Structure of Human Histocompatibility Leukocyte Antigen (Hla)-Cw4, a Ligand for the Kir2d Natural Killer Cell Inhibitory Receptor

The HLA-Cw4 heavy chain (residues A1–275) was overexpressed in the Escherichia coli strain BL21 (DE3) pLysS and purified as insoluble inclusion body protein 21. β₂m (microglobulin) inclusion body was produced in the E. coli strain XA90 as described 22. The HLA-Cw4 heavy chain and β₂m were reconstituted in the presence of the peptide, QYDDAVYKL, and the refolded complex was purified by gel filtration chromatography 21. Refolded HLA-Cw4 crystallized in 18% polyethylene glycol (PEG) 8000, 0.2 M Ca acetate, and 0.1 M Na cacodylate, pH 6.5. The crystals obtained initially were thin plates and were improved by streak seeding using 12% PEG 8000, 0.2 M Ca acetate, and 0.1 M Na cacodylate, pH 6.5, as the reservoir solution. The space group is

⁠.

The crystals were stabilized in a harvesting solution (22.5% PEG 8000, 0.2 M Ca acetate, and 0.1 M Na cacodylate, pH 6.5) for 2 h and then soaked in a cryoprotectant-containing solution (22.5% PEG 8000, 0.2 M Ca acetate, 0.1 M Na cacodylate, pH 6.5, and 25% glycerol) for 5 min before being flash-cooled with liquid nitrogen. X-ray diffraction data were collected to 2.9 Å with the ADSC 1K CCD detector at A-1 beamline of the Cornell High Energy Synchrotron Source (CHESS; Ithaca, NY). The diffraction was anisotropic, and the mosaicity of the crystal varied from 0.5 to 1.5° depending on the orientation of the crystal. Data were integrated and scaled (Table) using DENZO and SCALEPACK (HKL Research).

The HLA-Cw4 structure was determined by molecular replacement using the program AMoRe 23. Except for the peptide, the HLA-B27 molecule (Protein Data Bank [Brookhaven National Laboratory, Upton, NY] accession code 1HSA; reference 24) was used as the search model. Rotation and translation function searches using all data from 8 to 4 Å yielded a clear solution with a correlation coefficient of 44.9% and R factor of 44.4% (the next highest peak has a correlation coefficient of 26.7% and R factor of 51.0%).

The molecular replacement solution was used as the starting model for refinement, with residues that differ between HLA-Cw4 and HLA-B27 (34 total residues in the heavy chain) replaced by alanines and the peptide excluded. All data from 22 to 2.9 Å, with |Fo| > 0, were included for refinement; 10% of the reflections were omitted for the calculation of R_free. The model was subjected to an initial rigid body refinement in X-PLOR 25, where α1α2, α3, and β₂m were treated as individual domains. After two rounds of manual rebuilding in O 26 and refinement in X-PLOR, which involved simulated annealing with bulk solvent correction, clear electron density appeared in the 3Fo-2Fc map for the peptide region, including all of the side chains of the peptide. Subsequent refinement used the maximum likelihood method implemented in crystallography and nuclear magnetic resonance system (CNS) 27. The minimization procedure included positional refinement and simulated annealing with bulk solvent correction and initial overall anisotropic B factor correction, followed by group B factor refinement. The final model contains residues A2–274 of the heavy chain, B0–98 of β₂m (B0 corresponds to the initial methionine that was engineered for expression in E. coli), P1–9 of the peptide, and 35 water molecules (Table). The 35 water molecules were selected and refined based on peaks that were at least 2.0 σ in height in Fo-Fc and 3Fo-2Fc electron density maps. All φ and ψ angles lie in the allowed regions of the Ramachandran plot, with 86% in the most favorable regions. The NH₂ and COOH termini of the heavy chain (A1 and A275), as well as the COOH terminus of β₂m (B99), have no visible electron density. Side chains for residues A104–108 and A195–198 in the loop region of the heavy chain and residues B17–19 of β₂m have weak electron densities and B factors >80 Ų.

