Classical and nonclassical class I major histocompatibility complex molecules exhibit subtle conformational differences that affect binding to CD8alphaalpha.

The cell surface molecules CD4 and CD8 greatly enhance the sensitivity of T-cell antigen recognition, acting as "co-receptors" by binding to the same major histocompatibility complex (MHC) molecules as the T-cell receptor (TCR). Here we use surface plasmon resonance to study the binding of CD8alphaalpha to class I MHC molecules. CD8alphaalpha bound the classical MHC molecules HLA-A*0201, -A*1101, -B*3501, and -C*0702 with dissociation constants (K(d)) of 90-220 microm, a range of affinities distinctly lower than that of TCR/peptide-MHC interaction. We suggest such affinities apply to most CD8alphaalpha/classical class I MHC interactions and may be optimal for T-cell recognition. In contrast, CD8alphaalpha bound both HLA-A*6801 and B*4801 with a significantly lower affinity (>/=1 mm), consistent with the finding that interactions with these alleles are unable to mediate cell-cell adhesion. Interestingly, CD8alphaalpha bound normally to the nonclassical MHC molecule HLA-G (K(d) approximately 150 microm), but only weakly to the natural killer cell receptor ligand HLA-E (K(d) >/= 1 mm). Site-directed mutagenesis experiments revealed that variation in CD8alphaalpha binding affinity can be explained by amino acid differences within the alpha3 domain. Taken together with crystallographic studies, these results indicate that subtle conformational changes in the solvent exposed alpha3 domain loop (residues 223-229) can account for the differential ability of both classical and nonclassical class I MHC molecules to bind CD8.

against intracellular pathogens such as viruses. CD8 greatly enhances antigen recognition by CTL (1) and is referred to as a co-receptor, since it binds to the same peptide-major histocompatibility complex class I molecules as the T-cell receptor (TCR) (1). Generation of mature CTL also depends upon the presence of CD8 at their surface (2) and upon the interaction of CD8 with MHC (3,4). CD8 exists at the cell surface as a homodimer of two ␣ chains (CD8␣␣) and as a heterodimer or an ␣ and a ␤ chain (CD8␣␤). Expression of the heterodimeric form of CD8 is limited to the ␣␤ T-cell population, whereas homodimeric CD8␣␣ is present not only on ␣␤ T-cells but also on subsets of ␥␦ T cells and natural killer (NK) cells (5).
Direct binding of CD8␣␣ to classical MHC alleles was initially demonstrated using cell-cell adhesion assays (6). Mutagenesis data suggested that CD8 and the TCR bind to separate sites on the MHC molecule (7). Recently, crystal structures of human and murine CD8␣␣-MHC complexes (8,9) have shown that CD8␣␣ binds to an extended site on classical MHC molecules, which includes nonpolymorphic residues from the ␣2 and ␣3 domain as well as ␤ 2 -microglobulin. The two CD8␣ subunits contact the ␣3 domain in a manner resembling antibody-antigen interaction, with their complementarity determining region-like loops forming a pocket which accommodates an exposed loop (residues 223-229) linking the C and D strands. This mode of interaction prevents the simultaneous binding of a second MHC molecule to CD8␣␣. Furthermore, the crystal structure also indicates that the binding of CD8␣␣ does not induce any significant changes in the TCR-binding platform of the MHC. We have recently studied the interaction of soluble forms of these molecules by surface plasmon resonance (10). Despite the large binding interface (total surface area buried upon binding ϳ2000 Å 2 ), CD8␣␣ binds HLA-A2 with an extremely low affinity (K d ϳ200 M). In agreement with the structural studies, binding of CD8␣␣ to HLA-A2-peptide did not affect the binding of a TCR to the same peptide-MHC complex (10).
To investigate further the mechanism of co-receptor function, we have determined the extent to which the affinity of CD8/class I MHC interactions varies, by analyzing the binding of CD8␣␣ to HLA-A, -B, and -C molecules. The genes that encode these classical MHC molecules are the most polymorphic human genes described (11). While CD8␣␣ can bind to most classical MHC alleles, there is evidence that some alleles, such as HLA-A68 and -B48, possess polymorphisms in the ␣3 domain, which make them especially poor ligands for the co- receptor (12,13). Therefore CD8␣␣ binding to a number of different alleles, including HLA-A68 and -B48, was tested. Also, since our understanding of the interaction of CD8 with nonclassical MHC molecules is limited, the binding of CD8␣␣ to the nonclassical class I MHC molecules HLA-E and HLA-G was investigated. Site-directed mutagenesis was used to establish the molecular basis of the observed differences in CD8␣␣ binding.
