A Single Bottleneck in HLA-C Assembly*

Poor assembly of class I major histocompatibility HLA-C heavy chains results in their intracellular accumulation in two forms: free of and associated with their light chain subunit (β2-microglobulin). Both intermediates are retained in the endoplasmic reticulum by promiscuous and HLA-dedicated chaperones and are poorly associated with peptide antigens. In this study, the eight serologically defined HLA-C alleles and the interlocus recombinant HLA-B46 allele (sharing the HLA-C-specific motif KYRV at residues 66–76 of the α1-domain α-helix) were compared with a large series of HLA-B and HLA-A alleles. Pulse-labeling experiments with HLA-C transfectants and HLA homozygous cell lines demonstrated that KYRV alleles accumulate as free heavy chains because of both poor assembly and post-assembly instability. Reactivity with antibodies to mapped linear epitopes, co-immunoprecipitation experiments, and molecular dynamics simulation studies additionally showed that the KYRV motif confers association to the HLA-dedicated chaperones TAP and tapasin as well as reduced plasticity and unfolding in the peptide-binding groove. Finally, in vitro assembly experiments in cell extracts of the T2 and 721.220 mutant cell lines demonstrated that HLA-Cw1 retains the ability to form a peptide-receptive interface despite a lack of TAP and functional tapasin, respectively. In the context of the available literature, these results indicate that a single locus-specific biosynthetic bottleneck renders HLA-C peptide-selective (rather than peptide-unreceptive) and a preferential natural killer cell ligand.

Class I human leukocyte antigens (called HLA) are cell-surface heterotrimers formed by a highly polymorphic heavy (44 kDa) chain, a non-polymorphic light (12 kDa) chain subunit (␤ 2 -microglobulin (␤ 2 m) 3 ), and a short (8 -11-mer) peptide antigen derived from the degradation of intracellular proteins (1). The assembly pathway of most class I molecules involves an early interaction of the heavy chain, still free of ␤ 2 m, with cal-nexin, followed by association with ␤ 2 m and binding to the so-called peptide-loading complex. This is a supramolecular endoplasmic reticulum structure comprising, among others, two HLA-dedicated chaperones: TAP transporter associated with antigen processing) and the peptide editor/facilitator tapasin (1). Successful peptide loading results in tight association of the heavy chain with ␤ 2 m and the release of thermally stable, folded class I conformers (1)(2)(3). These are exported to the cell surface, where they activate and inhibit cytotoxic T lymphocytes expressing the rearranging T cell receptor and natural killer (NK) cells expressing non-rearranging receptors such as the killer immunoglobulin-like receptors, respectively (4).
There are Ͼ1000 class I molecules, encoded by three highly polymorphic allelic series: HLA-A, -B, and -C (www. anthonynolan.org.UK/HIG/index.html). They share a conserved general architecture, a common peptide-loading pathway, and a similar set of functions, but also display a number of alleleand locus-specific sequences responsible for distinctive structural properties and specialized functional features (reviewed in Ref. 1). In the case of HLA-C, there are three known locusspecific motifs, one of which (particularly extensive) involves heavy chain residues 66, 67, 69, and 76 (KYRV) in the ␣1-domain ␣-helix (5). Next to the KYRV 66 -76 motif, a functional dimorphism involving residues 77 and 80 regulates NK cell recognition by the KIR2DL1/KIR2DL2 inhibitory receptors (4). Because crystallographic data show that T cell receptor and killer immunoglobulin-like receptor footprints on HLA-C are largely overlapping and that both encompass the HLA-C-specific motif (6), it appears that a stretch of polymorphic residues on the ␣1-domain ␣-helix encodes a unique recognition structure promiscuously recognized by HLA-C-specific cytotoxic T lymphocytes and NK cells. Relevant to this tight packaging of distinct structural determinants within a narrow region, it appears that HLA-C, although an extremely good ligand for NK cells, is a much poorer antigen-presenting class I molecule than HLA-A and HLA-B (reviewed in Refs. 1, 5, and 7). Also in contrast to HLA-A and HLA-B, HLA-C is a rather poor assembler. This latter feature has been proposed to correlate with impaired intracellular transport and low surface expression (8,9), although this conclusion has been questioned (10).
