Abstract
HLA-G is a natural tolerogenic molecule and has the following unique features: seven isoforms (HLA-G1 to HLA-G7), formation of disulfide-linked homodimers, and β2-microglobulin (β2m)-free forms. Interestingly, individuals null for the major isoform, HLA-G1, are healthy and expressed the α2 domain–deleted isoform, HLA-G2, which presumably compensates for HLA-G1 function. However, the molecular characteristics of HLA-G2 are largely unknown. In this study, we unexpectedly found that HLA-G2 naturally forms a β2m-free and nondisulfide-linked homodimer, which is in contrast to the disulfide-bonded β2m-associated HLA-G1 homodimer. Furthermore, single-particle analysis, using electron microscopy, revealed that the overall structure and domain organization of the HLA-G2 homodimer resemble those of the HLA class II heterodimer. The HLA-G2 homodimer binds to leukocyte Ig-like receptor B2 with slow dissociation and a significant avidity effect. These findings provide novel insights into leukocyte Ig-like receptor B2–mediated immune regulation by the HLA-G2 isoform, as well as the gene evolution of HLA classes.
Introduction
HLA-G is a nonclassical MHC that is known as a natural tolerogenic molecule. HLA-G has seven isoforms; four are membrane-bound (HLA-G1 to HLA-G4), and three are soluble (HLA-G5 to HLA-G7) (Supplemental Fig. 1). The typical isoforms of HLA-G, HLA-G1 and HLA-G5 (hereafter referred to as HLA-G1, because they have essentially identical ectodomains), have a similar structure to classical HLA class I molecules (Fig. 1A). Additionally, HLA-G1 can form a disulfide-bonded homodimer through Cys42 located in the α1 domain (1, 2) (Fig. 1A). The HLA-G1 homodimer possesses two receptor binding sites and transmits the signals much more strongly than the monomer through the leukocyte Ig-like receptors (LILRs) B1 and B2 (2). Furthermore, they bind to the mouse LILRB ortholog, the paired Ig-like receptor B on APCs, and exhibit efficient anti-inflammatory effects on collagen-induced arthritis (CIA) in mice (3). As expected, the HLA-G1 homodimer showed greater functional effects than the monomer (3).
Dimer formation of HLA-G1 and HLA-G2. (A) Domain structure and crystal structures of the HLA-G1 monomer (PDB ID: 1YDP; left panel) and homodimer (PDB ID: 2D31; right panel). HLA-G1 is known to form a disulfide-linked homodimer through a Cys42. The Cys42 and the glycosylation residue, Asn86, are represented as green and black circles, respectively. HLA-G1 H chain (α1, red; α2, orange; α3, magenta); β2m, blue; peptide, black. (B) Domain structure of the HLA-G2 monomer and dimer. HLA-G2 possesses the same Cys42 (green circle) as HLA-G1, but it is unclear whether it forms a disulfide-linked homodimer or a noncovalently associated homodimer. (C) Preparation of rHLA-G2. Schematic representation of HLA-G2 refolded from protein expressed in E. coli (upper left panel). HLA-G2 and HLA-G2(C42S) were eluted before the 44-kDa marker molecule in gel filtration analysis (lower left panel). The eluted peak fraction of HLA-G2 did not form the disulfide-linked dimer, as assessed by nonreducing SDS-PAGE analysis (right panel).
Dimer formation of HLA-G1 and HLA-G2. (A) Domain structure and crystal structures of the HLA-G1 monomer (PDB ID: 1YDP; left panel) and homodimer (PDB ID: 2D31; right panel). HLA-G1 is known to form a disulfide-linked homodimer through a Cys42. The Cys42 and the glycosylation residue, Asn86, are represented as green and black circles, respectively. HLA-G1 H chain (α1, red; α2, orange; α3, magenta); β2m, blue; peptide, black. (B) Domain structure of the HLA-G2 monomer and dimer. HLA-G2 possesses the same Cys42 (green circle) as HLA-G1, but it is unclear whether it forms a disulfide-linked homodimer or a noncovalently associated homodimer. (C) Preparation of rHLA-G2. Schematic representation of HLA-G2 refolded from protein expressed in E. coli (upper left panel). HLA-G2 and HLA-G2(C42S) were eluted before the 44-kDa marker molecule in gel filtration analysis (lower left panel). The eluted peak fraction of HLA-G2 did not form the disulfide-linked dimer, as assessed by nonreducing SDS-PAGE analysis (right panel).
