Structural Characteristics of RAGE and Its Ligand-Binding Sites

1. Introduction

The Receptor for Advanced Glycation Endproducts (RAGE) is a transmembrane multi-ligand receptor, which serves as a progression factor amplifying immune and inflammatory responses in (1) pathogenesis and therapy of cancer, (2) neurological disorders, inflammation and apoptosis, including kidney disease and Alzheimer’s disease (AD), (3) diabetes-related cardiovascular complications, such as thrombosis and myocardial infarctions, and (4) in impairing the physiological function of stem cells, all due to the buildup of numerous Advanced Glycation Endproducts (AGEs), RAGE agonists and other RAGE ligands, during these conditions [1,2,3,4].

RAGE is expressed in many cell types and its expression increases as a direct result of ligand accumulation. In therapy of cancer invasion and metastasis, the role of various AGEs together with another group of RAGE ligands, the S100 proteins, is particularly important [5,6,7,8,9]. There, it was demonstrated that RAGE interactions with AGEs and S100 proteins in tumor cells increased activity of NF-κB and expression of proinflammatory cytokines TNF-α and IL-6 [10,11,12]. In inflammation and neurological disorders, RAGE is involved on multiple levels [13,14,15,16,17]. For example, during development of AD there is buildup of amyloid beta (Aβ) peptides, which is a known group of RAGE agonists [13,16]. Here, apart from being a multi-ligand receptor, RAGE also functions in tandem with other receptors, such as Toll-Like Receptors (TLR), where RAGE enhances TLR response [18,19]. In diabetes, RAGE signaling was recently shown to regulate progression of diabetic complications [20,21]. Independently from diabetes-induced cardiovascular complications, the AGE-RAGE axis was found to be involved in initiation and progression of other types of cardiovascular diseases [22]. Thus, it is not surprising that both RAGE and its various ligands show significant potential as drug targets.

The sequence-based domain structure of RAGE, which includes one Transmembrane domain (TM), one possible Immunoglobulin V-like (IgVRAGE) domain and two possible Immunoglobulin C-like (IgCRAGE) domains, had been described in Neeper et al. [23] even before the structural immunoglobulin fold was officially reviewed and defined by Bork et al. [24], and Harpaz and Chothia [25]. The RAGE domain structure can be schematically described as Nterminus-IgVRAGE-IgC1RAGE-IgC2RAGE-TM-Cterminus. Topologies of IgVRAGE, IgC1RAGE and IgC2RAGE domains within the Immunoglobulin-like fold (Ig-like fold; full name Immunoglobulin-like β-sandwich fold; SCOP ID: 2000051; InterPro ID: IPR013783) are as defined in Bork et al. [24] for the respective classical V-type Ig-like and C-type Ig-like domains, IgV and IgC domains, respectively [26]. Similar to classical IgV domains, the IgVRAGE domain contains a disulfide bridge between Cys38 and Cys99 [27]. The link between the IgVRAGE and IgC1RAGE domains is rigid, creating a “V-C1RAGE” structural unit, which is predominantly non-polar/positively charged, and thus can recognize negatively-charged ligands or surfaces [26,28,29]. The IgC2RAGE-domain is different. It carries a negatively-charged surface [30], and it is bound to the V-C1RAGE unit by a flexible seven amino acid linker [29,31].

Along with RAGE, there is a large variety of unrelated proteins that include one or several domains with the Ig-like fold. Antigen receptors, which include various antibodies, i.e immunoglobulins, and T-cell receptors (TCRs) constitute the most populated group of proteins of the immunoglobulin superfamily with domains having the Ig-like fold. Similar to antibodies, all TCR chains contain one IgV domain at the N-terminus. For both TCRs and immunoglobulins, the main antigen recognition element is a set of canonical Complementary-Determining Region (CDR) loops, CDR1, CDR2 and CDR3, which were already compared based on their sequence and structure, and classified for all known antigen receptors [32]. Analysis of CDR-equivalent loops in RAGE (CDRRAGE loops) would bring insight into its multi-ligand recognition abilities.

