A conserved energetic footprint underpins recognition of human leukocyte antigen-E by two distinct αβ T cell receptors

αβ T cell receptors (TCRs) interact with peptides bound to the polymorphic major histocompatibility complex class Ia (MHC-Ia) and class II (MHC-II) molecules as well as the essentially monomorphic MHC class Ib (MHC-Ib) molecules. Although there is a large amount of information on how TCRs engage with MHC-Ia and MHC-II, our understanding of TCR/MHC-Ib interactions is very limited. Infection with cytomegalovirus (CMV) can elicit a CD8+ T cell response restricted by the human MHC-Ib molecule human leukocyte antigen (HLA)-E and specific for an epitope from UL40 (VMAPRTLIL), which is characterized by biased TRBV14 gene usage. Here we describe an HLA-E–restricted CD8+ T cell able to recognize an allotypic variant of the UL40 peptide with a modification at position 8 (P8) of the peptide (VMAPRTLVL) that uses the TRBV9 gene segment. We report the structures of a TRBV9+ TCR in complex with the HLA-E molecule presenting the two peptides. Our data revealed that the TRBV9+ TCR adopts a different docking mode and molecular footprint atop HLA-E when compared with the TRBV14+ TCR–HLA-E ternary complex. Additionally, despite their differing V gene segment usage and different docking mechanisms, mutational analyses showed that the TCRs shared a conserved energetic footprint on the HLA-E molecule, focused around the peptide-binding groove. Hence, we provide new insights into how monomorphic MHC molecules interact with T cells.