Expression of HLA Heavy Chains-The extracellular portion (residues 1-276) of both classical and nonclassical HLA class I molecules were expressed in E. coli BL21pLysS(DE3) strain using the T7 promotor based vectors pET23d (Novagen) or pGMT7 (15). Expression of HLA-A*02, -B*35, and -C*07 and HLA-E heavy chains has been reported previously (16 -18). HLA-A*11, -A*6801, and -B*4801 heavy chains were cloned into pET23d using the unique restriction sites NcoI/HindIII. HLA-G heavy chains were cloned into pGMT7 with the unique restriction sites NdeI/HindIII. HLA heavy chains were purified from inclusion bodies and solubilized in denaturants as described previously (16).
Preparation of Biotinylated HLA Class I Complexes-Biotinylated soluble HLA class I complexes were prepared as described previously (10,19,20). Briefly, ␤ 2 -microglobulin was refolded and then chemically biotinylated using hydroxysuccinimidobiotin (Sigma, Gillingham, UK). HLA complexes were then prepared by refolding the relevant HLA heavy chain and peptide with the chemically biotinylated ␤ 2 -microglobulin in a standard manner (16). HLA-peptide complexes were subsequently purified by size exclusion using a Superdex 75 column (Amersham Pharmacia Biotech, Uppsala, Sweden). The elution volume of each complex was similar to that of the 43 kDa molecular mass standard ovalbumin.
Surface Plasmon Resonance-Surface plasmon resonance binding studies were performed at 25°C using a BIAcore TM 2000 (BIAcore AB, St. Albans, UK) in Hanks' balanced salt solution (BIAcore AB). Hanks' balanced salt solution contains 10 mM HEPES (pH 7.4), 150 mM NaCl, 3.4 mM EDTA, and 0.005% surfactant P20. Streptavidin (Sigma, Gillingham, UK) was covalently coupled to Research Grade CM5 sensor chips (BIAcore AB) via primary amines using the Amine Coupling Kit (BIAcore AB). Streptavidin was dissolved in 10 mM sodium acetate (pH 5.5) and injected over the sensor chips at 0.5 mg/ml during coupling. Biotinylated proteins (HLA complexes and the control protein OX68, an anti-rat CD4 mAb, IgG2a) were immobilized at 4000 -6000 RU (reso- nance units) by injection at 50 -150 g/ml for 0.5-10 min over the streptavidin-coupled surfaces. All CD8␣␣ injections were for 30 s, at a flow rate of 10 l⅐min Ϫ1. K d values were obtained either by Scatchard plots or by nonlinear curve fitting of the Langmuir binding isotherm (bound ϭ C*max/(K d ϩ C), where C is CD8␣␣ concentration and max is maximum CD8␣␣ binding) to the data using the Levenberg-Marquardt algorithm as implemented in the Windows 95 application Origin (version 5; Microcal Software, Northampton, MA).

RESULTS AND DISCUSSION
Classical HLA-A, -B, and -C Alleles Define a Standard CD8␣␣ Binding Affinity-We used SPR to directly measure the affinity of CD8␣␣ binding to HLA-A*1101, HLA-B*3501, and HLA-C*0702 alleles. Equilibrium affinity measurements (see Fig. 1) were carried out by injecting soluble recombinant CD8␣␣ (10, 14) over sensor chip surfaces coated with particular peptide-MHC complexes, or a control protein, as described under "Experimental Procedures." The difference in the responses obtained in the immobilized peptide-MHC and control surfaces represented specific binding to the peptide-MHC. The equilibrium dissociation constant (K d , or affinity) was measured by repeating such injections over a range of CD8␣␣ concentrations. As shown in Fig. 1, CD8␣␣ bound to HLA-A*1101, HLA*B3501 and HLA-C*0702 with affinities ranging from 90 to 220 M. These compare with an affinity of 140 M, which was previously determined for the CD8␣␣/HLA-A*0201 interaction 110 (n ϭ 1) A2-254AT 470 (n ϭ 1) c B48-245TA 120 (n ϭ 1) EloopA 160 Ϯ (n ϭ 2) AloopE Ն1000 (n ϭ 1) c a Mean Ϯ S.D. or, for n ϭ 2, mean Ϯ range of n independent determinations.
b Data from Wyer et al. (10). c Saturation was not reached in these experiments, making estimation of the affinity difficult. In each case, the maximum binding response predicted during nonlinear curve fitting did not exceed that predicted on the basis of the amount of specific HLA immobilized on the sensor surface. (10). Collectively, the exact ␣3 domain sequences of these molecules are representative of a large number of HLA alleles, and it is, therefore, likely that CD8␣␣ will bind to the great majority of HLA alleles with affinities in this range (K d 90 -220 M at 25°C).
