Structural basis for recognition of the nonclassical MHC molecule HLA-G by the leukocyte Ig-like receptor B2 (LILRB2/LIR2/ILT4/CD85d)

Abstract

HLA-G is a nonclassical MHC class I (MHCI) molecule that can suppress a wide range of immune responses in the maternal–fetal interface. The human inhibitory immune receptors leukocyte Ig-like receptor (LILR) B1 [also called LIR1, Ig-like transcript 2 (ILT2), or CD85j] and LILRB2 (LIR2/ILT4/CD85d) preferentially recognize HLA-G. HLA-G inherently exhibits various forms, including β₂-microglobulin (β₂m)-free and disulfide-linked dimer forms. Notably, LILRB1 cannot recognize the β₂m-free form of HLA-G or HLA-B27, but LILRB2 can recognize the β₂m-free form of HLA-B27. To date, the structural basis for HLA-G/LILR recognition remains to be examined. Here, we report the 2.5-Å resolution crystal structure of the LILRB2/HLA-G complex. LILRB2 exhibits an overlapping but distinct MHCI recognition mode compared with LILRB1 and dominantly recognizes the hydrophobic site of the HLA-G α3 domain. NMR binding studies also confirmed these LILR recognition differences on both conformed (heavy chain/peptide/β₂m) and free forms of β₂m. Binding studies using β₂m-free MHCIs revealed differential β₂m-dependent LILR-binding specificities. These results suggest that subtle structural differences between LILRB family members cause the distinct binding specificities to various forms of HLA-G and other MHCIs, which may in turn regulate immune suppression.

In the maternal–fetal interface, special immune suppression systems exist that allow the fetus to evade certain maternal immune responses (1). HLA-G is a nonclassical MHC class I (MHCI) molecule that is expressed only on placenta trophoblasts and thymic epithelial cells. Trophoblast cells do not express major classical MHCIs (HLA-A and -B) to avoid induction of T cell responses. Instead, these cells express HLA-G, -C, and -E to suppress the maternal immune responses by binding to inhibitory immune receptors. The reported receptors for HLA-G are limited but include leukocyte Ig-like receptors (LILRs, also called LIR, Ig-like transcript (ILT), or CD85), LILRB1 (LIR1/ILT2/CD85j), and LILRB2 (LIR2/ILT4/CD85d), as well as KIR2DL4 (killer cell Ig-like receptor 2DL4), expressed on natural killer cells and some T cells. Because the LILRBs are expressed on a wide range of leukocytes and mediate inhibitory signals, HLA-G is believed to have a pivotal role in a broad range of immune suppression functions in the placenta. Like other classical MHCIs, HLA-G comprises a heavy chain, β₂-microglobulin (β₂m), and an 8- to 10-aa peptide derived from proteolytically degraded proteins inside cells. In addition to the tissue expression pattern, HLA-G has a number of unique characteristics. HLA-G shows very limited polymorphism and exists in several forms of spliced variants, including soluble and domain-deleted forms. In addition, HLA-G forms a disulfide-linked dimer in solution and at the cell surface (2, 3). This dimer mediates much more efficient inhibitory signals than HLA-G monomers by binding to LILRB1 and LILRB2, as shown by cellular and biochemical studies (2–4). The recent crystal structures of the HLA-G monomer (5) and dimer (4) provided the structural basis for efficient signaling. The HLA-G dimer exposes LILR-binding sites that are easily accessible for receptor binding, enabling one HLA-G dimer to bind two receptors simultaneously, which positions the intracellular domains of the receptors close to each other, supporting high efficiency of signaling.

Interestingly, β₂m-free heavy chains (fHCs) of HLA-G are found in the placenta (6). fHCs of MHCIs are also expressed in some other tissues, including activated T cells and tumor cells. In addition, the pathological allele HLA-B27 exists as an unusual fHC homodimer form (7). Thus, fHCs are believed to have physiological roles in regulating immune responses. To date, only two receptors for fHC, have been shown to be members of the LILR family LILRB2 and LILRA1 (LIR6/CD85i), by elucidating the binding of HLA-B27 fHC tetramer to LILRB2- and LILRA1-transfected cells (8). Gonen-Gross et al. (6) showed that the fHC of HLA-G cannot bind to LILRB1, a close relative of LILRB2, but did not examine whether the fHC of HLA-G can bind to LILRB2.

