The Human Leukocyte Antigen G (HLA-G) is a nonclassical Major Histocompatibility Complex class I (MHC-I) molecule that possesses immunomodulatory properties (1). Its presence is tissue-restricted, being expressed in fetal tissues [trophoblast cells (2)] and constitutively expressed in adult thymic medulla (3), cornea (4), pancreatic islets (5), erythroid, and endothelial cell precursors (6). However, the expression of HLA-G can be induced in several conditions (1), including cancer (7, 8), transplantation (9), viral infections (10, 11), and autoimmune and inflammatory diseases (12, 13).

A well-recognized function of the HLA-G molecule in these pathological and physiological conditions is the inhibition of the cytotoxic activity of Natural Killer (NK) and CD8⁺ T lymphocytes. This function is mediated by interaction with leukocyte receptors, particularly with the Leukocyte Ig-like Receptors (LILRs), also known as Immunoglobulin-like Transcripts (ILT2, ILT4). ILT2 and ILT4 interact with several classical class I HLA molecules, but have higher affinity for HLA-G (14). ILT2 is expressed by B cells, some subtypes of T cells and NK cells, and all monocytes/dendritic cells (15). It is also described as a receptor for HLA-G associated with β2-microglobulin. On the other hand, ILT4 is myeloid-specific and only expressed by monocytes/dendritic cells (16), being capable of recognizing HLA-G free heavy chains (17, 18). Through these differentially expressed receptors, HLA-G can interact with all these different cell types, primarily inhibiting their functions. In addition, HLA-G may also generate regulatory/suppressor cells. For instance, human tolerogenic dendritic cells (DC-10) express high levels of membrane-bound HLA-G1 and are potent inducers of adaptive allospecific Type 1 regulatory T (Tr1) cells (19). The HLA-G gene is located within the MHC region, presenting low polymorphism, in contrast with the highly polymorphic classical class I genes, i.e., HLA-A, -B, -C (20). Geragthy et al. (21) first described the HLA-G gene in 1987, and its structure is homologous to other HLA class I genes. The HLA-G primary transcript may generate at least seven alternative splicing mRNAs that encode membrane-bound (HLA-G1, G2, G3, G4) and soluble (HLA-G5, G6, G7) protein isoforms (22–25). HLA-G1 may also be detected in plasma after proteolytic cleavage by metalloproteases, and presents the same domains (α1, α2, and α3) of classical class I molecules, being also associated with a β2-microglobulin. HLA-G2 is devoid of the α2 domain encoded by exon 3. HLA-G3 does not have the α2 and α3 domains encoded by exons 3 and 4, and HLA-G4 lost the α3 domain. The soluble HLA-G5 and HLA-G6 isoforms have the same extra globular domains as HLA-G1 and HLA-G2, respectively, and are generated by transcripts retaining intron 4, which block translation of the transmembrane domain (exon 5). The 5’ region of the intron, in the reading phase with exon 4, is translated into a stop codon and generates the HLA-G5 and HLA-G6 isoforms. These isoforms contain a specific 21 residues long tail involved in molecule solubility. The soluble HLA-G7 isoform is limited to the α1 domain and retains two intron 2 specific amino acids. All alternative transcripts are devoid of exon 7 (26, 27).

Sequence comparison of the HLA-G molecule to other HLA class I proteins reveals some interesting particularities. First, HLA-G has an unusually long half-life on the cell surface, resulting from the absence of an endocytosis motif in its truncated cytoplasmic domain (28). Second, HLA-G sequences have two unique Cysteine residues located at positions 42 and 147. Dimerization of HLA-G occurs through the creation of disulfide bonds between the two unique Cysteine residues at position 42 (Cys42-Cys42 bonds). Since all isoforms carry Cys42, all translated isoforms could potentially form membrane-bound homodimers, soluble homodimers, β2-microglobulin-free homodimers, and possibly homotrimers (associated or not to β2-microglobulin) (29, 30). Noteworthy, HLA-G dimers: i) do not induce significant structural changes to the main backbone of the protomers (i.e., chains forming the dimer) (17); ii) may exhibit distinct inhibitory functions as compared to monomers [e.g., dimers bind to ILT receptors with higher affinity in vitro (29) and in vivo (31)]; and iii) exhibit slower dissociation rates than monomers (17). ILT recognition of HLA-G dimers has a pivotal role on immune suppression at the maternal-fetal interface, possibly contributing to the prevention of pregnancy complications such as pre-eclampsia and recurrent miscarriages (17, 20).

