The Structure of the Interleukin-15α Receptor and Its Implications for Ligand Binding*

Interleukin (IL)-15 is a member of the small four α-helix bundle family of cytokines. IL-15 was discovered by its ability to mimic IL-2-mediated T-cell proliferation. Both cytokines share the β and γ receptor chains of the IL-2 receptor for signal transduction. However, in addition, they target specific α chain receptors IL-15Rα and IL-2Rα, respectively. The exceptionally high affinity binding of IL-15 to IL-15Rα is mediated by its sushi domain. Here we present the solution structure of the IL-15Rα sushi domain solved by NMR spectroscopy and a model of its complex with IL-15. The model shows that, rather than the familiar hydrophobic forces dominating the interaction interface between cytokines and their cognate receptors, the interaction between the IL-15 and IL-15Rα complex involves a large network of ionic interactions. This type of interaction explains the exceptionally high affinity of the IL-15·IL-15Rα complex, which is essential for the biological effects of this important cytokine and which is not observed in other cytokine/cytokine receptor complexes.

Two independent groups recognize that interleukin (IL) 2 -15 is able to mimic IL-2-mediated T-cell proliferation (1)(2)(3). This is due to both cytokines sharing the IL-2 receptor ␤ (IL-2R␤) and the common ␥ chain receptor for signal transduction (1,2,4,5). The IL-2R␤ and ␥ chain receptors are both members of the hematopoietic growth factor receptor superfamily containing the typical cytokine binding module that consists of two fibronectin type III domains. Besides these two receptors, both cytokines each use a specific ␣ receptor, the IL-15 receptor ␣ chain (IL-15R␣) and the interleukin-2 receptor ␣ chain (IL-2R␣) (6). The structure of the quaternary complex between IL-2 and its ␣, ␤, and ␥ receptors has very recently been solved by x-ray crystallography (7). Both IL-15R␣ and IL-2R␣ receptors are highly homologous and exhibit amino acid homology to a group of ␤ sandwich protein modules called sushi domains. These domains have been described previously in proteins of the complement system and the coagulation cascades (8). Sushi domains have typical sizes of 60 -70 residues and contain two cysteine bridges (9,10). The IL-2R␣ consists of two sushi domains, whereas the IL-15R␣ contains only one (11). These domains are structurally different from other cytokine receptors, which bind their ligands via cytokine binding modules (12). Both IL-2R␣ and IL-15R␣ use their sushi domains for ligand binding but with remarkably different affinities. Whereas the IL-2R␣ binds IL-2 with low affinity (K a value of ϳ10 8 M Ϫ1 ), the IL-15R␣ shows a 1000-fold higher binding affinity to IL-15 (K a value of ϳ10 11 M Ϫ1 ) (6,11). IL-2 and IL-15 not only differ in their affinity for their specific ␣ receptors but also in their tissue distribution. IL-2 is exclusively expressed by activated T-cells, whereas IL-15 has been described to be expressed in nonimmune (keratinocytes, skeletal muscle cells) and immune cells (monocytes and activated T-cells) (6,11). Although IL-2 and IL-15 share some function, such as T-cell activation, differences in their biological function have been clearly identified by the differences in the phenotypes of knock-out mice. The IL-15 and IL-15R␣ knock-out mice show a reduced number of CD8 ϩ memory, natural killer, and natural killer T-cells (13,14). In contrast, the IL-2 and IL-2R␣ knock-out mice exhibit autoimmune phenotypes and an increased number of T-and B-cells (15,16). Moreover, IL-2 promotes activation-induced cell death, which is suppressed by IL-15 (17). In a number of studies, it has been shown that the latter cytokine is proinflammatory and is associated with the pathogenesis of inflammatory diseases, such as rheumatoid arthritis and psoriasis (18,19). Because of its broad distribution and high inflammatory potential, the expression of IL-15 is believed to be tightly regulated (20,21).
Dubois et al. (22) has shown that the sushi domain of IL-15R␣ is essential for ligand binding and that eight different splice forms from the IL-15R␣ mRNA exist. The splice forms vary by either the presence or absence of exon 2 (which encodes for the sushi domain), exon 3 (which encodes for the linker region), and the alternative use of exon 7 and 7Ј (which encode the cytoplasmic part of the IL-15R␣). Interestingly, this receptor is not only located at the plasma membrane but also in the nuclear membrane. However, only the splice forms that contain exon 2 have been detected in the nuclear membrane (22). In line with this observation, Nishimura et al. (23) suggest a model of transcriptional regulation mediated via the IL-15 bound to its ␣ receptor (23).