LILRs are encoded within the human leukocyte receptor complex 19q13.4 and are closely related to the killer cell Ig-like receptors (KIR/CD158) (9). Although LILRB1 is widely expressed on monocytes, dendritic cells, B cells, and subsets of natural killer and T cells, LILRB2 is almost exclusively expressed on the myelomonocytic lineage. Both LILRB1 and LILRB2 have immunoreceptor tyrosine-based inhibitory motifs in their cytoplasmic tails to recruit the protein tyrosine phosphatase SHP-1, resulting in the inhibitory signaling. In addition to the maternal–fetus tolerance (10), LILRB2 is thought to be involved also in regulating the tolerogenic antigen-presenting cells (APCs) induced by regulatory T cells (11) and CD68⁺ macrophages and neutrophils in synovium from rheumatoid arthritis patients (12).

Both LILRB1 and LILRB2 have four tandem Ig-like domains in the extracellular region, and the N-terminal two Ig-like domains are responsible for MHCI recognition. Killer cell Ig-like receptor family members recognize the peptide-binding region of MHCIs and thus exhibit some peptide selectivity (13). By contrast, ligands for most LILR family members are yet to be identified, but LILRB1 and LILRB2 have been shown to bind to a broad range of MHCIs by recognizing the α3 domain and β₂m of MHCIs, both of which are conserved among classical and nonclassical MHCIs. As mentioned earlier, however, the ligand recognition of LILRB1 is different from LILRB2 in terms of β₂m dependency, even though the two receptors show high sequence identity (81%). On the other hand, our previous report showed that LILRB1 and LILRB2 show some allele-dependent recognition, such as preferential binding to HLA-G in comparison with other MHCIs (14).

The crystal structure of the LILRB1/HLA-A2 complex clearly showed that LILRB1 binds two sites, the α3 domain and β₂m of HLA-A2, simultaneously (15). As mentioned above, HLA-G has a number of unique characteristics compared with other classical MHCIs. In addition, LILRB2 exhibits different MHCI-binding characteristics than LILRB1. However, the structural characteristics of any MHCI complex with LILRB2 and the structural characteristics of any receptor complex of HLA-G have yet to be elucidated. Here, we report the crystal structure of the LILRB2/HLA-G complex. The structure demonstrated that the binding interface of the LILRB2/HLA-G complex shows some overlap with that of the LILRB1/HLA-A2 complex. However, the binding interface also shows some significant differences, which are observable in both the crystal structure and NMR binding studies using ¹⁵N-labeled β₂m. LILRB2 adopts a rather distinct binding mode by using a small number of different amino acids from LILRB1, resulting in predominant recognition of the α3 domain and providing the β₂m-independent MHCI recognition that is demonstrated by surface plasmon resonance (SPR) studies on LILRB2 binding to various forms of HLA-G and MHCIs. More hydrophobic interactions were observed between LILRB2 and the α3 domain of HLA-G compared with the LILRB1/HLA-A2 complex. It is reasonable to suggest that these interactions confer a higher affinity for HLA-G. Based on these structural and functional results, the physiological roles for LILR recognition of various forms of HLA-G as well as other MHCIs are discussed.

Results

Overall Structure of the LILRB2 Complex with HLA-G.

The crystal structure of the complex of HLA-G and LILRB2 was solved by molecular replacement [the program used was Amore; the search probes were the structures of HLA-G. Crystallographic data and refinement statistics are summarized in Table 1, which is published as supporting information on the PNAS web site, and the 2F_o − F_c map is shown in Fig. 1B. The overall complex structure demonstrated that LILRB2 recognizes the α3 and β₂m domains of HLA-G by using the N-terminal Ig-like domain (site 2) and the interdomain region (site 1), respectively (Fig. 1A). At first glance, this complex structure looks similar to the LILRB1/HLA-A2 complex reported by Willcox et al. (15). Unexpectedly, however, Fig. 2 shows that the recognition modes of LILRB1 and LILRB2 are not the same, which plausibly contributes to different ligand specificities as described in the next paragraph.

Fig. 1.

Overall structure of the LILRB2 complex of HLA-G. (A) Ribbon diagram of the LILRB2/HLA-G1 complex. Red, LILRB2; cyan, HLA-G heavy chain; green, β₂m; blue, nonapeptide. LILRB2 uses two binding interfaces: D2-β₂m (site 1) and D1-α3 (site 2) domain. The stick models of the amino acids involved in the complex formation are shown as follows. Site 1: orange, D2 domain of LILRB2 (Trp-67, Asp-177, Asn-179, and Val-183); sky blue, β₂m (Lys-6 and Lys-91). Site 2: magenta, D1 domain of LILRB2 (Arg-36, Tyr-38, Lys-42, Ile-47, and Thr-48); blue, α3 domain (Phe-195, Tyr-197, and Glu-229). The descriptions of site 1 and site 2 apply to Figs. 1–4. (B) 2F_o − F_c map (green mesh) contoured at 1σ onto the stick model of the LILRB2/HLA-G complex (around Phe-195 and Tyr-197 of HLA-G).