The ectodomain of the HLA-Cw4 heavy chain was overexpressed as inclusion bodies in E. coli and reconstituted in the presence of β₂m and a nonameric peptide (QYDDAVYKL) that contains the consensus peptide binding motif for HLA-Cw4 21,28,29. The consensus peptide motif was determined by pool sequencing and included an aromatic residue or proline for P2, a hydrophobic COOH-terminal anchor, a hydrophobic auxiliary anchor at P6, and the frequent use of glutamic acid and aspartic acid at P4 29. Refolded HLA-Cw4 binds directly to KIR2DL1, as shown by a native gel shift assay 21. The crystal structure of HLA-Cw4 was determined by molecular replacement, using HLA-B27 24 as the search model, and refined at 2.9 Å

⁠. The final model contains heavy chain residues A2–274, β₂m residues B0–98, peptide residues P1–9, and 35 water molecules. The loop regions at residues A104–108 and A195–198 in the heavy chain and residues B17–19 in β₂m have weak electron density and high B factors (>80 Ų), indicating that these regions are disordered. Clear electron density is observed for the entire peptide (Fig. 1).

The overall structure of HLA-Cw4 is similar to that of HLA-A and -B molecules 24,30,31,32,33,34, as expected from the high sequence homology (∼85% sequence identity) among HLA-A, -B, and -C heavy chains. The root mean square (rms) deviation between HLA-Cw4 and HLA-A2 is 0.8 Å for 368 C_α atoms, and the rms between HLA-Cw4 and B53 is 0.8 Å for 380 C_α atoms. (HLA-A2 [2.5 Å] and -B53 [2.3 Å] are chosen for comparison because the structures of these molecules bound to single nonameric peptides are available at high resolution.) The structure differences among HLA-A, -B, and -C molecules are due to the relative orientations of the individual α1α2, α3, and β₂m domains. The α1α2 domains of HLA-Cw4 and HLA-A2 can be superimposed with an rms deviation of only 0.6 Å for 172 C_α atoms. As shown in Fig. 2 a, when the α1α2 domains of HLA-Cw4 and HLA-A2 are superimposed, the α3 domain adopts a different position relative to α1α2 in each structure.

In the α1α2 domain, the structure of HLA-Cw4 differs from many human and mouse class I MHC structures near the COOH-terminal portion of the α1 helix (residues A67–77), widening the peptide binding groove in this region by up to 2.4 Å in comparison with HLA-A2 (reference 32; Fig. 2 b). HLA-Cw4 also differs dramatically (up to 4.2-Å shift in C_α atom positions) from HLA-A2 in the loop region connecting S1 and S2 of the α1 domain β sheet (residues A14–20) (Fig. 2 b). Relative to HLA-A2, the S1-S2 loop protrudes up and toward the α1 helix (Fig. 2 a). As a result, Arg17 in HLA-Cw4 partially replaces Arg14 found in HLA-A2 in forming a hydrogen bond with Glu19 and a salt bridge to Asp39 (not shown). Furthermore, two salt bridges between Glu19 in the loop and Arg75 on the α1 helix are lost. As the S1-S2 loop is involved in crystal packing and may be an artifact, the biological significance of the loop difference remains to be explored. Other regions in the α1α2 domain that differ between HLA-Cw4 and HLA-A2 include the loop connecting S3 and S4 of the α1 domain β sheet (residues A38–45, up to 2.8-Å difference in C_α positions) (Fig. 2 b). The COOH-terminal end of the α1 helix is implicated in HLA-C recognition by KIR (for review see references 9 and 10). The α2 helix has been observed to affect binding of Ly-49 to mouse class I MHC molecules 35,36,37. It is unknown whether the α2 helix may also participate in the KIR–HLA-C interaction. The COOH-terminal end of the α1 helix, together with its adjacent loop region, may define a KIR binding site on HLA-C that is different from the Ly-49 binding site on mouse class I MHC molecules.

