Structural Definition of the H-2Kd Peptide-binding Motif*

Classic major histocompatibility complex (MHC) proteins associate with antigen- and self-derived peptides in an allele-specific manner. Herein we present the crystal structure of the MHC class I protein H-2Kd (Kd) expressed by BALB/c mice in complex with an antigenic peptide derived from influenza A/PR/8/34 nucleoprotein (Flu, residues 147-155, TYQRTRALV). Analysis of our structure in conjunction with the sequences of naturally processed epitopes provides a comprehensive understanding of the dominant Kd peptide-binding motif. We find that Flu residues TyrP2, ThrP5, and ValP9 are sequestered into the B, C, and F pockets of the Kd groove, respectively. The shape and chemistry of the polymorphic B pocket make it an optimal binding site for the side chain of TyrP2 as the dominant anchoring residue of nonameric peptides. The non-polar F pocket limits the amino acid repertoire at P9 to hydrophobic residues such as Ile, Leu, or Val, whereas the C pocket restricts the size of the P5-anchoring side chain. We also show that Flu is accommodated in the complex through an unfavorable kink in the otherwise extended peptide backbone due to the presence of a prominent ridge in the Kd groove. Surprisingly, this backbone conformation is strikingly similar to Db-presented peptides despite the fact that these proteins employ distinct motif-anchoring strategies. The results presented in this study provide a solid foundation for the understanding of Kd-restricted antigen presentation and recognition events.

Class I major histocompatibility complex (MHC) 2 proteins serve a critical role in the adaptive immune response by binding short peptide fragments intracellularly and presenting them at the cell surface for surveillance by cytotoxic T lymphocytes (1)(2)(3)(4). Structural studies of human and murine MHC class I proteins in complex with a variety of peptides have revealed conservative structural elements that promote efficient binding and presentation of peptide epitopes (5)(6)(7)(8)(9). Peptides of 8 -10 residues are bound in a predominantly extended conformation within a narrow groove formed by two antiparallel ␣-helices positioned above an eight-strand ␤-sheet platform. Conservative hydrogen bonding networks are established in the binding groove with peptide mainchain and terminal atoms that enable largely sequence-independent ligation.
Although low affinity, kinetically short-lived peptide-MHC complexes can be established by highly diverse epitope sequences, stable association requires the anchoring of peptide side chains into specific pockets in the MHC groove. MHC polymorphisms are clustered in these pockets (3,10,11), and their shape and chemistry impose constraints that are reflected by allele-specific motifs found in the sequences of naturally processed peptides (12,13). For example, H-2 d cell lines (P815) and H-2 b cell lines (EL4) infected with the same influenza virus present different antigenic peptides for CTL recognition (14). Thus, outbred populations that express varied MHC proteins can survey diverse peptide fragments from a given pathogen despite specificity constraints imposed by each individual allele.
To resolve the structural underpinnings of the dominant K d -binding motif we have undertaken crystallographic studies of K d in complex with the antigenic peptide from influenza virus nucleoprotein (Flu). The 2.6-Å resolution structure of K d -Flu provided an excellent framework to delineate the role of polymorphic anchoring pockets in determining K d -specific peptide binding. To extend our understanding of the overall binding motif to a broad population of K d epitopes, we analyzed 95 naturally processed K d peptides in conjunction with our structural data. Comparisons of K d -Flu to other class I peptide-MHC complexes reveal that the conformation of Flu in the K d groove is similar to that of peptides associated with D b despite differences in anchoring strategies between the two MHC proteins. Lastly, our structural studies provide a detailed framework for understanding the role of individual peptide residues in T-cell recognition events.
The purified, detergent-free, inclusion bodies were solubilized overnight in 6 M Gdn⅐HCl, 10 mM Tris, pH 8.0, and 10 mM ␤-mercaptoethanol. To form the K d -Flu complex, murine ␤ 2 m, and heavy chain were refolded under oxidative conditions in the presence of 10 molar excess of Flu (influenza A/PR8/34 nucleoprotein residues 147-155, TYQRTRALV). Refolding was performed at 4°C using a rapid dilution method. Briefly, Flu was diluted to 15 M in 500 ml of refolding buffer (100 mM Tris, pH 8.0, 400 mM L-Arg, 2.0 mM EDTA, 0.5 mM GSSG, 5.0 mM GSH, and protease inhibitors). Murine ␤ 2 m was injected into the refolding reaction to concentration of 4.5 M. Following 30 min of incubation the heavy chain (final concentration 1.5 M) was injected in three separate batches spaced over a 24-h period. The final concentration of Gdn⅐HCl in the refolding reaction did not exceed 100 mM. After an overnight incubation the refolding reaction was concentrated to 4 ml using an Amicon ultrafiltration device (Millipore, Billerica, MA).