Fig. 2.

Structural comparison of the LILRB2/HLA-G and LILRB1/HLA-A2 complexes. (A) The LILRB2/HLA-G complex is superimposed onto the LILRB1/HLA-A2 complex around the MHCI regions. The molecular surface of HLA-G is shown. Cyan, HLA-G heavy chain; green, β₂m; red, LILRB2; yellow, LILRB1. The D1 domain of LILRB2 has more binding interface on the α3 domain (site 2). The different interactions are observed in D2–β₂m interfaces (site 1). (B) This complex image is produced by rotating the image in A around the horizontal axis ≈90°. (C and E) The buried surface areas of LILRB2/HLA-G (C) and LILRB1/HLA-A2 (E). The buried surface areas were calculated by using the program SURFACE (CCP4 suite) with a 1.4-Å probe radius. Cyan, HLA heavy chain; green, β₂m; red, LILRB1 or LILRB2. The buried surfaces are shown in yellow. (D and F) NMR analysis of LILRB2–HLA-Cw7 and LILRB1–HLA-Cw7 interactions. The model orientation is similar to that of C and E. Mapping of amino acids whose ¹H-¹⁵N heteronuclear sequential quantum correlation peaks were perturbed upon complex formation is shown (orange, >0.09 ppm chemical shift changed; magenta, >50% intensity reduced; red, disappeared; green, <0.09 ppm chemical shift changed; white, unassigned). (D) Interaction of conformed (heavy chain/peptide/β₂m) (Left) and free (Right) forms of β₂m with LILRB2. (F) The same binding analysis as in D with LILRB1.

Structural Comparison of the LILRB2/HLA-G and LILRB1/HLA-A2 Complexes.

There are some major differences in the contact interfaces with both the α3 domain and the β₂m (details of the contact residues are summarized in Table 2 and Fig. 5, which are published as supporting information on the PNAS web site). LILRB2 forms a more intimate contact interface with the α3 domain (red arrows in Fig. 2 A and B). Fig. 2 C–F shows that LILRB2 interacts with a much larger surface area of the α3 domain of HLA-G (site 2, 460 Ų) and slightly larger surface area of β₂m (site 1, 610 Ų) compared with the LILRB1/HLA-A2 complex (α3, 280 Ų; β₂m, 570Ų). This finding indicates that LILRB2 predominantly binds to the α3 domain, whereas LILRB1 preferentially binds to β₂m, which explains why LILRB2 can bind to MHCIs with much less β₂m dependency than LILRB1 (as described under Different β₂m Dependency of MHCI Recognition of LILRBs).

In terms of specific amino acid interactions at the LILRB2–β₂m interface (site 1), Val-183 of LILRB2 forms a hydrophobic interaction with Trp-67, whose aromatic ring also forms a typical π–cation interaction with Lys-91 of β₂m (Fig. 6A, which is published as supporting information on the PNAS web site). By contrast, Glu-184 of LILRB1 (the equivalent position of Val-183 of LILRB2) forms a salt bridge with Lys-91 of β₂m, causing the hydrophobic face of Trp-67 to rotate around the C^α–C^β bond to a different orientation to form the π–cation interaction with Lys-91 of β₂m as well as interactions with Ile-92 and Val-93 (Fig. 6B). It causes the distinct binding interfaces on β₂m between LILRB1 and LILRB2: LILRB2 interacts with the region closer to the bottom of the α2 domain, whereas the LILRB1-binding interface is closer to the membrane-proximal site (Fig. 2 C and E). This finding is consistent with the NMR studies described in a later section (Fig. 2 D and F). Notably, previous work has shown that LILRA1 (LIR6/CD85i), like LILRB2, can bind to both conformed (heavy chain/peptide/β₂m) and β₂m-free HLA-B27 (8, 16) and also has Val-183, suggesting that LILRA1 likely has a binding mode that is similar to that of LILRB2. All other “group 1” LILRs except for LILRB1 contain Val-183 and are thus likely to adopt a LILRB2-type MHCI recognition mode if they can bind to any MHCIs.

