Structural insights into the interaction of IL-33 with its receptors

Overall Structure.

The complex structure of active IL-33_112–270 (Ser112 to Thr270) with ST2 ectodomain (Lys19 to Lys321) was determined at a resolution of 3.27 Å by the molecular replacement method (Table S1 and Fig. 1). The final model includes IL-33 residues Ser117 to Asn171, Gly179 to Val252, and Glu261 to Ser268 and ST2 residues Ser21 to Ser51, Ser56 to Lys223, Asn232 to Gly271, and Gly279 to Arg317. Three N-acetyl-D-glucosamine (NAG) glycans are linked to Asn95, Asn140, and Asn191 of ST2, respectively (Fig. 1). Other missing residues were not built because of weak electron densities. The ectodomain of ST2 consists of three IgG-like domains (D1 to D3) as expected (Fig. 1). The D1 and D2 domains pack together to form a single D1D2 module, to which the D3 domain is connected through a long linker (Fig. 1). In the complex, the ST2 wraps around IL-33 like a grasping hand with all three domains (Fig. 1).

IL-33 adopts a 12-stranded β-trefoil fold (Fig. S1). The crystal structure of IL-33 in the complex superimposes well with solution structure of IL-33 determined by NMR method with an r.m.s.d. of 1.2 Å for 123 matched Cα atoms (30). Twelve β-strands forming the β-trefoil fold are conserved in both structures, whereas conformational variations reside in the loops connecting the strands, especially in loops β2-β3, β3-β4, β4-β5, β10-β11, and β11-β12 involved in receptor interaction (Fig. S1). The β4-β5 and β11-β12 loops are disordered in the crystal structure, and in the NMR structure, the β4-β5 loop is also the region with the most diverse conformation (Fig. S1). These two loops are involved in the recruitment of accessory receptor IL-1RAcP after the binding of IL-33 with ST2, and their roles are discussed below.

All members except SIGIRR in the IL-1 receptor family have three Ig-like domains (D1 to D3) in the extracellular region (26). The packing and orientation between D1 and D2 domains in the D1D2 module are conserved among primary receptors ST2, IL-1RI, and IL-1RII, the structures of which have been determined (Fig. S2) (25, 29, 31, 32). The linker connecting the D3 domain to the D1D2 module is flexible, resulting in different orientations of the D3 domain among these three primary receptors when they bind their respective ligands (Fig. S2).

IL-33/ST2 Interface and Binding Assay.

IL-33 interacts with all three domains of ST2, and a large surface area of ∼1,798 Ų on IL-33 and ∼1,785 Ų on ST2 are buried upon binding. This large binding interface is composed of two separate binding sites (Fig. 2). At site 1, thirteen IL-33 residues, from β3-strand, β1-β2, β2-β3, β3-β4, and β10-β11 loops, have contacts with the D1D2 module of ST2 receptor (Fig. S3). Among these residues, Glu144, Glu148, Asp149, and Asp244 form salt-bridge interactions with ST2 residues Arg38, Lys22, Arg198, and Arg35, respectively (Fig. 2). Glu144 and Asp149 also form hydrogen-bonding interactions with ST2 receptor main-chain atoms (Fig. 2). At site 2, eight IL-33 residues from β1-, β4-, β5-strands and β8-β9 loop have contacts with the D3 domain of ST2 (Fig. S3). The chemical nature here is a mixture of hydrophobic and hydrophilic interactions. Among the IL-33–interacting residues, four of them are hydrophobic and a significant hydrophobic cluster is formed by IL-33 residues Tyr164 and Leu182 and ST2 residues Leu246, Leu306, and Leu311 (Fig. 2). Acidic residue Glu165 has salt-bridge interaction with Arg313 of ST2 (Fig. 2).

Previous efforts to identify “hot spot” residues of IL-33 for ST2 interaction were based on NMR chemical-shift perturbation data and subsequent mutagenesis and affinity measurements (30). Twenty-eight IL-33 mutants were examined, and only one mutant (E144K) had a pronounced effect on the interaction between IL-33 and ST2 by decreasing the affinity sevenfold from 0.45 to 3.15 nM (30). We measured the ST2-binding affinities of IL-33 mutants identified by structural analysis of the IL-33/ST2 interface with surface plasmon resonance (SPR) method (Table S2). These mutation sites were not examined in the previous report except the Glu144 position. In our measurements, wild-type IL-33 bound to ST2 ectodomain with an affinity of 0.74 nM (Table S2 and Fig. S4), comparable to the previously reported 0.45 nM (30). All IL-33 mutations of acidic residues at site I (E144K, E148K, D149K, and D244K) decreased the ST2-binding affinity. E144K and D244K mutants exhibited ∼two- to threefold reduction from 0.74 to 1.89 and 1.65 nM, respectively (Table S2 and Fig. S4). Mutation E148K had a more pronounced effect on the binding with ∼sevenfold affinity reduction from 0.74 to 5.15 nM (Table S2 and Fig. S4). Mutation at Asp149 position (D149K) reduced the binding of IL-33 with ST2 to an undetectable level by SPR method (Table S2 and Fig. S4). At site 2, mutation of acidic residue Glu165 (E165K) also decreased the binding affinity ∼eightfold from 0.74 to 6.07 nM, whereas mutations of hydrophobic residues Tyr163 and Leu182 had less effect on the binding (Table S2 and Fig. S4).