Poor assembly of HLA-C is strongly supported by the intracellular accumulation of two broad classes of folding intermediates: ␤ 2 m-free and ␤ 2 m-associated class I heavy chains. HLA-C heavy chains free of ␤ 2 m (free heavy chains) were the first folding intermediates to be identified and extensively char-acterized (8) based on their reactivity with antibodies that bind linear epitopes accessible on unfolded but hidden in ␤ 2 m-associated heavy chains. More recently, taking advantage of two such antibodies (HC10 and L31) to mapped, nearly contiguous linear epitopes spanning the entire N-terminal region of the ␣1-domain ␣-helix (L31 binds an epitope centered on an aromatic amino acid at position 67), we have provided evidence that free HLA-C heavy chains bear an extensive local un/misfolding involving the HLA-C-specific 66 -76 motif and crucial peptide-anchoring positions in the so-called B-pocket of the antigen-binding groove (11). We have also shown that free HLA-C heavy chains are thermally unstable, peptide-free, and essentially unreceptive to peptides in in vitro assembly assays, i.e. their binding groove is in an "open" (as defined in Refs. 2 and 3) conformation. Consistent with these biochemical features, ␤ 2 m-free HLA-C heavy chains associate primarily with calnexin and, in much smaller amounts, with members of the peptide-loading complex, including TAP and tapasin (11). Thus, it appears that ␤ 2 m-free HLA-C heavy chains accumulate not only at the calnexin/␤ 2 m checkpoint in ␤ 2 m-defective mutants (as would be expected if they just failed to bind ␤ 2 m), but also at subsequent steps, on several distinct chaperones acting in sequence, and even in the absence of obvious defects in the antigen-processing/peptide-loading machinery (11,12).
The second class of incompletely folded class I heavy chain intermediates are ␤ 2 m-associated. Like free heavy chains, they are thermally unstable and peptide-free (i.e. they also meet the definition of open conformers), but unlike free heavy chains, they are associated primarily with TAP and are peptide-receptive. It has been proposed that these HLA-C/␤ 2 m dimers are intermediates awaiting peptide loading and that they exhibit an exceedingly long "standby" TAP association because the mechanism of in vivo peptide loading of HLA-C is particularly selective (9).
One may therefore ask whether the "backward" accumulation of multiple folding intermediates scattered on distinct chaperones might result from a single downstream bottleneck (i.e. selective peptide loading). Furthermore, one may ask whether a common structural feature, e.g. the KYRV motif (5), underlies the holdback of un/misfolded intermediates from the productive assembly pathway. If this is the case, all the heavy chain alleles sharing a given motif would be expected to accumulate as unstable/open conformers. In addition, the same HLA-C-specific motif(s) might confer local and generalized unfolding and promote strong association with TAP as well as tapasin in the peptide-loading complex. To address these issues, we have assessed the accumulation of open heavy chain intermediates of the eight serologically defined HLA-C alleles, of selected HLA-A and HLA-B alleles, and of the interlocus recombinant HLA-B46 (B*4601) allele, in which the HLA-Cspecific KYRV motif from HLA-Cw1 (Cw*0102) has been neatly transferred into an HLA-B (B*1501) heavy chain background (13) as the result of a gene conversion event.
The accumulation of open conformers has been compared, taking advantage of an additional feature of antibody L31 (14): its ability to bind all the HLA-C alleles and a few HLA-B alleles (HLA-B7, -B8, -B35 -B51, -B54, and -B56) displaying different abilities to assemble (8,15). Biochemical experiments, molecu-lar dynamics simulation studies of the packing forces of conformed class I heavy chains differing in the KYRV motif, and expression of one HLA-C allele (HLA-Cw1) in TAP-and tapasin-defective cells indicate that old and novel (described herein) features of HLA-C have a common denominator in the architecture of the binding groove.

