CD94-NKG2A recognition of human leukocyte antigen (HLA)-E bound to an HLA class I leader sequence

To establish how HLA-E, when bound to the leader sequence peptide of HLA-G (residues 3–11, VMAPRTLFL, HLA-E^VMAPRTLFL), was specifically recognized by CD94-NKG2A, we expressed and purified both components, crystallized the complex in space group I4₁32 with unit cell dimensions a = b = c = 345.2 Å, and determined the structure of the CD94-NKG2A-HLA-E^VMAPRTLFL complex to 3.4 Å resolution with an R_factor and R_free of 24.8 and 27.9%, respectively (see Materials and methods and Table I). Although the CD94-NKG2A–HLA-E^VMAPRTLFL structure was determined to moderate resolution, we had previously determined the structures of the nonligated CD94-NKG2A (25) and HLA-E^VMAPRTLFL molecules (26) to 2.5 Å resolution that enabled the structure to be solved readily by molecular replacement, and permitted the ternary complex to be refined well at this resolution. For example, the initial experimental phases clearly showed unbiased electron density for the VMAPRTLFL peptide (Fig. S1 A), confirming the correctness of the molecular replacement solution. In addition, the electron density at the CD94-NKG2A–HLA-E^VMAPRTLFL interface was unambiguous, and side chains at the interface could be modeled into the electron density (Fig. S1, B and C), thus allowing contacts at the interface to be interpreted readily. The crystal contained two virtually indistinguishable ternary complexes within the asymmetric unit (root mean square deviation 0.7 Å). Consequently, structural analyses will be confined to one CD94-NKG2A–HLA-E^VMAPRTLFL complex.

The CD94-NKG2A docked toward the C-terminal end of the HLA-E antigen-binding cleft, binding at an angle of ∼70° (Fig. 1, A and B, and Figs. S2 and 3, which are available) in a manner that permitted CD94-NKG2A to sit across both the α1 and α2 helices of HLA-E (Fig. 1). A comparison of the CD94-NKG2A–HLA-E^VMAPRTLFL complex with the nonligated CD94-NKG2A (25) and HLA-E^VMAPRTLFL structures (26) revealed no significant conformational change in either HLA-E or CD94-NKG2A upon complex formation (root mean square deviation 0.5 and 0.4 Å for CD94-NKG2A and HLA-E, respectively). The one disordered loop of the nonligated CD94-NKG2A heterodimer (residues 199–204 in NKG2A) became ordered in the complex (not depicted), although this observation was attributable to crystal-packing effects, as this loop did not contact HLA-E^VMAPRTLFL. Only one residue, Gln 112 of CD94, was re-orientated upon ligation to maximize the complementarity at the interface. Accordingly, this “lock and key” engagement between HLA-E^VMAPRTLFL and CD94-NKG2A exemplified the “innate characteristic” of this interaction.

The CD94 and NKG2A subunits lay across the antigen-binding cleft, with CD94 and NKG2A almost exclusively interacting with the α1 and α2 helices of HLA-E, respectively. Analysis of the electrostatic surfaces of HLA-E and CD94-NKG2A highlighted a role for charge complementarity at the CD94-NKG2A–HLA-E interface (Fig. 2 A). Namely, a basic region on the α1 helix of HLA-E interacted with an acidic region on CD94 and, conversely, an acidic region on the HLA-E α2 helix docked with a distinct patch of basic charge on NKG2A (Fig. 2 A). The buried surface area (BSA) at the interface was extensive (∼2,100 Ų), with CD94 and NKG2A contributing ∼69 and 31%, respectively (Fig. 2 B). Consistent with the electrostatic complementarity, the CD94-NKG2A–HLA-E interface was characterized by a large number of polar interactions, including 8 salt bridges and 19 H bonds. There was also a small and focused patch of hydrophobic contacts at the interface (Table II). Somewhat surprisingly, the overall shape complementarity (SC) at the interface was relatively poor (SC = 0.63), although the SC for CD94 alone was markedly better (SC = 0.68) compared with NKG2A (SC = 0.31), which further highlighted the greater contribution that CD94 makes to the docking onto HLA-E^VMAPRTLFL compared with NKG2A.