The K d -Flu complex was purified from protein aggregates and other impurities on a Superdex75 (GE Healthcare, Piscataway, NJ) size exclusion column using a running buffer containing 20 mM HEPES, pH 7.5, 150 mM NaCl, and 0.01% NaN 3 . The fractions containing K d -Flu were pooled, diluted 3-fold in buffer containing 20 mM Tris, pH 8.5, loaded on an anion exchange Mono Q column (GE Healthcare), and eluted with a NaCl gradient (0 mM to 400 mM NaCl over 30 ml). Prior to crystallization the pure K d -Flu complex was exchanged in buffer containing 20 mM HEPES, pH 7.5, and 150 mM NaCl. Typically 400 g of purified complex were obtained from a 500-ml refolding reaction.
Crystallization and Data Collection-Diffraction-quality crystals of K d -Flu were obtained by hanging-drop vapor diffusion method. Protein at 6 -8 mg/ml was equilibrated at 20°C against 12% (w:v) polyethylene glycol 2000, 5% (w:v) 2-methyl-2,4-pentanediol, and 100 mM HEPES, pH 6.9. Small crystals obtained in these drops were used to microseed protein hanging drops equilibrated against similar conditions with marginally lower concentration of polyethylene glycol 2000. Larger crystals appeared overnight and grew over 3-4 weeks. As the crystallization conditions were cryoprotective, no additional compounds were added for liquid nitrogen flash cooling. X-ray diffraction data for K d -Flu were collected on a Raxis IV detector (Rigaku/MCS, The Woodlands, TX) to 2.6-Å resolution. A total of 249 frames was collected each representing a 1.5°oscillation range. These data were indexed and integrated using DENZO (HKL Suite, HKL Research, Inc., Charlottesville, VA) in the primitive orthorhombic lattice with cell dimensions a ϭ 117.2 Å, b ϭ 85.3 Å, c ϭ 42.6 Å and scaled and merged using SCALEPACK (HKL Suite, HKL Research, Inc., Charlottesville VA). Wilson scaling was applied to the final output structure factor amplitudes (Collaborative Computing Project 4 (CCP4), Daresbury Laboratory, Warrington UK (32)).
Structure Determination and Refinement-The structure of K d -Flu was determined by molecular replacement using AMoRe (CCP4). The coordinates of K b -Ova (PDB 1VAC) with the Ova peptide and water molecules omitted were used as a search model. The rotation search yielded a single, distinct solution, and translation searches were run in all possible orthorhombic symmetry groups. The highest signal (correlation coefficient ϭ 33.1% and R value ϭ 49.4% for all 15-4.0 Å data) was obtained in the P2 1 2 1 2 1 space group, consistent with the systematic absences in the data. Rigid body refinement of the three domains of the search model (␣ 1 ␣ 2 , ␣ 3 , and murine ␤ 2 m) against the K d data yielded an R value of 41.8% for the 20-to 2.6-Å resolution data. Extensive model building was performed with the macromolecular modeling program O (O version 6.22, Uppsala Software Factory, Sweden) using 2F o Ϫ F c , F o Ϫ F c , and 2F o Ϫ F c composite omit maps (CNS, Yale University, New Haven CT (33)). Atomic refinement was done employing simulated annealing, energy minimization, and restrained B-factor refinement protocols as implemented in CNS. The final model includes a total of 383 residues (residues A1 to A275 for the heavy chain, B1 to B99 for murine ␤ 2 m, and P1 to P9 for Flu) and 114 water molecules. Atomic coordinates were not assigned to residues 276 -283 from the heavy chain and the N-terminal methionine of ␤ 2 m as no interpretable electron density was seen for these regions of K d -Flu. Refinement of this final model against the 20-to 2.6-Å resolution data converged to an R value of 21.6% with an R free of 26.9% (4.3% test set) with good geometry (Table 1).