Furthermore, the C′E loop (residues 153 and 154) of LILRB2 makes contact with the N-terminal site of β₂m, but that of LILRB1 forms a 3₁₀ helix that is not involved in MHCI binding (Fig. 6 C and D). The FG loop (Asp-177 and Asn-179) of LILRB2 can recognize the N-terminal site of β₂m together with the BC loop (residues 124–126). In contrast, LILRB1 uses only the BC loop for MHCI recognition (Fig. 6 C and D). Thus, this interface on β₂m likely contributes to stability of the recognition mode of LILRB2 in a manner distinct from that of LILRB1.

On the other hand, the α3 domain interacts with different faces of LILRB1 and LILRB2 (site 2, Figs. 1 and 2). LILRB1 mainly uses the C strand for recognition of the α3 domain (e.g., Tyr-38, Arg-36, etc.), with additional interactions formed by the F strand (Tyr-76) (Fig. 3B and D; see also Fig. 7, which is published as supporting information on the PNAS web site). In contrast, LILRB2 accommodates the membrane-proximal AB loop, including Phe-195 and Tyr-197 of the HLA-G α3 domain, in a large cleft between the 3₁₀ helices and the C strand (Fig. 3 A and C), a space unoccupied in the LILRB1 complex (Fig. 3D).

Fig. 3.

LILR binding interfaces (site 2) of the α3 domain of the LILRB2/HLA-G and LILRB1/HLA-A2 complexes. (A and C) LILRB2/HLA-G complex. (B and D) LILRB1/HLA-A2 complex. Cyan, HLA-G heavy chain; light blue, HLA-A2 heavy chain; green and light green, β₂m; magenta, LILRB2; yellow, LILRB1. (A and B) The binding interface around the 195–197 loop of HLA-G. (C and D) The binding interface around the cleft between the first 3₁₀ helix and the C strand of LILRBs.

Previous binding studies and structural reports of HLA-G proposed that the hydrophobic amino acids Phe-195 and Tyr-197 form hydrophobic interactions with LILRB1 and LILRB2 (4, 5, 14). The current structure clearly showed that Phe-195 is located close to the hydrophobic faces of Thr-48 and Ile-47 of LILRB2 and that Tyr-197 directly interacts with Arg-36 and Tyr-38 of LILRB2 (Fig. 3 A and C). In contrast, LILRB1 uses a different set of residues on the C and F strands, Tyr-76 and Tyr-38, to make contacts with the 195–197 loop of HLA-A2 (Fig. 3 B and D). Assuming that HLA-G recognizes LILRB1 in a similar way to HLA-A2, Phe-195 is likely to interact with the hydrophobic faces of Tyr-76 and Tyr-38, whereas Tyr-197 may be too far from LILRB1 to be involved in direct contact. Therefore, Phe-195 and/or Tyr-197 induce the higher binding affinity of HLA-G to LILRB2 and LILRB1, even though the mechanisms may be different.

An additional HLA-G-binding site composed of Lys-42 on LILRB2 is observed at the interface with the α3 domain (Glu-229). In the LILRB1/HLA-A2 complex, there is no observable interaction between Lys-42 and the α3 domain at this location; instead, Lys-42 forms a salt bridge with Asp-96 of β₂m (Table 2 and Figs. 5 and 8, which are published as supporting information on the PNAS web site). Thus, this interaction could contribute to the different recognition modes of LILRB2 and LILRB1.

Structural Changes Upon Complex Formation.

Although LILRB1 shows some interdomain angle changes (14–19°) upon complex formation, LILRB2 maintains the same interdomain angle (Figs. 9 and 10, which are published as supporting information on the PNAS web site). Two 3₁₀ helices (residues 46–50 and 53–57) between the C and E strands of LILRB2 are formed upon HLA-G complex formation, whereas free LILRB2 shows only one 3₁₀ helix (residues 52–55) (17) (Fig. 9 A and B). This conformation in the complexed state of LILRB2 resembles that of LILRB1, even though their MHCI-binding sites are not the same (Fig. 9C). In contrast, for bound HLA-G, the positional changes of only side chains and loops without significant conformational differences are observed in comparison with the free HLA-G structure (4, 5) (Fig. 11, which is published as supporting information on the PNAS web site). Our preliminary thermodynamic data showed that the LILRB2–HLA-G interaction tends to be entropic- and enthalpic-driven (K. Kuroki, M.S., L.R., K.T., I.K., D.K., and K.M., unpublished data), whereas the LILRB1–HLA-G interaction appears to be entropic-driven, as determined by van't Hoff analysis and titration calorimetry (18). The LILRB1 free state already exhibits two 3₁₀ helices and thus exhibits less conformational change in the local structure being induced upon complex formation. Therefore, the conformational rearrangements of LILRB2 upon complex formation may reflect some entropic loss in comparison with LILRB1.