In addition, it has been shown that a natural form of a soluble IL-15R␣ exists. This soluble form is generated by a process called shedding and involves the cleaving of IL-15R␣ by a metalloproteinase, namely the tumor necrosis factor-␣-converting enzyme (TACE/ADAM17) (24,25). The soluble IL-15R␣ has been shown to elicit antagonistic properties (24) and thereby prevent collagen-mediated arthritis (26), short term carrageenan-induced inflammation (8), and the induction of allergic inflammation (27). Moreover, in a mouse model, it has been shown to enhance cardiac allograft survival (28). All of these data suggest a promising therapeutic potential for the soluble IL-15R␣ receptor.
Many cytokine⅐cytokine receptor complexes have been analyzed with respect to their interaction mechanisms. Two different modes of interaction between a cytokine and its receptor have been described so far. The growth hormone/growth hormone receptor complex was the first three-dimensional structure of such a complex. Alanine scanning mutagenesis has revealed that the two molecules interact in a "hot spot" manner (29), whereas the IL-4/IL-4R has been described to interact in an "avocado-like" manner (30).
Here we present the three-dimensional structure of IL-15R␣ solved by multidimensional NMR spectroscopy. This structure has enabled us to build a model of the IL-15⅐IL-15R␣ complex that reveals a new mode of cytokine/cytokine receptor interaction.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, and Sample Preparation-The mature coding sequence of IL-15R␣ was originally cloned in pGEX2T (Amersham Biosciences). This clone was used as a template for amplification of the IL-15R␣ sushi domain coding region by PCR. The PCR products were purified, digested using BamHI and HindIII restriction enzymes and ligated into pQE30 (Qiagen). The plasmid was transformed into Escherichia coli SG13009 cells.
Fifty milliliters of minimal medium (6 g of Na 2 HPO 4 /liter, 3 g of KH 2 PO 4 /liter, 2 g of NaCl/liter, 0.24 g of MgSO 4 /liter, 2 g of 15 NH 4 Cl/ liter, 1.86 g of 13 C-glucose or 2 g of 12 C-glucose*H 2 O) containing 50 g/ml ampicillin and 30 g/ml kanamycin was inoculated with 500 l of culture of a single clone and cultured overnight at 37°C. 350 ml of minimal medium was inoculated with 3.5 ml of the overnight culture and incubated at 37°C. Expression of the recombinant protein was induced by adding isopropyl-␤-thiogalactopyranoside (final concentration of 1 mM) when the cell culture had reached an OD of ϳ0.5. The growth of the culture was continued for a further 4 h.
The cells were harvested by centrifugation and resuspended in a lysis buffer (50 mM Tris/HCl, pH 8.0, 1 mM EDTA). Lysis of bacteria was achieved by sonification, and the inclusion bodies were isolated by centrifugation. Inclusion bodies were washed three times with a buffer containing Tween 20 (50 mM Tris/HCl, 1 mM EDTA, 2 mM urea, and 0.5% Tween 20, pH 8.0) and three times with the same buffer containing no detergent. Inclusion bodies were then solubilized in 6 M guanidine hydrochloride, 20 mM sodium phosphate, 20 mM imidazole, 150 mM sodium chloride, and 1 mM dithiothreitol at pH 7.4.
A nickel-nitrilotriacetic acid-agarose column (Qiagen, Hilden, Germany) was equilibrated with the same buffer as before, but instead of 1 mM dithiothreitol, the buffer contained 1 mM reduced glutathione and 0.2 mM oxidized glutathione. The solubilized inclusion bodies were loaded onto the column, and nonspecific protein was removed by washing. The protein was refolded via a gradient of from 6 to 0 M guanidine hydrochloride and then eluted with 250 mM imidazole (36). The imidazole was removed using a NAP column (Amersham Biosciences). The sample was concentrated via an AMICON ultrafiltration cell (Millipore, Eschborn, Germany) to ϳ1 ml and the buffer exchanged to a buffer containing 7% D 2 O and 0.03% NaN 3 . The protein sample was further concentrated to an approximate concentration of 1 mM using VIVA spin 500 concentrators (Vivascience, Hannover, Germany).