HLA-A*6801 and HLA-B*4801 Bind CD8␣␣ with a Severely Reduced Affinity-As shown in Fig. 2, A and C, CD8␣␣ bound to HLA-A*6801 and HLA-B48*01 with a significantly lower affinity than the alleles described above. Despite the injection of CD8␣␣ at high concentrations (Ն1800 M), binding did not reach saturation, making precise estimates of affinities difficult. However, the affinities of the HLA-A*6801 and HLA-B48*01 interactions were both Ն1 mM (see Table I). Previously reported cell binding experiments and mutational studies have indicated that for both alleles; the polymorphism at position 245 accounts for these reduced affinities (12,13). We sought to confirm this by mutating 245 to the consensus Ala and testing whether affinity was increased. As can be seen in Fig. 2, B and D, CD8␣␣ bound to the two mutant proteins A68-V245A and B48-T245A with affinities comparable with HLA-A*0201. This is strong evidence confirming that polymorphism at residue 245 is indeed responsible for the low affinity of CD8␣␣ binding to HLA-A*6801 and HLA-B*4801. In support of this, introduction of reciprocal mutations to Val or Thr at position 245 of HLA-A*02 (mutants A2-A245V and A2-A245T, respectively) led to a significant drop in affinity for CD8␣␣ (see Table I).
The Nonclassical MHC Molecules HLA-E and HLA-G Have Different Affinities for CD8␣␣-HLA-E and HLA-G are similar to classical MHC molecules in terms of homology, secondary structure, and association with ␤ 2 -microglobulin, but exhibit only very limited polymorphism (21). Within the ␣3 domain, both molecules contain differences from the consensus sequence of classical HLA molecules. As can be seen in Fig. 4, these occur at positions 219, 223, 224 in the HLA-E sequence and 214 and 228 in the HLA-G sequence. All of these changes are located extremely close to the HLA/CD8␣␣ interface in or around the 223-229 loop (see Figs. 4 and 5) and could poten-tially affect binding to CD8␣␣.
Despite the presence of such amino acid differences, CD8␣␣ bound HLA-G (see Fig. 3A) with an affinity (K d ϳ150 M) comparable with that of CD8␣␣/HLA-A*02 interaction. In contrast, HLA-E bound poorly to CD8␣␣ (Fig. 3B), and even at high concentrations (Ͼ 800 M) binding responses were small and did not reach saturation. The affinity of CD8␣␣ for HLA-E was estimated to be Ͼ1 mM (see Table I), a figure comparable with interaction with HLA-A*6801 and -B*4801.
Conformational Changes in the ␣3 Domain 223-229 Loop Can Account for Variation in CD8␣␣ Binding-Direct interaction of CD8␣␣ with the 223-229 loop of the HLA ␣3 domain was predicted from mutagenesis studies (7). Similar studies also indicated that polymorphism at position 245 in the ␣3 domain of HLA-A*68 resulted in a loss of CD8␣␣-mediated cell-cell adhesion (12). The crystal structure of an HLA-A*02-CD8␣␣ complex confirmed that this loop lies at the center of the interface, making direct contacts to both CD8␣ subunits (8). Comparison of HLA-A*6801 and HLA-A*0201 structures showed that despite the similarity in the overall fold of the molecules, the introduction of a bulky Val side chain at position 245 in HLA-A*6801 causes a steric conflict with Thr 228 , triggering a small but significant distortion of the 223-229 loop in the ␣3 domain (8) (see Fig. 6). This distortion is sufficient to reduce binding to CD8␣␣ severely.
Our data show that CD8␣␣ also binds HLA-B*4801 with an affinity similar to that of the CD8␣␣/HLA-A*6801 interaction and confirms that this lower affinity is due to polymorphism at residue 245 (see Fig. 2). The two amino acids implicated, Val (HLA-A*6801) and Thr (B*4801), have similar sized side chains, and this would suggest that in HLA-B*4801, steric conflict between Thr 245 and Thr 228 is likely to induce structural perturbation of the 223-229 loop via a similar mechanism to that operating in HLA-A*6801. It has been suggested that B*8101, which like B*4801 has a Thr at position 245, will also have a reduced affinity for CD8␣␣ (13).