Since HLA-G5 isoform has the same extra globular domains as HLA-G1, it could potentially be recognized by the same receptors. In fact, it has been reported that ILT2 can interact with β2-microglobulin-associated HLA-G5, while ILT4 could be able to recognize isoforms that are not associated to β2-microglobulin (17, 32). Such β2-microglobulin-free heavy chain has been detected in cell culture supernatants expressing HLA-G5 (33). It has also been shown that the expression of soluble HLA-G5 could inhibit the cytotoxicity of NK cells, and that the degree of inhibition was more evident when induced by HLA-G5, as compared to the membrane-bound HLA-G1. Most importantly, it was shown that the combination of HLA-G1 and HLA-G5 leads to significantly greater suppression than the effects of HLA-G1 or HLA-G5 alone (34). The direct involvement of HLA-G5 in inducing graft acceptance in vivo after human transplantation was provided by the observation that HLA-G5 purified from the plasma of transplanted HLA-G-positive patients suppressed alloproliferation of T cells in vitro (35).

Considering all the aforementioned structural diversity of known HLA-G isoforms, and the multiple roles of HLA-G in different immunological pathways, it is astonishing how little is known about the structure and dynamics of these molecules. As of today, there are only seven crystal structures of HLA-G receptors in the Protein Data Bank (PDB) (36). Note that these structures are limited to HLA-G1, and that even for this particular isoform they do not capture the full molecule (see Supplementary Table 1 ). In addition, there is only so much that can be understood from a static crystal structure in which relates to the dynamic behavior of these molecules. For instance, previous analysis of the membrane-bound HLA-G1 has indicated an oblique orientation of the protomers. Such orientation makes the ILT2 and ILT4 binding sites slightly more accessible to the interaction with these receptors (17). However, it does not tell us if this oblique orientation is stable in the soluble HLA-G1 dimer, or if other arrangements are possible. Finally, available structural data cannot inform us about the structure and dynamics of all other HLA-G alleles and isoforms.

As a step forward in addressing all these open questions, the present work reports for the first time the complete structure and dynamic behavior of the membrane-bound HLA-G1 model. In addition, it also characterizes the dynamics of the soluble HLA-G1 dimer. These efforts allowed for the first time the observation of a tilting movement of the membrane-bound HLA-G1 monomer, and the total rotational freedom of the HLA-G1 dimer in solution ( Figure 1 ). Finally, it investigates the stability of three different proposed structures for the soluble HLA-G5 isoform.

HLA-G plays an important role on the suppression of immune responses, and both membrane-bound and soluble isoforms may exert this function. As of September 2020, the IMGT-HLA database includes 80 HLA-G alleles, encoding 21 complete and 4 truncated proteins (HLA-G1*01:05N, G*01:13N, G*01:21N and G*01:25N) (68, 69). All alleles encoding the complete protein have the potential to i) form dimers through the conserved Cysteine at position 42, ii) form the seven commonly described HLA-G isoforms (HLA-G1 to HLA-G7), and iii) interact with the leukocyte receptors (26). This remarkable structural diversity must be studied in detail in order to clarify the diverse roles played by HLA-G molecules in both physiological and pathological conditions. Unfortunately, many questions remain unanswered about the structure, dynamics, expression and interaction patterns of different HLA-G alleles and isoforms. For instance, previous experimental studies have provided structures for the membrane-bound HLA-G1, and HLA-G1 dimer, either alone or interacting with other receptors. However, even in these cases the structures were incomplete. In addition, there was no available information on the dynamics of membrane-bound and soluble isoforms. Our goal was to conduct accurate structural modeling and molecular dynamics analysis of i) the complete membrane bound HLA-G1, ii) the soluble HLA-G1 dimer, and the soluble HLA-G5 monomer. This work moves the field forward, providing both insights on the dynamics of these complexes and complete 3D models that can now be used by other groups for further analysis.

Our complete model of HLA-G1 encompasses the heavy-chain (α1, α2, and α3 domains), connecting peptide, transmembrane portion and cytoplasmic tail of the most frequently observed HLA-G*01:01 molecule. The HLA-G*01:01 allele group encompasses 25 synonymous substitutions, as reported for the HLA-G*01:01:01:01 to HLA-G*01:01:25 alleles (68, 69). The associated light-chain (β2-microglobulin) was also included in our complete model. Finally, the RIIPRHLQL peptide, derived from histone H2A, was selected to be used in this study since (i) it is known to confer stability to the HLA-G molecule (64), (ii) is one of the most abundant peptides displayed by HLA-G (70), and (iii) was present in the cell cultures used for previous HLA-G X-ray diffraction studies (70). Note that an additional structural analysis comparing the binding of RIIPRHLQL with another HLA-G-binder, RLPKDFRIL, suggested that both peptides should provide similar level of stability to HLA-G complexes.