Computational Analysis-Graphical structure representations were primarily created using Ribbons (34). Molecular surfaces of the peptidebinding groove (Figs. 4A and 5) were generated using InsightII (Biosym Technologies, San Diego CA). r.m.s.d. values between the different MHC proteins were calculated using an incremental combinatorial extension algorithm (35). r.m.s.d. values between the different MHC peptides were calculated using Lsqkab (CCP4). HBPLUS (36) was used to catalogue contacting atoms and putative hydrogen bonds. Shape complementarity scores (37) were calculated using CCP4. Atomic accessible surfaces were calculated using the program NACCESS. 3

RESULTS
Structure Determination-The extracellular domains of K d (heavy chain, residues 1-283 and murine ␤ 2 m, residues 1-99 plus an N-terminal methionine) were expressed separately in Escherichia coli as insoluble inclusion bodies. The K d -Flu complex was formed in vitro under oxidative refolding conditions in the presence of excess peptide and was purified using size exclusion and anion exchange chromatographies. Electrospray mass spectral analysis of the complex confirmed the presence of abundant peaks at 32,865.69 Da and 11,817.81 Da corresponding to the predicted molecular weights of the heavy chain and murine ␤ 2 m, respectively. Further inspection of the mass spectrum over lower mass ranges revealed a monoisotopic, singly charged peak at 1,106.7 Da corresponding to Flu.
The K d -Flu complex crystallized in the orthorhombic space group P2 1 2 1 2 1 with one complex per asymmetric unit. Initial phase estimates were obtained by molecular replacement. After initial refinement, easily interpretable electron density was seen for the bound peptide that improved upon further building and refinement cycles. Diffraction data to 2.6-Å resolution were used for refinement of the final atomic model, which has an R factor of 21.6% (R free ϭ 26.9%) with good angle and bond geometry ( Table 1). The electron density maps for the whole complex were of good quality (Fig. 1A). No ambiguities were seen for the main chain and the side chains of the bound peptide except for a small break in the electron density between the C␤ and C␦ carbons of the exposed Arg P6 side chain (Fig. 1A).
Overall Structural Features of the Complex-K d -Flu is very similar to the structures of other MHC class I proteins (Fig. 1B). Minor differences were observed for the conformations of solvent-exposed loops and the terminal regions of the complex. Flu is bound in the K d groove between the ␣ 1 and ␣ 2 helices and on top of the ␤-sheet platform ( Fig. 1) in canonical manner (7). Structural alignment of K d -Flu to the structures of other murine MHC class I complexes yielded overall pairwise r.m.s.d. values of 1.14 Å (D d , 80.4% sequence identity) to 2.15 Å (D b , 80.7% sequence identity). An alignment of the ␣ 1 ␣ 2 domains alone yielded pairwise r.m.s.d. values of 0.63 Å (K b , 80.4% sequence identity) to 1.01 Å (L d , 81.1% sequence identity) reflecting the high degree of general similarity among these proteins.
Backbone Conformation of K d -presented Flu-Like most MHCbound nonameric peptides, the backbone of Flu assumes a predominantly extended conformation with a bulge at residues P6 and P7. Surprisingly, the main-chain kink adopted by Flu results from an infrequently observed, unfavorable conformation of Arg P6 that is well supported by our experimental data (Fig. 1A). The P6 -P7 bulge is associated with a hydrophobic ridge formed by polymorphic residues Tyr 156 and Trp 73 in the K d groove. While favorable Ramachandran angles would be observed if the P6 -P7 peptide bond were flipped, this conformation would preclude several favorable interactions and engender steric clashes with K d (see Fig. 4A).
We compared the conformation of Flu to that of nine-residue peptides bound to K b (38,39), L d (40), and D b (41)(42)(43)(44)(45). The main-chain conformations of K b -presented peptides vary significantly from that of Flu with overall r.m.s.d. values ranging from 1.43 to 1.47 Å. The greatest differences were observed in the region between P5 and P7 where the r.m.s.d. values for the C␣ atoms ranged from 1.22 to 3.66 Å (Fig. 2A). The conformation of the L d peptide resembles more closely that of Flu with r.m.s.d. 0.95 Å, but nevertheless differs significantly between P4 and P6 (Fig. 2A). Comparison of Flu with D b -presented peptides reveals that they adopt nearly identical main-chain conformations all the way from P1 to P9 with r.m.s.d. values ranging from 0.62 to 0.81 Å (Fig. 2A).
Interestingly, D b has a similar hydrophobic ridge as K d located beneath the P6 -P7 kink (43).