Comparison of Binding Site 1 in Different Ligand–Receptor Complexes.

Previous crystal structures revealing ligand recognition by primary receptor are only available for IL-1β–IL-1RI, IL-1β–IL-1RII, and IL-1Ra–IL-1RI pairs (25, 29, 31, 32). Structural comparison of IL-33–ST2 with them revealed a similar architecture with the involvement of all three Ig-like domains (D1 to D3) and two separate sites in ligand binding (Fig. 3A). Site 1 is more structurally conserved than site 2 in different ligand–receptor pairs because of rigidity of the D1D2 module in different receptors. However, a more close examination revealed differences of IL-33 from IL-1β and IL-1Ra in receptor binding at site 1. We have shown that IL-33 acidic residues Glu144, Glu148, Asp149, and Asp244 form salt-bridge interactions with side chains of basic residues of ST2 receptor (Fig. 2), which is also evidenced by the complementary electrostatic potential on the surfaces of IL-33 and ST2 at site 1 (Fig. 3B). The surface of IL-33 at site 1 exhibits strong negative charge, and the corresponding surface of ST2 is dominated by positive charge (Fig. 3B). In contrast, IL-1β–critical residues Arg21, Gln15, and Gln32 at site 1 interact with IL-1RI and IL-1RII mainly through van der Waals contacts and hydrogen bonds with receptor main-chain atoms (25, 29, 31). This phenomenon has also been observed at the binding site 1 between IL-1Ra and IL-1RI (32). Surface electrostatic-potential analyses also showed that the surfaces of IL-1β and IL-1Ra at site 1 are not dominated by negative charge, and the corresponding surfaces of IL-1RI and IL-1RII have no remarkable electrostatic properties (Fig. 3C). Although the crystal structure of IL-18 in complex with IL-18Rα is not available, IL-18 has an obvious patch with negative charge at potential binding site 1 (Fig. 3D) (33, 34), similar to that of IL-33.

Solution X-Ray–Scattering Studies of Receptors and Ligand–Receptor Complexes.

Currently, all available structures of the IL-1 primary receptors are in ligand-bound state and have flexibly linked domains. However, the degree of flexibility between domains of the receptors correlated with their dynamic nature relevant to molecular recognition has not been investigated. Using a combination of crystallography and SAXS (35), we characterized the functional conformations of the receptors in solution and revealed conformational changes upon complex formation with the ligand. ST2 and IL-1RAcP SAXS data were collected (Fig. S5) and further analyzed to determine whether the receptors exist as well-defined conformations or as an assembly of different conformers in solution by using the program BILBOMD (36), in which the D1D2 module and D3 domain were treated as two separate rigid bodies connected by a flexible linker. In BILBOMD, 10,000 distinct conformers were generated by constrained molecular dynamics (MD) simulation, theoretical scattering profiles were calculated by FoXS (37), and then Minimal Ensemble Search (MES) were applied to select a subset containing one to four conformers that best fit the experimental data (36). The full-atomic model is crucial to match the experimental data (38); therefore, missing loops and residues in ST2 and IL-1RAcP ligand-bound crystal structures were constructed with MODELLER (39). For ST2, the theoretically calculated SAXS profile from initial model does not agree well with experimental data (χ = 5.0) (Fig. 4A). An ensemble of three conformers fits the data significantly much better (χ = 1.7) than the single best-fit conformer (χ = 3.2) (Fig. 4A). For IL-1RAcP, the single best-fit model with a small movement in D3 domain compared with crystal structure improves the fit (χ = 1.1) than the initial full-atomic model derived from crystal structure (χ = 3.9), and MES approach does not improve the fit (χ = 1.1) (Fig. 4B).

The crystal structure of IL-33–ST2–IL-1RAcP complex is still not available. To resolve its molecular architecture, we generated a model of IL-33–ST2–IL-1RAcP by fitting IL-33–ST2 crystal structure into the previously determined IL-1β–IL-1RII–IL-1RAcP ternary complex and replacing IL-1β and IL-1RII with IL-33 and ST2, respectively. To validate this model and to compare the receptors in ligand-bound and ligand-unbound state, we collected the SAXS data on IL-33–ST2 and IL-33–ST2–IL-1RAcP complexes (Fig. S5). Theoretical SAXS profiles from initial full-atomic models of IL-33–ST2 and IL-33–ST2–IL1RAcP complexes fit the experimental data well, with the χ value of 1.5 and 1.8, respectively (Fig. 5 A and B).