Structural and Functional Basis for LILRB Immune Checkpoint Receptor Recognition of HLA-G Isoforms Free

The leukocyte Ig-like receptor (LILR; also known as LIR, Ig-like transcript [ILT], CD85) family is a group of immune checkpoint receptors. LILR molecules are expressed on the surfaces of lymphoid and myeloid cells. They are categorized as a paired receptor family, consisting of the activating and inhibitory receptors responsible for finely tuning immune responses by regulating cell signaling (1, 2). LILRs are closely related to the killer cell Ig-like receptors (KIR; CD158) expressed on NK and T cells and recognize a wide variety of physiological and microbial ligands (3). The ligands of LILRs are very diverse and include angiopoietin-like proteins (ANGPTLs), neurologic proteins, Nogo, myelin-associated glycoprotein (MAG), β-amyloid, semaphorin-4A (Sema4A), and microbial molecules (3–5). Thus, LILRB1 and B2 regulate a broad range of biological functions in immunological, neurologic, developmental, and infectious responses.

The inhibitory LILRs, LILRB1 and LILRB2, reportedly recognize classical and nonclassical MHC class I molecules (MHCIs) to maintain immune self-tolerance, with a range of affinities mediated through two membrane-distal extracellular domains (D1 and D2) (6–8). Whereas LILRB1 is widely expressed in monocytes, dendritic cells, B cells, and subsets of NK and T cells, LILRB2 expression is restricted within the myelomonocytic lineage. Emerging evidence has highlighted the important regulatory roles of LILRB1 and B2 with other immunotherapy targets, such as programmed cell death 1 (PD-1), B- and T-lymphocyte attenuator (BTLA), and inhibitory KIRs.

We previously reported that LILRB1 and LILRB2 preferentially bind to HLA-G (8). HLA-G has unusual and distinct characteristics from classical MHCIs, such as low polymorphism, restricted tissue distribution, and several different isoforms (9, 10). In pregnancy, the semiallogenic fetus must evade the maternal immune responses. The trophoblast cells of the placenta do not express the major classical MHCIs (HLA-A and -B) for immune regulation. Instead, these cells express HLA-G, -C, -E, and -F, which suppress the maternal immune responses by binding to LILRB1, LILRB2 and other immune receptors. In addition to its natural immunoregulatory function, the abnormal expression of HLA-G on many kinds of human solid and hematological tumors and cells contributes to their evasion from immune surveillance. Therefore, the HLA-G/LILRB interaction is a promising target for cancer immunotherapy (11) and immune modulation during pregnancy (12).

Various forms of HLA-G1, such as Cys42-dependent disulfide-linked dimers of HLA-G1 and disulfide-linked dimers or multimers of β2-microglobulin (β2m)–free HLA-G1 H chains, have also been found in cell lines and in vivo (13–17) (Fig. 1). The placental villous cytotrophoblast (vCTB) cells, comprising the placental villi, lack β2m expression and exclusively express β2m-free disulfide-linked HLA-G1 and -G5 (17). In addition to HLA-G, the β2m-free HLA class I H chains (fHCs) are present in activated and neoplastic cells. Interestingly, the serum levels of fHCs are increased in multiple myeloma and correlated with disease activity (18). LILRB2 can recognize the fHCs, whereas LILRB1 cannot, even though the two receptors show high sequence identity (81%) (7). HLA-G2, which only contains α2 domain–deleted H chains, was also specifically recognized by LILRB2 (19). Therefore, the β2m-free MHCIs, including HLA-G1 and HLA-G2, selectively confer an inhibitory effect on LILRB2-expressing monocytes, macrophages, and dendritic cells.

Cellular and biochemical studies revealed that the disulfide-linked HLA-G1 dimer triggered much more efficient inhibitory signals through LILRB1 and B2 than the monomer (14, 16, 20). The crystal structures of the HLA-G1 monomer (7, 21) and dimer (20) provided the structural basis for this efficient signaling. The HLA-G1 dimer adopts an oblique configuration, exposing the sites for LILRB binding and enabling one HLA-G1 dimer to bind two receptors simultaneously (20). In this mode, the intracellular domains of the receptors are close to each other, thus facilitating highly efficient signaling. In collagen-induced arthritis (CIA) model mice, the administration of the recombinant HLA-G1 protein and its disulfide-linked dimer induced the sustained suppression of immune responses (22). Specifically, the disulfide-linked HLA-G1 dimer reduced the CIA scores more effectively than the monomer.