We also compared the dihedral angles of the aligned peptides (Fig.  2B). This analysis revealed that the backbones of the D b peptides adopt a P6 -P7 bulge associated with unfavorable dihedral angles for their P6 residues similar to the one in Flu (Fig. 2B). This bulging at P6 -P7 was absent in the L d peptide (Fig. 2B) despite the presence of a similar hydrophobic ridge in the same region of the L d groove (40).
Hydrogen Bonding to the Flu Backbone-The Flu main chain has 19 nitrogen and oxygen atoms, 14 of which hydrogen bond with K d either directly or through water-mediated networks. Of the 17 MHC amino acids that participate in hydrogen bonding 8 are invariant among MHC class I proteins. These residues anchor the N-and C-terminal regions of Flu through highly conservative hydrogen bonding networks at each end of the binding groove (Fig. 3).
Eight main-chain nitrogen and oxygen atoms between Gln P3 and Val P8 mediate hydrogen bonds with polymorphic groove residues (Fig.  3). Of particular note are the hydrogen bonds to the main-chain oxygen of Arg P6 and the main-chain oxygen and nitrogen of Ala P7 . The carbonyl oxygen atoms of both residues participate in a bifurcate hydrogen bonding network with the N⑀1 nitrogen of Trp 73 , whereas on the opposite side of the Flu backbone the amide nitrogen atom of P7 hydrogen bonds with Asp 152 (Fig. 3). This hydrogen bonding arrangement can only form as a result of the unfavorable turn in the Flu main chain at Arg P6 .
Binding Pockets in the K d Groove-Specificity of peptide-MHC association is imparted through a myriad of interactions with peptide anchor side chains, which are sequestered in distinct pockets of the MHC groove. To visualize these pockets in K d we calculated a solventaccessible surface (46) for a spherical probe with a radius of 1.4 Å for the ␣ 1 ␣ 2 domain. Five distinct pockets are clearly apparent in the K d groove, which correspond to pockets A, B, C, D, and F according to the nomenclature of Matsumura et al. (47) (Fig. 4A). Pocket E, which is the most variable between the different MHC proteins, is absent in our structure. Instead, the polymorphic residues Trp 73 and Tyr 156 fill the E pocket location creating the hydrophobic ridge across the K d groove that accommodates the P6 -P7 turn in Flu (Fig. 4A). Interestingly, this ridge is absent in K b , which preferentially binds 8-residue peptides.
To systematically identify the MHC residues that make up the K d pockets, we calculated solvent-accessible surfaces of the ␣ 1 ␣ 2 domain with the use of probes of different radii ranging from 1.4 to 5.0 Å. Using this method we were able to assign residues to pockets when they are accessible to a small, 1.4-Å probe but are inaccessible to a probe larger than 3.5 Å. The   Fig. 1 with side chains beyond the C␤ atom omitted. Water molecules are displayed in gray. Putative hydrogen bonds are shown as small silver balls connecting atoms Ͻ3.5 Å apart with reasonable geometry. The complex is oriented as in Fig. 1A.
larger pockets A, B, and F are created by 8, 9, and 8 residues, respectively (Table 2), whereas the smaller C and D pockets are each formed of 5 amino acids ( Table 2). Each pocket contains residues accessible only to a probe no larger than 2.5 Å ( Table 2) indicating that all five K d pockets are deep. This is in contrast, for example, to the open nature of the pockets in the K b groove, but is similar to the discrete nature of the pockets in D b .
The adjoining pockets B and C are highly polymorphic. Seven nonconserved and two invariant residues create the B pocket, and four non-conserved residues and one invariant residue create the C pocket ( Table 2). In contrast, pockets A and F are composed of mostly conserved residues, which is consistent with their roles in anchoring of the N and C termini of the bound peptide.
Anchoring of Flu in the K d Groove-In the complex Flu participates in 147 van der Waals contacts, 55 hydrophobic contacts, 20 direct, and 3 water-mediated hydrogen bonds with K d . The majority of the hydrogen bonds (19 out of 23) are directed to the main chain of the bound peptide. In contrast, most of the hydrophobic contacts associated with ligand binding (51 out of 55) are established with peptide side chains. A total of 1236 Å 2 (79%) of the Flu solvent-accessible surface is buried in the MHC groove, whereas 753 Å 2 of K d becomes solvent-inaccessible upon complex formation.