␣␤ T cell receptors (TCRs) interact with peptides bound to the polymorphic major histocompatibility complex class Ia (MHC-Ia) and class II (MHC-II) molecules as well as the essentially monomorphic MHC class Ib (MHC-Ib) molecules. Although there is a large amount of information on how TCRs engage with MHC-Ia and MHC-II, our understanding of TCR/ MHC-Ib interactions is very limited. Infection with cytomegalovirus (CMV) can elicit a CD8 ؉ T cell response restricted by the human MHC-Ib molecule human leukocyte antigen (HLA)-E
and specific for an epitope from UL40 (VMAPRTLIL), which is characterized by biased TRBV14 gene usage. Here we describe an HLA-E-restricted CD8 ؉ T cell able to recognize an allotypic variant of the UL40 peptide with a modification at position 8 (P8) of the peptide (VMAPRTLVL) that uses the TRBV9 gene segment. We report the structures of a TRBV9 ؉ TCR in complex with the HLA-E molecule presenting the two peptides. Our data revealed that the TRBV9 ؉ TCR adopts a different docking mode and molecular footprint atop HLA-E when compared with the TRBV14 ؉ TCR-HLA-E ternary complex. Additionally, despite their differing V gene segment usage and different docking mechanisms, mutational analyses showed that the TCRs shared a conserved energetic footprint on the HLA-E molecule, focused around the peptide-binding groove. Hence, we provide new insights into how monomorphic MHC molecules interact with T cells.
Major histocompatibility complex class I (MHC-I) molecules perform critical roles in regulating innate and adaptive immune responses. These molecules have evolved to bind both self-and pathogen-derived peptides and present them for recognition by diverse populations of cells, prominently cytotoxic T cells and natural killer (NK) 8 cells (1). There are two main classes of MHC-I molecules: classical MHC-I (MHC-Ia) molecules that constitute the major ligands for CD8 ϩ or cytotoxic T cells and non-classical MHC-I (MHC-Ib) molecules whose functions are less understood but, like their classical counterparts, are recognized by both the adaptive and innate immune systems (2).
The genes encoding MHC-Ia proteins are highly polymorphic with the polymorphisms being primarily clustered around six pockets located within the antigen-binding cleft. As peptide binding to MHC-Ia proteins is dependent on the accommodation of peptide side chains in these pockets, variation therein creates allotype-specific peptide binding properties (3). In contrast, MHC-Ib genes are far less polymorphic and in some cases, such as human leukocyte antigen (HLA)-E, essentially dimorphic (4). This restricted polymorphism may be associated with the additional distinct functions aside from the presentation of peptides to cytotoxic T cells. Nevertheless, the only identified function of many MHC-Ib molecules is the presentation of peptides for T cell recognition or the regulation of NK cell activity.
T cell recognition of MHC-I proteins and any associated peptide is mediated by the ␣␤ T cell receptor (TCR) via three loops on the ␣and ␤-chains termed complementarity-determining regions (CDRs). However, the majority of the current knowledge of peptide recognition by the TCR is focused on MHC-Ia molecules (3). Nevertheless, there is evidence for distinct roles for T cells restricted by MHC-Ib. For example, in mice, there are numerous MHC-Ib loci, and the MHC-Ib molecules Qa-1 b , Q9, and H2-M3 have been implicated in responses against Salmonella enterica serovar Typhimurium (5-7), polyoma virus (8) ,and Listeria monocytogenes (9), respectively. Similarly, populations of regulatory cells have been shown to recognize Qa-1 b (10). However, it is unclear whether these T cells recognize their MHC-Ib ligands in an analogous manner to MHC-Ia and whether they elicit qualitatively distinct responses to pathogens. There is evidence that H2-M3-restricted T cells are functionally distinct from those restricted by MHC-Ia both in terms of thymic phenotype and their requirements for activation. For example, H2-M3-restricted T cells appear to be important in the primary immune response following Listeria infection but are not expanded upon rechallenge. Furthermore, unlike MHC-Ia-restricted T cells, H2-M3-restricted T cells can be selected on cells of hematopoietic origin (11).
There are far fewer MHC-Ib molecules in humans compared with mice (HLA-E, -F, and -G). Of these, the function of HLA-E is perhaps best understood. It is the least polymorphic of all HLA genes with 17 alleles described to date but with only two alleles dominating, each being present at frequencies of nearly 0.5 (4). Crystal structures of HLA-E have revealed a highly restrictive peptide-binding grove, ideal for accommodating nonamer peptides derived from the leader sequence of other HLA-I proteins (12)(13)(14). Indeed, the primary function of HLA-E is to present these peptides to CD94-NKG2 receptors, thus regulating the activity of NK cells. However, HLA-E can also present pathogen-derived peptides from S. enterica Typhimurium (6), Mycobacterium tuberculosis (15), and human CMV (16,17) to CD8 ϩ T cells. Moreover, infection of macaques with engineered CMV vectors that expressed HIV type 1 proteins results in HIV-specific, Mamu-E-restricted CD8 ϩ T cell responses (18).
Interestingly, a CMV epitope from the protein UL40 shares an identical amino acid sequence with that found in many but not all HLA-C-derived leader peptides (VMAPRTLIL, or "LIL") and consequently when bound to HLA-E can also promote interactions with CD94-NKG2 receptors (19). In most individuals, the complete sequence identity between these HLA-C-and UL40-encoded peptides likely results in deletional tolerance (20). However, in individuals who lack HLA-C alleles that encode the LIL determinant, robust UL40-specific, CD8 ϩ T cell responses have been observed (17,21). This UL40specific T cell response to the LIL epitope appears restricted with the TRBV14 gene segment utilized in a number of T cell clones isolated from several unrelated donors (17,22).
The structure of a TRBV14 ϩ TCR, called KK50.4, was solved in complex with HLA-E-LIL and showed that the interaction was broadly similar to TCR recognition of MHC-Ia (22). There were, however, a number of distinct features of the TCR/ HLA-E interaction. In particular, the CDR2␤ loop dominated the contacts made with HLA-E. Furthermore, each of the CDR loops of TCR␤ made direct contacts with the Ile at position 8 of the peptide (the only peptide residue that differed from selfencoded peptides). To date, the investigation of T cell responses to UL40 have been limited to the prototypical TRBV14 ϩ TCRs that recognize the LIL epitope. However, it remains unclear whether such features are general to TCR recognition of HLA-E-UL40 peptide complexes or to any HLA-E-restricted TCR.
Previous data suggested that the fine specificity of the UL40 T cell response was modified by the presence or absence of HLA-A alleles in the donor that encoded the sequence VMAPRTLVL ("LVL") (17). Notably, in individuals lacking LVL as a self-peptide (e.g. donors lacking HLA-A2-type leader peptides), the UL40 T cell response cross-reacted with the LVL peptide. Furthermore, T cells expressing TRBV2 (V␤22) or TRBV9 (V␤1), rather than the typical TRBV14, characterized this response. Here, we cloned and expressed a TRBV9 ϩ TCR isolated from such a donor and report two structures of this TCR bound to HLA-E (presenting the LIL and LVL peptides). Using mutagenesis, we determined the energetic contribution of individual HLA-E residues interacting with both the TRBV9 ϩ and TRBV14 ϩ TCRs. Our data suggest that although different TCRs adopt distinct docking modes on HLA-E the energetic basis of the TCR interaction is defined by a set of conserved HLA-E residues.