However, the molecular basis for the binding of HLA-G isoforms by LILRBs, especially the dependence on β2m for binding and the effect of the disulfide-linked dimerization of β2m-free HLA-G1, are still poorly understood. In this article, we report the binding characteristics of the recombinant β2m-free disulfide-linked homodimer of HLA-G1 to LILRBs. We also report the crystal structure of the LILRB1/HLA-G complex, in comparison with our previous LILRB2/HLA-G1 structure. These findings provide, to our knowledge, novel insights into the recognition of HLA-G isoforms by LILRs and the regulation of diverse immune responses.

The wild type and mutants of the two N-terminal Ig-fold domains (D1D2) of LILRB1 and LILRB2 were expressed in Escherichia coli and purified as previously described (8). The refolding mixture was concentrated and applied to a Superdex 75 column (GE Healthcare), equilibrated with 20 mM Tris-HCl (pH 8.0) and 100 mM NaCl, for size-exclusion chromatography. For final purification, LILRB2 was subjected to cation exchange chromatography on a Resource S column (GE Healthcare) equilibrated in 20 mM MES (pH 6.5). Soluble HLA-G1 proteins, including the Cys42Ser mutant, were prepared by refolding from inclusion bodies of the HLA-G1 H chain expressed in E. coli, with or without human β2m and the synthetic peptide (RIIPRHLQL). The refolding mixture of the β2m-free HLA-G1 WT dimer was concentrated and applied to a Superdex 200 column (GE Healthcare), equilibrated with HBS-EP buffer (10 mM HEPES, pH 7.4, 150 mM NaCl, 3.0 mM EDTA, 0.005% Surfactant P20), for surface plasmon resonance (SPR). To compare the properties of the refolded β2m-free HLA-G1 dimer, the recombinant FLAG/β2m-free HLA-G1 protein was generated in human embryonic kidney cells (HEK293) stably transfected with the construct encoding HLA-G5 (sG1) (17). The β2m-conformed HLA-G1C42S monomer for the structural analysis and the mutants for the binding analysis were purified on a Superdex 75 column (GE Healthcare), equilibrated with 20 mM Tris-HCl (pH 8.0) and 100 mM NaCl, and further purified by anion exchange chromatography on a Resource Q column (GE Healthcare) in 20 mM Tris-HCl (pH 8.0). The proteins with a C-terminal biotinylation enzyme (birA) recognition tag (GSLHHILDAQKMVWNHR) were prepared in the same manner as those without the tag and enzymatically biotinylated.

SPR experiments were performed using a BIAcore2000 or BIAcore3000 (GE Healthcare). Refolded biotinylated HLA-G1 dimers (20) were coupled onto SA-immobilized CM5 chips. The HLA-G1 H chain dimer remaining on the sensor chip after washing, by an injection of glycine (pH 2.0) for 30 s as described (7), was used as the β2m-free HLA-G1 homodimer protein. The β2m-free HLA-G1 expressed by HEK293 cells was directly immobilized by amine-coupling on the CM5 chips. For binding assessments and equilibrium binding assays, the WT and mutant LILRB proteins were injected over immobilized HLA-G1 proteins. The affinity constants of LILRB2 binding to the immobilized β2m-free HLA-G1 homodimer were determined by an equilibrium binding analysis.

To confirm the avidity binding effects, the biotinylated LILRBs were immobilized (2000 RU for binding check and 250 RU for kinetic experiments), and the refolded β2m-free HLA-G1 and the β2m-free HLA-G1 expressed by HEK293 cells (17) in HBS-EP buffer were injected. As a control protein, chemically biotinylated BSA was used in this study. Data were analyzed using the BIAevaluation software, version 4.1 (GE Healthcare), and Origin7 (Origin Lab).