Flu binding is associated with the complete burial of Tyr P2 , Thr P5 , and Val P9 in pockets B, C, and F, respectively (Fig. 4A). These side chains account for 44% of the total peptide buried surface area. The interaction surface between Flu and K d is characterized by a shape complementarity Putative hydrogen bonds connecting atoms Ͻ3.5 Å apart with reasonable geometry are also shown and represented as small gray balls. The heavy chain is represented as a ribbon tube and colored in cyan. The peptide is rendered as a stick model and colored as in Fig. 1 Note that the side chains of P1, P3, P4, P6, and P8 are omitted beyond their respective C␤ carbon atoms. score of 0.76 comparable to that of other peptide-MHC interfaces and slightly greater than antigen-antibody interfaces (37).
B Pocket-Our structural data reveal that Tyr P2 fits snugly in the B pocket where it makes three edge-to-face -stacking interactions with Tyr 7 , Phe 45 , and Phe 99 and mediates large hydrophobic contacts with Val 9 and the aliphatic portion of Arg 66 (Fig. 4B). The hydroxyl group of Tyr P2 hydrogen bonds directly to the O␦1 oxygen of Asp 70 and makes favorable electrostatic contacts with the guanidinium group of Arg 97 (Fig. 4B). A number of binding studies of P2 peptide variants (48,49) reveals that Tyr P2 is the dominant anchoring residue. Conservative changes at that position result in two orders of magnitude decrease in binding affinity, whereas non-conservative mutations of the P2 Tyr completely abrogate binding.
C Pocket-Thr P5 of Flu is buried in the C pocket where it makes favorable hydrophobic contacts with the indole ring of Trp 73 (Fig. 4B). The hydroxyl group of Thr P5 mediates a direct hydrogen bond to the O␦2 oxygen of Asp 70 and participates in favorable van der Waals contacts with the guanidinium group of Arg 97 (Fig. 4B). The C pocket appears larger than the small Thr residue anchored in it (Fig. 4A) indicative of a capacity to accommodate both small and medium size side chains. Indeed, binding studies indicate that only variants with large and/or charged amino acids at P5 exhibit decreased binding affinity (50).
F Pocket-The Flu terminal residue, Val P9 , is fully buried in the F pocket and makes a number of favorable hydrophobic contacts with Trp 73 , Tyr 84 , Phe 95 , Tyr 123 , Thr 143 , and Trp 147 (Fig. 4C). The Val P9 side chain appears smaller than the F pocket indicating that larger hydrophobic residues such as Ile and Leu would bind well in this region of the K d groove. Mutations at the P9 position show that the F pocket tolerates broad amino acid substitutions as peptides with non-conservative P9 substitutions still retain, albeit low, binding capacity (48,49).

DISCUSSION
To structurally define peptide binding and presentation by K d we crystallized it in complex with the well characterized antigenic epitope derived from influenza nucleoprotein. Tyr P2 , Thr P5 , and Val P9 of Flu are completely buried in pockets B, C, and F, respectively. There they make a myriad of favorable hydrophobic and electrostatic contacts arguing that these polymorphic pockets are the dominant structural determinants of peptide binding specificity for K d (Fig. 5).
The K d Peptide-binding Motif-To define the peptide-binding motif of K d in the context of relevant physiological data, we analyzed the sequences of 95 naturally processed, nonameric K d epitopes recently identified by Unanue and colleagues (60) and further available in MHC peptide databases (51)(52)(53). Although peptides of alternate length have been shown to associate with K d , the resulting complexes are short-lived and are unlikely to be physiologically dominant (60).
As expected from the exquisite binding chemistry and geometry, tyrosine is found almost exclusively at the P2 position (96%) of the naturally processed nonameric peptides (Fig. 5). The size, shape, and chemical nature of the B pocket clearly dictate the stringent selection of tyrosine at this position. Aromatic residues are flawlessly poised for stacking interactions with the tyrosine phenyl ring; the guanidinium group of Arg 97 neutralizes the negative electrostatic potential of the tyrosine hydroxyl moiety, whereas Asp 70 hydrogen bonds directly to the hydroxyl oxygen.
Second only to the P2 position, the C-terminal position in the K d peptides exhibits high levels of amino acid restriction with preference for Leu, Ile, and Val. Among these, Ile and Leu are found at nearly equal frequencies and account for Ͼ80% of the amino acids found at the P9 position in the K d peptide pool (Fig. 5). This preference toward larger hydrophobic side chains likely results from a higher kinetic stability (lower k off ) of the peptide-MHC complex when a larger hydrophobic area is buried in the F pocket (54,55). In addition, the chymotrypsin-like activity of the proteasome generates cleavage products with hydrophobic C-terminal residues further contributing to the observed hydrophobic side chain prevalence at the C termini of MHC-I-presented epitopes (56,57).