The GF4 TCR, a non-canonical, TRBV9 ؉ , HLA-E-restricted TCR
Previous studies of a canonical TRVB14 ϩ TCR, KK50.4, typical of several HLA-E-restricted UL40-specific T cell clones expanded in vitro from unrelated donors (17,22) demonstrated it to be highly specific for the peptide LIL (17). In contrast, UL40-specific cells isolated from donor GF did not utilize TRBV14 and recognized target cells pulsed with both LIL and LVL peptides with high avidity (17). To assess whether the molecular recognition of HLA-E was conserved between TRBV14 ϩ -and non-TRBV14 -expressing T cells, we sequenced and cloned the TCR from a T cell clone, GF4, obtained from this donor. The GF4 TCR utilized the TRAV35 and TRBV9 gene segments, and sequence analysis showed the GF4 and KK50.4 TCR CDR3 regions have few elements in common (Table 1). Moreover, the GF4 TCR possessed a longer CDR3␤ loop than the KK50.4 TCR (15 versus 12 amino acids) and ␣␤ TCR recognition of HLA-E lacked the conserved Arg 110␤ characteristic of TRBV14 ϩ UL40 -specific T cell clones (22).
We then determined the affinity of the GF4 TCR for the HLA-E bound to the LIL and LVL peptides by surface plasmon resonance (SPR). GF4 TCR bound to HLA-E-LIL with an affinity of Ϸ37 M (Fig. 1A) and surprisingly exhibited a 10-fold higher affinity for HLA-E-LVL with a K D of Ϸ3 M (Fig. 1B). The higher affinity of the GF4 TCR for the LVL peptide compared with LIL correlated with a slower dissociation rate ( Fig.  1). As we had previously determined that the KK50.4 TCR was highly specific for the presence of Ile at P8 (22), the sequence differences between the GF4 and KK50.4 TCRs impacted their peptide specificity.

GF4 TCR binds to HLA-E molecule with an orthogonal docking angle
To define the molecular basis of the interaction between HLA-E and the GF4 TCR, we determined the structures of the GF4 TCR in complex with HLA-E presenting both LIL and LVL peptides ( Table 2). The structures of the two GF4 TCR-HLA-E-peptide complexes were similar and aligned with an overall root mean square deviation of 0.49 Å (Fig. 2, A and B). Here, the GF4 TCR docked centrally on top of the HLA-E cleft with a docking angle of 88°whereupon the TCR ␤-chain and TCR ␣-chain were positioned above the ␣1 and ␣2 helices, respectively ( Fig. 2, C and D). The buried surface area (BSA) upon complexation was ϳ2,100 Å 2 , which falls within the range of previously observed TCR-peptide-MHC-I structures (3).
As the GF4 complexes with both peptides were very similar, we analyze below the GF4 TCR-HLA-E-LVL complex determined at higher resolution ( Table 2). The GF4 TCR contacted two large stretches of the ␣1 and ␣2 helices from residues 65-80 and 145-158 (Table 3 and Fig. 2C). The interaction with the HLA-E ␣1 helix was largely mediated by the GF4 TCR ␤-chain (Table 3). Here, all three CDR loops along with Arg 66␤ , a framework residue, made direct contacts. The CDR1␤ and

Table 2 Data collection and refinement statistics
Values in parentheses are for the highest-resolution shell. r.m.s.d., root mean square deviation.

