CD8αβ Has Two Distinct Binding Modes of Interaction with Peptide-Major Histocompatibility Complex Class I*

Interaction of CD8 (CD8αα or CD8αβ) with the peptide-major histocompatibility complex (MHC) class I (pMHCI) is critical for the development and function of cytolytic T cells. Although the crystal structure of CD8αα·pMHCI complex revealed that two symmetric CD8α subunits interact with pMHCI asymmetrically, with one subunit engaged in more extensive interaction than the other, the details of the interaction between the CD8αβ heterodimer and pMHCI remained unknown. The Ig-like domains of mouse CD8αβ and CD8αα are similar in the size, shape, and surface electrostatic potential of their pMHCI-binding regions, suggesting that their interactions with pMHCI could be very similar. Indeed, we found that the CD8α variants CD8αR8A and CD8αE27A, which were functionally inactive as homodimers, could form an active co-receptor with wild-type (WT) CD8β as a CD8αR8Aβ or CD8αE27Aβ heterodimer. We also identified CD8β variants that could form active receptors with WT CD8α but not with CD8αR8A. This observation is consistent with the notion that the CD8β subunit may replace either CD8α subunit in CD8αα·pMHCI complex. In addition, we showed that both anti-CD8α and anti-CD8β antibodies were unable to completely block the co-receptor activity of WT CD8αβ. We propose that CD8αβ binds to pMHCI in at least two distinguishable orientations.

Both the extracellular domain and the cytoplasmic tail of the CD8␤ subunit have been implicated in providing increased efficiency of CD8␣␤ (12)(13)(14)(15)(16). We reported previously that the extracellular domain of the CD8␤ subunit is critical for this enhanced efficiency (14) and that introduction of the CD8␤ stalk region is sufficient to confer a CD8␣␤-like co-receptor efficiency to the CD8␣␣ homodimer (17). In addition, the sialylation of the O-linked glycans in the CD8␤ stalk region is differentiation stage-dependent and may modulate the intrinsic activity of CD8␣␤ during the transition from double-positive (CD4 ϩ CD8 ϩ ) to single-positive (CD8 ϩ ) T cells (18,19). It was also reported that palmitoylation of a membrane-proximal cysteine residue in the cytoplasmic tail of CD8␤ during T-cell activation facilitates partition of CD8␣␤ heterodimers into lipid rafts, where it associates with the CD3 component of TCR complexes (20). Lastly, the kinase activity of p56 lck , which is associated only with the cytoplasmic domain of CD8␣, can be enhanced by CD8␤ (21). It is conceivable that the extracellular domain and the cytoplasmic tail of CD8␤ may independently contribute to the enhanced co-receptor activity of CD8␣␤.
The interaction of CD8␣␣ and pMHCI has been elucidated through x-ray analyses of the crystal structures of human CD8␣␣⅐HLA-A2 and mouse CD8␣␣⅐H2-K b (22,23). This structural information confirms the results of many mutational studies. The complementarity-determining region-like loops of CD8␣␣ and the MHCI ␣3 domain CD loop (residues 220 -228) are the most critical for CD8␣␣⅐pMHCI interaction (24 -26). In both human and mouse CD8␣␣⅐pMHCI complexes, two symmetric CD8␣ subunits clamp asymmetrically onto the relatively rigid MHCI ␣3 domain CD loop. One subunit, CD8␣1, is involved in more than 70% of the total interaction surface between CD8␣␣ and the MHCI ␣3 domain and ␤ 2 -microglobulin (␤ 2 M). The second subunit, CD8␣2, is less engaged and interacts only with the MHCI ␣3 domain (22,23). To simplify reference to these interaction sites, we designated the site occupied by the CD8␣1 subunit of CD8␣␣⅐pMHCI as site 1 and the site occupied by the CD8␣2 subunit as site 2.