The CD94 footprint on HLA-E was broad, with residues within loops 2, 3, and 5, and β strands 6 and 7 from CD94 interacting with a region spanning residues 65–89 from the α1 helix of HLA-E (Fig. 2 C). Arg65 of HLA-E is situated at the periphery of CD94 subunit and formed van der Waals (VDW) interactions with Arg171. Asp69 of HLA-E appeared to play a central role in the interaction with CD94, as it salt bridged to Arg171 and H bonded to Gln113 (Fig. 2 C). Arg75 formed a salt bridge with Asp163 of CD94, whereas Arg79 of HLA-E H bonded to Ser143 (Fig. 2 C). Glu89 from HLA-E, which is located at the loop C-terminal to the α1 helix, H bonded to Thr146. Gln72 H bonded to Glu164 and Asn170 from CD94, whereas the aliphatic moiety of Gln72 packed against the aromatic ring of Phe114 (Fig. 2 C). Phe114 of CD94 sat in a hydrophobic niche that is formed by Ile73 and Val76 from HLA-E and by P8-Phe from the peptide. Moreover, the neighboring Leu162 of CD94, which formed VDW contacts with Val76 and packed against the aliphatic moiety of Arg75 from HLA-E, extended this hydrophobic patch (Fig. 2 D).

When compared with that of CD94, the footprint of NKG2A on HLA-E was markedly smaller and more focused with loop 3 and β strands 2, 5, and 6 interacting with residues 151–162 of the α2 helix of HLA-E (Fig. 2 E). Nevertheless, analogous to the CD94–HLA-E interactions, the NKG2A–HLA-E contacts were dominated by polar residues. For example, Asp162 from HLA-E formed a salt bridge with Lys217 of NKG2A, whereas Arg137 of NKG2A salt bridged to Glu154 and H bonded to Ser151 of HLA-E. Of the VDW-mediated contacts, the aliphatic moiety of Glu154 interacted with Gln212 of NKG2A. His155 was nestled between Pro171 and Ser172 of NKG2A, and Ala158 interacted with Gln212 and Lys217 of NKG2A (Fig. 2 E).

Accordingly, the large and predominantly polar network of interactions between CD94-NKG2A and HLA-E underscored the specificity of this interaction and highlighted the dominant role of the CD94 subunit with respect to the NKG2A subunit.

Overall, the peptide contributed 23% of the BSA at the HLA-E interface, in which the invariant CD94 subunit played a much more marked role (80%) in interacting with the peptide compared with the NKG2A chain (20%). CD94 sat over the P8-Phe position of the peptide while NKG2A was adjacent to P5-Arg, with its contact being limited to a VDW interaction with Pro171 (Fig. 3, A and B). In contrast, CD94 interacted with residues P5-Arg, P6-Thr, and P8-Phe, with Gln112 intercalated between the bulky P5-Arg and P8-Phe and hydrogen bonded to the main chain of P6-Thr. Unexpectedly, three polar residues (Asn156, Asn158, and Asn160) surrounded and contacted the hydrophobic P8-Phe, the latter of which also interacted with Phe114 and Gln112 (Fig. 3 B).

The P5-Arg protruded into a cavity between the NKG2A and CD94 subunits (Fig. 3 A), with its aliphatic side chain flanked in an anti-parallel manner with Gln112 of CD94, whereas its guanadinium group H bonded to the main chain of Ser110 of CD94 (Fig. 3 B). However, the P5-Arg did not appear to be well complemented by any charged interactions with the CD94-NKG2A, which may be attributable to it salt bridging to Glu152 of HLA-E.

Accordingly, the interactions with the peptide are dominated by the invariant CD94 subunit, in which the major determinants of the peptide, namely positions P5 and P8, appear poorly matched in terms of the chemical and electrostatic complementarity to CD94-NKG2A.

Before determining the structure of the CD94-NKG2A–HLA-E ternary complex, we conducted an alanine-scanning mutagenesis and surface plasmon resonance study at this interface (25). Accordingly, this not only provided us with an opportunity to independently verify the structural footprint from the low-resolution structure reported here, but it also allowed us to address the energetic basis of this specific interaction. First, of the 16 mutations that were made in HLA-E, 8 residues were shown to be important in the interaction. All of these HLA-E residues (Arg65, Arg75, Arg79, Val76, Gln72, Glu152, Asp162, and Glu166), which encompassed residues from the α1 and α2 helices, made contacts with CD94-NKG2A in the ternary complex (Fig. 4 A). Second, of the 21 alanine-scanning mutations made in CD94-NKG2A, 10 residues were shown to be important for the interaction. Again, the structural data indicated that these CD94-NKG2A residues (CD94: Gln112, Phe114, Asn160, Leu162, Asp163, Glu164, Asp168; NKG2A: Gln212, Arg137, Arg215) made contacts with HLA-E, thereby further confirming the interpretation of the CD94-NKG2A–HLA-E footprint (Fig. 4 B). Nevertheless, as expected, there were residues that made contact at the interface that did not contribute significantly to the energetics of the interaction as assessed by surface plasmon resonance (see Discussion).