NMR Analysis of LILRB1 and LILRB2 Binding to MHCI and β₂m.

To further confirm the differences in MHCI binding of LILRB1 and LILRB2 directly in solution, NMR measurements were performed. We focused on a small subunit of MHCIs, β₂m, because x-ray crystallographic data indicate that these two receptors embody distinct modes of interactions with β₂m in the MHCI complex. The β₂m was expressed as a ¹⁵N-labeled protein and refolded with or without the nonlabeled heavy chain of MHCI and its cognate peptide for identification of the LILR-binding interface. For NMR measurements, the HLA-Cw7 heterotrimer was used, because it is much more stable and less readily aggregated than HLA-G, and therefore it is suitable for solution NMR analyses. ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) peaks originating from 87 and 67 of 93 backbone amide groups of β₂m were used as spectroscopic probes for the free and conformed (heavy chain/peptide/β₂m) forms, respectively. We successfully observed spectral changes upon LILRB1 and LILRB2 complex formation. Fig. 2 D and F Left show the amino acid residues of the conformed form of β₂m that exhibited HSQC peaks significantly perturbed upon LILRB1 or LILRB2 complex formation. As expected from the crystal structures of the LILR complexes, LILRB1 and LILRB2 recognize β₂m in solution, but their binding sites are different. The area of the LILRB1-binding site is widely distributed from the α2–β₂m interface to the membrane-proximal site. In contrast, the LILRB2-binding site is located close to the α2 helix of MHCI, consistent with the crystal structure of the LILRB2 complex described above. Furthermore, we also examined LILRB binding to the free form of β₂m. The results showed that the LILR-binding sites of β₂m are very similar to those of the conformed form (Fig. 2 D and F Right), which indicates that LILRB1 and LILRB2 have the ability to recognize β₂m in a similar fashion either with or without heavy chain and peptide and suggests that the differential LILR recognition modes are common to MHCIs and determined, at least partly, by LILR–β₂m interactions.

Different β₂m Dependency of MHCI Recognition of LILRBs.

Our previous report of SPR analysis for LILRB–MHCI interactions showed that, unexpectedly, immobilization of MHCI (HLA-Cw7) by using a chemically biotinylated form of β₂m inhibits binding to LILRB1 but not LILRB2 (14). Previous cellular-based assays have showed that the β₂m-free homodimer form of recombinant HLA-B27 can bind to LILRB2-transfected but not to LILRB1-transfected cells (19). These results clearly indicated that LILRB1 and LILRB2 have different ligand recognition modes in view of β₂m dependency. Furthermore, the current crystallographic and NMR analyses clearly support this idea.

Here, detailed binding studies using a series of β₂m-free MHCI forms were performed. The fHCs of HLA-G and HLA-Cw4 were prepared following the standard refolding method for the β₂m-associated MHCIs with the exception that no β₂m was added. SPR analysis showed that the immobilized fHCs of HLA-G1 and HLA-Cw4, which are not recognized by anti-β₂m mAb BBM.1 or by anti-MHCI mAb w6/32, can bind to LILRB2 but not to LILRB1 (Fig. 4 and Fig. 12, which is published as supporting information on the PNAS web site). The β₂m-associated MHCIs were also immobilized by means of the biotinylation tag of the heavy chain and acid-treated with a solution of 10 mM Gly/HCl (pH 2) to remove β₂m. This method is common in the preparation of fHCs on the cell surface. The acid-treated fHC was also able to bind to LILRB2 but not to LILRB1 (data not shown). These results are consistent with the cellular-based and SPR data described above.

Fig. 4.

SPR analyses. Binding of LILRB2 (Left) and LILRB1 (Right) to HLA-G heterotrimer (red lines) and β₂m-free HLA-G heavy chain (green lines). Heterotrimers and β₂m-free forms of MHCIs were immobilized on the sensor chip at ≈2,000 response units (RU). Black lines show the responses to the control protein (BSA).