Purified HLA-G1 and LILRB1 were mixed at a 1:1 molar ratio for crystallization (23 mg/ml total protein concentration). Microcrystals were initially obtained under conditions with 0.2 M sodium iodide/0.1 M Bis Tris propane (pH 6.5)/25% PEG 1500 at 293 K, by the sitting drop vapor diffusion method. Subsequently, crystals suitable for data collection were obtained under conditions with 0.2 M sodium iodide/0.1 M Bis Tris propane (pH 8.5)/20% PEG 3350 at 293 K, by the hanging drop vapor diffusion method. Streak seeding with a whisker was necessary to obtain well-formed crystals suitable for x-ray diffraction. The crystals were transferred to a solution containing 25% (v/v) glycerol for a few seconds for cryoprotection and then flash cooled. All data sets were collected at 100 K. The final data set was collected at beamline BL5A at the Photon Factory. The crystals belonged to the space group H32 (a = b = 164.38 Å, c = 326.84 Å, α = β = 90°, γ = 120°) and contained two HLA-G1/LILRB1 complexes per asymmetric unit. The data were processed and scaled with the HKL2000 program package (23).

The structure was determined by molecular replacement, using Molrep (24) in the CCP4 suite (25). The coordinates of HLA-G1 and LILRB1 were used as the search probes and were obtained from the structures of the LILRB2/HLA-G (Protein Data Bank identifier [PDB ID] 2DYP) and LILRB1/UL18 (PDB ID 3D2U) complexes, respectively. One clear solution was found by using the reflections in the range of 20–3.5 Å, and the initial model was refined using phenix.refine (26) and refmac5 (27), alternating with manual rebuilding using Coot (28). The R_free and R_work of the final model were 28.88 and 26.13%, respectively. Atomic coordinates and structure factors were deposited in the Protein Data Bank (https://www.rcsb.org/) under the accession code 6K60. Buried surface areas and intermolecular contact atoms were calculated by the PISA server (29) and the CONTACT program in the CCP4 suite. The shape complementarity of the two interacting molecular surfaces was quantified with the program SC (30) in the CCP4 suite. Structural figures were prepared using PyMOL (https://pymol.org/2/).

HLA-G1 forms a β2m-free disulfide-linked isoform in placental trophoblast cells (17) (Fig. 1). The recombinant β2m-free disulfide-linked HLA-G1 proteins expressed by HEK293 cells were previously characterized (17). These molecules tend to dimerize, as illustrated by PAGE analyses under nonreducing conditions (17). Although these proteins had LILRB2 binding activity, they were not suitable for the determination of binding characteristics. Therefore, we employed the refolding method using HLA-G1 H chain inclusion bodies, without any peptides or human β2m, to prepare the β2m-free disulfide-linked HLA-G1 isoform. A gel filtration analysis revealed that the refolded HLA-G1 H chains migrated as dimers or multimers (Fig. 2A), and SDS-PAGE analyses, under reducing and nonreducing conditions, demonstrated that the dimeric and multimeric species were separated (Fig. 2A). No β2m-free HLA-G1 H chain monomer fractions were detected in the gel filtration chromatogram.

Previous SPR studies showed that the β2m-free HLA-G1 H chains immobilized on the sensor chip could bind to LILRB2 but not to LILRB1 (7). To determine the effect of the disulfide-linked dimerization of the HLA-G1 H chain on the LILRB2 binding, SPR analyses were performed using the immobilized β2m-free HLA-G1 H chain dimer treated with an acidic solution. This method is commonly used for the preparation of fHCs on the cell surface. LILRB1 (∼52 μM) showed little binding responses to the refolded β2m-free HLA-G1 isoform, whereas LILRB2 bound in a dose-dependent fashion (Fig. 2B, 2D). By contrast, both LILRB1 and LILRB2 bound to the β2m-conformed HLA-G1 dimer (Fig. 2B). This binding specificity was consistent with the results obtained with the β2m-free disulfide-linked HLA-G1 proteins expressed by HEK293 cells, as shown in Fig. 2C. The LILRB2 binding to the β2m-free HLA-G1 dimer was weak and exhibited fast dissociation. The equilibrium analysis using a simple 1:1 (Langmuir) binding model (Fig. 2D inset) showed a calculated dissociation constant of K_d >50 μM, which is more than 10-fold weaker than the binding affinity of LILRB2 to β2m-conformed HLA-G1 (8).