RAGE undergoes oligomerization with its transmembrane and extracellular regions prior to ligand binding and signaling activity [33,34]. For example, it has been demonstrated that in the soluble version of RAGE (sRAGE) the V/V-domain homodimerization is required for AGE and S100B binding [35,36]. At the same time, recognition of S100A12 requires formation of the C1/C1-domain interface [37], and existence of the C2/C2-domain disulfide crosslinked dimer is needed for receptor stability [38]. For the binding of S100 proteins, it was concluded that they may require RAGE oligomerization for signal transduction [39], and as shown by the example of S100B, an S100 protein might need to be Ca2+-loaded [26]. The NMR structure of the RAGE-S100B complex showed a bound Zn2+ ion at the C1/C1-domain interface coordinating conserved residues from both sides, implying the possibility of a divalent cation requirement for RAGE multimerization [26]. Additionally, given multitude of bound ligands, existence of several RAGE oligomerization schemes should not be ruled out [34]. Taking into consideration all oligomerization data shown above, it is not difficult to see structural similarity between RAGE and heavy chain-only antibodies, with their standalone IgV domains mimicking nanobodies [40], which gives not only a better understanding of the multiligand functionality of RAGE, but also opens opportunities for search of novel therapeutic molecules [3,7].

Here, we aim to structurally align RAGE with the rest of the classical Ig-like domains, to determine and compare structural elements between RAGE and the other Ig-like domains, which are responsible for ligand recognition, and consequently protein function in these proteins.

2. Materials and Methods

The Protein Data Bank (PDB, http://www.rcsb.org/; 15 March 2025 [41]) and the PDBsum database (PDBsum, https://www.ebi.ac.uk/thornton-srv/databases/pdbsum/; 15 March 2025 [42]) were used to retrieve Ig-like proteins, including RAGE, and also PDB entries of RAGE complexes with ligands. The SCOP classification database (SCOP2, https://www.ebi.ac.uk/pdbe/scop/; 15 March 2025 [43]) and InterPro classification of protein families (https://www.ebi.ac.uk/interpro/; 15 March 2025 [44]) were used to determine whether the 3D structures of the analyzed proteins belong to the Ig-like beta-sandwich fold. Detailed structural information from the above set of PDB files are given in Section 3.1 and Table S3.

Structure visualization and structural analysis of interactions between amino acids in proteins (hydrogen bonds, hydrophobic, other types of weak interactions) were conducted using PyMOL molecular graphics system (https://pymol.org/; 17 March 2025), and software [45] to determine interatomic contacts, i.e., of ligand–protein contacts (LPCs) and contacts of structural units (CSUs). Alignment of primary structures was carried out through the Dali software [46]. The localization of the complementary determining regions (CDRs) of the light chain fragment of the Bence-Jones protein (Rhe) was determined using the PyIgClassify2 server (https://dunbrack2.fccc.edu/PyIgClassify2/default.aspx; 17 March 2025) [47].

Weak hydrogen bonds were identified based on geometrical criteria [48]. Figures were drawn with PyMOL molecular graphics system (https://pymol.org/; 17 March 2025).

3. Results and Discussion

As it is described in Introduction, the immunoglobulin-like (Ig-like) domains show several key structural features, such as the overall β-strand topology, existence of S-S bonds and CDR regions that define ligand recognition by these domains. The aim of our analysis was to take different Ig-like domains as templates, divide into conserved and variable structural elements as bricks of a puzzle, derive their functional purpose, and then “wrap” respective RAGE domains around templates to describe functional attributes of different structural units of RAGE.