As expected, the MD simulations ( Figure 2B ) showed a stable membrane-bound HLA-G1 molecule, without evidence of unfolding of secondary structures (i.e., α-helices and β-sheets) ( Figure 2D ). In addition, both the β2-microglobulin and the coupled peptide ligand remained stably-bound during all simulations. Interestingly, we observed for the first time the natural “tilting” movement of the membrane-bound HLA-G1 in solution ( Supplementary Figure 1 ). This motion, in addition to the lateral swinging movement of the transmembrane portion in the lipid bilayer (spanning 22.5 Å), is reflected on the observed RMSD values. However, the PCA analysis shows great agreement between simulations in terms of sampled conformations. The superposition of frames from the beginning and end of the simulation ( Figure 2D ) also shows that all secondary and tertiary structures were preserved during MD. These results confirms the stability and compactness of the complete membrane-bound HLA-G1 model generated ( Figure 2C ), which could now be used for additional structural analyses.

We also report for the first time the complete model of the soluble HLA-G1 dimer ( Figure 4A ), and the great conformational flexibility of this molecule in solution ( Figure 4B ). While the disulfide-linked dimer is locked into a 45° angle between the protomers ( Figure 1 ), our simulations demonstrate that the soluble dimer is able to explore the full rotational flexibility enabled by the disulfide bridge ( Figure 4C ). Note that the secondary and tertiary structures of each protomer were very stable in solution ( Figure 4D ), despite overall dimer flexibility. The peptide-ligands also remained stably-bound to the respective clefts ( Figures 4A, B ). The PCA analysis of the three independent simulations of the soluble HLA-G1 dimer ( Figure 4B ) demonstrated that every dimer explored a different region of the conformational space, while still presenting similar collective motions, as demonstrated by the residue-residue distance comparisons ( Figures 5A, B ). A direct comparison between the PCAs for the membrane-bound HLA-G1 monomer and soluble HLA-G1 dimer is not possible, since the principal components used in each case reflect features that better capture the movements observed in each system. However, it is possible to observe that the soluble HLA-G1 dimer PCA captures a much greater conformational freedom.

Our HLA-G1 dimer model includes residues located at positions 195, 196, 197, 266, and 267, which were missing in the available crystal structure of the disulfide-linked dimer (17). All these residues are located in the α3 domain, where major leukocyte receptor binding sites are located. For instance, they include the putative binding sites for ILTs (residues 195 and 197) and CD8 (residues 223-229 residues). Note that these sites are very close to membrane in the membrane-bound HLA-G1 monomer ( Figure 3B ), which might limit interaction with these protein-ligands. It has indeed been observed that HLA-G1 dimers display higher affinity for leukocyte receptors than monomers (17, 20). This advantage has been associated with the 45° angle of the protomers in the disulfide-linked dimer, which can help exposing these sites for interaction (17, 67). Note that the free rotation of the protomers in the HLA-G1 soluble dimer, as observed in our simulations, would enable even greater exposure of these biding sites. In order to further explore our models and investigate the interaction with ILT4, we decided to conduct a protein-protein docking experiment with ClusPro. As expected, ClusPro successfully identified binding modes in which the D1 domain of ILT4 interacts with the α3 domain of HLA-G1, specifically involving F195 and Y197. Some of the predicted binding modes displayed ILT4 approaching HLA-G from the “top,” as previously described by Wang et al. (67). The authors of that study discuss the limited flexibility of ILT4 in terms of bending between Ig-like domains, and describe this “top-down” binding mode as the only interaction possible for membrane-bound forms of HLA-G1. Interestingly, in the absence of the membrane, ClusPro predicted better binding modes in which ILT4 approaches HLA-G1 from the “bottom,” while still preserving interactions between D1 and F195/Y197. Based on these results, we can speculate that higher affinity of soluble HLA-G1 dimer for ILT2/ILT4 ligands could be explained by the possibility of using this alternative “bottom-up” binding mode. Further computational and experimental studies would be necessary to investigate the occurrence and stability of alternative binding modes involving ILT2 and ITL4.

The interaction of HLA-G with T CD8⁺ cells may induce FasL up-regulation, soluble FasL secretion and CD8⁺ cell apoptosis by Fas-FasL interaction, whose binding sites have not been determined yet (71). Compared to classical class I molecules (e.g., HLA-A, -B, -C), HLA-G binds to CD8α/α loop (residues 223-229) with medium affinity (63, 64), thus inhibiting the T CD8⁺ cytotoxic function. Although little is known regarding the HLA-G dimer interaction with CD8, it is possible that the interaction confers increased avidity in a proper structural orientation, permitting an efficient signaling to CD8 as well as it does for ILT2/ILT4 (17, 38). The freedom of rotation reported in this study for the soluble HLA-G1 dimer, exposing two easily accessible binding sites for ILTs/CD8 receptors, corroborates the potential for multiple orientations of the dimer. Considering that these major leukocyte receptors are adjacent to each other, it is possible the formation of complexes containing multiple combinations of one HLA-G dimer and two leukocyte receptors (ILT2/ILT2, ILT4/ILT4; ILT2/ILT4, ILT2/CD8, ILT4/CD8, CD8/CD8).

