peptides, tapasin has the same effect on B27 as on A2 in the absence of TAP. Relatively efficient generation of stable HLA-A2 and B27 dimers in Sf9 cells co-expressing tapasin and (in the case of B27) TAP suggests that peptide generation both in the cytosol and the ER of infected insect cells supplies a sufficient number of high affinity ligands for both HCI molecules, or at least their precursors. Thus, not only are insect cell proteasomes capable of generating selected HCI ligands (46,52), but also both their cytosolic and luminal protein degradation machineries provide a sufficient amount of ligands for two human MHC class I molecules.
HLA-A2 and B27 are known to depend to a different degree on TAP for peptide supply (37,38,53). The relative TAP independence of HLA-A2 is at least partly due to its capacity of binding signal peptide-derived peptides (54) which are also likely to represent the major TAP-independent peptide source in Sf9 cells. Nevertheless, in view of the reduction of A2 surface expression in TAP-deficient human cells by at least 50% (53), it was surprising that even co-expression of TAP with tapasin had no effect on the formation of stable HLA-A2 molecules. This phenomenon may be related to the ligand preferences of the HCI molecules. Although the HLA-B27 preferences are very well adapted to those of human TAP, HLA-A2 prefers ligands that are poorly adapted to TAP (25). Therefore, HLA-A2 ligands may frequently need to enter the ER as longer precursors that require processing in the ER (55), possibly by a mechanism absent from insect cells. In contrast, potential HLA-B27 ligands are much more likely to be translocated from the cytosol into the ER (25). Moreover, B27 is known to be capable of binding longer peptides and may therefore depend less on a specific trimming activity in the ER (56). It is also possible that A2 depends more than B27 on additional processing cofactors, some of which remain to be identified (7,57,58). In any case, the two HCI molecules studied here may bind insect cell peptides derived from TAP-dependent (B27) or -independent (A2) sources, respectively, with exceptional efficiency since peptide filled H-2 K b molecules have been reported to be generated at a much lower level in similarly reconstituted Drosophila cells (46).
The relatively small effects of calnexin co-expression in our system are compatible with a role in preventing aggregation and degradation of free heavy chains, thereby increasing the amount of HCI molecules available for assembly with ␤ 2 m and peptide (59). Thus, we do not find evidence for a role of calnexin in folding and assembly of HCI dimers (60). Moreover, despite high expression levels and very efficient co-precipitation of calnexin with free heavy chains in our system, only very small quantities associated with dimers, suggesting that its described association with HCI dimers possesses very low efficiency (61).
Our findings on calreticulin associations are in conflict with some but not all previous reports. Calreticulin has been reported to associate not at all (4) or very poorly (62) with free heavy chains in cells lacking ␤ 2 m. In contrast, we find that it associates with equally high efficiency with free heavy chains and (presumably empty (4)) dimers. Several studies suggest that preferential calreticulin binding to free dimers in human cells may be related to HCI conformation rather than glycan modification (3,63), so that its binding to insect cell-expressed free heavy chains may reflect an altered conformation. However, two very recent studies in which the chaperone co-precipitated with HCI molecules recognized by mAb HC10 (64,65) suggest that calreticulin can associate with free heavy chains in human cells. In support of this conclusion, we have been able to co-precipitate calreticulin with HC10-recognized heavy chains expressed by human cells cultured at 27 and 37°C (not shown). In any case, the moderate but reproducible effect of calreticulin on cell surface expression and assembly of the TAP-independent peptides with HCI complexes is compatible with preferential stabilization of empty dimers by this chaperone, as previously suggested (4).
We have also found strong evidence for an association of free HCI heavy chains with the TAP complex whose detection was likely to be facilitated by protein overexpression. Although two initial studies did not detect an association of HCI heavy chains with TAP in ␤ 2 m-deficient cell lines (66,67), a more recent study reported co-precipitation of TAP complexes with HCI heavy chains in two ␤ 2 m-deficient cell lines (62). Very recently, Cresswell and associates (10) also reported weak but significant association of free heavy chains with TAP in ␤ 2 m-deficient cells. Our results provide evidence for a direct interaction between HCI heavy chains and the TAP complex that is formed in the absence of ␤ 2 m and tapasin. In accordance with results obtained in mutant human cell lines (68,69), this interaction involves mainly TAP1 and weakly TAP2.