Because the LILRBs bound to the β2m-conformed HLA-G1 homodimer and the HLA-G2 homodimer with a strong avidity effect (19, 20), we next performed SPR analyses in the opposite direction, using the diluted SEC fraction mainly containing the β2m-free HLA-G1 H chain dimer (fraction B in Fig. 2A), without any concentration or buffer exchange step, to avoid multimerization or aggregation. Although LILRB1 showed no binding, LILRB2 bound the β2m-free disulfide-linked HLA-G1 isoforms with slow dissociation rate (Fig. 2E).

These binding analyses in both orientations indicated that the disulfide-linked β2m-free HLA-G1 dimer binds to LILRB2 with large avidity effect. The kinetic analysis revealed that apparent K_d calculated by the simple 1:1 binding model (local fitting, χ² value is 2.2) is 2.4–4.6 nM (Fig. 2F), indicating the much higher K_d of the monomeric form (K_d > 50 μM, Fig. 2D) and apparent K_d of the β2m-conformed HLA-G1 dimer form (∼750 nM), reported previously (20).

We previously determined the crystal structure of the HLA-G1/LILRB2 complex. However, the structural bases for the β2m dependency and the avidity effects of the LILR recognition of HLA-G isoforms remained largely unknown. In this study, we performed a crystallographic study of LILRB1 complexed with an HLA-G1 isoform to compare the molecular bases for the binding of the same HLA ligand to different LILRBs. The two N-terminal Ig-like domains (D1–D2) of the LILRB1 extracellular region and the HLA-G1 complexes with β2m and the RIIPRHLQL peptide were prepared by standard refolding methods, as previously reported (8) (Supplemental Fig. 1A, 1B). A 1:1 (LILRB1/HLA-G1) mixture was concentrated to 23 mg/ml in 20 mM Tris-HCl (pH 8.0) with 100 mM NaCl. Crystals of the complex between HLA-G1 and LILRB1 were successfully obtained by the hanging drop vapor diffusion method, under conditions with 0.2 M sodium iodide/0.1 M Bis Tris propane (pH 8.5)/20% PEG 3350 at 293 K (Supplemental Fig. 1C). The complex structures were solved by molecular replacement. The crystallographic data and refinement statistics are summarized in Table I, and the 2mFo-DFc map of the interface between HLA-G1 and LILRB1 is shown in Supplemental Fig. 2A.

The asymmetric unit of the crystals includes two complexes. The structure of one of the complexes (complex 1) was fully determined, but the other (complex 2) showed poor electron density for the regions of LILRB1 distant from the interface with HLA-G1 (Supplemental Fig. 2B). The overall complex structures demonstrated that LILRB1 recognizes both the α3 domain and β2m of HLA-G1, which are far from the peptide-binding region (Fig. 3A). LILRB1 binds to β2m via the interdomain region (site 1) and to the α3 domain by the N-terminal D1 domain (site 2) (Fig. 3B, 3C).

In the complex structures, the bottom side of the α3 domain and the top side of LILRB1 showed high B factors. The electron densities for these regions were much worse than those for the D1–D2 interdomain region. This finding indicates that this site in the complex is flexible and/or adopts multiple conformations, although the overall binding mode, especially the center of the LILRB1/HLA-G1 interface, is maintained. Interestingly, complexes 1 and 2 were similar, although their binding modes were shifted (described in detail later, Fig. 3). Specifically, the relative orientation between the HLA-G1 α3 domain and the D1 domain of LILRB1 (site 2) was different (Fig. 3A). Complex 1 exhibited similar surface complementarity but a much lower (2-fold) interface area in site 2 (0.512 and 143 Ų) than complex 2 (0.524 and 268 Ų). In addition, although a wide range of interactions between LILRB1 and β2m were maintained in both complexes 1 and 2 (site 1, Fig. 3B), fewer interactions were maintained with the α3 domain (site 2, Fig. 3C, Supplemental Fig. 2C). This finding clearly reveals that the main interaction site for LILRB1 is β2m (site 1) and the binding to the α3 domain is additional and flexible (site 2).