Data collection statistics
Resolution ␣␤ TCR recognition of HLA-E CDR3␤ loops formed hydrophobic interactions with the HLA-E heavy chain ( Fig. 3A) with Leu 37␤ interacting with Val 76 of HLA-E and Pro 110␤ binding Arg 79 and Thr 80 of the HLA-E heavy chain. The GF4 TCR also made extensive use of CDR2␤ loop with four of the six residues ( 56 YYNGEE 65 ) alongside Arg 66␤ , an adjacent FW residue, all being involved in contacting HLA-E. This CDR2-FW␤ segment abuts an 11-residue-long stretch of HLA-E (residues 65-76), engaging directly with seven of these residues (Table 3). This region of the HLA-E ␣1 helix is highly charged with four Arg and one Asp residue interacting with the two Glu residues and one Arg of the CDR2-FW␤ segment, thereby forming an extensive salt-bridging network (Table 3 and Fig. 3B). A large network of hydrogen bonds and hydrophobic contacts further strengthened the GF4/ HLA-E interaction (Table 3). Here, the side chain of Tyr 57␤ of the CDR2␤ loop was lodged between the Val 76 and Ile 73 of HLA-E, forming a peg/notch interaction (Fig. 3C). Similarly, the side chain of the FW residue Arg 66␤ inserted its guanidinium group between the side chains of Asp 69 /Gln 72 and the CDR3␣ loop (Fig. 3D). Altogether, this highlights the important contribution of the GF4 TCR ␤-chain in interacting with the HLA-E molecule (Fig. 2C). TCR recognition of the HLA-E ␣2 helix was mainly mediated by the TCR ␣-chain (Table 3). All three CDR␣ loops along with a framework residue, Arg 84␣ , contacted the ␣2 helix ( Table 3). The GF4 TCR ␣-chain contacted the center of the ␣2 helix either side of the hinge region spanning residues 150 -155, highlighting the central docking mode of the TCR onto the HLA-E molecule (Fig. 2C). The CDR1␣ formed hydrophobic interactions with Ala 150 , Glu 154 , and His 155 as well as being within hydrogen-bonding distance to the main chain of Ser 150 (Fig. 3E). Most of the CDR2␣ contacts were mediated via the Tyr 57␣ whose aromatic side chain lay flat on the hinge of the ␣2 helix, whereas the FW␣ residue Arg 84␣ formed a salt bridge with Glu 154 (Fig. 3E). The interactions between the CDR3␣ loop and HLA-E were focused around the His 155 residue. The 107 QPLGG 111 residues of the CDR3␣ loop surrounded the His 155 side chain, which was further caged by the Asn 37␣ and the P5 Arg from the peptide. The "cage" surrounding His 155 is closed by interaction between Gln 107␣ and Glu 152 of HLA-E, which fully buried the ␣2 helix residue upon binding of the GF4 TCR (Fig. 3F). Hence, our structural analysis showed that the GF4 TCR used all six CDR loops as well as FW residues from both chains to engage the two helices of the HLA-E in an orthogonal docking mode.

The GF4 TCR interaction with the LVL and LIL peptides is driven by the CDR3␣ loop
The GF4 TCR bound to a large stretch of the peptide, ranging from P4 Pro to P9 Leu ( Table 3). The majority of the contacts were made by the CDR3␣ loop (87% BSA), whereas the CDR3␤ loop made a smaller contribution (13% BSA) ( Table 3). The CDR3␤ formed a salt bridge with the P5 Arg via Glu 116␤ , whereas Pro 110␤ and Asn 115␤ both contacted the P8 Val residue