The K d peptide comparisons reveal that the buried P5 position tolerates a variety of amino acid residues with restriction mainly based on their size (Fig. 5). In addition to Ser and Thr, which account for 31% of the residues at P5, small and medium size hydrophobic residues are also prevalent (Fig. 5). We believe that this broader tolerance is due to the larger size of the C pocket and the ability of Asp 70 to hydrogen bond to the P5 main chain nitrogen facilitating its burial in the complex in the absence of a hydrogen bond donor at the P5 position.
Implication for TCR Recognition-Our structural data and peptide analysis predict that P4 and P6 are dominantly involved in TCR recognition. These side chains extrude prominently from the K d groove in Flu (Figs. 1B and 4A) and show the greatest amino acid variability in the K d peptide pool. Indeed, P4 and P6 have been shown to be critical for T-cell recognition in several K d -restricted model systems (Fig. 5) (20,22,23). For example, the recognition of a peptide derived from dengue type 2 NS3 protein by the 2D42 CTL clone is compromised by P6 substitutions. The recognition of NS3 peptides from all three dengue serotypes by the E10.6 clone is diminished by P4 substitutions (22). Another example is the cross-reactive CTL clone B7-B7. It recognizes HA peptides derived from two strains of influenza, A/Jap and A/PR/8, and its activity is sensitive to amino acid-substituted peptides at P4 and P6 (23). Lastly, the G9 -C8 CTL clone that recog- nizes an insulin-derived peptide is similarly dependent on the P4 and P6 peptide residues for K d -restricted cytotoxic activity (20).
The T-cell recognition data in the dengue virus and diabetes model systems also implicate residues P1, P3, P7, and P8 as additional recognition elements. The 2D42 clone distinguishes between type 2 and type 1 and 3 NS3 peptides based on a single Glu to Gly substitution at P8, whereas its activity is greatly diminished by amino acid substitutions at P7 and P8 within the type 2 NS3 peptide (22). In addition, the cytotoxic activity of the E10.6 clone is weakened by the presentation of NS3 peptides substituted at P3 (22). Likewise, alanine substitutions at P1, P3, and P8 compromise recognition of the insulin peptide by the G9 -C8 clone (20). The P1 residue has also been shown to be critical for TCR recognition in the C18 T-cell rejection of transplantable fibrosarcomas. In this model, tumor elimination is dependent on the differential recognition of a variant ERK2 kinase peptide that differs from wildtype only at the P1 position (15,17). These results are consistent with Aligned under the rendering are the sequences of Flu (boxed) and six other antigenic peptides representative of the K d peptide-binding motif as well as three antigenic peptides for which TCR contact residues (denoted by the asterisk) have been determined. Pocket B and the P2 residue are highlighted in red while the secondary anchors P5 and P9 and their respective pockets C and F are highlighted in green. Percent values in parentheses reflect the frequencies at which each amino acid is found at the indicated position; only the most prevalent amino acids are denoted. The shorthand above each aligned residue denotes the orientation of each residue of Flu in the complex. The anchor symbol denotes anchoring residues that point down toward the peptide binding platform; 1 denotes residues pointing away from the peptide binding platform and toward solvent; 3 denotes residues pointing toward the ␣ 1 helix; and 4 denotes residues pointing toward the ␣ 2 helix (see Fig. 4A).
the structure of K d -Flu in which these auxiliary contact residues are appreciably solvent-accessible (Fig. 4A).
In K d -Flu the P5 anchor residue is buried in the C pocket. Interestingly, Wong et al. have shown that substitutions at P5 in the insulin-derived peptide reduce or completely abolish G9 -C8 cytotoxic activity, which the authors interpreted as P5 being able to directly interact with the TCR (20). It is conceivable that the insulin peptide binds to K d with an exposed P5 residue. However, it is also feasible that reorganization events related to anchoring of different residues in the C pocket result in concerted changes in the peptide-MHC conformation. In fact, buried secondary anchor residues are known to be capable of indirectly modulating T-cell recognition events by buttressing the overall peptide-MHC surface conformation (58,59). Taken together, the results of our study enable a detailed understanding of K d -restricted antigen processing, presentation, and recognition in a number of murine model systems involving pathogen recognition, tumor rejection, and induction of diabetes.