To further assess the footprint of CD94-NKG2A on HLA-E, an additional nine mutations were made in HLA-E, and the impact of these mutations was examined using surface plasmon resonance. These mutations (Fig. 4 C) were not alanine-scanning mutations and included charge-reversal mutations (Asp69 to Arg; Glu154 to Arg), which enabled the specificity of the interaction to be further addressed. Only one mutation (Glu152 Val) had a significant effect on the affinity of the interaction, which highlighted that it is the negative charge of Glu152, and not the aliphatic moiety per se, that is a requisite for this interaction. The charge-reversal mutations had no effect on the affinity of the interaction, consistent with the corresponding alanine-scanning mutations at these positions, thereby further confirming that positions 69 and 154 are not critical for determining the specificity of the interaction. Position 155 represents a polymorphic site between HLA-E (His155) and other HLA-Ia molecules (Gln155); surprisingly, substitution of His155 to Gln or Ser had a differential effect, with the former having a moderate effect on the affinity of the interaction, whereas the latter had no significant impact. Moreover, positions 65, 69, and 155, which make contacts with the TCR in all TCR–MHC-I structures determined to date (27, 28), all appear to be dispensable in the interaction with CD94-NKG2A (Fig. 4 C), further highlighting the differing requirements for adaptive and innate receptors interacting with MHC molecules.

The structure of the CD94-NKG2A–HLA-E^VMAPRTLFL complex provided an opportunity to compare it to (a) the αβTCR–HLA-E^VMAPRTLIL complex (Fig. 3 C and Figs. S2 and S3) (29), thereby allowing us to address how both innate and adaptive receptors can recognize the same ligand, and (b) the homodimeric NKG2D–MHC-like ligand complexes (11, 12, 14), which enabled us to address the central issue of promiscuous versus specific NK receptor recognition. First, the position of the CD94-NKG2A footprint on HLA-E^VMAPRTLFL was similar to the footprint of an HLA-E–restricted TCR (termed KK50.4; Figs. S2 and S3) (29). Similar to CD94-NKG2A, the KK50.4 TCR formed an extensive footprint on HLA-E^VMAPRTLIL, focusing on the C-terminal end of the antigen-binding cleft with the TCR docked in a diagonal mode; however, in contrast to CD94-NKG2A recognition of HLA-E, each chain of the αβTCR contributed roughly equally to the interface (Figs. S2 and S3). The KK50.4 TCR interacted predominantly with positions P5-Arg and P8-Ile of the peptide, analogous to that of CD94-NKG2A, suggesting that both innate and adaptive receptors are sensitive to changes to positions P5 and P8 of the peptide. Nevertheless, although the interactions between the peptide and CD94-NKG2A were CD94 centric, both chains of the TCR made substantial contacts with the peptide, with the TCRα chain focused on the peptide positions P4-P6, and the TCRβ chain interacting with P8-Ile (Fig. 3 C). Notably, the interactions between the αβTCR and the peptide exhibited excellent chemical complementarity, which contrasted markedly with the CD94 peptide-mediated interactions (Fig. 3 C).

Second, the location of the CD94-NKG2A footprint on HLA-E^VMAPRTLFL was analogous to the positioning of the footprint of human NKG2D on its MHC-like ligands, MICA and, to a lesser extent, ULBP3 (Fig. 5, A and B). Nevertheless, there were notable differences between these footprints, which are attributable to the narrower cleft between the α helices of MICA and ULBP3 compared with HLA-E, and the structural differences between CD94-NKG2A and NKG2D. CD94-NKG2A possessed a flat interacting surface with HLA-E, whereas that of NKG2D was more “saddle-like,” which enabled it to clamp around both α helices of the MHC-like ligands (Fig. 5 A) (11, 12, 14). Moreover, the closer juxtapositioning of the α1 and α2 helices of MICA and ULBP3 enabled each NKG2D subunit within the homodimer to interact with both α helices to some extent, whereas the interactions with the CD94 and NKG2A subunits are virtually polarized to each α helix of HLA-E (Table II). Moreover, although CD94-NKG2A is structurally homologous to NKG2D, different regions within these NK receptors were used to interact with their respective ligands (Fig. 5 C). In contrast to the predominantly electrostatic nature of the CD94-NKG2A–HLA-E^VMAPRTLFL interaction, the polar interactions contribute to but do not dominate the interaction between NKG2D and its ligands (Fig. 5, B and C). Specifically, the nonpolar interactions at the NKG2D-MHC–like and CD94-NKG2A–HLA-E^VMAPRTLFL interfaces contribute ∼60 and 40% to the BSA at the interface, respectively. This differential usage of polar-mediated contacts may account for the promiscuity in NKG2D recognition compared with a high degree of specificity for ligand displayed by CD94-NKG2A.