3.1. Creating a Dataset of the Analyzed Ig-Like Proteins

Extracellular part of the human RAGE structure (V and C1 domains, PDB ID: 6XQ1 [49]; C2 domain, PDB ID: 4P2Y [50]) has been compared with topologies of five different types of Ig-like domains, the V-type (IgV), C-type (IgC), S-type (IgS), H-type (IgH) and I-type (IgI), where the IgV domain topology is represented by the light chain fragment of the Bence-Jones protein (Rhe), PDB ID: 2RHE [51]; the IgC by the IgG2a-kappa 4-4-20 Fab (heavy chain), PDB ID: 4FAB [52]; the IgS by the tenansin (fibronectin type III domain), PDB ID: 1TEN [53]; the IgH by the galactose oxidase, PDB ID: 1GOF [54]; and the IgI by telokin, PDB ID: 1FHG [55]. In addition to these six known Ig-like types, five Ig-like structures of unknown type exist and have been investigated too: 1) chaperone protein PapD, PDB ID: 2J2L [56]; 2) cytochrome f, PDB ID: 1CTM [57]; 3) arthropodan haemocyanin, PDB ID: 1HC1 [58]; 4) superoxide dismutase [Cu-Zn], PDB ID: 1Q0E [59] and 5) synaptotagmin I, PDB ID: 1RSY [60].

3.2. Structural Homology between IgV_RAGE and IgV, and the Overall Topology of Ig-Like Domains

As it is described in the Introduction section, the extracellular part of human RAGE protein consists of three Ig-like domains, IgV_RAGE, IgC1_RAGE and IgC2_RAGE, with the total length of about 300 amino acids. Recently, nine related X-ray complexes of the same two-domain IgV_RAGE-IgC1_RAGE (or V-C1_RAGE) configuration, but with nine different bound inhibitors, were reported [49]. We chose one of these nine structures, PDB ID: 6XQ1 (resolution 1.51 Å), as the RAGE domain representative of the V-C1_RAGE complex. From this complex we extracted representative structure of the IgV_RAGE domain.

Generally, the IgV domain is the main ligand binding domain in any antigen receptor and also in RAGE. As the first step, we checked if the IgV_RAGE domain indeed, by homology, is similar to the other known IgV domains, and also, we created a structural superposition of different IgV domains that could be used as the background superposition for comparative purposes. The V-type domain topology of the Ig-like domains [24] was historically taken from the light chain fragment of the Bence-Jones protein (Rhe; PDB ID: 2RHE [51]). A structural alignment between IgV_RAGE (6XQ1_A) and the IgV domain from the Bence-Jones protein (2RHE_A) has been done using the Dali software [46] (Figure 1). The Dali superposition resulted indeed in suggested structural homology between IgV_RAGE and IgV with the Z-score of 9.6 and RMSD of 2.9 Å over 114 residues.

Sixteen alignment positions were found to be identical, of which three positions contained two key cysteines and a tryptophan. The two key cysteines form S-S bond in both IgV_RAGE and IgV, linking the two β-sheets of the standard IgV topology together, and the key tryptophanes form similar stacking interactions with the S-S bonds in both types of structures. The Dali comparison also demonstrated similar arrangement of eight β-strands (A-G) in both domain types, underlying similar secondary structure topology (Figure 1). Earlier, it was suggested that in the immunoglobulin fold, a prototype Ig-like domain gene may had been derived from gene duplication, which in turn suggested possible existence of a prototype “half-domain” structure formed by either β-sheet ABCC’ or β-sheet GFED [61]. From the structural point of view, however, this hypothesis might work only for IgV domains, because of the resulting significant change in structure moving from two four-stranded β-sheets in IgV to “four-stranded plus three-stranded” β-sheets for IgC and IgS, where IgC lacks the C’ β-strand and IgS lacks the D β-strand (Figure 2) [24]. Thus, we did consider structures of the entire domains. We further “disassembled” different types of Ig-like domains into logical structural units in order to identify conserved structural blocks common to all types (IgV, IgC, IgS and IgH) of the Ig-like domains and variable structural blocks that are specific only to certain domains. Then, we determined roles of the conserved and variable structural blocks in ligand binding.