Discussion

HLA-G is expressed at the maternal–fetal interface, where the expression levels of classical MHCIs responsible for T cell responses are reduced (20). There are several reports showing that soluble and membrane-bound forms of HLA-G induce a strong immunosuppressive effect on a wide range of immune cells by binding to inhibitory immune cell receptors. To date, the reported receptors for HLA-G are LILRB1, LILRB2, KIR2DL4, and CD8. Our present structural study provides a description of the complexed states of both HLA-G and LILRB2. The structure showed that LILRB2 uses two sites, the tip of N-terminal domain and the interdomain region, for binding to the α3 domain and β₂m of HLA-G, respectively. This binding mode is similar to that of LILRB1 (Fig. 1); however, LILRB2 recognizes HLA-G in a distinct recognition mode from that seen for the LILRB1/HLA-A2 complex (Fig. 2). The difference is derived from a surprisingly small number of amino acid differences. LILRB2 uses the cleft between the C strand and the 3₁₀ helices for binding to the 195–197 loop of HLA-G, whereas LILRB1 uses the protruding site of the C strand with a much smaller interface (Fig. 3). This structural characteristic explains why LILRB2 binding to MHCI fHCs is of a sufficient affinity to maintain the interaction between the two proteins, whereas the LILRB1 binding affinity is too low (Fig. 4). NMR binding studies using ¹⁵N-labeled β₂m in free and conformed (heavy chain/peptide/β₂m) forms demonstrated that the LILRB1- and LILRB2-binding sites on β₂m are conserved in both forms, consistent with the complexed crystal structures (Fig. 2). This result indicated that the LILR–β₂m associations at least partly determine or support the variable recognition modes of the LILR/MHCI complex, achieved by the rearrangement of the interactions caused by only a few amino acid substitutions (e.g., Val-183) (Fig. 6).

HLA-G can bind to both LILRB1 and LILRB2 more strongly than other classical and nonclassical MHCIs. Tyr-197 of HLA-G deeply intrudes into the hydrophobic pocket composed of Tyr-36 and Arg-38 of LILRB2, forming typical π–π and π–cation interactions (Fig. 3) and providing a stronger and tighter association than His-197 of HLA-A2, which may form a somewhat weaker hydrophobic interaction with Tyr-36 (and Arg-38), explaining the higher affinity of LILRB2 for HLA-G. Phe-195 of HLA-G faces the hydrophobic area comprising Ile-47 and Thr-48, whereas the equivalent Ser-195 of other classical MHCIs possibly excludes such hydrophobic interactions at this location. However, the same scenario cannot be valid for the LILRB1/HLA-G interaction, because LILRB1 shows a different MHCI-binding interface compared with LILRB2. Based on the crystal structure of the LILRB1/HLA-A2 complex (15), hydrophobic interactions can also be formed between Tyr-38 and Tyr-76 of LILRB1 and Phe-195 of HLA-G, supporting a higher affinity for HLA-G. Our previous functional and structural report on the disulfide-bonded dimer form of conformed HLA-G showed that both LILRB1 and LILRB2 can provide significant high affinity by avidity effects (4). Both the present LILRB2/HLA-G complex structure and the previously reported LILRB1/HLA-A2 complex structure ensure 1:2 (dimer:receptor) binding stoichiometry and structural orientation of LILRB/HLA-G dimer complexes, which facilitates the close location of its intracellular domain, which is responsible for signal transduction, resulting in efficient signaling.

The human cytomegalovirus MHCI homologue UL18 can bind to LILRB1 with a remarkably high affinity (range of 3–20 nM) (21) that is ≈1,000 times higher than HLA-G/LILRB1-binding affinity. A mutational study introducing LILRB1 residues into LILRB2 showed that the Gln-76-Tyr/Arg-80-Asp/Trp-83-Arg mutant LILRB2 binds UL18 with a K_d of 0.5–1 μM, 10–30 times higher affinity than the wild-type LILRB2 (K_d ≈ 12–14 μM) but much lower than LILRB1 (22). These mutation sites are not directly involved in the MHCI binding of LILRB2, as shown by the present complex structure. Thus, they may have an indirect effect and/or different binding mode that results in the preferable UL18 binding. In contrast, both the Tyr-76-Ala/Asp-80-Ala/Arg-80-Ala and Tyr-38-Ala mutant LILRB1s show lower affinity (K_d ≈ 40 nM). However, the affinities are much higher than that describing the LILRB2/UL18 interaction, even though Tyr-38 is directly involved in the MHCI binding. Thus, the other LILR-binding sites of UL18, such as the β₂m interface, may dominantly contribute to high affinity to the LILRB1 template structure. Moreover, the relative orientation between α3 and β₂m sites should fit the binding interface of LILRs to show high affinity. Therefore, UL18 can provide more appropriate positioning and surface area of the α3 and β₂m sites for LILRB1 than for LILRB2.