By comparison, details of the functions and structures of other HLA-G isoforms (Supplemental Fig. 1B) remain largely unknown. The functional importance of domain-deleted HLA-G isoforms is indicated by the existence of an HLA-G–null allele (HLA-G*0105N) that cannot encode functional HLA-G1. Nevertheless, healthy female individuals who are homozygous for the null allele exhibit normal reproductive success (4, 5). Thus, the HLA-G1 isoform does not appear to be essential for fetal survival, and other isoforms, including HLA-G2 and HLA-G6 (hereafter referred to as HLA-G2 because they have essentially identical ectodomains) are functionally adequate (6). HLA-G2–transfected cells were protected from killing by NK cells and CTLs in vitro (7). Recently, we reported the significant immunosuppressive effect of rHLA-G2 protein in CIA mice (8). Because it lacks an α2 domain, the HLA-G2 isoform is expected not to associate with β2-microglobulin (β2m) and to form a disulfide-linked homodimer via Cys42 (9), like HLA-G1; however, structural data on HLA-G2 that could confirm this have not been available. Furthermore, the Fc-fusion proteins of HLA-G2 bound to membrane-bound LILRB2 but not to LILRB1 (10); however, the comprehensive binding and structural characteristics of these interactions have not been studied.
In this study, surprisingly, biochemical analysis and single-particle reconstruction from electron microscopic images revealed that HLA-G2 naturally forms a nondisulfide-bonded β2m-free homodimer. This is completely different from the hypothesized dimer but is similar to HLA class II molecules. The formation of this novel structure by a nonclassical class I molecule could be relevant to ways in which HLA class I and class II molecules may have evolved from an ancestral MHC molecule. Furthermore, surface plasmon resonance (SPR) analysis revealed that HLA-G2 bound to LILRB2 at nanomolar-order affinity with avidity effects but did not bind to LILRB1. Taken together, these results provide novel functional and structural insights into the role of domain-deleted HLA-G2 (and HLA-G6) isoforms in the immune tolerance system and the high affinity of these ligands for LILRB2 expressed on APCs.
Materials and Methods
Production of recombinant proteins
DNA encoding the structured core α1–α3 domain region of HLA-G2 (and HLA-G6) were amplified, using 5′-GGAATTCCATATGGGCTCCCACTCCATGAGG-3′ as the forward primer and 5′-CCCAAGCTTACTGCTTCCATCTCAGCATGAGGGGC-3′ as the reverse primer, from pFLAG–rsHLA-G2 (9) and were ligated into the pGMT7 vector (HLA-G2–pGMT7). HLA-G2 and C42S mutant [HLA-G2(C42S)] polypeptides were expressed as inclusion bodies in Escherichia coli BL21(DE3)pLysS cells, refolded by the dilution method, and purified by size-exclusion chromatography using a Superdex 200 column (GE Healthcare). The N-terminal two ectodomains of LILRB1 and LILRB2 and biotinylated proteins were prepared as described previously (2).
Electron microscopy
For negative staining, purified HLA-G2 (10 μg/ml) was adsorbed by thin carbon films supported by copper mesh grids, which were rendered hydrophilic in advance by glow-discharge under low air pressure. Samples were negatively stained with 2% uranyl acetate solution for 30 s twice, blotted, and dried in air. Micrographs of negatively stained particles were taken using a JEOL 100CX or JEOL 1230 transmission electron microscope at ×60,000 with an acceleration voltage of 100 kV. A total of 8057 particles was selected by the program e2boxer, and contrast transfer function correction was applied to the images using the e2ctf program in the EMAN2 suite (11). Multireference alignment was applied for two-dimensional averaging and classification into 86 classes. Reconstruction and refinement were performed with EMAN2. All 86 classes were used for the three-dimensional (3D) reconstruction. Resolution was assessed by dividing the data into two subsets and was calculated using the independent 3D reconstructions of each, which were compared by Fourier shell correlation at a threshold of 0.5. In these experiments, a 3D structure of 2.5 nm resolution was obtained. For rotary shadowing of HLA-G2, protein solution supplemented with ammonium acetate (0.5 M) and glycerol (50% by volume) was sprayed onto freshly cleaved mica surfaces, which were subsequently dried under vacuum. Rotary shadowing with platinum/carbon (thickness, 2 nm) was performed at an angle of 8° using a Bal-Tec BAF 060 Freeze-Fracture System (Bal-Tec, Balzers, Lichtenstein), and the replicas were observed with a JEOL 1230 transmission electron microscope.