The α3 and β₂m domains of HLA-Cw4 are very similar to those of HLA-A2. One of the regions that varies among the different HLA molecules in the α3 domain is the loop consisting of the acidic residues A223–229. Similar to other class I MHC molecules, residues A225–227 in HLA-Cw4 form a turn of 3₁₀ helix. The loop is involved in the binding of CD8 to class I MHC 38.

Structures of HLA-A and -B have demonstrated that the ends of the peptides (P1–P2, P8–P9) are similarly bound in the cleft through conserved hydrogen bonds, whereas the structural variations occur in the central portion of the peptides. Fig. 2 c compares the conformation of peptides bound to HLA-Cw4 (QYDDAVYKL), HLA-A2 (Tax peptide, LLFGYPVYV; reference 32), and HLA-B53 (epitope gag peptide from HIV2, TPYDINQML; reference 33). As in other HLA structures, the peptide termini in HLA-Cw4 are anchored in the cleft by a number of contacts between the peptide main chain atoms and conserved MHC side chains (Table; Fig. 3 a). At the NH₂ terminus, the P1Gln main chain atoms hydrogen-bond to three tyrosine residues from HLA-Cw4 (Tyr7, 159, and 171). A hydrogen bond between the main chain NH₂ group of P2Tyr and the side chain carboxylate of Glu63, which is observed in HLA-A2 and -B27 but absent in HLA-B53, is found in the HLA-Cw4 structure. The conserved hydrogen bonds at the COOH terminus include the ones from the terminal carboxylate oxygen of P9Leu to the side chains of Thr143 and Lys146. The invariant hydrogen bond from the carbonyl oxygen at P8Lys to the pyrrole nitrogen of Trp147 is also present. In addition to the conserved hydrogen bonding network found at the peptide termini, extensive interactions also occur between HLA-Cw4 and the central portion of the peptide. For peptide residues P2–P8, there are five hydrogen bonds from the peptide main chain to HLA-Cw4 side chain atoms and two from the peptide side chain to HLA-Cw4 side chain atoms. We have not been able to observe any water molecule in the region of the peptide, probably due to the limitation of the resolution.

The solvent-accessible surface area buried upon peptide binding for HLA-Cw4 is greater than that found in most peptide–MHC complexes. In the case of HLA-A2 bound to the Tax peptide (LLFGYPVYV), the total solvent-accessible area buried is 1,723 Ų (calculated using the program SURFACE, reference 39; probe radius, 1.4 Å); for HLA-B53 bound to the gag epitope of HIV2 (TPYDINQML), the area buried is 1,647 Ų. In contrast, a total 1,914 Ų of solvent-accessible surface area is buried (1,109 Ų on the peptide and 805 Ų on the MHC heavy chain) upon peptide binding to HLA-Cw4. As shown by the plot in Fig. 4, the three NH₂-terminal residues (P1–P3) and the COOH-terminal residue (P9) of the HLA-Cw4 peptide are almost completely buried (with >93% of the surface area buried for each residue). The two residues in the central region that are mostly exposed are P4Asp and P8Lys. P4Asp is the highest point of the peptide; together, P4Asp and P3Asp form the kink in the peptide main chain conformation that has been observed for many known structures of peptides bound to class I MHC molecules. P8Lys, on the other hand, has been implicated in the binding of HLA-Cw4 to KIR 40.

The HLA-Cw4 structure differs from most known structures of class I MHC molecules in that the peptide conformation includes a pattern of internal hydrogen bonding within the peptide (Fig. 3 b). The main chain–main chain hydrogen bond between the carbonyl oxygen of P2Tyr and the NH₂ group of P4Asp causes P4Asp to adopt φ and ψ angles that are found in a left-handed helix and usually observed in residues forming tight turns and kinks

⁠. All other residues of the peptide have φ and ψ angles that are typical for an extended β strand. In addition, two hydrogen bonds are also formed between a side chain carboxylate oxygen atom of P3Asp and the main chain amino groups of P4Asp and P5Ala. The side chain of P3Asp can not fit into a small D pocket 41 underneath the α2 helix and formed by the side chains of Arg97, Phe99, Arg156, and Tyr159 (Fig. 5 c). As a result, although the C_β atom of P3Asp points toward the α2 helix, its side chain carboxylate is turned back toward the peptide, forming hydrogen bonds with the peptide main chain atoms and the side chain atoms of Arg156 (Fig. 3 b; Table). An internal hydrogen-bonded type I turn has previously been identified in the structure of an HIV gp120 peptide bound to murine H-2D^d 42,43.