␣␤ TCR recognition of HLA-E
in the LVL peptide (Fig. 4A) and formed similar contacts with the P8 Ile with the LIL peptide (Table 3). In addition, Pro 110␤ made contacts with P9 Leu ( Table 3). The CDR3␣ loop made an extensive series of contacts whereupon P5 Arg of the peptide was encompassed by the 110 GGSNYK 115 region of the CDR3␣ loop and further encaged at the top by Gln 107␣ that was within hydrogen-bonding distance with P5 Arg. The CDR3␣ loop was positioned above the P4 Pro and the backbone of the P6 Thr and P7 Leu. Moreover, Tyr 114␣ contacted the side chain of the P8 Val of the LVL peptide, which was also contacted by Asn 113␣ (Fig. 4B). As such, the P8 Val was fully buried upon binding of the GF4 TCR. Analysis of GF4 bound to the HLA-E-LIL complex showed a similar set of contacts, although the larger P8 Ile created a ripple of small structural adjustments within the CDR3 loops (Fig. 4C) to accommodate the additional methyl group, which might be sufficiently less favorable for binding to the GF4 TCR and account for the affinity differences (Fig. 1). Hence, the GF4 TCR engaged the peptide with a large footprint driven by the CDR3␣ loop.

Structural comparison of KK50.4 TCR and GF4 TCR in complex with HLA-E
Despite the GF4 and KK50.4 TCRs (22) engaging with the same monomorphic HLA-E molecule, the differing V gene segment usage (Table 1) between the TCRs was associated with differences in the molecular architecture of the TCR/HLA-E interaction (Fig. 4D). Notably, the KK50.4 TCR is heavily reliant (60% of contacts) on the TCR ␤-chain to engage with the HLA-E-LIL (Fig. 4E) (22), whereas the GF4 TCR uses both chains to a similar extent (Fig. 4F). The BSA is larger in the complex with the GF4 TCR (2,100 Å 2 ) compared with the KK50.4 TCR (1,800 Å 2 ). Moreover, the docking angles are markedly different with a diagonal docking for the KK50.4 TCR (56°) (Fig. 4E) and an orthogonal docking for the GF4 TCR (88°) (Figs. 2, C and D, and  4F). Although the center of mass of the ␤-chains from each TCR aligned well (Fig. 4, E and F), the ␣-chain of the GF4 TCR made a ϳ30°shift toward the C-terminal end of the antigenbinding cleft compared with the KK50.4 TCR. In addition, although both the ␣and ␤-chains of KK50.4 TCR contacted the LIL peptide and the ␣2 helix of the HLA-E (Fig. 4E), the GF4 TCR largely used the TCR ␣-chain to interact with these parts of the HLA-E-peptide complex (Fig. 4F).
The GF4 and KK50.4 TCRs contacted 22 and 19 HLA-E residues, respectively, 15 of which were shared, albeit interacting with different TCR residues (Table 3). Here, the CDR2␤ loops were shifted by ϳ5 Å between the two TCRs. Despite this docking difference, the side chains of two residues (57 and 66) from the CDR2␤-FW␤ segment made similar contacts with the HLA-E molecule; Val 57␤ and Tyr 57␤ of KK50.4 and GF4 TCRs, respectively, lodged their side chains between Ile 73 and Val 76 of the HLA-E molecule (Fig. 5C), whereas the FW␤ residue at position 66 (Arg and Gln from GF4 and KK50.4 TCRs, respectively), pointed its side chain toward Asp 69 , Gln 72 , and Ile 73 (Fig. 5C).
Although the CDR3␣ loops of GF4 and KK50.4 TCRs were located similarly above P4 Pro and surrounded the P5 Arg side chain (Fig. 5A), the GF4 TCR CDR3␣ loop (BSA of ϳ30%) contributed double the BSA of the KK50.4 TCR CDR3␣ loop (BSA of 14%) (22). This higher contribution of the GF4 TCR CDR3␣ ␣␤ TCR recognition of HLA-E loop was associated with Tyr 114␣ sitting above the P8 Val residue of the LVL peptide (Fig. 5B). However, the area surrounding the P8 residue of the peptide formed by the CDR␤ loops of KK50.4 was larger than that in the GF4 TCR and therefore might accommodate a P8 Ile more favorably than a P8 Val (Fig.  5B). The comparison of the GF4 and KK50.4 structures in complex with HLA-E-peptide showed that despite disparate TCR gene usage and differing docking strategies a similar set of HLA-E and peptide residues formed the TCR recognition site.