SPR
SPR experiments were performed using a Biacore 2000 or Biacore 3000 system (GE Healthcare). For binding analyses, the biotinylated LILRBs and control chemically biotinylated BSA were immobilized onto an SA-coupled CM5 chip (GE Healthcare). Soluble rHLA-G2 (rsHLA-G2) expressed in HEK293 cells, rHLA-G2, or C42S mutant proteins produced by E. coli were injected over the immobilized LILRBs in HBS-EP Buffer (GE Healthcare). For kinetic measurements, LILRB2 was immobilized at 200 or 500 response units. The data were analyzed using BIAevaluation version 4.1 (GE Healthcare). To record avidity effects, biotinylated rHLA-G2 was immobilized onto an SA-coupled CM5 chip.
Results and Discussion
HLA-G2 protein forms a nondisulfide-linked homodimer
We successfully produced the ectodomain of HLA-G2 by adding Lys-Gln to the C terminus to facilitate refolding (Fig. 1C). The free cysteine, Cys42, located in the α1 domain, forms an intermolecular disulfide bond in HLA-G1 homodimers (2). In contrast, the other free cysteine residue within the peptide binding grove, Cys147, in HLA-G1 does not exist in HLA-G2. Therefore, HLA-G2 is thought to exist as a disulfide-linked homodimer of α1–α3 domains through Cys42 (9) (Fig. 1B). Unexpectedly, however, refolded HLA-G2 and HLA-G2(C42S) proteins eluted at an elution volume indicative of a dimer species (Fig. 1C). This indicated that the intermolecular Cys42–Cys42 bond is not necessary for homodimer formation in solution. Furthermore, the HLA-G2 protein resolved as a monomer (21 kDa) by SDS-PAGE under reducing and nonreducing conditions (Fig. 1C). This further indicates that, in contrast to previous assumptions about its structure based on the HLA-G1 homodimer, HLA-G2 forms a nondisulfide-linked homodimer.
Single-particle analysis reveals the 3D structure of HLA-G2
We performed negative stain electron microscopy (EM) of refolded HLA-G2 proteins and successfully reconstructed the 3D structure of HLA-G2 at 2.5-nm resolution using the single-particle reconstruction technique. Additional size-exclusion chromatography was applied to the specimen just prior to grid preparation to eliminate denatured or aggregated protein. The peak top fraction that eluted at 1.59 ml was used for analysis (Supplemental Fig. 2A). From the negatively stained EM images, we found that the size and shape of HLA-G2 particles were less homogeneous than those of HLA-G1 (Supplemental Fig. 2A). Particle images of HLA-G1 show that it is a heart-shaped relatively rigid structure and the images are consistent with those from the crystal structure (i.e., 5 nm in width and 7 nm in height [Protein Data Bank (PDB) ID: 1YDP]). In contrast, HLA-G2 particles were somewhat swollen and heterogeneous, although the majority conformed to the reconstructed 3D structure (Supplemental Fig. 2A). This lower homogeneity of HLA-G2 may be caused by the relaxed structure of the top domains and the flexible foot domains. Notably, the poor electron density in the reconstructed 3D structure indicates a more flexible orientation of α3 domains. In agreement with these structural characteristics, the purified rHLA-G2 protein was found to degrade within the α1 domain, as determined by N-terminal protein sequencing (each N terminus residue of the fragments was Gly26, Phe33, Ser40, Arg75, or Thr80), when stored for 3 mo at 4°C.