P3Asp and P4Asp are important for the binding of the consensus peptide to HLA-Cw4. A peptide containing a single amino acid substitution from P3Asp→P3Ala fails to increase the level of assembled HLA-Cw4 on the surface of the TAP-deficient RMA-S cells that are transfected with the HLA-Cw4 cDNA and human β₂m 40. The P3Asp→ P3Ala mutation results in the loss of three hydrogen bonds mediated by the P3Asp side chain, two internal hydrogen bonds and one hydrogen bond with the side chain of Arg156. Substituting P3Asp and P4Asp with P3His and P4Pro also abolishes peptide binding to the cleft 40. These substitutions would have eliminated all of the internal hydrogen bonds that stabilize the peptide conformation. The side chain of P3His would not be able to fit into the small D pocket, which remains unoccupied even in the case of a smaller side chain of P3Asp. Furthermore, electronic repulsion between the side chains of Arg97, Arg156, and P3His would have greatly destabilized the peptide–MHC complex.

The HLA-Cw4 peptide binding groove is characterized by four specificity pockets (Fig. 5, a and b). The P1 pocket forms the NH₂-terminal boundary of the peptide binding groove and is located in the region of the A pocket 41. The peptide binding groove is completely blocked at this end by the residues Arg62 and Trp167, which point toward each other across the cleft (not shown). The P1 pocket includes the highly conserved tyrosine residues 7, 159, and 171, which hydrogen-bond to the peptide NH₂ terminus (Fig. 3 a). As in HLA-A2, the pocket is lined with the rather polar residues Tyr59, Glu63, Lys66, Tyr159, Thr163, Cys164, and Tyr171, and its floor is formed by Met5 and Tyr7. The P1Gln side chain is firmly positioned by two hydrogen bonds from its side chain amide group to Glu63 and Lys66. Substitution of P1Gln by P1Ser induces a similar level of HLA-Cw4 expression on the cell surface 40, indicating that the P1 pocket is accommodating to medium sized polar residues.

The HLA-Cw4 structure possesses a P2 pocket that is highly specific for tyrosine. The P2 pocket is formed by a cluster of aromatic residues, including Tyr7, Phe22, Tyr67, and Phe99. The hydroxyl group of P2Tyr points toward two polar residues, Arg97 and Gln70, which separate the P2 pocket from the neighboring P7 pocket. The P2 pocket is rather spacious and is not completely filled even by the bulky side chain of P2Tyr. It is conceivable that with minor adjustments, a water molecule could bridge the interaction between the hydroxyl group of P2Tyr and the polar side chains of Arg97 and Gln70. The role of water molecules in mediating the interaction between the peptide and the cleft has been observed in the structures of both human and mouse class I MHC molecules, including HLA-B53, HLA-B27, H-2K^b, H-2D^b, H-2D^d, and H2-M3 24,33,42,44,45,46.

Proline was identified by pool sequencing to be an alternative anchor residue at P2 for HLA-Cw4 29. Cellular binding assays indicate that substitution of P2Tyr with P2Pro abolishes peptide binding to cell-surface HLA-Cw4 40. The structure of HLA-Cw4 also predicts that proline at P2 would be destabilizing. Substitution of P2Tyr with P2Pro would eliminate the hydrogen bond from the main chain amino group of P2 to the side chain carboxylate oxygen of Glu63 and leave the entire P2 pocket vacant. It is possible that the pool sequencing signal for P2Pro was due to the low levels of HLA-B35 expressed on the surfaces of the cells used for pool sequencing.