The LILRB1/HLA-G complexes clearly showed higher structural similarity to LILRB1/HLA-A2 (root-mean-square deviation 1.058 Å with 404 Cα atoms) than the LILRB2/HLA-G complex (root-mean-square deviation 1.254 Å with 475 Cα atoms) (Fig. 4). Therefore, the binding modes of the LILR–MHCI (and HLA-related molecules) complexes are determined solely by the LILRs. The interface area of β2m and its surface complementarity (site 1) were relatively similar in all of the complexes: 569 Ų/0.677 (complex 1), 673 Ų/0.599 (complex 2), and 635 Ų/0.698 (LILRB2 complex). A comparison of the HLA-G complexes between LILRB1 and LILRB2 revealed a distinct LILRB binding mode (5–10 Å toward the bottom of the α3 domain) (Fig. 5A). This results in a smaller binding area and less surface complementarity on the α3 domain (site 2, 143 Ų/0.512 [complex 1] and 268 Ų/0.524 [complex 2]) in the LILRB1 complex than in the LILRB2 complex (630 Ų/0.538). Therefore, this finding indicates that LILRB1 predominantly recognizes β2m, as compared with the α3 domain, thus accounting for the β2m dependence for binding.

Supplemental Table I provides a summary of the intermolecular contacts in the two complex states in the asymmetric unit. There are many hydrogen bond interactions in the β2m binding site (site 1), such as 1) main chain–main chain, Ser88/Ala98, Thr86/Ile100, Gln2/Gln125, and Gln2/Ala127 (β2m/LILRB1); and 2) others, such as Ser88/Gly97, Ser88/Ala98, and Gln89/Gln18 (β2m/LILRB1), together with hydrophobic interactions (Fig. 3B). Notably, in LILRB1, the Nε of the indole ring of Trp67 formed a hydrogen bond with the main chain carbonyl group of Ile92 of β2m. However, in the LILRB2/HLA-G complex, the indole ring of Trp67 in LILRB2 formed a π-cation interaction with Lys91 of β2m (Supplemental Fig. 3A) (7). Because the side chain of Val183 is small, Trp67 interacted with Lys91 of β2m in the LILRB2 complex (Supplemental Fig. 3A). Glu184 (LILRB1), corresponding to Val183 (LILRB2), has a larger side chain that formed a salt bridge with Lys91 and further filled the space, causing Trp67 to flip to the opposite side, close to the main chain of Ile92 (β2m), in the LILRB1/HLA-G complex (Supplemental Fig. 3). These interactions were maintained in all of the LILRB1 complexes. To determine whether these differences contribute to the distinct LILRB binding modes, we prepared the LILRB1W67Y and W67A mutants to assess the effect of Trp67 on the unique recognition modes of LILRB1 and LILRB2. The SPR analysis revealed that the K_d values for LILRB1W67Y and LILRB1W67A binding to HLA-G1 are marginally smaller than that for LILRB1WT (Table II). These results suggested that Trp67 does not contribute to the HLA-G1 binding as much as the β2m dependence, and the other interactions at sites 1 and 2 could maintain the LILRB/HLA-G1 binding mode.

LILRB1 uses the D1 domain to contact the α3 domain of HLA-G. We previously reported that HLA-G shows a slightly higher (two- to several-fold) affinity toward LILRB1 and LILRB2 than other classic MHCIs and that residues Phe195 and Tyr197 on HLA-G (site 1) could account for this difference in binding affinity (7). In complex 2, Tyr38 in LILRB1 forms a π−π interaction with Phe195, whereas Tyr197 is located far from the interface with LILRB1 (Fig. 5B). No electron density for either Phe195 or Tyr197 was observed in complex 1, suggesting that the region around Tyr38 of D1 in LILRB1 does not form a strong binding surface (Fig. 3C, Supplemental Fig. 2C). SPR analyses of LILRB1 binding to the HLA-G F195A and Y197A mutants showed slightly lower affinities (Table II), suggesting that these amino acids are somewhat involved in LILRB1 binding. In contrast, LILRB2 used Arg36 together with Tyr38 to form a π-cation interaction with Tyr197 in HLA-G in the complex between the two molecules (Fig. 5B).