The two-dimensional averaged images of HLA-G2 were classified into 86 classes and used for the reconstruction. We started the analysis with 8057 particles. However, because of the strict selection of good-quality particle images, a total of 2027 particles was included in the final reconstruction (Fig. 2A). Superimposition of the α1–α3 atomic coordinates of HLA-G1 onto the reconstructed EM volume at 2.5-nm resolution revealed that the HLA-G2 dimer has high similarity to the image for the HLA class II molecule, which is composed of two polypeptide chains (Fig. 2B). In the fitted image, the α1 domains coupled to each other, and the β-sheets at the bottom of the groove showed no steric hindrance. We applied the rotary shadowing EM technique to validate the structure of HLA-G2 obtained from the single-particle analysis. The replica images confirmed that the structural feature of individual HLA-G2 molecule fits well with the reconstructed 3D structure (Fig. 2C). In accordance with the results of biochemical analyses, Cys42 residues were located at the surface of the homodimer and were not juxtaposed for disulfide bond formation (Fig. 2B). The glycosylation site at Asn86, which is identical to HLA-G1, was also located at the surface, in accordance with the capacity for sugar modification (Fig. 2B), which may contribute to protein stability in tissues or blood.
3D reconstruction of the HLA-G2 homodimer. (A) Particle images and reconstruction of HLA-G2. Negatively stained EM images of HLA-G2 (upper panel). Projections from the reconstructed 3D volume (P) are presented with corresponding averaged images (A) and raw images (middle panel). The elucidated resolution was 2.5 nm by the forward scatter > 0.5 criterion (lower panel). Scale bar, 10 nm. (B) Reconstructed 3D volume of HLA-G2 (top panel) and superimposition of the α1α3 atomic coordinates of HLA-G1 onto the reconstructed EM volume by manual docking (middle panel). The positions of Cys42 and Asn86 are indicated in red and green, respectively. The crystal structure (bottom left panel) and the domain structure (bottom right panel) of HLA class II heterodimer (PDB ID: 3C5J). α-chain (brown), β-chain (green), peptide (black). (C) Replica images of the rotary shadowed HLA-G2 molecule (indicated by arrows in the left panel and enlarged in the right panel) fit well with the reconstructed π-shaped structure. Scale bar, 10 nm.
3D reconstruction of the HLA-G2 homodimer. (A) Particle images and reconstruction of HLA-G2. Negatively stained EM images of HLA-G2 (upper panel). Projections from the reconstructed 3D volume (P) are presented with corresponding averaged images (A) and raw images (middle panel). The elucidated resolution was 2.5 nm by the forward scatter > 0.5 criterion (lower panel). Scale bar, 10 nm. (B) Reconstructed 3D volume of HLA-G2 (top panel) and superimposition of the α1α3 atomic coordinates of HLA-G1 onto the reconstructed EM volume by manual docking (middle panel). The positions of Cys42 and Asn86 are indicated in red and green, respectively. The crystal structure (bottom left panel) and the domain structure (bottom right panel) of HLA class II heterodimer (PDB ID: 3C5J). α-chain (brown), β-chain (green), peptide (black). (C) Replica images of the rotary shadowed HLA-G2 molecule (indicated by arrows in the left panel and enlarged in the right panel) fit well with the reconstructed π-shaped structure. Scale bar, 10 nm.
HLA-G2 proteins bind to LILRB2 with a notable avidity effect
HLA-G2 bound to LILRB2, but not LILRB1, on transfected cells (10). HLA-G1 dimer exhibits remarkable avidity for LILRB1 and LILRB2 binding (2). In this study, we analyzed the LILR-binding characteristics of the HLA-G2 dimer using SPR. First, we compared the binding activity of rsHLA-G2 expressed by HEK293 cells (9) and refolded HLA-G2 to immobilized LILRB1 and LILRB2. Both HLA-G2s functioned equally well and bound to LILRB2 specifically with a slow dissociation rate, whereas no clear binding response was observed with LILRB1 (Fig. 3A). The binding activity of immobilized LILRBs was confirmed by checking the binding of HLA-G1 monomer and dimer (data not shown). In addition, HLA-G2(C42S) showed significant and similar binding activity to LILRB2 as did the wild-type (Supplemental Fig. 2B), which is in accordance with the above-mentioned biochemical results indicating the formation of a nondisulfide-bonded dimer. We next performed SPR analyses with