Shared energetic footprint between KK50.4 and GF4 TCRs onto the monomorphic HLA-E molecule
Although both the GF4 and KK50.4 TCRs interacted with a similar set of residues on HLA-E, the relative energetic contributions of these HLA-E residues toward TCR recognition was unclear. Consequently, based on the structures of the TCR-HLA-E-peptide complexes, we mutated the HLA-E residues engaged by both TCRs (Table 3) (with the exception of Lys 146 that is also critical for peptide binding) (23). Each residue was substituted to an alanine with the exception of Ala 150 and Ala 158 , which were mutated to glycine. In addition to the 14 HLA-E residues mutated, we also mutated a residue outside of the binding cleft, Thr 216 , as a negative control. We tested the affinity, as determined by SPR, of these 15 HLA-E mutants for both the GF4 and KK50.4 TCRs in the presence of their highaffinity peptides LVL and LIL, respectively (Fig. 5, D and E). The impact of each mutation was classified as follows: no effect if the affinity was decreased less than 3-fold, a moderate effect if the affinity was decreased by 3-5-fold, or a dramatic impact if the affinity was decreased by more than 5-fold compare with the wild-type HLA-E. First, the T216A mutant, used as negative control, did not impact the affinity of either the GF4 or KK50.4 TCRs. The affinity between wild-type HLA-E and KK50.4 was ϳ25 M (mean of three independent experiments, 26.7 Ϯ 1.3 M; data not shown), similar to results reported previously (22).
Of the 15 substitutions tested, seven (R65A, Q72A, R75A, R79A, A150G, E154A, and R157A) had no significant effect on the affinity of interaction with the KK50.4 TCR, whereas mutation of seven residues (D69A, I73A, V76A, T80A, E152A, H155A, and A158G) had a marked effect upon recognition by KK50.4 TCR (Fig. 5D). Similarly, seven of these substitutions (R65A, D69A, R75A, R79A, A150G, E154A, and R157A) had little impact on the GF4 TCR binding with one having a moderate impact (A158G) and six (Q72A, I73A, V76A, T80A, E152A, and H155A) dramatically impacting the affinity of GF4 TCR for the HLA-E-LVL complex (Fig. 5E). Noticeably, 11 of ␣␤ TCR recognition of HLA-E the 15 substitutions on the HLA-E molecule impacted the affinity in the same fashion despite the difference in the interactions with the respective TCR-HLA-E interfaces (Fig. 4, F and E). Of the substitutions that equally impacted TCR affinity, five decreased the affinity by more than 5-fold, two on the ␣2 helix (E152A and H155A) and three on the ␣1 helix (I73A, V76A, and T80A) (Fig. 5, D and E), thereby representing a common "hot spot" for recognition by the two TCRs with divergent gene usage.