Unlike CD8␣␣⅐pMHCI, the structure of CD8␣␤⅐pMHCI is not known. Two different binding models have been proposed for the interaction of pMHCI with mouse and human CD8␣␤ (22,23). Considering that the stalk region of CD8␣ is longer than that of CD8␤ (44 versus 35 amino acids), which allows the Ig-like domain of CD8␣ to extend farther from the T-cell surface, we proposed a binding mode for mouse CD8␣␤⅐pMHCI interaction in which CD8␤ and CD8␣ occupy site 1 and site 2, respectively (23). Consistent with this model is our observation that co-expression of WT CD8␤ with the CD8␣ R8A variant resulted in co-receptor activity indistinguishable from that of WT CD8␣␤ (17). Because mouse CD8␣ R8A lacks the side chain of Arg 8 critical for engaging in extensive interaction with pMHCI (site 1 interaction), the functional CD8␣ R8A ␤ heterodimer must bind to pMHCI with its ␤ subunit occupying site 1 and its ␣ subunit (CD8␣ R8A ) occupying site 2. On the other hand, based on electrostatic interactions between modeled human CD8␣␤ and pMHCI, Gao et al. (22) proposed that in the human CD8␣␤⅐pMHCI complex, the CD8␣ subunit occupies site 1, and the CD8␤ subunit occupies site 2. Interestingly, like mouse CD8␣ R8A , human CD8␣ R4K and CD8␣ L25A are not able to form functional homodimeric co-receptor (34). Yet, unlike the mouse CD8␣ R8A variant, neither of these human CD8␣ variants is able to form functional heterodimeric co-receptor with WT CD8␤ (34). This observation led to the conclusion that in the human CD8␣␤⅐pMHCI complex, the CD8␣ subunit occupies only site 1 (34). Thus, the interactions between CD8␣␤ and pMHCI in mouse and human systems appear to be different. The molecular basis for this difference remains unclear.
To further understand the interaction between CD8␣␤ and pMHCI and possibly resolve the apparently conflicting observations in human and mouse systems, we asked, in the context of CD8␣␤⅐pMHCI complex, whether mouse CD8␤ could also occupy site 2. Among a panel of CD8␤ variants that were unable to occupy site 1, we identified two CD8␤ variants that were capable of forming functional co-receptor with WT CD8␣. Therefore, the mouse CD8␤ subunit can also occupy site 2 in the CD8␣␤⅐pMHCI complex. In addition, we showed that an anti-CD8␣ antibody only blocked activity of co-receptors in which CD8␣ was occupying site 1 but not site 2. These results support the notion that CD8␣␤ interacts with pMHCI in at least two distinguishable binding orientations.

EXPERIMENTAL PROCEDURES
mAbs and Flow Cytometry-All of the mAbs used in the indirect flow cytometric analyses and mAb inhibition assays are described previously (17,23).

Structural Basis for CD8␤ Occupancy of Both Site 1 and Site 2-
The overall protein architecture of CD8␣␤ is similar to that of CD8␣␣ (22,23) and resembles the Ig-variable domain fragment of an antibody (Fig. 1A). Despite the low degree of protein sequence identity between CD8␣ and CD8␤ (Fig. 1C), superimposing their structures reveals only limited differences in the loop regions, including the CDR1-and CDR2-like loops (Fig.  1B). The CDR-like loops are involved in ligand binding and are expected to be more variable than other regions. However, the nine residues that are identical in human and mouse CD8␣ and CD8␤ are important for maintaining an Ig-fold structure and dimeric interface (7). This conservation ensures similarity in the overall structure of ␣ and ␤ subunits (Fig. 1B). The apparent structural similarity between CD8␣ and CD8␤ raises the possibility that CD8␣␤⅐pMHCI interactions may be very similar to those of CD8␣␣⅐pMHCI, with the ␤ subunit replacing either one of the ␣ subunits in CD8␣␣⅐pMHCI complex.