immobilized HLA-G2. Similar to previous observations with HLA-G1 dimer (2), binding of LILRB2 to immobilized HLA-G2 was very weak, and dissociation was rapid (Fig. 3B). Equilibrium analysis showed that the conventional Scatchard plots of LILRB2 binding data were consistent with a simple 1:1 (Langmuir) binding model (Fig. 3B), and Kd was calculated to be 43 μM. This is 10-fold weaker than the binding affinity of LILRB2 to HLA-G1 (Kd = 4.8 μM) (Table I). This means that the effect of bivalent binding on affinity was not observed because the HLA-G2 homodimer was immobilized on the chip. Moreover, loss of the interaction with β2m likely is responsible for the lower-affinity binding of LILRB2 to immobilized HLA-G2 compared with its binding to HLA-G1. Therefore, HLA-G2 dimerization was expected to give a sensorgram pattern of high affinity with LILRB2 owing to the avidity effect (Fig. 3A), as shown by the HLA-G1 homodimer (2). The kinetic analysis of HLA-G2 binding to immobilized LILRB2 fitted reasonably well to the bivalent analyte model (Fig. 3C). In comparison with the affinity with the HLA-G1 homodimer, HLA-G2 showed a higher apparent affinity with slower dissociation (Table I). The apparent Kd calculated by the 1:1 Langmuir binding model was 1.7 nM. The avidity effects are likely to involve two distinct LILRB2 binding sites of the HLA-G2 dimer. In contrast, deletion of the α2 domain and β2m component presumably contributes to the change/shift of the LILRB2 binding site on HLA-G2, resulting in the acquisition of receptor specificity. The contact region of HLA-G1 with LILRB2 (12) is located at the exposed surfaces of the EM structure of the HLA-G2 homodimer, suggesting that two LILRs can simultaneously bind to the HLA-G2 dimer with bivalent binding (Fig. 3D, Supplemental Fig. 2C). To exert a notable avidity effect for immune suppression through LILRB2, the receptors may need to be expressed on immune cells at high density, such as may occur at immunological synapses. Future investigations of the structure of the LILRB2–HLA-G2 complex will provide a detailed molecular basis for this interaction.
Binding properties of HLA-G2. (A) Binding of rsHLA-G2 expressed in HEK293 cells (left panel) and refolded HLA-G2 from E. coli (right panel) to LILRs. Both HLA-G2s specifically bound to LILRB2 (black line) but not to B1 (gray line). rsHLA-G2 (1.1 μM) or HLA-G2 (1.1 μM) was injected over the control (BSA) and LILRBs. Representative data with the control response subtracted are shown. (B) Binding of LILRB2 to immobilized HLA-G2. LILRB2 (22 μM) was injected through the flow cell with control (BSA, dashed line) and HLA-G2 (solid line) (left panel). Representative data for the equilibrium binding of LILRB2 to HLA-G2 (right panel). Plots of the equilibrium binding responses of LILRB2 (●) versus concentration were fitted by the 1:1 Langmuir binding model (inset). Representative data are shown. (C) Kinetic analysis of HLA-G2 binding to immobilized LILRB2. HLA-G2 was injected at the indicated concentrations. Response curves (gray line) were fitted locally using a bivalent analyte model (black line). (D) Contacting residues in stick model (cyan) of HLA-G1 to LILRB2 (12) for the EM structure of HLA-G2. The arrows indicate presumable receptor binding sites.
Binding properties of HLA-G2. (A) Binding of rsHLA-G2 expressed in HEK293 cells (left panel) and refolded HLA-G2 from E. coli (right panel) to LILRs. Both HLA-G2s specifically bound to LILRB2 (black line) but not to B1 (gray line). rsHLA-G2 (1.1 μM) or HLA-G2 (1.1 μM) was injected over the control (BSA) and LILRBs. Representative data with the control response subtracted are shown. (B) Binding of LILRB2 to immobilized HLA-G2. LILRB2 (22 μM) was injected through the flow cell with control (BSA, dashed line) and HLA-G2 (solid line) (left panel). Representative data for the equilibrium binding of LILRB2 to HLA-G2 (right panel). Plots of the equilibrium binding responses of LILRB2 (●) versus concentration were fitted by the 1:1 Langmuir binding model (inset). Representative data are shown. (C) Kinetic analysis of HLA-G2 binding to immobilized LILRB2. HLA-G2 was injected at the indicated concentrations. Response curves (gray line) were fitted locally using a bivalent analyte model (black line). (D) Contacting residues in stick model (cyan) of HLA-G1 to LILRB2 (12) for the EM structure of HLA-G2. The arrows indicate presumable receptor binding sites.