Discussion
Of the MHC-Ib molecules found in humans, only HLA-E has been shown to bind to the ␣␤ TCR expressed by CD8 T cells. Indeed there has been significant interest in HLA-E-restricted T cells as a result of CMV vector-based vaccination strategies that elicit protective Mamu-E-restricted T cell responses specific for pathogens such as HIV in macaques (18). However, because there is only one crystal structure of HLA-E in complex with a TCR, our understanding of TCR recognition of HLA-E and MHC-Ib molecules more broadly remains extremely limited. Here, we describe two additional TCR-HLA-E-peptide structures and have assessed the energetic contribution of individual HLA-E residues to TCR binding. Our findings show that different TCR gene usage results in distinct TCR docking mechanisms onto HLA-E.
We had previously observed that the affinity of KK50.4 TCR for HLA-E was relatively low compared with many other MHC-Ia-virus-specific TCRs (22). We proposed that this in part may have been due to the near sequence identity between the cognate CMV-derived peptide-ligand-and self HLA-I-leader sequence-encoded peptides expressed in the donor thymus that differed by only a single methyl group (VMAPRTLIL versus VMAPRTLVL). Donor GF was somewhat distinct, lacking HLA-I alleles that encode both VMAPRTLIL and VMAPRTLVL. We hypothesized that in such donors there is greater scope to select TCRs with specificity for CMV-encoded peptides. This would potentially increase the repertoire of TCR and the likelihood of TCR with higher affinity for antigen being selected as a result of more relaxed negative selection. Consistent with this, the GF4 TCR bound the HLA-E-LVL complex with significantly higher affinity than the KK50.4 bound to the HLA-E-LIL complex.
A number of studies have shown polymorphisms in this region of the UL40 gene. We reported that only half of the UL40 sequences isolated from a cohort of hematopoietic stem cell transplant recipients possessed the canonical VMAPRTLIL epitope, and the variants VMAPRTLLL and VMAPRTLVL were found in 16 and 12% of patients, respectively (15). Indeed, although the vast majority of research on UL40-specific T cells has focused on those that recognize the canonical VMAPRTLIL peptide, the polymorphism in this region of UL40 suggests that the capacity to generate HLA-E-restricted UL40-specific T cells may not be strictly limited to individuals with specific HLA-C types. The ability of individuals to produce such cells will depend not only on the HLA-Ia alleles present that likely shape the HLA-E-restricted, UL40-specific repertoire but also on the UL40 sequence present within the CMV strain.
The different TRAV and TRBV usage as well as the altered docking orientations between KK50.4 and GF4 indicates that HLA-E is capable of selecting T cells that express distinctly different TCRs into the immune repertoire. Interestingly, despite the sequence and structural differences, these TCRs utilized a similar energetic hot spot of recognition on HLA-E. This is in contrast to previous finding of TCR recognition of MHC-Ia where different structural footprints can result in altered energetic footprints (23). Significantly, here we have demonstrated that approximately half of the HLA-E contact residues are critical for binding to both the KK50.4 and GF4 TCRs. This, in general, is in contrast with the smaller energetic footprints that can underpin TCR-MHC-I recognition (24,25). Nevertheless, previous studies have identified position 155 in the MHC-Ia heavy chain as well as the MHC-II-equivalent residue 70␤ as often (but not always) representing a critical position for TCR recognition of MHC (26 -29). The loss of KK50.4 and GF4 recognition of HLA-E caused by the H155A mutation is consistent with the importance of this position.
HLA-E is more conserved in evolution than classical HLA-I with the gene arising prior to primate evolution, presumably becoming fixed as a consequence of acting as a ligand for CD94-NKG2 receptors (30,31). We have previously assessed the ability of mutant HLA-E molecules to interact with the inhibitory NK cell receptor CD94-NKG2A and shown that this innate recognition of HLA-E was also characterized by a broad energetic footprint with seven of 11 of the contact residues contributing to the binding energy (32,33). Given that a large number of HLA-E residues are critical to the interaction with both NK cell receptors and TCRs, it is possible that the energetic principles governing immune recognition of MHC-Ib may be more stringent. In any case, our data indicate that the rules governing TCR recognition of MHC-Ia molecules may not always be applicable to the less polymorphic MHC-Ib molecules.

Generation of recombinant HLA-E
cDNA encoding the extracellular domain of wild-type HLA-E*0103 and human ␤ 2 -microglobulin were cloned into the pET-30 expression vector. The proteins were refolded with the synthetic peptide VMAPRTLIL (LIL) or VMAPRTLVL (LVL) (GenScript) as described (22). The resulting complexes were purified by anion-exchange and gel-filtration chromatography. Mutations were introduced into the HLA-E heavy chain by the QuikChange site-directed mutagenesis kit (Stratagene La, Jolla, CA), and the sequence of the mutated HLA-E cDNA was verified by DNA sequencing. HLA-E mutants were expressed in Escherichia coli with yields similar to the wild-type HLA-E and an appearance identical to the wild-type HLA-E heavy chain when inclusion bodies were separated by SDS-PAGE (data not shown). Moreover, mutants eluted from size-exclusion columns at the same volume as wild-type HLA-E and were each recognized by the pan-HLA class I mAb w6/32 and an antihuman ␤ 2 -microglobulin antibody in a capture ELISA (data not shown) (34).