NMR Spectroscopy-NMR experiments were recorded at 298 K on Bruker DRX600 and Varian INOVA 600-MHz spectrometers equipped with 5-mm z-gradient 1 H/ 15 N/ 13 C probes optimized for 1 H detection. Backbone and side chain resonance assignments of the sushi domain of the IL-15R␣ were established from a series of three-dimensional triple-resonance experiments (i.e. HNCA, HNCO, (H)C(CO)NH, H(CCO)NH)) and 15 N-and 13 C-edited total correlation spectroscopy experiments. Distance restraints were obtained from three-dimensional 15 N-and 13 C-edited nuclear Overhauser effect (NOE) spectroscopyheteronuclear single quantum correlation experiments, all recorded with a mixing time of 100 ms. A number of hydrogen bond restraints were obtained from a two-dimensional long range HNCO experiment (37) with 13 C-␣ decoupling during the long 15 N-13 CЈ defocusing/refocusing period (38). Proton chemical shifts were referenced to 3-(trimethylsilyl)-propionic acid, whereas 15 N and 13 C chemical shifts were indirectly referenced (39). All spectra were processed with the NMRPipe program (40) and analyzed with NMRview software (41). The ratio of the intensity of the water and the corresponding 1 H/ 15 N diagonal peak (Ͻ0.1) in the 15 N-edited NOE spectroscopy with water flipback is an indication of slow amide proton exchange. Therefore, these amide groups are potential proton donors involved in hydrogen bonding.
Structure Calculation-Structure calculations were performed using the Dyana software (42). The structure calculations were based on 1123 NOE distance restraints extracted from the 15 N-and 13 C-edited NOE spectra. Distances of 1.8 and 5.0 Å were taken as lower and upper limits for all NOE cross-peaks, respectively. In addition, 2 *10 hydrogen bonds, classified by a distance of 1.8 -2.4 Å between hydrogen and oxygen and 2.8 -3.4 Å between nitrogen and oxygen, were introduced. After preliminary structure calculations, these low resolution structures were searched for potential hydrogen bond acceptors, which then were included in further structural refinement. Distances representing the two disulfide bonds with restraints 2.03-2.15, 3.03-3.13, and 2.97-4.49 Å for SG/SG, SG/CB, and SG/CA distances, respectively, were also included in the final structural refinement calculations. One-hundred structures were initially calculated, of which, 20 structures with the lowest target function were selected. An averaged structure was calculated from the ensemble of the 20 structures and energy-minimized using the steepest descent algorithm implemented in the GROMOS program package (43). The chemical shifts have been deposited in the Biological Magnetic Resonance Bank under accession number 6882, and the coordinates are deposited at the Protein Data Bank under accession code 2ERS.
Model of the IL-15⅐IL-15R␣ Complex-To construct a model of the IL-15⅐IL-15R␣ complex, the recently published structure of the IL-2⅐IL-2R␣ complex was used as a template (31). The IL-15R␣ was superimposed with the IL-2R␣, and our recently published model of the IL-15 (32) was superimposed onto the IL-2 structure. The program WHATIF (44) was used to calculate the solvent accessibilities of the bound as well as the free proteins. For graphical representations, the programs RIB-BONS (45) and GRASP (46) were used.

RESULTS AND DISCUSSION
Structure of IL-15R␣-The two-dimensional 1 H-15 N heteronuclear single quantum correlation spectrum exhibits well resolved resonances. Sequential assignment information was obtained from the three-dimensional experiments and includes the three asparagine side chain amide-proton pairs (Fig. 1). The absence of the 1 H-15 N amide frequencies for Gly-68 and Ser-71 is presumably because of chemical exchange at pH 7.4. The 12-residue N-terminal His tag is missing in all 15 N-edited spectra and was therefore not included in the structure calculations. Residue Ile-31 represents the first residue of the mature protein domain. The aliphatic side chains of Gly-68 and Ser-71 and Pro-34 -36 and 89 were assigned using the 13 C-edited spectra. The ensemble of 20 superimposed structures is depicted in Fig. 2B and displays an overall root mean square deviation of 1.48 Ϯ 0.29 for the backbone, whereas the secondary structure elements exhibit a root mean square deviation of 0.82 Ϯ 0.20 (Table 1). Interestingly, large regions of the domain exhibit no regular secondary structure elements. Only ϳ29% of the residues are involved in the assembly of the five ␤-strands. The residues Val-39, Glu-40 (loop A), and Thr-87 (loop E) display backbone dihedral angles in the disallowed areas of a Ramachandran plot (data not shown) and are located in the less structured parts of the molecule.