3.3. The Cross-β Zone of the IgV_RAGE Domain

While different domains within the Ig-like fold contain 3x4, 4x4 or 5x4 β-sheet sandwiches, they all contain four central β-strands B, C, E and F (Figure 2) [24]. The topology of these four strands is known as the “cross-β” structural motif [62]. In IgV_RAGE the respective cross-β motif is shown in Figure 3. Because IgV_RAGE is structurally homologous to IgV, all the respective β-strands exist also in IgV_RAGE, however their length is different, as it is seen from Figure 1. In Figure 3, we have accepted five amino acids as boundaries for these four β-strand segments in IgV_RAGE, because (1) five is the average length of β-strands in IgV_RAGE [49], and (2) five is the minimum length to keep the entire hydrophobic core of the cross-β Motif in IgV and IgV_RAGE (Figure 3).

In IgV_RAGE, fragments Leu₃₄-Leu₅₃ and Gly₈₂-Met₁₀₂ contain βB-loop-βC and βE-loop-βF super-secondary structures, where βB, βC, βE and βF designate β-strands B, C, E and F, respectively (Figure 3). β-Strands B and C lie in the opposing β-sheets of the β-sandwich and are antiparallel to each other (Figure 2). The same is true for β-strands E and F. In order to create the cross-β structural motif, the βB-loop-βC and βE-loop-βF super-secondary structures are put together antiparallel to each other, so that the loop regions reside on the opposite sides of the motif (Figure 4A).

Similar to IgV, in IgV_RAGE the cross-β structural motif takes form of an antiparallel β-sheet sandwich, where β-strands B and E and β-strands C and F become connected by characteristic “antiparallel bridges” (Table S1). An “antiparallel bridge” is a known binding pattern, observed in the secondary structure of proteins, and specifically β-sheets [64], where pairs of hydrogen bonds occur between main chain atoms of β-strands within a β-sheet [65]. In IgV_RAGE, β-Strands B and E are connected by two bridges, and β-strands C and F - by three (Table S1). Taken all the structural features together, we come to the description of the “cross-β structural zone” in IgV and IgV_RAGE. The cross-β structural zone consists of approximately 40 amino acids, which incorporate four strands of two β-sheets totaling to 20 residues, the BC loop (in IgV_RAGE) or the CDR1 loop (in IgV) of about 10 residues each, and the EF loop (in both IgV and IgV_RAGE) of approximately 10 residues long (Table S2). Amino acids of the zone form a compact standalone sub-structure, and interact with each other by means of antiparallel bridges and a network of other weak interactions. Two interacting β-sheets form hydrophobic core and two different outer surfaces. The cross-β zone of the IgV_RAGE domain plays central role for protein function.

3.4. Hydrophobic Core of Ig-Like Structures

Comparison between IgV and three other types of Ig-like domains, IgC, IgS and IgH, is given in Bork at al., 1994 [24], where the authors concluded that the cross-β motif was structurally conserved in all the four Ig-like domains. Here, we showed that the cross-β zone of the IgV_RAGE domain does also belong to that group (Figure 3), and the same is also true for the two C-type domains of RAGE, IgC1_RAGE and IgC2_RAGE, where two β-loop-β motifs form a cross-β substructure with the hydrophobic core similar to that seen in the other Ig-like domains (Table 1). Even the length of the respective β-strand and loop regions of the two β-loop-β units in the cross-β motif are equivalent (Table S2), which supports the hypothesis of gene duplication in a prototype Ig-like domain [61]. Also, same arrangement can be seen in the I-type Ig-like (IgI) domain (PDB ID: 1FHG, telokin).