Accordingly, the innate NK receptors and the αβTCR focus on a similar region of the HLA-E or MHC-like ligand, regardless of whether the ligand presents peptide or not. However, the characteristics of the footprint deviate between these receptors, thereby providing a basis for the NKG2D promiscuity versus the CD94-NKG2A specificity.

We have determined the structure of the CD94-NKG2A–HLA-E^VMAPRTLFL complex and thereby provide a structural basis for the specificity of this interaction, which contrasted with the promiscuous recognition of MHC-like ligands by the homodimeric NKG2D receptor. There was little evidence of conformational adjustments of either CD94-NKG2A or HLA-E^VMAPRTLFL in the formation of the CD94-NKG2A–HLA-E^VMAPRTLFL complex, which is consistent with previous thermodynamic measurements that suggested a “rigid body” mode of recognition (22). This lock and key interaction contrasts with the conformational plasticity that is exemplified by antigen recognition by the adaptive immune system (30). Although CD94 and NKG2A are structurally similar, the distribution of surface electrostatics of CD94-NKG2A differed markedly, dictating the docking orientation of these subunits on HLA-E. A surprising feature of the CD94-NKG2A–HLA-E^VMAPRTLFL complex was the degree to which CD94 dominated the interaction with HLA-E and the peptide, as NKG2A was limited to contacting a small region on the α2 helix of HLA-E and a single VDW contact with the peptide.

Our previous alanine-scanning mutagenesis data that assessed the energetic contribution of HLA-E and CD94-NKG2A residues at the interface validated our observed positioning of CD94-NKG2A on HLA-E^VMAPRTLFL (25). Virtually all of the HLA-E residues that were critically important to the affinity of the interaction interacted with CD94-NKG2A and vice versa. The role of Glu152 from HLA-E remains unclear because although this residue directly interacted with Gln112 of CD94, the Glu152 Ala and Glu152 Val mutations may also affect the conformation of P5-Arg, which is a critical determinant for CD94-NKG2A recognition (31). Other HLA-E residues that contacted CD94-NKG2A either had a moderate role (Arg65, Arg79, and Glu166) in the energetics interaction or no role at all (Asp69 and Glu154). The alanine-scanning mutagenesis also revealed that CD94 played a notably more dominant role in the energetics of the interaction compared with the NKG2A subunit. Specifically, the CD94 energetic hotspot comprised residues Gln112, Phe114, Asn160, Leu162, Asp163, and Glu164, all of which interacted with HLA-E or the peptide, whereas surprisingly no NKG2A residues were identified as being critical to the interaction. Collectively, these observations also reveal that the energetic footprint of the CD94-NKG2A–HLA-E^VMAPRTLFL interaction is smaller than the corresponding structural footprint. Such an observation is consistent with that observed in adaptive immunity (32, 33) and contrasts the more synonymous structural and energetic footprints of the innate NKT TCR–CD1d interaction (34, 35).

Despite the dominant role of CD94, paradoxically there is an approximate sixfold difference in the affinity of CD94-NKG2A (0.7 μM) versus CD94-NKG2C (4.4 μM) for HLA-E^VMAPRTLFL (22, 24). The structure of the CD94-NKG2A–HLA-E^VMAPRTLFL complex allowed us to begin to address the basis of this observation. Surprisingly, of the sequence differences between NKG2A and NKG2C, none directly contacts HLA-E^VMAPRTLFL. However, a region of NKG2A-2C sequence difference spans residues 166–170, which lies at the interface of the CD94-NKG2A heterodimer and has been shown to modulate the affinity for HLA-E (26). This region is in close vicinity to the P5-Arg binding pocket, and accordingly it is conceivable that alterations in the heterodimer interface may impact on the affinity of the interaction. Because none of the residues that was exchanged directly contacts HLA-E, it is also possible that these differences are indirect, namely impacting on the conformation of proximal residues in CD94.