EXPERIMENTAL PROCEDURES
Cell Lines-The 721.221 and 721.220 cell lines (referred to as 221 and 220 hereafter, respectively) were obtained in the same ␥-irradiation/mutagenesis experiment from parental 721 lymphoblastoid cells (16). The former are completely defective in HLA-A, -B, and -C, but retain HLA-E expression (17,18); the latter are defective in HLA-A and HLA-B, but retain ϳ50% of HLA-Cw*0102 (Cw1) expression and, in addition, lack functional tapasin (16). The TAP-defective 174ϫCEM.T2 cells (referred to as T2) express HLA-A2, -B51, and -Cw1 (19) heavy chains in the absence of TAP as a result of somatic hybridization between the TAP-defective 174 cell line (also a derivative of 221) and the T lymphoid cell line CEM. The HLA-A, -B, and -C transfectants in 221 were obtained through the courtesy of several investigators (see "Acknowledgments") and are referenced elsewhere (14). Epstein-Barr virus-transformed, HLA-typed, and homozygous B cells are also described elsewhere (12,14).
Immunochemical Methods-Cells were metabolically labeled with [ 35 S]methionine (9.25 MBq/ml) as described in the figure legends and solubilized with either 1.0% Nonidet P-40 or 0.5% CHAPS in phosphate-buffered saline (0.01 M phosphate (pH 7.0) and 0.15 M NaCl). For immunoprecipitation, purified antibodies were covalently linked to Affi-Gel (Bio-Rad). The isoelectric focusing (IEF) and Western blot techniques (reducing conditions in all cases) used have been described (11,12,14). In vitro assembly experiments were performed exactly as described in the presence of the NCPERIITL Cw1 ligand (11).
Molecular Dynamics Simulations-Models of the isolated heavy chains of HLA-B15 (B*1501), HLA-B27 (B*2705), HLA-B46 (B*4601), and HLA-Cw1 (Cw*0102) were created, and molecular dynamics simulations were performed and analyzed using InsightII, Biopolymer, Discover3, and Analysis (Accelrys, San Diego, CA). No explicit water molecules were included, and electrostatic forces were modeled using a distance-dependent dielectric constant (4*r). The combined valence force field was used with cell multipole summation, fourth order Taylor series expansions, and 1.0-Å update width. Initially, heterotrimer models were created either directly from or by homology with appropriate published crystallographically resolved structures in the Macromolecular Structure Database: HLA-B15 and HLA-B46 from HLA-B*1501 (Protein Data Bank code 1XR9) (25), HLA-B27 from HLA-B*2705 (code 1JGE) (26), and HLA-Cw1 from HLA-Cw4 (code 1IM9) (27). Unwanted peptide chains, crystallographic water molecules, and heteroatoms were removed, hydrogen atoms were added, and the models were energy-minimized with the Polak-Ribiere conjugate gradient method. Peptide and ␤ 2 m chains were then removed, and the isolated heavy chain models were again energy-minimized as described above. Molecular dynamics simulations were carried out with velocity Verlet integration at 1.0-fs intervals using NVT ensembles at the temperatures specified for 100,000 iterations. Initial heating steps of 1000 iterations were used, after which trajectories were saved every 500 iterations.

An Abundant Pool of Free HLA-C Heavy Chain Conformers in B Lymphoid
Cells-It was shown in a previous study (11) that the ␤ 2 m-free heavy chain conformers reactive with antibody L31 are unable to stably assemble with peptides and ␤ 2 m, whereas the ␤ 2 m-associated conformers reactive with F4/326 and W6/32 do assemble with peptides in vivo as well as in vitro.
The combined pulse-chase/thermal stability experiment shown in Fig. 1A was carried out in stable Cw1 transfectants of the 221 cell line (221.Cw1) to estimate the turnover of the two conformers and the kinetics of acquisition of thermal stability by ␤ 2 m-associated conformers. During the first part of the chase (until the 105-min point), L31 conformers slowly and progressively declined, whereas F4/326 conformers gradually increased (lanes 1-5 and 7-11, respectively), as expected (11,14), although thermal stability (indicative of stable peptide binding) was acquired not earlier than 45 min from synthesis (compare lanes 7-9 and 13-15 and lanes 10 and 16). From this point on, accumulation of F4/326 conformers and acquisition of thermal stability simultaneously increased, peaked at 105 min (lanes 11 and 17), and then simultaneously declined (lanes 12 and 18).