We previously demonstrated that LILRB1/MHCI recognition shows fast kinetics and is remarkably entropy driven (31). In this study, we determined the structure of the LILRB1/HLA-G1 complex. The absence of significant hydrophobic interactions at the interface presumably contributes a favorable entropic effect upon binding. The two distinct structures represent snapshots of the LILRB1/HLA-G1 complex and clearly show the ensembles of the MHCI complex states, consisting of the flexible and plastic recognition of the α3 domain of HLA-G1 by the LILRB1 D1 domain, and the maintenance of the β2m recognition in both complexes. This finding reveals the large number of possible configurations in the complex states, resulting in entropic-driven binding. These results indicate that LILR–MHCI recognition involves rough discrimination, ensuring fast kinetics, low affinity, and approximate specificity. This is beneficial for the efficient surveillance of LILR-expressing immune cells and is typical of immune checkpoint cell-surface receptor ligand interactions.

We previously reported the crystal structure of the disulfide-linked HLA-G1 homodimer, showing that the unique dimer orientation is reasonable to permit efficient signaling (20). The crystal structures of the LILRB1/HLA-G1 complex from this study were superimposed on the HLA-G1 dimer structure, as illustrated in Fig. 6. The locations of the D1 domains of LILRB1 and LILRB2 on HLA-G1 were distinct, but the locations of their D2 domains were similar. This finding suggests that the β2m-conformed HLA-G1 dimer enables efficient LILRB1- and LILRB2-mediated signaling in similar manners. We also showed that LILRB2 had remarkable avidity for binding to the β2m-free HLA-G isoforms. The dimerization of β2m-free HLA-G1 did not inhibit LILRB2 binding but instead maintained the LILRB2 specificity with significant avidity (Fig. 2). This binding characteristic is similar to that observed with the β2m-conformed disulfide-linked HLA-G1 homodimer. Notably, despite the significant reduction of the LILRB2 binding surface to HLA-Gs in the absence of β2m, LILRB2 still bound to β2m-free HLA-Gs. This raises the possibility that LILRB2 might use another site for binding to the newly exposed surfaces of β2m-free HLA-Gs. Furthermore, β2m-free HLA-B27 homodimers, which may play a pathogenic role in ankylosing spondylitis, exhibited similar binding properties to LILRB2, as well as KIR3DL2 and KIR3DL1 (32, 33). Interestingly, another nonclassical MHCI, HLA-F, is also expressed in placenta and exists as open conformers, which sometimes associate with other MHCIs and could potentially bind LILRB1 and B2 (34). It is interesting that the β2m-free HLA-G1 dimer has the ability to cis-associate with other MHCIs or receptors, such as HLA-F. Future investigations will clarify the functional similarities and differences between HLA-G and other open MHCI conformers, such as HLA-B27 and HLA-F, and their relationships to immune tolerance and disorders.

The present crystal structure of the LILRB1/HLA-G1 complex provides the set of complex structures with the same MHCI ligand and with the ligand binding domains of different LILRs. Glu184 in LILRB1, corresponding to Val183 in LILRB2, is located in the loop exposed on the kinked D1–D2 interdomain angle and at the center of the interface with HLA-G1 (Fig. 3, Supplemental Fig. 3). The substitution of Val183 (LILRB2) with Glu184 (LILRB1) changes the positioning of Trp67, which serves as a fulcrum for determining the overall orientation of LILR–MHCI recognition (Supplemental Fig. 3). These observations are consistent with the previously proposed grouping of LILR members (groups 1 and 2) (35). However, the mutations of Trp67 did not significantly affect the HLA-G1 binding (Table II). Our preliminary SPR experiments with the LILRB1E184V and LILRB2V183E mutants showed that these mutations cause subtle reductions in the binding to β2m-conformed HLA-G1 but still maintain the binding to the β2m-free HLA-G1 dimer, indicating that these structural features are not the main driving forces for the binding specificities. In contrast, LILRB1 clearly uses a wide area of the interdomain region for binding to β2m and the D1 domain for the small and flexible interfaces with the α3 domain. This finding explains the β2m-dependent MHCI binding of LILRB1.