We performed detailed binding studies using fHCs of HLA-G and HLA-Cw4 to reveal the importance of β₂m and the α3 domain for binding to LILRB1 and LILRB2. The result showed that LILRB1 cannot bind to the fHCs but that LILB2 can, consistent with several previous reports. The physiological roles of fHC are not yet fully understood. However, it is of note that fHCs are inducibly expressed on activation of T cells. LILRB2 on APCs can recognize fHCs, thus raising the possibility that LILRB2 can inhibit the APC function by binding to fHCs on activated T cells. Furthermore, because the LILRB2 expression is up-regulated on tolerogenic APCs by regulatory T cells (11), it may be interesting to examine their fHC expression. The fHC of HLA-G is observed in the placenta (6) and exhibits the same biochemical and structural characteristics as fHCs of other MHCIs demonstrated in the present study, and, thus, it likely regulates APC function at the maternal–fetal interface.

The fHC of HLA-B27 is proposed to be directly associated with spondyloarthropathies. HLA-B27 forms a disulfide-linked (Cys-67-Cys-67) homodimer of fHC, which can bind to LILRB2- and LILRA1-transfected cells but not to LILRB1-transfected cells (19). The fHC dimer of HLA-B27 is likely to have a similar recognition mode to the HLA-G dimer, possibly eliciting high affinity to efficiently transduce signaling (4). Further investigation is required, but it is plausible that the fHC regulation system may have significant physiological relevance.

Paired Ig-like receptor (PIR) B, a mouse homologue of LILRBs, can bind β₂m with remarkably high affinity (approximately nanomolar range) and β₂m-associated MHCIs with low affinity (approximately micromolar range) (23). PIR-B can bind to the β₂m-free fHC homodimer form of HLA-B27 (24), suggesting that PIR-B likely has a function similar to LILRB2, although it is currently unknown whether PIR-B can bind to β₂m-free fHC of mouse MHCIs.

Group 1 LILRs can be further categorized into LILRB1 and LILRB2 types with regard to MHCI-binding modes. The cleft between the C strand and the 3₁₀ helices is used as the MHCI-binding site on LILRB2 (site 2) but not on LILRB1. As described above, LILRA1 has similar amino acid composition to LILRB2 in the cleft area and Val-183 regions, which are supposed to determine the binding modes. Thus, LILRA1 is likely to belong to the LILRB2 type of receptors, which is supported by the result that, like LILRB2, LILRA1 can bind to the β₂m-free B27 homodimer (19). The other group 1 LILRs also have the LILRB2-like amino acids in these regions, similar to LILRA1. However, our recent data for the structure of one of the “group 2” LILRs, LILRA5 (LIR9/ILT11/CD85f), showed that this cleft is filled by a C′ strand that is changed from the 3₁₀ helices (LILRB1 and LILRB2) to form the β-sheet with the C strand (25). Consistently, LILRA5 cannot bind to classical or nonclassical MHCIs (25). Furthermore, previous preliminary reports demonstrated that the other group 2 LILRs, LILRB3 (19, 26), LILRB4 (27), and LILRB5 (28), cannot bind to MHCIs. Our speculation about the evolutionary aspect of the LILR family is that, initially, the ancestor receptor of PIRs/LILRs acquired MHC-binding function and subsequently evolved to discriminate between β₂m-free (LILRB2 and LILRA1) and β₂m-associated (LILRB1) MHCIs and to recognize unknown non-MHCI ligands (group 2 LILRs).

In conclusion, the LILRB2/HLA-G complex structure reveals that LILRB2 shows remarkably distinct MHCI-binding recognition from LILRB1, binding more to the α3 domain than β₂m. This finding could explain the β₂m-independent MHCI binding of LILRB2 demonstrated by the SPR binding studies, which contrasts with LILRB1. LILRB2 uses the cleft of D1 to tightly associate with the membrane-proximal loop of the HLA-G α3 domain, including the aromatic amino acids Phe-195 and Tyr-197, which likely explains the high affinity of LILRB2 binding to HLA-G. The slight structural differences between members of the LILRB family achieve distinct ligand-binding specificities, including recognition of β₂m-free MHCIs, which might regulate immunologically relevant events. Moreover, because the MHCI-binding interfaces of LILRB2 are different from those of LILRB1, LILRB1- or LILRB2-specific inhibitors can be designed for blocking either interaction.