One of the distinct features of HLA-Cw4 is a P7 pocket located on the side of the α1 helix, formed mostly by residues from the α1 helix (Gln70, Asp74, and Asn77) and the β sheet platform (Ser9, Phe22, Leu95, Arg97, and Phe116). The P7Tyr side chain is secured in the specificity pocket by two hydrogen bonds from the P7Tyr hydroxyl group to the carboxylate oxygen atoms of Asp74. In the known structures of nonameric peptide–MHC complexes, the P5 and P7 side chains are generally oriented toward the α2 helix, whereas the P4 and P6 side chains point toward the α1 helix 32. A well-defined P7 pocket has not been observed in HLA-A and -B structures. In the structure of HLA-E bound to a nonamer derived from the signal peptide of HLA-B8 47, P7 side chain fills a single pocket down toward the α2 helix, which coincides with the E pocket identified by Saper et al. 41. In HLA-Cw4, the C_β atom of the P7Tyr residue points toward the α2 helix; however, the large side chain of Arg156 forces the P7 side chain to turn and point its phenyl ring toward the α1 helix. The P7 pocket in HLA-Cw4 is at the location of the C pocket 41. The presence of Asp9 and Ala73 in the C pocket of some HLA-C alleles has been linked to increased susceptibility to psoriasis vulgaris 48,49. The distinct features of the C pocket in HLA-Cw4 may also be important in its association with type 2 diabetes 7.

The C pocket in HLA-Cw4 is adjacent to the B pocket that hosts the P2 side chain and is separated from the B pocket by a polar ridge formed by the side chains of Ser9, Arg97, and Gln70 (Fig. 5 a). A mouse-specific hydrophobic ridge formed by Trp73, Tyr156, and Trp147 has been found in H-2D^b and H-2L^d 45,50,51. A tryptophan wall created by residues Trp97 and Trp114 is located in the middle of the H-2D^d cleft 42.

The P9 pocket at the COOH-terminal end of the peptide binding groove is hydrophobic, formed by the side chains of Leu81, Leu95, Phe116, Tyr123, Ile124, and Trp147. The pocket forms the COOH-terminal boundary of the cleft. The pocket is not completely filled by the side chain of P9Leu. It can host an even larger hydrophobic residue, such as phenylalanine. Hydrophobic PΩ pocket is characteristic of HLA-A2, HLA-E, and all known mouse class I MHC structures 30,41,42,44,45,47,50,52.

The KIR binding site on HLA-C is located on the α1 domain and includes residues 73, 76, 77, and 80 at the COOH-terminal end of the α1 helix and residue 90 on the loop following it in sequence 14,53,54,55,56. Studies of the KIR3D receptors (e.g., NKB1) that specifically recognize the HLA-Bw4 family indicate that the same region, residues 77–83 in the α1 domain of HLA-B molecules, participates in the interactions with KIR 53,57. Peptide residues P7 and P8 have also been observed to affect the binding of KIRs to HLA-C and -B molecules 40,58,59,60.

In HLA-Cw4, the region surrounding the COOH-terminal end of the α1 helix and residue P8 of the peptide has an electropositive polar surface (Fig. 6 c, blue). The elbow region of KIR2DL1 contains residues involved in HLA-C binding (Met44, Phe45, and Thr70) 18,19,20,61; it has an electronegative polar surface (Fig. 6 e, red). The ligand binding site on KIR2DL1 and the receptor binding site on HLA-Cw4 are complementary in their polarity, and recognition of HLA-Cw4 by KIR2DL1 is possibly mediated by the polar interactions between the oppositely charged surfaces on the two molecules.

Among the residues implicated in KIR binding, residue 80 on HLA-C molecules determines the specificity of HLA-Cw4 for KIR2DL1 and, similarly, that of HLA-Cw3 for KIR2DL3 55. The structure of HLA-Cw4 confirms the importance of residue 80 in mediating the interaction between HLA-C and KIR2D. As shown in Fig. 6 a, the residues that differ between HLA-Cw4 and HLA-Cw3 are concentrated along the peptide binding groove. Once the peptide is bound, however, many of the residues, including residue 77, are buried in their interactions with the peptide (Fig. 6 b). Residue 80 is highly exposed and located at the center of the electropositive surface on HLA-Cw4 that forms the potential KIR binding site (Fig. 6b and Fig. c).