Based on these kinetics and turnover profiles, a 2-h continuous metabolic labeling can be predicted to provide a cumulative estimate of the stoichiometry of ␤ 2 m-free and ␤ 2 m-associated conformers throughout the crucial period of productive class I assembly, before the beginning of disposal. Conformer ratios (a measure of stable folding/assembly) were therefore evaluated under these metabolic labeling conditions in a panel of 221 cells transfected with single class I alleles, including seven serologically defined HLA-C alleles, two HLA-B alleles (HLA-B7 and HLA-B51) representative of good and poor assembly efficiencies, respectively (15), two alleles (HLA-A2 and HLA-B15) lacking the crucial residues forming the optimal L31 epitope (14), and the interlocus recombinant HLA-B46 allele. Representative results of immunoprecipitation experiments with L31 and W6/32 (F4/326 gave similar results; data not shown) are depicted in Fig. 1B.
Impaired and Unstable Assembly Is a Locus-specific Feature of HLA-C-The data in Fig. 1 provide initial evidence that the accumulation of L31 conformers is a feature of HLA-C shared with HLA-B46. It remains to be established whether this accumulation results from (a) poor assembly, (b) increased dissociation, or (c) both. In addition, trivial causes of inaccuracy must be excluded, such as (d) the inclusion of non-classical class I HLA molecules (expressed by 221 cells) in the W6/32 conformer pool and (e) different integrated copy numbers of the transfected class I alleles in distinct 221 transfectants. Different copy numbers may result in different levels of class I heavy chain expression against a background of similar (and possibly limiting) class I chaperoning.
To distinguish among features a, b, and c and to rule out features d and e, we determined the extent as well as the kinetics of the accumulation of the different heavy chain conformers by pulse-chase experiments and resolved the individual class I alleles coexpressed in different cell lines by IEF. In addition to 221 transfectants, HLA homozygous B lymphoid heavy chains (17 alleles altogether), three of these (JY, CJO, and EDR) simultaneously express one L31-reactive HLA-B allele and one L31reactive HLA-C allele, making them suitable for stringent allele  comparisons. Representative results in two cell lines are shown in Fig. 2, and a synopsis of the densitometric data of HLA-Cand HLA-B-specific bands for all the tested Y/F67 ϩ HLA-C and HLA-B alleles is shown in Fig. 3. Consistent with feature a (poor HLA-C assembly), all the eight tested HLA-C alleles and the poor assembler HLA-B51 (15) reacted more intensely with L31 than with W6/32 at either or both the 0-and 30-min chase points, whereas HLA-B7, -B35, and -B56 displayed the opposite pattern (W6/32 Ͼ L31). HLA-B46 was similarly reactive with the two antibodies. Consistent with feature b (increased dissociation of HLA-C assemblies), the W6/32-reactive conformers of the eight HLA-C alleles, HLA-B46 and HLA-A31, were less stable (half-lives of ϳ30 min) than the W6/32 conformers of HLA-B7, -B35, -B51, -B56, -A2, -A32, -B44-, -A11, -A30, and -B13 (half-lives exceeding 300 min) (Figs. 2 and 3 and data not shown). Although neither feature a nor b by itself is HLA-C-specific, only HLA-C alleles displayed both, and HLA-B46 resembled HLA-C more closely than HLA-A or HLA-B.