CD8␤ Subunit of Heterodimeric CD8␣ E27A ␤ Occupies Site 1-We recently showed that CD8␣ R8A ␤ is functionally active (17,28). This suggests that Arg 8 residue of the CD8␣ subunit is dispensable for the co-receptor function of CD8␣ R8A ␤. The Arg 8 is not only critical for the ability of CD8␣ to occupy site 1 but is also the antigenic epitope recognized by mAb H59 (17). Hence, it is possible that mAb H59 blocks the interaction of pMHCI and CD8␣1 but not that of pMHCI and CD8␣2. If so, mAb H59 would be expected to block the co-receptor activity of CD8␣␣ but not that of CD8␣ R8A ␤. However, as the binding epitope of mAb H59 is lost in CD8␣ R8A (17), this variant is not suitable for performing the antibody blocking experiment. Because both Arg 8 and Glu 27 form hydrogen bonds with Lys 58 of ␤ 2 M in the crystal complex of CD8␣␣⅐H-2K b (Fig. 1, D and  E), we engineered a substitute variant, CD8␣ E27A , and tested whether, like CD8␣ R8A ␤, CD8␣ E27A ␤ is functionally active. Fig.  2A shows that despite comparable surface expression levels of CD8 and TCR among the CD8 transfectants, cells expressing CD8␣ E27A ␣ E27A exhibit co-receptor activity about 1000-fold less efficiently than cells expressing WT CD8␣␣ (Fig. 2B). Thus, like CD8␣ R8A ␣ R8A , CD8␣ E27A ␣ E27A is incapable of interacting productively with pMHCI. In addition, the co-receptor activity of CD8␣ E27A ␤ is indistinguishable from that of WT CD8␣␤ (Fig. 2C). Therefore, similar to CD8␣ R8A ␤, CD8␣ E27A ␤ can  productively engage with pMHCI. mAb H59 binds to cells expressing WT CD8␣␤ and CD8␣ E27A ␤ variants ( Fig. 2A) indicating that the binding epitope of mAb H59 is retained. Fig. 2E shows that mAb H59 completely inhibits the co-receptor activity of CD8␣␣, yet it does not affect that of CD8␣ E27A ␤ (Fig. 2D). We also tested the co-receptor activity of CD8␣ E27A ␤ in the presence of an anti-CD8␤ mAb YTS156. In contrast to mAb H59, mAb YTS156 completely blocks the activity of CD8␣ E27A ␤ (Fig. 2D).
The fact that mAb H59 has no effect on the co-receptor activity of CD8␣ E27A ␤ is consistent with the prediction that the interaction between the MHCI ␣3 domain and site 2-occupying CD8␣ E27A is minimal, and the binding of mAb H59 does not compromise the productive interaction between CD8␣ E27A and pMHCI. Conversely, the complete inhibition of the coreceptor activity of CD8␣ E27A ␤ by mAb YTS156 indicates that heterodimeric CD8␣ E27A ␤ is responsible for all observed coreceptor activity. As CD8␣ E27A is no longer capable of occupying site 1, these results indicate that in the CD8␣ E27A ␤⅐pMHCI complex, the CD8␤ subunit occupies site 1. Significantly, the "all-or-none" inhibitory effect of mAb YTS156 and mAb H59 on the co-receptor activity of CD8␣ E27A ␤ suggests that the occupation of site 1 by CD8␤ is the only binding mode in the interaction between CD8␣ E27A ␤ and pMHCI.
CD8␤ Subunit of Heterodimeric CD8␣␤ var Occupies Site 2-Because CD8␣ R8A ␤ and CD8␣ E27A ␤ interacted with pMHCI in only one orientation, the CD8␣ R8A co-transfection system also provided a measure to identify residues of CD8␤ that were crit-ical for CD8␤ being able to occupy site 1. We recently reported that the Lys 55 in the CDR2-like loop and also Ser 101 and Lys 103 in the CD8␤ CDR3-like loop is critical for the coreceptor activity of CD8␣ R8A ␤, as co-expression of each of these CD8␤ variants with CD8␣ R8A does not lead to detectable co-receptor activity (28). In light of the observation that heterodimeric CD8␣ R8A ␤ is functionally active (17), we questioned whether each of these CD8␤ variants can form a functional coreceptor with WT CD8␣. The coreceptor activity of the CD8␣␤ var would indicate that each of these CD8␤ variants could occupy site 2 and form active co-receptors with site 1-occupying WT CD8␣. We found that cell lines expressing similar levels of CD8␣ and CD8␤, CD8␤ K55A , CD8␤ S101A , or CD8␤ K103A proteins on the cell surface (data not shown) exhibited coreceptor activity (Fig. 3A). These results suggest that the heterodimeric CD8␣␤ var can productively interact with pMHCI.
However, WT CD8␣ can form functional CD8␣␣ homodimers in these transfectants (17) and contribute to the observed co-receptor activity (Fig. 3A). We reported previously that the co-receptor activity of WT CD8␣␤ and CD8␣␣ can be triggered by peptide antigen at concentrations of 10 Ϫ9 and 10 Ϫ7 M, respectively (14). Thus, the CD8␣␤ var heterodimers are likely to be functionally active if cells expressing CD8␣ and CD8␤ var are responsive to peptide antigen at a concentration lower than 10 Ϫ7 M. Fig. 3A shows that cells expressing CD8␣ and CD8␤ K55A or CD8␤ K103A is responsive to peptide antigen at a concentration of 10 Ϫ8 M. This result suggests that heterodimeric CD8␣␤ K55A and CD8␣␤ K103A are functionally active. Because cells expressing CD8␣ and CD8␤ S101A exhibited a CD8␣␣-like antigen sensitivity of ϳ10 Ϫ7 M, it appears that CD8␣␤ S101A was not functionally active.