It has been proposed that HLA-G could interact with the killer cell immunoglobulin-like receptor KIR2DL4 (25), and that such interaction could induce both inhibitory as well as activating signals (28, 72, 73). Although the inhibition of the innate and adaptive immune response is the most accepted role of HLA-G, activating responses have also been reported (74). The soluble form of HLA-G could be the natural KIR2DL4 ligand, since it accumulates in KIR2DL4⁺ endosomes and induces endosome signaling (75). In fact, structural representations (17) indicate that steric constraints would prevent KIR2DL4 from interacting with HLA-G dimers (65, 66, 75). Considering that KIR2DL binding residues are located in the α1 domain of the HLA-G molecule, and considering the HLA-G1 dimer rotation presented here, it is possible that the KIR2DL4 binding area would be more accessible in the soluble HLA-G1 dimer as opposed to the membrane-bound HLA-G1 dimer. However, it is important to stress that we have not tested this interaction in our study, and that recent studies have not found evidence of HLA-G/KIR2DL4 interaction. Once again, the models produced in this work can now be used to further investigate this potential interaction.

Contrary to what was observed for the aforementioned systems, greater instability was observed in two of our HLA-G5 models. Specifically, the only stable system was the one containing both the nonapeptide RIIPRHLQL and ß2-microglobulin ( Figure 6 and Supplementary Material – HLA-G5 Monomer, nonapeptide, and coupled β2-microglobulin Simulation Video). Previous studies have reported HLA-G5 isoforms both with and without β2-microglobulin (33, 76). However, our results suggest that the monomeric form of HLA-G5 would not be stable without these other chains ( Supplementary Material – HLA-G5 Monomer Simulation Video). On the other hand, it is possible that HLA-G5 dimers could be stable in the β2-microglobulin-free form, which was not tested here. For instance, the intronic tails of both protomers could interact with each other, not causing the effect of cleft deformation observed in our simulations ( Supplementary Material – HLA-G5 Monomer and nonapeptide Simulation Video). Such dimeric structures for HLA-G5 without β2-microglobulin could be similar to the dimers composed by the α1–α3: α1–α3 domains, as in the work published by Kuroki et al., in which HLA-G2 isoform (membrane bound α1–α3) naturally formed a β2-microglobulin-free homodimer which did not have disulfide bridges keeping the structures in place (77). Electron microscopy revealed that the general structure and domain organization of such HLA-G2 homodimers resembled those of class II HLA heterodimers (α1–α2: β1–β2) (20). Published data (77) described the binding of β2-microglobulin and β2-microglobulin-free forms of HLA molecules to members of ILT receptor family and demonstrated that, in addition to ILT4, β2-microglobulin-free structures are recognized by several other members of this receptor family. In fact, the “activating” members of the ILT family showed a preference for such structures. Therefore, it is possible that this could also be the case for HLA-G. This would support the notion that structural variations of HLA-G may be relevant in the modulation of biological function (32). It’s also intriguing to consider that, similar to classic class I HLA molecules, HLA-G may have activating receptors. In fact, it is possible that there are other receptors for HLA-G, specific for isoforms or not, and the study of HLA-G structures other than HLA-G1 and HLA-G5 may allow us to identify them (32).

In conclusion, the present study describes for the first time the complete membrane-bound HLA-G1 3D structure and its dynamic behavior in solution. Our study also described the dynamics of the soluble HLA-G1 dimer. Our simulations highlighted the great flexibility enabled by the disulfide bridge, which could even promote alternative binding modes with ILT2/ILT4 receptors. Our study of the HLA-G5 isoform and its structural alternatives demonstrated greater structural instability when the peptide or β2-microglobulin were absent. More comprehensive structural studies will be necessary to verify the existence of other structural conformations for HLA-G5. This work produced insights on the structure, dynamics, and interaction patterns of important HLA-G variants. It also produced 3D models that can now be used to further investigate these and other HLA-G molecules, to identify new HLA-G ligands, and to design potential pharmacological interventions.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material .

TA, DA and ED contributed to the conception and design of the study. TA generated all structural models. TA and DA performed simulations. TA, DA, MR and JA performed the analysis. SG and LK supervised the analysis. TA, DA and ED wrote the article. All authors contributed to the article and approved the submitted version.

This work was supported in part by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES). This work was also partially supported by the Cancer Prevention & Research Institute of Texas (CPRIT), through award number RP170508, and through a fellowship from the Gulf Cost Consortia on the Computational Cancer Biology Training Program (grant number RP170593). Finally, this work was also partially supported by a training fellowship from the National Library of Medicine Training Program in Biomedical Informatics (grant number T15LM007093), and by Rice University funds.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.