Unlike classical V-, C-, S-, H- and I-types of Ig-like domains, several unclassified Ig-like domains taken from the work of Halaby et al., 1999 [66]: 2J2L, chaperone protein PapD; 1CTM, cytochrome f; 1HC1, arthropodan haemocyanin; 1Q0E, superoxide dismutase [Cu-Zn]; and 1RSY, synaptotagmin I (Table S2), show variations in length of the β-loop-β fragments BC and EF, and the long loop CE. Nevertheless, according to the SCOP2 database [43], three out of the five structures, 2J7L, 1HC1, and 1Q0E, still have the Ig-like beta-sandwich fold (SCOP ID: 2000051), and 1RSY has fold, which is topologically similar to Ig-like beta-sandwich. Because all five of them have similar hydrophobic cores of the cross-β zone, their 3D structures can be potentially put together as new M(Mixed)-type of Ig-like domains.

3.5. Two Types of Ligand Binding in RAGE

We have analyzed sixteen available ligand-bound complexes of RAGE [27,49,67,68,69,70,71,72] for surfaces and amino acids involved in ligand binding using the PDBsum database [42]. All residues involved in ligand recognition in these structures of IgV_RAGE are summarized in Table S3. Additionally, Table S3 contains highlighted information whether the ligand-interacting residues belong to the cross-β zone strands and loops (marked in red and yellow color, respectively) or to the rest of the IgV_RAGE domain, such as the CE-loop, etc. The summary figure of all amino acids involved in ligand binding in RAGE is shown in Table S4. All ligand-interacting amino acids in RAGE (the IgV_RAGE) domain can be roughly divided into two spatial surface-exposed clusters.

(1) β-Sheet based ligand-binding cluster: The core of the first ligand-binding cluster in RAGE consists of six large polar amino acids, Glu₅₀, Lys₅₂, Arg₉₈, Gln_100, Lys₁₁₀ and Asn₁₁₂, which reside on the outer surface of the larger β-sheet of the IgV_RAGE domain β-sandwich (β-strands C, F and G; Figure 4A). β-Strands C and F belong to the cross-β zone. β-Strand G does not belong to the cross-β zone, but in all five standard Ig-like domain types the β-strand G forms a conserved β-hairpin with the β-strand F of the cross-β zone, creating an antiparallel β-sheet.

(2) CE-loop based ligand-binding cluster: The core of the second ligand-binding cluster in RAGE consists of seven amino acids: Asp₇₃, Ser₇₄, Arg₇₇, Val_78, Leu₇₉, Phe₈₅ and Pro₈₇, which are predominantly located in the CE-loop of RAGE that lies outside the cross-β zone and on the side of the central β-sheet (Figure 4B).

Amino acids of the two cluster above are the most frequent amino acids repeatedly found in different RAGE complexes to bind different ligands (see numerical values in Table S4). At the same time the full set of ligand-interacting residues is listed in Table S4, where it is seen that almost all amino acids of the long CE-loop do interact with ligands in one complex or the other, which makes CE-loop a unique ligand-binding substructure.

The structure of the β-sheet based ligand-binding cluster is more rigid, because it is formed by conserved secondary structure elements of the cross-β motif, and its ligand-binding properties depend on the type of side-chain groups of interacting amino acids. The CE-loop based ligand-binding cluster is more flexible, and its ligand binding capabilities are determined by side-chain and main-chain groups of interacting residues, and by their secondary structure/conformation. The two ligand-binding clusters of RAGE shown above recognize both large molecules (proteins) and small molecule ligands (Table S3). At the same time, these ligand-binding clusters in RAGE do not coincide as a whole or in part with the three CDR peptides of the IgV domain (PDB ID: 2RHE), thus functionally distinguishing RAGE from immunoglobulins.

3.6. RAGE Dimerization and Binding of DNA

In the introduction section we showed that RAGE oligomerization is known to be required for binding a multitude of different ligands. If we look from the other side, it was also said that there may be several oligomerization techniques that RAGE employs to recognize those ligands [34]. Here, we demonstrate that IgV_RAGE homodimer uses sites different from either the C-F-G β-sheet or the CE-loop described above (section 2.5) to bind nucleic acids, and thus can be roughly considered as the third type of ligand binding by IgV_RAGE, complementary to the two types of binding shown above.