The interactions made by CD94-NKG2A with the HLA-E–bound peptide were suboptimal in chemical, shape, and electrostatic complementarity, in that the P8-Phe and the P5-Arg are accommodated within a polar and a noncharged pocket, respectively. Nevertheless, CD94-NKG2A recognition is sensitive to substitutions at both P5 and P8 positions of the peptide (25, 31). For example, P8-Tyr abrogated recognition, which may be attributed to unfavorable interactions with Phe114 of CD94, and substituting the P8 position to smaller residues, such as P8-Leu, Val, and Ala, also reduced the affinity of the interaction, presumably by creating a cavity at the interface. Similarly, the conservative P5-Arg to P5-Lys mutation abrogated recognition, suggesting either that the P5-Lys-Glu 152 salt bridge is lost, thereby introducing a buried positive charge at the interface, or highlighting the importance of the P5 residue interacting with the main chain of Ser110. Nevertheless, this lack of complementarity is atypical in innate receptor–ligand interactions and is perhaps more characteristic of the “imperfect” interfaces that have been observed in adaptive immunity (27).

Several proteins other than MHC-I molecules have also been shown to encode peptides that can bind to HLA-E. Perhaps the best characterized of these is encoded by UL40 protein of CMV that contains VMAPRTLIL, an exact mimic of a sequence found in most HLA-C alleles, and both binding and functional studies have demonstrated an interaction with CD94-NKG2A (24, 26, 36, 37). Based on the structure of the CD94-NKG2A–HLA-E^VMAPRTLFL complex, the presence of an Ile rather than a Phe at P8 would not be predicted to improve the chemical or SC around P8. Consistent with this, binding data indicate that the substitution of Ile for Phe results in a threefold reduction in the affinity of the interaction. In addition, peptides derived from HCV core antigen (YLLPRRGPRL) (38), the ATP-binding cassette transporter multidrug resistance–associated protein 7 (ALALVRMLI) (39), and gliadin (SQQPYLQLQ) (40) have also been shown to bind HLA-E and protect otherwise susceptible target cells via a CD94-NKG2–dependent mechanism. Given the marked differences of these peptides from the canonical MHC-I leader sequence peptide, it will be of interest to establish how these peptides, when bound to HLA-E, interact with CD94-NKG2A.

In addition to being a ligand for the CD94-NKG2 receptors, HLA-E is recognized by CD8 T cells expressing conventional αβTCRs (41–44). The data presented here reveal that the footprint of CD94-NKG2A on HLA-E closely overlaps with that of an HLA-E–restricted TCR (KK50.4) (29). Interestingly CD94-NKG2 receptors are also expressed on these HLA-E–specific T cells (unpublished data) (45). Given the differing affinities between the TCR and CD94-NKG2 receptors for HLA-E, it will be of interest to dissect the biological outcome with these cells. In vitro data suggests that expression of the cognate ligand of the HLA-E–restricted TCR can result in T cell activation despite the expression of the inhibitory CD94-NKG2A receptor (45).

Given the sequence and structural conservation between CD94-NKG2A and NKG2D, as well as their ligands HLA-E, MICA, MICB, and ULBP1-4, it is perhaps not surprising that their footprints were broadly similar. Nevertheless, the molecular basis for ligand recognition differed dramatically with CD94-NKG2A recognition of HLA-E being dominated by charged interactions, whereas hydrophobic interactions played a more significant role in ligand recognition by NKG2D. This dependence on charge in the CD94-NKG2A–HLA-E interaction bears notable similarity with the interaction between KIRs and their class I ligands where disruption of salt bridges abrogated recognition of HLA-C (46, 47).

Although the evolutionary origins of the CD94-NKG2 receptors are not entirely clear, comparison of CD94 and NKG2 sequences in primates and rodents further emphasized the importance of CD94. Although there is significant cross-species variation in the residues that correspond to those in NKG2A that contact HLA-E, 5 of the 13 CD94 residues that contact HLA-E are completely conserved across both primates and rodent species (48–52). Thus, although CD94 by itself is unable to bind HLA-E, it nevertheless dictates the specificity of the interaction, whereas the primary role of the NKG2 subunit is to determine the outcome of ligand recognition.