HLA-Cw4 loaded with peptides containing the negatively charged glutamic acid or aspartic acid at P8 are not recognized by a KIR2DL1–Ig fusion protein 40. Peptide residue P8 is important for KIR binding, partly due to the fact that P8 is highly exposed (with 68% of its surface area exposed), with its side chain forming a protrusion on the HLA-Cw4 surface (Fig. 6b and Fig. c). As the P8Lys protrusion is one of the highest points on the HLA-Cw4 surface, it will be readily contacted by KIR2DL1 that approaches HLA-Cw4 from the top of the Cw4 cleft. The positively charged P8Lys in the consensus peptide contributes greatly to the electropositive surface around the COOH-terminal end of the α1 helix. The negatively charged glutamic acid or aspartic acid side chain would result in electronic repulsion between the electronegative surface at the KIR2DL1 elbow and the P8 residue of HLA-Cw4 peptide, thereby abolishing receptor–ligand binding.

A single substitution of tyrosine by glutamic acid at P7 in the peptide also disrupts the interaction between KIR2DL1 and HLA-Cw4 40. The P7 side chain is buried in a pocket under the α1 helix (compare Fig. 6 a and b). The P7 pocket is acidic, formed in part by the residues Gln70, Asp74, and Asn77. Charge–charge repulsion between the P7Glu side chain and the acidic pocket would destabilize the structure. The effect of the P7 residue on the binding of KIR to HLA-Cw4 is likely to be mediated through the conformational changes of the peptide main chain α1 and α2 helices that are necessary to accommodate the P7 side chain.

The specificity of HLA-Cw4 and KIR2DL1 interaction is also mediated by hydrophobic interactions. As shown by Fig. 6 d, residues that differ between KIR2DL1 and KIR2DL2 form hydrophobic patches adjacent to the electronegative surface at the KIR2DL1 elbow (Fig. 6 e). Among these residues, Met44 determines the specificity of KIR2DL1 for HLA-Cw4 18, and Thr70 affects their binding affinity 20. A single mutation from Met44 in KIR2DL1 to Lys44 found in KIR2DL2 switches the specificity of KIR2DL1 from HLA-Cw4 to HLA-Cw3. Lys44 may disrupt the hydrophobic interactions mediated by Met44. Furthermore, its electropositive side chain may be repelled by the electropositive surface on HLA-Cw4.

The structure of HLA-Cw4 allows us to predict the structural features of the closely related HLA-Cw3 molecule. First, HLA-Cw3 and HLA-Cw4 have different specificities for the P2 pocket. The residues that line the P2 pocket include Ser9 and Phe99 in HLA-Cw4 but Tyr9 and Tyr99 in HLA-Cw3. These large side chains found in HLA-Cw3, if oriented like those in HLA-Cw4, would reduce the size of the P2 pocket and preclude any bulky side chain for the P2 residue. This structural argument is consistent with pool sequencing results, which indicate that the dominant signal for P2 in HLA-Cw3 is alanine 29.

Unlike most known class I MHC structures that possess a P7 pocket coincident with the E pocket under the α2 helix, the HLA-Cw4 structure reveals a P7 pocket on the side of the α1 helix. The P7 pocket in HLA-Cw4 is located in the same area as the C pocket identified by Saper et al. 41. The side chain of P7Tyr in HLA-Cw4 fits into a pocket adjacent to that for P2, partly because the internal hydrogen bonds cause the peptide residues from P4 to P7 to shift toward its NH₂ terminus (up to 1.1 Å) and deeper into the cleft (up to 1.6 Å) relative to the Tax peptide bound to HLA-A2. The HLA-Cw3–specific peptides do not contain a P3Asp, which is crucial in forming the internal hydrogen bonds within the HLA-Cw4–bound peptide. Peptides bound to HLA-Cw3 therefore are more likely to adopt conformations similar to those found in HLA-A and -B peptides. In particular, there is not likely to be a shift of the central part of the peptide (P4–P7) toward the peptide NH₂ terminus. The large side chain of Arg156 precludes a P7 pocket on the side of the α2 helix in HLA-Cw4. In HLA-Cw3, Arg156 is substituted with the smaller residue Leu156, creating a P7 pocket in the area of the E pocket under the α2 helix. Instead of tyrosine at P7 for HLA-Cw4, pool sequencing indicates that HLA-Cw3 has a strong signal for phenylalanine or tyrosine at P6. As the residues that form the C pocket are conserved between HLA-Cw4 and HLA-Cw3, it is possible that the C pocket is filled with the aromatic side chain of P6 in HLA-Cw3.