To further verify that CD8␣␤ K55A and CD8␣␤ K103A are indeed functionally active, we also examined their co-receptor activity in the presence of anti-CD8␤ mAb YTS156 or anti-CD8␣ mAb H59. Because mAb YTS156 is ␤ chain-specific, it is expected to block the activity of the CD8␣␤ var heterodimeric co-receptor. Conversely, mAb H59, which is specific for the ␣ chain, would block the co-receptor activity associated with CD8␣␣ homodimers and CD8␣␤ var heterodimers in which CD8␣ subunits occupy site 1. Fig. 3B shows that mAb YTS156 was able to block some co-receptor activity associated with cells transfected with WT CD8␣ and CD8␤ K55A , indicating that CD8␣␤ K55A is indeed a contributor to that activity. Similar results were obtained from the cell line expressing WT CD8␣ and CD8␣␤ K103A (Fig. 3C). Thus, although CD8␤ K55A and CD8␤ K103A are unable to occupy site 1 (7), they are capable of forming functional co-receptors with WT CD8␣ by occupying site 2. Surprisingly, very little, if any, co-receptor activity associated with cells expressing CD8␣ and CD8␤ S101A is inhibited by mAb YTS156 (Fig. 3D), indicating that CD8␣␤ S101A does not contribute to the co-receptor activity shown in A. This is consistent with the observation that these cells exhibit CD8␣␣-like antigen sensitivity (10 Ϫ7 M). In contrast to mAb YTS156, mAb H59 completely blocks the co-receptor activity in all three cell lines (Fig. 3, B-D) suggesting that all functional co-receptors expressed must interact with pMHCI with a CD8␣ subunit occupying site 1 and that the portion of co-receptor activity that cannot be blocked by mAb YTS156 is contributed by CD8␣␣. These results show that the ␤ subunit of heterodimeric CD8␣␤ K55A or CD8␣␤ K103A occupies site 2.
Wild-type CD8␣␤ Heterodimer Has Two Distinct Binding Modes-Using variants of CD8␣ or CD8␤, we are able to restrict the interaction between each of these heterodimeric variants and pMHCI in only one orientation. It would be important to determine whether the WT CD8␣␤ heterodimer indeed interacts with pMHCI in both orientations. We analyzed the inhibitory effect of mAb H59 and mAb YTS156 on the co-receptor activity of WT CD8␣␤. Three possible CD8⅐pMHCI complexes were expected: CD8␣␣⅐ pMHCI, CD8␣␤⅐pMHCI with the ␣ subunit occupying site 1, and CD8␣␤⅐pMHCI with the ␤ subunit occupying site 1 (Fig.  4A). mAb H59 was expected to block the first two complexes but not the third. mAb H59 did not block all of the co-receptor activities (Fig. 4B), indicating that CD8␣␤⅐pMHCI interaction with CD8␣ occupying site 2 contributed at least some co-receptor activity. Because mAb YTS156 inhibited the activity of CD8␣␤ complexed with pMHCI in either orientation (Figs. 2 and 3), the CD8␣␣ homodimers were responsible for the coreceptor activity that was not affected by this antibody. These results are in agreement with our proposal that WT CD8␣␤ can bind to the pMHCI complex in at least two distinct modes, one sensitive and the other insensitive to inhibition by mAb H59.

DISCUSSION
The structure of the CD8␣␣⅐pMHCI complex reveals that the two ␣ subunits of the CD8␣␣ homodimer bind to the peptide-MHCI complex asymmetrically, with one subunit interacting extensively with pMHCI (site 1) and the other subunit interacting only with the ␣3 domain of pMHCI (site 2). Based on this information, we identified residues Arg 8 and Glu 27 as being critical to the ability of CD8␣ to occupy site 1. Using CD8␣ R8A and CD8␣ E27A , we found that CD8␤ could form a functional co-receptor with either variant. Therefore, in the context of CD8␣␤⅐pMHCI interaction, CD8␤ can occupy site 1. We also used the CD8␣ R8A variant to identify three CD8␤ variants incapable of occupying site 1. Two of these three CD8␤ variants could form a functionally active co-receptor with WT CD8␣ indicating that although, CD8␤ var could not occupy site 1, it could still occupy site 2. The concept that CD8␣␤ and pMHCI can form a complex in different orientations is also supported by the observations that anti-CD8␣ mAb H59 completely blocks the co-receptor activity of CD8␣␤ var , yet it does not affect that of CD8␣ var ␤ and only partially blocks that of WT CD8␣␤. Our results support a two-binding-mode model in which heterodimeric CD8␣␤ can interact with pMHCI with two distinct binding orientations, one sensitive to mAb H59 and the other not.