Residues 66 -76 in the ␣1-Domain ␣-Helix Reduce the Flexibility of the Binding Groove in Vivo-From the above data, it is conceivable that KYRV residues 66 -76 introduce a major bottleneck preventing stable class I assembly. An unexpected conformational flexibility of the ␣1and ␣2-domains has been inferred by a molecular dynamics simulation study of amino acids 1-181 of HLA-A2 heavy chains (28). We then tested whether the KYRV motif modifies this flexibility. Models of the isolated HLA-B15, -B27, -B46, and -Cw1 heavy chains (amino acids 1-276) were created and energy-minimized. At this stage, there were no significant differences apparent in the binding groove between the models of the different alleles, which were all essentially similar to the models with peptide and ␤ 2 m present. The stability of HLA-B15, -B46, and -Cw1 heavy chains was then comparatively tested in these models by molecular dynamics simulations. As shown in Fig. 4, at 300 K, the HLA-B15 heavy chain progressively deviated from the starting position, folding downwards with significant helix unfolding in the C-terminal region of the ␣1-helix. In parallel, the N-terminal region of the ␣2-domain helix also folded down while maintaining helical structure, accentuating the kink between the ␣2-1and ␣2-2-helix segments. This did not occur over the time scale of the simulations with either of the models containing the HLA-C motif (Cw1 and B46) where the helical structure of the ␣1-helix was maintained. In a model of HLA-B27, which differs from HLA-B15 at multiple positions (including region 66 -76) but does not include any amino acid of the KYRV motif, the behavior was essentially similar to that of HLA-B15. The results with HLA-B15 and HLA-B27 were thus similar to those obtained in the extended molecular dynamics simulation reported above of HLA-A2 (28), whereas the results with HLA-B46 and HLA-Cw1 indicated a significantly more stable, less flexible binding groove structure. Similar results were obtained with all the alleles tested at a higher temperature of 320 K.     Class I heavy chain components were assigned to specific alleles based on previous IEF comparisons with HLA-C transfectants (12,14). IEF bands were scanned, and densitometric values of all the bands of a given allele in a given immunoprecipitate and at a given time point were summed. This resulted in 104 cumulative estimates of free and ␤ 2 m-associated heavy chains accumulating over time, sorted by allele (13 panels, one for each L31-reactive allele) and antibody (L31 (f) and W6/32 (E)).
The KYRV 66 -76 Motif in the ␣1-Domain ␣-Helix Is Involved in Strong TAP/Tapasin Association-Neisig et al. (9) have shown that HLA-C molecules accumulate, free of peptide, in association with TAP. We therefore tested whether the KYRV motif is involved in determining this feature. To this end, we used antibodies to TAP1 to compare the amounts of co-immunoprecipitated heavy chains in CHAPS extracts of 221.Cw1, 221.B46, and 221.B15 transfectants (Fig. 5). In parallel, co-immunoprecipitation was also carried out with antibodies to tapasin. To detect heavy chains, we used three distinct antibodies, including HC10 and Q1/28, which bind conserved ␣1and ␣3-domain epitopes, respectively. HLA-Cw1 and HLA-B46 (but not HLA-B15) heavy chains were detectably associated with TAP as well as tapasin (compare lanes 2/3 with lanes 7/8 and 12/13 and lanes 32/33 with lanes 37/38 and 42/43; also see lanes 22/23 and 27/28). Thus, the KYRV motif promotes the accumulation of HLA-C heavy chains in association with the two HLA-dedicated chaperones of the peptide-loading complex.
In Vitro Assembly of HLA-Cw1 Heavy Chains in Cells Defective in Tapasin or TAP-Although they are unable to become stably assembled and peptide-filled in cells lacking TAP (T2) or functional tapasin (220), HLA-A and HLA-B alleles do retain peptide receptivity, i.e. the ability to assemble in vitro upon incubation with exogenously added peptides (16, 29 -35). Because of its peptide selectivity (9), one may hypothesize that HLA-C is more crucially dependent than HLA-A and HLA-B on TAP and tapasin. Because the available information is limited (36), we evaluated the peptide receptivity of HLA-Cw1, an allele that displays a particularly deep folding impairment (Fig. 3) and a rigid binding groove (Fig. 4). In vitro assembly experiments were performed on soluble extracts of the partially isogenic (see "Experimental Procedures") T2 and 220 cells, both of which naturally express HLA-Cw1, following metabolic labeling and two different chase periods (15 and 120 min) in the absence and presence of a specific Cw1 ligand (11). Regardless of the cell line and length of chase, all the heavy chains, except a minor fraction of HLA-A2 components (presumably stabilized by endogenous signal peptides (37)), melted at 37°C (Fig. 6, compare lanes 2 and 3, 5 and 6, 9 and 10, and 12 and  13), as expected. However, HLA-Cw1 (arrowheads) was specifically stabilized by its ligand (compare lanes 2 and 4, 5 and  7, 9 and 11, and 12 and 14). We conclude that although a The starting conformations are shown on the left as ribbon diagrams. Conformations from the final stages of the simulations were energy-minimized, and the resulting models are shown on the right. The gray regions of the ribbons correspond to amino acids 66 -76, which contain, in HLA-B46 and HLA-Cw1, the KYRV motif. The potential energy variations observed during the time scale of the molecular dynamics simulations are plotted over the arrows and indicate that at 300 K, but also at 320 K (data not shown), the HLA-B46 and HLA-Cw1 models achieved a stable overall conformation at intermediate potential energy, which did not occur with the HLA-B15 or HLA-B27 model. quantitative assessment of peptide receptivity in vivo is beyond the scope of in vitro assembly experiments, no absolute folding impairment prevents HLA-Cw1 from becoming peptide-receptive.