Immobilized . | LILRB1 . | LILRB2 . |
---|---|---|
HLA-G1 monomer | 3.5 μM (2) | 4.8 μM (13) |
HLA-G1 dimer | 6.7 nM (2) | 0.75 μMa (2) |
HLA-G2 | No bindingb | 1.7 nMab |
Immobilized . | LILRB1 . | LILRB2 . |
---|---|---|
HLA-G1 monomer | 3.5 μM (2) | 4.8 μM (13) |
HLA-G1 dimer | 6.7 nM (2) | 0.75 μMa (2) |
HLA-G2 | No bindingb | 1.7 nMab |
Apparent affinity because of bivalent binding.
Determined in this study.
The first structure of the domain-deleted isoform of HLA-G, HLA-G2 (and HLA-G6) revealed a noncovalent homodimer that is quite different from the expected disulfide-linked homodimer structure. It has been reported that some disulfide-linked dimers of rHLA-G2 and rHLA-G6 expressed in transfected cells were observed by SDS-PAGE under nonreducing conditions (9, 10). We found a minor disulfide-linked dimer population by nonreducing SDS-PAGE (Fig. 1C); however, refolded HLA-G2 eluted as a dimer species under reducing conditions with 1 mM DTT showed only the monomer population (21 kDa) after SDS-PAGE (data not shown). Our result suggests that the extra dimer molecules were formed during the concentration and denaturation steps of the experiment and that the disulfide bond is not necessary to form a functional HLA-G2 molecule. In this study, we prepared functional rHLA-G2, which exhibited specific binding to LILRB2 in vitro (Fig. 3) and an anti-inflammatory effect in vivo (8). Furthermore, the EM structure of HLA-G2 is in good agreement with the fit for the HLA-G1 α1 and α3 domains without the Cys42–Cys42 disulfide bond. Therefore, HLA-G2 essentially resembles an HLA class II molecule. Although it is uncertain which class of MHC had evolved first at this moment, the structural similarity between HLA-G2 (class I isoform) and HLA class II could suggest that the evolutional advantage for the α2 domain deletion to facilitate unexpected formation of a homodimer might reflect the formation of class II proteins (or vice versa for the addition of the α2 domain).
Soluble forms of HLA-G at the maternal–fetal interface are believed to be particularly important in the suppression of maternal immune responses. HLA-G2 dimerization to increase the avidity of binding to the inhibitory LILRB2 receptor on maternal cells might be important for efficient immunosuppressive effects. This is supported by the observations of the existence of healthy individuals who do not have HLA-G1 and who, therefore, would be predicted only to have HLA-G2 forms capable of binding to LILRB2. In a previous study, we demonstrated the significant immunosuppressive effects of the HLA-G1 monomer/homodimer and HLA-G2 in a CIA mice model (3, 8). In terms of receptor distribution, paired Ig-like receptor B (mouse) and LILRB2 (human) are mainly restricted to APCs, whereas LILRB1 (human) is broadly expressed on immune cells. Thus, specific regulation through LILRB2 signaling would be more suitable to avoid side effects, such as infectious diseases and tumorigenesis, in humans. Therefore, HLA-G2 may represent a good candidate for use as an immunosuppressive reagent in medical treatments in which immune suppression of LILRB2-expressing cells, such as APCs, could be beneficial.
Acknowledgements
The rsHLA-G2 sample was kindly provided by J. S. Hunt and P. J. Morales (University of Kansas Medical Center).
Footnotes
This work was supported in part by the Platform for Drug Discovery, Informatics, and Structural Life Science and Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research KAKENHI (Grants 23770102, 25870019, 16J05871, and 22121007) and by grants from the Ministry of Education, Culture, Sports, Science and Technology of Japan and the Ministry of Health, Labour and Welfare of Japan, including the Program for Advancing Strategic International Networks to Accelerate the Circulation of Talented Researchers, and CREST, Japan Science and Technology. K.K. is supported by the Naito Foundation Subsidy for Female Researchers after Maternity Leave and the Support Office for Female Researchers at Hokkaido University.
The online version of this article contains supplemental material.
References
Disclosures
The authors have no financial conflicts of interest.