Equilibrium binding measurements clearly indicated that HLA-E bound CD8␣␣ with an affinity far lower than the majority of classical HLA alleles (Ͼ1 mM, see Fig. 3). While the overall conformation of HLA-E is similar to HLA-A*0201 (22), detailed comparisons of the HLA ␣3 domains indicate that the 223-229 loop adopts a conformation distinct from that of HLA-A2 and similar to that of HLA-A68 (Fig. 6). In contrast to HLA-A*6801, this difference is most likely due to amino acid changes adjacent to the loop itself (at positions 219, 223, and 224 in the HLA-E sequence, see Figs. 4 and 5). Mutation of these residues to the HLA-A2 sequence (mutant EloopA) resulted in CD8␣␣ binding with an affinity comparable with that of CD8␣␣/HLA-A*02 interaction (K d ϳ140 M, see Fig. 3C). In the converse experiment, the introduction of the HLA-E amino acids at these positions into the HLA-A*02 sequence (mutant AloopE), severely reduced the affinity of the interaction with CD8␣␣ (see Table I). These results demonstrate that amino acid differences adjacent to the ␣3 domain loop are responsible for the decreased affinity of HLA-E for CD8␣␣. We suggest that the structural basis of this effect is likely to be conformational changes within the loop itself (see Fig. 6).
CD8␣␣ Interaction with Classical MHC Molecules-One critical feature of CD8 ϩ function in both thymic selection and peripheral activation of T-cells is a requirement that CD8 and TCR bind to the same MHC molecule (4,(23)(24)(25). This raises the question of how CD8/MHC interaction is able to increase the sensitivity of antigen recognition, and contribute to thymic selection events, without compromising specificity. To resolve this question it is important to know the physical properties of CD8/MHC interaction and how they relate to TCR/MHC binding.
Our results suggest that most HLA-A, -B, and -C alleles are likely to interact similarly with CD8␣␣. These interactions are characterized by a fairly narrow range of affinities (90 -220 M), lower than typical adhesion molecule interactions (e.g. FIG. 4. Polymorphisms in the ␣3 domain of class I HLA molecules. HLA-A2 residues, which hydrogen-bond to CD8, are shaded blue, those making non-hydrogen bonding contacts of Ͻ4 Å are shaded gray. Residues 223-229, comprising the exposed loop, are shaded yellow. The ␤ strand regions are marked with horizontal bars, and are labeled A-G. Amino acid changes in HLA-E (shaded red) and HLA-G (green) correspond to the colored spheres in Fig. 5. Residues are numbered according to the mature protein sequence. Protein sequences were obtained from the SWISS-PROT and GenBank TM data bases. In HLA-E the conformation of the 223-229 loop differs from the consensus (CD8 binding) conformation (see Fig. 6). A probable cause of this is subtle adjustments in the packing of the 219 and 224 side chains against conserved core residues of the ␣3 domain centered on Tyr 257 . In contrast, the Thr to Val change at 228 in the HLA-G sequence lies directly within the loop (see Fig. 5). However, in HLA-A2 this side chain does not interact directly with CD8. Also, both Thr and Val are of a similar size, and the substitution of a ␥ oxygen for a ␥ carbon atom would appear unlikely to perturb the structure of the ␣3 domain.
CD2-CD48, 30 -50 M at 25°C). The standard CD8␣␣/HLA interaction is also considerably weaker than TCR/peptide-MHC interaction (1-50 M at 25°C). Although this study has been focused upon the affinity of CD8␣␣/HLA interactions, it is highly likely that the extremely fast dissociation kinetics meas-ured for HLA-A*02/CD8␣␣ engagement (10) are typical of CD8␣␣ interactions with other class I MHC molecules. Consequently, when both TCR and CD8␣␣, which may be physically associated, engage the same peptide-MHC complex during antigen recognition, the binding is likely to be dominated by the TCR, which imparts antigen specificity. In comparison, CD8␣␣/ MHC interaction is both weaker, and remarkably short-lived, with CD8␣␣-MHC complexes up to 100-fold less stable than TCR-MHC complexes (10). Furthermore, the low solution affinity of CD8/class I MHC interaction suggests that in general, significant engagement only occurs following recruitment of CD8 into the TCR-CD3 complex, as suggested previously (10).