DISCUSSION
Several explanations have been offered for the low surface expression of HLA-C: (a) transcript instability (10); (b) poor assembly with ␤ 2 m, resulting in the accumulation of ␤ 2 m-free heavy chains (8); and (c) selective peptide binding resulting in prolonged TAP association of the heavy chain, followed by endoplasmic reticulum degradation (9). Our results provide further evidence for explanation c and imply additional redundancy in the down-regulation of HLA-C expression, i.e. (d) increased dissociation of HLA-C/␤ 2 m complexes. We also show that explanation d, like explanation c, depends on an HLA-C-specific KYRV motif (5) of the ␣1-domain ␣-helix.
Additionally, the KYRV motif may impair the flexibility of the class I heavy chain, as measured by molecular dynamics simulation, and confers binding not only to TAP but also to tapasin, although the acquisition of a peptide-receptive state by HLA-Cw1 requires neither TAP nor tapasin. Below, we argue that the KYRV motif introduces a single bottleneck impairing HLA-C assembly.
Impaired Assembly, Increased Disassembly, and the KYRV 66 -76 Motif of HLA-C-Early studies by Neefjes and Ploegh (8) identified an abundant accumulation of free HLA-Cw2, -Cw3, and -Cw4 heavy chains. In contrast, McCutcheon et al. (10) did not find evidence for impaired assembly of HLA-Cw3, -Cw6, and -Cw7 complexes and found that assembled HLA-A2, -B27, and -Cw1 complexes are equally long-lived. Taking advantage of the distribution of the L31 epitope on a set of alleles that includes one member from each of the eight serologically defined HLA-C specificities and five HLA-B alleles, we have been able to carry out a homogeneous comparison of the accumulation of the different unfolded heavy chains and to estimate their assembly abilities. This panel is representative of the different assembly efficiencies of class I alleles because it includes HLA-B51, known as one of the poorest class I assemblers (15). At variance with McCutcheon et al., we have found (Figs. 1-3) that HLA-C heavy chains are poor assemblers and that few HLA-C heavy chain/␤ 2 m complexes (eight alleles tested), but most HLA-A and HLA-B complexes (11 alleles tested, including the HLA-A2 and HLA-B27 alleles tested by McCutcheon et al.), survive a 5-h chase. The reasons for these discrepancies are unclear. Our results are more similar to those of Neefjes and Ploegh (8) and Gillet et al. (38). The latter group observed a greater dissociation rate of HLA-B and HLA-C compared with HLA-A molecules, but because of the lack of suitable reagents such as HLA-C transfectants and antibodies, could not detect differences in dissociation rates between HLA-C and HLA-B.
HLA-B*4601 results from a gene conversion event that replaced the ␣1-domain ␣-helix of HLA-B*1501 with that of HLA-Cw*0102 (this includes both an HLA-C-specific motif and the L31 epitope). Analysis of the published (13) high pressure liquid chromatograms of eluted peptides reveals that HLA-B*4601 lost the "humped" profile of the natural HLA-B*1501 peptide ligands and acquired a "flat" profile dominated by a limited number of distinct prominent peptide peaks more similar to HLA-Cw*0102. These results are consistent with selective peptide binding by HLA-C (9) being due, at least in part, to HLA-C-specific sequences.