The idea that CD8␣␤ can interact with pMHCI in two orientations prompted questions regarding whether the antigen presentations mediated by these two distinct CD8␣␤⅐pMHCI complexes are qualitatively and/or quantitatively different. Note that the p56 lck kinase is only associated with the cytoplasmic domain of CD8␣ but not with that of CD8␤ (21). Thus, the distance between p56 lck and the chains in a TCR complex that has CD8␣␤ engaged in one orientation is expected to be different from that in another TCR complex that has CD8␣␤ engaged in the opposite orientation. In addition, a specific pMHCIa or pMHCIb, such as H-2K, H-2D, H-2L, or Qa-1, Qa-2 plus H2-M3 (31), might favor the interaction with CD8␣␤ in one orientation over the other. It is also possible that one specific binding mode might be better accommodated because of the particular position at which a specific TCR docks with pMHCI (29 -31). Furthermore, the differentiation stage-related sialylation of the O-glycans associated with the stalk regions of CD8 proteins (18,19) may be compatible with one, but not another, binding orientation. Therefore, a given TCR⅐pMHCI complex may preferentially or exclusively interact with CD8␣␤ in one binding mode but not the other. Such an asymmetric binding mode might facilitate a better engagement between a given TCR complex and the heterodimeric co-receptor that is not possible for its homodimeric counterpart.
Both CD8 and Ly49A NK cell receptors are shown to interact with pMHCI on the same cell (cis interaction) (32,33). Specifically, the inhibitory Ly49A NK cell receptor can bind to its pMHCI ligand, H-2D d , expressed on either the target cells (trans interaction) or the same NK cell (cis interaction) (33). Cis interaction of Ly49A with pMHCI reduces the number of Ly49A available for binding of pMHCI ligand on the target cells. By attenuating the inhibition on NK cells, Ly49A can fine-tune an optimal sensitivity for NK cells to discriminate normal versus abnormal host cells. Similarly, interaction of CD8␣␤ to pMHCI in cis may be compatible with one, but not another, binding mode. Future studies are needed to better understand the potential regulatory role of heterodimeric CD8␣␤ on antigen presentation.
The structural comparison of human and mouse CD8␣ shows that the corresponding residue of mouse Arg 8 is human Arg 4 . Similar to the mouse CD8␣ R8A ␣ R8A , human CD8␣ R4K ␣ R4K homodimer cannot productively interact with pMHCI (34). However, unlike its mouse counterpart, human CD8␣ R4K cannot form a functional co-receptor with WT human CD8␤ (34). Judging from the overall structural similarities between human and mouse CD8 as well as pMHCI proteins, it is difficult to understand the molecular basis for such a discrepancy. However, close examination reveals some differences that may explain this discrepancy. For example, human CD8␤ can be expressed on the cell surface as functionally inactive homodimers CD8␤␤, whereas mouse CD8␤ cannot (27). This may drastically reduce the surface abundance of heterodimeric CD8␣␤ and may therefore explain why no apparent co-receptor activity was detected in the human experimental system. In addition, transient expression cultures, rather than established cell lines, were used for experimentation in the human system. The background noise inherently associated with transient expression systems may compromise the detection of any functional activity of heterodimers in a cell adhesion assay. Thus, it is conceivable that the observed discrepancy was because of differences in experimentation systems and that human CD8␣␤, like mouse CD8␣␤, also interacts with pMHCI with two distinct binding orientations.
Our results may also explain the apparent difficulty in obtaining protein crystals for structure determination of the CD8␣␤⅐pMHCI complex, as WT protein mixtures may contain two types of complexes preventing formation of ordered crystals. In this regard, our CD8 heterodimeric variants capable of only one binding mode may provide an alternative to obtain the protein crystal of CD8␣␤⅐pMHCI complexes and facilitate future efforts to elucidate their structures by crystallographic or other biophysical approaches. By generating transgenic mice expressing one heterodimeric variant, it will be possible to investigate the physiological significance of one binding mode without the complications associated with the second binding mode.