␣␤ TCR recognition of HLA-E Generation of recombinant TCR
Soluble, recombinant forms of the UL40-specific TCRs KK50.4 and GF4 were generated essentially as described previously (22). Briefly, the ␣and ␤-chains of the TCRs were cloned separately into the pET-30 expression vector, expressed and purified from inclusion bodies, and then refolded by dilution in a buffer containing 5 M urea. The resulting ␣␤ TCR complex was purified by anion-exchange, gel-filtration, and Mono Q chromatography.

Surface plasmon resonance
SPR experiments were performed as described previously (22). Experiments were conducted at 25°C either on a BIAcore 3000 or Bio-Rad ProteOn instrument at a flow rate of 20 (BIAcore) or 30 l/min (ProteOn). Both the running and sample buffers were 10 mM HEPES (pH 7.4), 150 mM NaCl, and 0.05% Tween 20 plus 1% bovine serum albumin to inhibit nonspecific binding. The machines yielded consistent binding affinities across the different platforms. The antibody 12H8 that is specific for the constant region of TCR ␣␤ (25) was coupled to all flow cells of a CM5 (BIAcore) or GLC (ProteOn) chip. For each experiment, two different preparations of either KK50.4 or GF4 TCR were passed over two different flow cells, and ϳ300 -500 response units of the TCR was captured by the antibody. The other flow cells served as control cells for the experiments. Wild-type HLA-E or HLA-E mutants were injected over all flow cells at a concentration range of 1.8 -100 M. The antibody surface was regenerated between each analyte injection with ActiSep (Sterogene) or glycine HCl (pH 3) (Bio-Rad). All measurements were minimally done in duplicate. BIAevaluation version 3.1 or ProteOn Manger version 2.1 was used for data analysis, and data were fitted using the 1:1 Langmuir binding model.

Crystallization, data collection, and structure determination
Crystals of the GF4 TCR in complex with HLA-E-LVL (10 mg/ml) or with HLA-E-LIL (10 mg/ml) in 10 mM Tris-HCl (pH 8) and 150 mM NaCl buffer were grown at 20°C by the hangingdrop, vapor-diffusion method with a protein/reservoir drop ratio of 1:1. Crystals of GF4 TCR-HLA-E-VMAPRTLVL and GF4 TCR-HLA-E-VMAPRTLIL were obtained in 0.2 M potassium sodium tartrate, 0.1 M Bistris propane (pH 8.0), and 20% PEG3350 (w/v). Crystals were soaked in a cryoprotectant solution containing mother liquor solution supplemented with 20% PEG3350 (w/v) and flash frozen in liquid nitrogen. All data sets were collected on the MX2 beamlines at the Australian Synchrotron, Clayton, Australia using the Area Detector Systems Corp. Quantum 315r charge-couple device detectors (at 100 K). Data sets were processed with MOSFLM software (35) and scaled using SCALA software (International Standard Serial Number 0907-4449) from the CCP4 suite (36). Both GF4 TCR-HLA-E-LVL and GF4 TCR-HLA-E-LIL structures were determined by molecular replacement using the PHASER program with the LC13 TCR (Protein Data Bank code 1KGC (37)) and unliganded HLA-E-VMAPRTLIL (Protein Data Bank code 2ESV (22)) as search models.
Manual model building was conducted using Coot software (38) followed by maximum-likelihood refinement with BUSTER (39). The GF4 TCR was numbered according to the IMGT unique numbering system (40). The final models were validated using the Protein Data Base validation web site, and the final refinement statistics are presented in Table 1. Coordinates for GF4 TCR-HLA-E-LVL (code 5W1W) and GF4 TCR-HLA-E-LIL (code 5W1V) were submitted to the Protein Data Bank. All molecular graphics representations were created using PyMOL (41).
Author contributions-L. C. S. cloned TCRs and mutant HLA-E, made recombinant proteins, conducted the SPR and mutagenesis experiments, analyzed the results, and wrote the manuscript. N. G. W. made recombinant proteins and conducted data collection, structural refinement, and analysis. C. F. and S. G. conducted structural refinement and structural analysis and drafted the paper. G. P.