Recognition of DNA and RNA and the subsequent activation of signaling receptors by nucleic acids is a mechanism of detection of infection and tissue damage. DNA recognition by RAGE affects immune activation of Toll-like receptor 9 (TLR 9), the principal DNA-recognizing signaling receptor [63]. The binding of a nucleic acid requires homodimerization of the IgV_RAGE domain by “stacking” two A-B-E β-sheets, which are opposite to the ligand-binding C-F-G β-sheets, from two respective IgV_RAGE domains, as it is shown by the crystal structure PDB ID: 4OI7 [63] in Figure 4C. The stacking of β-sheets from two IgV_RAGE domains happens through interactions of two hydrophobic clusters, Leu₇₉, Pro₈₀, Phe₈₅ and Pro₈₇, around the Phe₈₅-Phe₈₅ hydrophobic link from each IgV_RAGE domain of the dimer. The DNA recognition comes with the help of electrostatic attraction by means of the triple-lysine Lys₃₇-Lys₃₉-Lys₄₃ cluster on each of the IgV_RAGE domain [63].

Such nucleic acid-binding mechanism leaves the CE-loop and the C-F-G β-sheet exposed for binding other macromolecules or small-molecule inhibitors.

3.7. Antiparallel Homodimerization of Two β-Loop-β Fragments to Create Functional Core of RAGE

Formation of a protein structural catalytic core (SCC) or a core zone of a protein through antiparallel dimerization of two similar fragments of protein structure has already been seen by us in structural analysis of the EF-hand calcium-binding proteins [73] and acid proteases [74]. In the EF-hand proteins [73], two “helix – binding loop – strand – helix” segments were locked together in an antiparallel manner by the two “locks”, called the “black” and “grey” clusters, which situated at the α-helical regions on both sides of the motif. The third “lock” was the resulting two-stranded antiparallel β-sheet in the middle of the motif. Here, in Ig-like domains, two “strand – loop – strand” segments are locked together in an antiparallel manner by the two “locks” of formed antiparallel β-sheets on the two ends, thus forming the cross-β functional motif. In acid proteases [74], the dimer is formed by two sets of equivalent structural segments, the DD-link, D-loop, and G-loop, around two catalytic aspartates in each individual chain. In immunoglobulins, such organization created an argument in favor of gene duplication [61]. However, as we have shown above, RAGE does have homologous IgV_RAGE domain with the cross-β motif, but it is not immunoglobulin with respect to CDR loops or the other ligand recognition or functional elements, thus giving possibility that if gene duplication in a prototype Ig-like domain existed, it could not be an attribute of immunoglobulins, but something larger.

4. Conclusions

Receptor for Advanced Glycation Endproducts (RAGE) is a transmembrane multi-ligand receptor, which incorporates one Transmembrane domain (TM), one possible Immunoglobulin V-like (IgV_RAGE) domain and two Immunoglobulin C-like (IgC_RAGE) domains. Here, we showed that the ligand-binding IgV_RAGE domain of RAGE is structurally homologous to the standard V-type Ig-like domain, and similarly contains respective disulfide bridge and the cross-β motif. Moreover, there is a wider conserved structure, which we call the cross-β zone. The cross-β zone in its inner side contains the hydrophobic core of the domain, and in its outer side contains main ligand-binding sites of the protein: one ligand-binding site on the surface of the larger β-sheet C-F-G of the β-sheet sandwich; and the other on the long loop fragment CE. Based on the data observed, we could speculate that the β-sheet based binding site could be more “universal” for Ig-like domains, while the CE-loop based was more “specific” for a particular domain. As opposed to the C-F-G β-sheet, the A-B-E β-sheet is used for RAGE dimerization through the hydrophobic cluster on its surface. Oligomerization allows RAGE to bind even larger variety of ligands, such as for example nucleic acids, and participate in even more immune signaling mechanisms. However, regardless of structural similarities and features mentioned above, RAGE does not show ligand-binding regions analogous to the Complementary-Determining Regions (CDR) in immunoglobulins, thus making RAGE and immunoglobulins functionally different.