Methods

Preparation of Recombinant Proteins.

The refolded protein fragment of the two N-terminal extracellular Ig-like domains (residues 1–196) of LILRB2 was prepared as described in ref. 14. Soluble HLA-G Cys-42 Ser mutant was refolded from inclusion bodies of heavy chain and β₂m and the synthesized peptide (RIIPRHLQL) and purified as described in ref. 14. The purified HLA-G and LILRB2 were mixed at a 1:1 molar ratio at low concentrations (≈1 mg/ml of each) and concentrated by using an Amicon Ultra concentrator (Millipore, Billerica, MA) to 7–10 mg/ml. The buffer was finally exchanged to 20 mM Tris·HCl (pH 8.0)/100 mM NaCl for further experiments.

Crystallization, Data Collection, and Structural Determination.

Crystals suitable for data collection were obtained in the condition of 0.1 M Tris·HCl (pH 8.0)/45–50% PEG 400 at 20°C by the hanging drop vapor-diffusion method. All of the data sets were collected at 100 K. A 5-day soak in dehydration buffer (0.1 M Tris·HCl, pH 8.0/60% PEG 400) yielded crystals that diffracted to 2.5 Å. The final data set was collected at beamline BL41XU at SPring-8. Data were processed and scaled with the HKL2000 program package (29). The crystals belonged to the space group P3₁21 (a = b = 81.4 Å, c = 186.7 Å, γ = 120°). The structure was determined by molecular replacement by using Amore in the CCP4 suite (30) with coordinates of HLA-G (PDB ID code 2D31) and LILRB2 (PDB ID code 2GW5) as the search probes. One clear solution was found by using the reflections in the range of 20–3.5 Å. The structure was further refined with the individual B factor and positional refinements of CNS (31) by using the 50- to 2.5-Å reflections and alternating with manual rebuilding in the interactive graphics program O (32). The R_free and R_work of the final model were 27.9% and 23.5%, respectively. Intermolecular contact atoms were identified by using the program CONTACT (CCP4 suite) (30). The distances of each pair of atoms were determined by following the report of Sheriff et al. (33). Buried surface areas were calculated by using NACCESS with a 1.4-Å probe radius. The figures of structural information were prepared by using PyMOL (35) and GRASP (36).

Binding Analysis Using NMR.

Uniformly ¹⁵N-labeled β₂m ([¹⁵N]β₂m) was prepared from Escherichia coli BL21 cells carrying an expression plasmid grown in M9 minimal medium containing 1 g/liter ¹⁵NH₄Cl and purified by the same method as the unlabeled protein. HLA-Cw7 associated with ¹⁵N-labeled β₂m (HLA-Cw7/[¹⁵N]β₂m) was prepared in the same way as normal MHCIs. [¹⁵N]β₂m and HLA-Cw7/[¹⁵N]β₂m dissolved in the buffer (10 mM sodium phosphate, pH 6.5/10% ²H₂O) were concentrated to 0.1 mM.

LILRB1 and LILRB2 were dissolved in the same buffer used for β₂m. For the direct observation of LILR binding interfaces of β₂m as either the free or conformed (heavy chain/peptide/β₂m) form, a series of ¹H-¹⁵N heteronuclear sequential quantum correlation spectra at 2:1 (LILRB:β₂m) or 1:1 (LILRB:HLA-Cw7) ratios were measured at 30°C with a DMX500 spectrometer (Bruker, Billerica, MA) equipped with a cryogenic probe.

Binding Analysis Using SPR.

The soluble form of LILRB1 and C-terminal biotinylated MHCIs were prepared as described in refs. 14 and 34. The soluble LILRB1 and LILRB2 were dissolved in HBS-EP buffer (10 mM Hepes, pH 7.4/150 mM NaCl/3.4 mM EDTA/0.005% Surfactant P20) (Biacore, Uppsala, Sweden). SPR experiments were performed by using a BIAcore2000 system (Biacore AB). All of the biotinylated proteins were immobilized on the CM5 sensor chip, on which streptavidin was covalently immobilized by amine coupling. Chemically biotinylated BSA was used as a control protein. The data were analyzed by using Biaevaluation 4.1 (Biacore AB) and ORIGIN 5 (Microcal, Northampton, MA) software.

Abbreviations

MHCI

MHC class I

LILR

leukocyte Ig-like receptor

SPR

surface plasmon resonance

β₂m

β₂-microglobulin

fHC

β₂m-free heavy chain

APC

antigen-presenting cell

PIR

paired Ig-like receptor