The LILRB1 and LILRB2 receptors are readily detectable on both blood monocytes and decidual macrophages. The placental β2m-free HLA-G1 probably diffuses into the decidua from invasive vCTB cells and the β2m-free HLA-G1 dimers or multimers circulating in maternal blood bind to blood monocytes. Our kinetic analysis revealed that the apparent K_d of the binding of the disulfide-linked β2m-free HLA-G1 dimer fraction to immobilized LILRB2 showed much higher affinity than that of the β2m-conformed HLA-G1 reported previously (Fig. 2) (20). Therefore, these results indicated that the soluble disulfide-linked β2m-free HLA-G1 protein, rather than the β2m-conformed homodimer, would dominate the LILRB2-mediated immune cell functions. To exert this dominant immune suppression through LILRB2, the receptors may need to be expressed on immune cells at a relatively high density. Future investigations of the expression profiles of each HLA-G form and LILRB2 will clarify the mechanism of immune regulation in the placenta. Thus, the β2m-free HLA-G1 dimers are likely to be critically important for the establishment of tolerance during human pregnancy by promoting an immunosuppressive profile in monocytes/macrophages.

Recently, we reported that HLA-G2, a β2m-free and α2-domain deleted HLA-G isoform, unexpectedly formed non–disulfide-linked homodimers similar to those of MHC class II molecules (19). The β2m-free HLA-G2 homodimer exhibited the highest affinity toward LILRB2, with remarkable avidity effects (19). The β2m-free HLA-G1 homodimers and the β2m-free HLA-G2 isoform both showed augmented avidity effects for LILRB2 binding. Thus, our results suggest that efficient LILRB2 binding requires the plasticity and/or flexibility of the complex states to produce a high avidity effect and enable efficient signaling (Fig. 6). The administration of HLA-G2 homodimers in the CIA mouse model produced a remarkable anti-inflammatory effect, which lasted for several months (36). Mouse PIR-B and human LILRB2 are mainly restricted to APCs, whereas human LILRB1 is expressed on a broad range of immune cells. This suggests that the specific regulation through LILRB2 signaling would be effective and suitable in vivo to avoid unexpected side effects. Therefore, β2m-free HLA-G homodimers, including HLA-G2 and other fHCs specific to LILRB2, could be biologic candidates for therapeutic medical applications. In contrast, besides the placenta, the impaired expression of β2m and the upregulation of MHCI fHC are also observed in tumor cells, raising the possibility that they contribute to the LILRB2-mediated immune escape (18, 37, 38). Immune checkpoint drugs designed to inhibit these interactions could be important for cancer therapy. Future structural studies of the LILRB2 complex of β2m-free HLA-G homodimers will provide important insights for the design of immune checkpoint drugs.

In summary, the crystal structures of the LILRB1/HLA-G1 complex clearly revealed the flexible states upon binding. These results provide novel functional and structural insights into LILR–MHCI recognition and the binding characteristics of the LILR family of immune checkpoint cell-surface receptors, which exhibit fast kinetics and low affinities. Furthermore, the strong LILRB2-specific binding capacity of β2m-free HLA-G1 homodimers, revealed by the SPR analysis, suggests the functional importance of the β2m-free HLA-G isoforms for the treatment of autoimmune diseases and cancers.

Note added in proof. While this manuscript was under revision, Wang et al. (39) reported the crystal structure of an HLA-G/LILRB1 (four domain) complex for which the binding mode is similar to those determined in this study.

We thank the beamline scientists at the Photon Factory and SPring-8 for their assistance with x-ray diffraction data collection. We also thank Y. Kasai for technical support and T. Tadokoro, D. L. Langat, D. Wheaton, B. Ding, and S. Kollnberger for helpful discussions.

The authors have no financial conflicts of interest.