In the structure of the human TCR–HLA-A2 complex 62,63, two different TCRs contact the entire length of the bound peptide (residues P1, P2, and P4–P8). The sequence and conformation of antigenic peptide determines the specificity of the T cell recognition. In mice, the extensive interaction between the TCR and the peptide across the cleft is observed in the structures of the mouse TCR 2C–H-2K^b and TCR N15–H-2K^b complexes 64,65,66. Unlike the case with T cells, which are activated by viral antigens presented on class I MHC molecules, recognition of properly processed self class I MHC–peptide assembly by KIR inhibits target cell lysis by NK cells. In mice, Ly-49⁺ NK cells bind to assembled peptide–class I MHC complexes, but a diverse array of peptides are capable of inducing inhibition 67, indicating that Ly-49 recognition of mouse class I MHC is peptide independent. In humans, both KIR2D and KIR3D contact the COOH-terminal end of the α1 helix (for review see references 9 and 10). Peptides also play a role in the recognition of the HLA-B and -C molecules; peptide residues P7 and P8 appear to affect the binding of KIR to class I MHC molecules directly 40,58,59,60. For HLA-Cw4, negatively charged residues at P7 and P8 of the peptide abolish HLA-Cw4–KIR2DL1 binding 40. For HLA-Cw3, two different peptides confer protection of Cw3-bearing target cells from P58.2⁺ NK cells 68. For HLA-B27, two NK cell clones discriminate among HLA-B27 loaded with four different peptides 59. Therefore, NK cell recognition of class I MHC molecules in humans is peptide dependent but not as specific as T cell recognition, as a much more diverse collection of peptides can confer protection.

The structure of HLA-Cw4 reveals features that may be involved in the recognition of the class I MHC molecule by KIR2DL1. The specific interaction of the receptor–ligand pair appears to involve complementary charged surfaces, with the KIR binding site on HLA-Cw4 being electropositive and the ligand binding site on KIR2DL1 electronegative. Peptide residue P8Lys contributes to specific binding in a unique way in that its side chain is exposed on top of the HLA-Cw4 binding surface, forming a projection that will inevitably be “touched” by the receptor. The structural features of HLA-Cw4 that mediate its interaction with KIR2DL1 may be conserved in the related HLA-C allotypes that are recognized by the same inhibitory NK cell receptor.

We thank L. Mosyak for advice on structure determination, E.O. Long for the cDNA clone of the HLA-Cw4 heavy chain and continuing collaboration on the NK cell receptor, A. Haykov, N. Sinitskaya, and R. Crouse for technical support, D.N. Garboczi, L. Mosyak, Y.H. Ding, and the staff at Cornell High Energy Synchrotron Source (CHESS) for help with data collection, E.O. Long, D.N. Garboczi, J.L. Strominger, and L. Mosyak for critical reading of the manuscript, and H. Huang and R. Gaudet for discussion.

Q.R. Fan is a recipient of a National Science Foundation predoctoral fellowship. D.C. Wiley is an investigator of the Howard Hughes Medical Institute.

Coordinates have been deposited in the Research Collaboratory for Structural Bioinformatics (RCSB) with the RCSB ID 009143 and Protein Data Bank accession code 1QQD. Coordinates are available before their release by e-mail (fan@xtal200.harvard.edu).