This study establishes that HLA-A*6801 and HLA-B*4801 are able to bind CD8␣␣, albeit with affinities that are markedly lower (Ն1 mM) than the majority of other classical alleles. This is consistent with cell binding experiments, which have shown that interaction of CD8␣␣ with either allele is insufficient to mediate cell-cell adhesion (12,13). Furthermore, our measurements confirmed the conclusion of these studies that polymorphism at position 245 is responsible for the lower affinities. Our data are also consistent with the finding that killing of target cells by HLA-A*6801-restricted CTL is functionally CD8-independent (26). One question that this observation prompted was how such CTL are positively selected in the thymus in the absence of CD8/HLA-A*6801 engagement. Our measurements suggest a possible explanation, that the relatively lower affinity of HLA-A68 for CD8␣␣, despite being too weak to enhance killing by mature CTL, may be sufficient to affect positive selection of such CD8 ϩ cells in the thymus. One idea, which is currently being investigated, is whether such cells compensate for suboptimal CD8/HLA-A68 interactions by selecting TCRs that bind with a proportionally higher affinity to HLA-A68.
Unlike CTL recognition of HLA-A*6801, alloreactive recognition of HLA-B*4801 by CTL has been shown to be CD8-dependent (13). This is surprising since CD8␣␣ bound each of these alleles with an affinity Ն1 mM. However, because of the difficulties associated with measuring such low affinity interactions, it is possible that CD8␣␣ binds significantly stronger to HLA-B*4801 than to HLA-A*6801. A different explanation (13) of this discrepancy is that alloreactive CTL recognition differs from conventional CTL recognition. Alternatively, it could be that HLA-B*4801-restricted CTL killing is enhanced by CD8␣␤, but not CD8␣␣, and HLA-A*6801-restricted killing enhanced by neither. This might imply that HLA-B*4801 binds preferentially to CD8␣␤, as suggested previously (13), and that HLA-A*6801 binds significantly to neither. If so, this would predict that preferential binding to CD8␣␤ could be mediated by small shifts in the ␣3 domain loop and that the shifts induced by the polymorphism at residue 245 in HLA-B*4801 and HLA-A*6801 are subtly, but significantly, different.
CD8␣␣ Interactions with HLA-E and HLA-G-Complexes of HLA-E bound to the conserved leader peptides of a range of class I HLA molecules (including that of HLA-G) are known to mediate protection from lysis by NK cells, by interacting directly with inhibitory CD94/NKG2A or B receptors on their cell surface (27). Most NK cells do not express CD8 at their cell surface. Consequently, our finding that HLA-E binds CD8␣␣ extremely weakly due to local conformational differences in the ␣3 domain presumably reflects the fact that HLA-E recognition by CD56 ϩ ,CD8 Ϫ NK cells does not require CD8.
HLA-G is mainly expressed in the placenta, but also in a number of other tissues, including the thymus. Its tissue distribution within the placenta is consistent with a role in the interface with the maternal immune system (21). It is expressed at a time when classical class I molecules are largely absent from the placenta. The expression of HLA-G has been shown to play a role in inhibiting lysis of target cells by NK cells derived from placenta (28,29), including clones that bear CD94/NKG2 receptors at their cell surface (30).
Since studies have shown that HLA-G is competent to present a range of peptides of intracellular origin in a manner similar to classical class I molecules (31,32), HLA-G may therefore have the potential to play a role in protecting the fetus from viral infection by presenting viral peptides on infected placental tissue to maternal CTL during a period when classical HLA molecules are not expressed on the placenta. Consistent with this hypothesis is the fact that HLA-G is expressed in thymic epithelial cells and could therefore induce tolerance to the maternal repertoire. Furthermore, since HLA-G has a limited polymorphism, alloreactive maternal Tcell responses directed toward placental HLA-G would be unlikely.
Our results show that HLA-G is able to bind CD8␣␣ with an affinity (150 M), which is in the middle of the range of classical HLA/CD8␣␣ affinities we have measured (90 -220 M). These measurements are consistent with experiments that indicate that HLA-G is able to mediate cell-cell adhesion by interacting with CD8␣␣ (33). Therefore it is possible that HLA-G/CD8 interaction may mediate the positive and negative selection of class I-restricted CTL, and facilitate HLA-G-restricted antigen recognition by such cells, in a manner similar to conventional classical class I-restricted CTL (3,4,34).