HLA-E was expressed as inclusion bodies in Escherichia coli and refolded with the VMAPRTLFL peptide according to the previously described method (29). The refolded HLA-E^VMARPTLFL complex was further purified using anion exchange and gel filtration chromatography. The extracellular domains of CD94 and NKG2A (CD94 residues 52–179, NKG2A residues 110–223) were expressed as inclusion bodies in E. coli and refolded as described previously (25). The refolded CD94-NKG2A heterodimer was purified by anion exchange and gel filtration chromatography, followed by an additional purification step using anion exchange. The CD94-NKG2A–HLA-E^VMARPTLFL complex was generated by mixing the CD94-NKG2A heterodimer with HLA-E^VMARPTLFL at a molar ratio of 1.2:1, respectively. The complex was purified via size exclusion chromatography for use in crystallization. The mutagenesis and surface plasmon resonance–based experiments were conducted as described previously (25).

Large crystals (>0.3 mm) of CD94-NKG2A–HLA-E^VMARPTLFL complex were grown at 5 mg/ml in 2.0 M NH₄SO₄, 0.1 M Tris-HCl, pH 7.7–7.9. Drops were made at a ratio of 2:1 (1 μl protein/0.5 μl mother liquor) via the hanging drop vapor diffusion method, and crystals were grown at 18°C. Crystals were flash frozen at 100 K using 20% glycerol as a cyroprotectant. A 3.4 Å dataset was collected at the X25 beamline (Brookhaven) using an ADSC Q315 detector. The data were processed and scaled using the CCP4 suite. The crystals belong to space group I4₁32, with a = b = c = 345.2 Å, with two complexes per asymmetric unit, with a high solvent content of 80%. Although the crystals diffracted only to 3.4 Å resolution, the I/σ_I in the highest resolution bin (3.5–3.4 Å) was 3.0, and the overall R_PIM (53) value to this resolution was 4.6%. Attempts to dehydrate the crystals or to find an alternative crystallization condition failed to yield better diffracting crystals. The crystal structure of the complex was solved by molecular replacement implemented by Phaser (54) using CD94-NKG2A and HLA-E (minus the bound peptide) as the search models. First, we used CD94-NKG2A (PDB code: 3bdw) as a search model and readily located two molecules within the asymmetric unit. We then fixed this initial solution and searched for the corresponding HLA-E molecules in which the peptide had been deleted (PDB code: 3bze) (unpublished data). Again, two HLA-E molecules were located within the asymmetric unit, with the HLA-E molecules sitting “beneath” the CD94-NKG2A solutions, indicating that we had located two ternary complexes in the asymmetric unit. Searching for a third ternary complex failed to yield a solution. To verify that the molecular replacement solution was correct, a 2 F_o-F_c electron density map was calculated and revealed unbiased electron density for the omitted peptide chain. The progress of refinement was monitored by the R_free value (5% of the data) with neither a σ nor a low-resolution cutoff being applied to the data, and no water molecules were modeled in the structure. The structure was refined using Phenix (55) interspersed with rounds of model building in “Coot” (56). Given the limited resolution of the data, the refinement proceeded cautiously, with tight noncrystallographic restraints applied to the independent copies within the asymmetric unit. Refinement of B factors led to an ∼4% drop in the R_factor and R_free values, thereby justifying the refinement of this parameter. As such, the complex refined very well, presumably aided by the fact that we had previously determined the structures of the nonligated NKG2A and HLA-E components to a 2.5 Å resolution; moreover, there were few conformational changes that took place upon complexation. For the data collection and refinement statistics, see Table I. See Fig. S1 for quality of electron density. All BSA calculations were undertaken using MSD PISA (http://www.ebi.ac.uk/msd-srv/prot_int/pistart.html). The PDB of the CD94-NKG2A–HLA-E complex has been deposited in the RCSB protein data bank, and the coordinate accession code is 3CDG.

Fig. S1 shows regions of unbiased and final electron density for the complex. Fig. S2 shows a comparison of structural footprints, and Fig. S3 shows the docking mode of CD94-NKG2A and other receptors. Figs. S1–S3 are available.

We would like to thank the staff at the Brookhaven Synchrotron for assistance with data collection.

This work was supported by grants from the National Health and Medical Research Council (NHMRC), the Australian Research Council (ARC), an ARC QEII fellowship (to C.S. Clements), Doherty Fellowships from the NHMRC (to L.C. Sullivan and T. Beddoe), an NHMRC Senior Research Fellowship (to M.C.J. Wilce), and an ARC Federation Fellowship (to J. Rossjohn).

The authors have no conflicting financial interests.