In this study, using the same 221.B*4601 transfectants, we have shown that the presence of an HLA-C-specific sequence confers to the interlocus recombinant several characteristics typical of HLA-C heavy chains, including local ␣1-domain unfolding, impaired assembly, strong binding to TAP and tapasin, and post-assembly instability. Thus, the HLA-C-specific KYRV motif (5) of the ␣1-domain ␣-helix is involved in determining a single locus-specific biosynthetic bottleneck responsible for several biochemical features of HLA-C. In contrast, the poor assembly of HLA-B51 has been mapped to the ␣2-domain, to residues for the most part outside the binding groove (15). Substitutions at residues 74 and 116 and at or around resi- . In vitro assembly of HLA-Cw1 in the absence of TAP and tapasin. 220 and T2 cells, lacking functional tapasin and TAP, respectively, were pulsed for 10 min and chased for either 15 or 120 min. At the end of the chase, Nonidet P-40 extracts were prepared; divided in three parts for incubation (4 h) at 4, 37, or 37°C in the presence of a Cw1 peptide ligand (20 g/ml; see "Experimental Procedures"); immunoprecipitated with the indicated antibodies or in the absence (Ϫ) of specific antibodies; and run on an IEF slab under reducing conditions. dues 122, 134, and 136 in the ␣1and ␣2-domains are known to affect the association of heavy chains with members of the peptide-loading complex and transport to the cell surface (39 -44). In the context of the available information, our results indicate that TAP and tapasin (presumably when associated within the peptide-loading complex) either bind an extended region encompassing the top of the ␣1and ␣2-domain ␣-helices or, irrespective of their precise docking sites, sense global changes in the conformation of the entire binding groove. Selective Peptide Loading of HLA-C-At least three groups have proposed that HLA-C selectively binds peptides (5,9,45). An indirect argument supporting this view is based on the observation that peptides containing certain HLA-C motifs are poorly transported by TAP, presumably resulting in the availability of a restricted pool of HLA-C ligands in the endoplasmic reticulum (discussed and referenced in Ref. 9). An alternative, although nonexclusive, mechanism for peptide selectivity was suggested by in vitro assembly experiments in which HLA-C, compared with HLA-A and HLA-B, required 10-fold greater concentrations of an equimolar mixture of completely degenerated random nonamer peptides to become peptide-bound and to be released from TAP (9). This latter experiment implies that intrinsic features of the HLA-C heavy chains restrict its ability to bind peptides, although the exact mechanism constraining ligand/acceptor combinations remains unclear.
Molecular dynamics simulation experiments (in the absence of peptide and ␤ 2 m) suggest that KYRV renders the binding groove more rigid. This is in agreement with the entire set of experimental data shown here. The persistence of free heavy chains bearing linear epitopes in the ␣1-domain ␣-helix and in the ␣3-domain is consistent with unfolding proximal and distal to the KYRV motif (Figs. 1B and 5). Thus, local folding coupled to binding, as is believed to occur in HLA class I molecules (46), may be more demanding in KYRV-containing alleles. On this basis, the simplest interpretation of selective peptide binding is that the KYRV motif, by limiting the plasticity of the HLA-C heavy chain, restricts the range of acceptable ligands and causes the retention by TAP and tapasin of peptide-receptive heavy chains awaiting the "right" peptides. This single biosynthetic bottleneck would then cause a backward jam of irreversibly (11) unfolded free heavy chain intermediates.
It should be noted that this bottleneck does not involve an absolute impairment in the ability of HLA-C to acquire a generic fold compatible with peptide binding. On the contrary, in vitro assembly experiments show that this property is conserved even in the absence of TAP or tapasin in all class I heavy chains (16, 29 -35), including HLA-C (Fig. 6). Thus, our results favor a stringent constraint/selection imposed by the KYRV motif on peptide selection.
It may be of interest to determine whether or not the locusspecific biosynthetic bottleneck, described here, favors the loading of a dedicated set of peptides in the trophoblast, where HLA-C is selectively expressed in the absence of HLA-A and HLA-B, activates decidual NK cells, and favors implantation through blood vessel remodeling (47).