Comparison with IL-2R␣-The structure of the IL-2R␣ in complex with its ligand has been solved by x-ray crystallography (31). As such, the solution structure of the IL-15R␣ with the IL-2R␣ structure were superimposed in an effort to gain insight into any significant structural differences. Because IL-2R␣ contains two sushi domains but ligand binding is mediated primarily by the N-terminal domain (IL-2R␣-D1), the superimposition was performed using this domain. As shown in Fig.  3, the central regions of both sushi domains appear rather similar. The strands C, CЈ, D, and EЈ in the sushi domain of IL-15R␣ are in the same position as the corresponding ␤-strands C*, CЈ*, D*, and EЈ* of IL-2R␣-D1. Whereas the E*-loop of the IL-2R␣-D1 domain is not resolved in the x-ray structure, the structure of the IL-15R␣ exhibits a fourth ␤-strand in this region.
The two domains of the IL-2R␣ show a ␤-strand exchange. Loop A* and ␤-strand B* are part of D2, whereas loop F* and strand G* are part of D1. In the structure of the IL-15R␣ sushi domain, loop F* of IL-2R␣ corresponds to loop A. Because of this ␤-strand exchange between the domains, the orientation of ␤-strand G* in the IL-2R␣ is different from the corresponding loop B in the IL-15R␣. The ␤-strand G* of the IL-2R␣ is part of the ␤-sheet consisting of the three strands D*, C*, and G*, whereas loop B of IL-15R␣ lies behind strand C and runs perpendicular to loop A. This rearrangement has notable implications for ligand binding between the two receptors.
Model of the IL-15⅐IL-15R␣ Complex-With the help of the experimentally derived structure of the IL-15R␣ sushi domain and the previously published model of IL-15 (32), a model of the IL-15⅐IL-15R␣ complex was built. To orient the two molecules relative to each other, the recently published structure of the IL-2⅐IL-2R␣ complex was used as a template (31). The three-dimensional model of the IL-15⅐IL-15R␣ complex is shown in Fig. 4A. In this complex, the sushi domain is located on top of IL-15 and utilizes the C-, CЈ-, and D-strands to form the interface for ligand interaction. The binding epitope of IL-15 consists of the ABloop, the B-helix, and regions of the CD-loop.

IL-15 Receptor Structure
The AB-loop (amino acid residues 23-33) of IL-15 has been proposed to be involved in binding to high affinity IL-15R␣ (33). On the basis of peptide scanning experiments and site-directed mutagenesis studies, Bernard et al. (32) suggests the B-and C-helices of IL-15 are involved in receptor binding. Whereas the participation of the AB-loop and the B-helix could be confirmed by this model of the IL-15⅐IL-15R␣ complex, the interaction of helix C of IL-15 with the sushi domain do not fit the model because of the limited size of the sushi domain and the location of the C-helix not proximal to the binding interface. Fig. 4B shows a close-up of the IL-15/IL-15R␣ binding interface. Most of the residues participating in the interaction are charged residues. Here, the binding epitope of the IL-15R␣ sushi domain exhibits almost exclusively basic residues, whereas the IL-15 epitope comprises acidic residues. Two arginines (Arg-54 and Arg-56) and two lysines (Lys-64 and Lys-66) from the IL-15R␣ and two aspartic acids (Asp-30 and Asp-56) and four glutamic acids (Glu-53, Glu-87, Glu-89, and Glu-93) from IL-15 form a large surface network of charge/charge interactions, rather than specific ion pairs. In contrast to other cytokine⅐cytokine receptor complexes, only a few hydrophobic residues (Leu-72 and Ala-67 of the sushi domain and Leu-52 and Leu-45 from IL-15) (see Fig. 4B) are located at the binding interface. In addition, these hydrophobic residues are located at the edges of the binding surface and are, at least partially, accessible by surrounding water molecules. Exchange of two hydrophobic residues Leu-52 and Leu-45 in IL-15 to aspartic acid residues leads to mutants with an increased affinity for IL-15R␣ and greater proliferation potential (32). The IL-15⅐IL-15R␣ model supports this observation, as the complex reveals that the introduction of additional negative charges, which are located in close vicinity to the network of interacting charges, (see Fig. 4B) leads to an extension of the negatively charged surface area of IL-15 and thereby a potential increase in the stability of the complex.
Comparison between the IL-15⅐IL-15R␣ and the IL-2⅐IL-2R␣ Complexes-The kinetic binding constants (k on and k off ) and affinities (K D ) of IL-15 and IL-2 to their respective ␣ receptors have been reported ( Table 2). The K D of IL-15 is three orders of magnitude higher compared with that of IL-2. To examine the potential structural origin for this difference in affinities, we analyzed and compared the interaction areas of the complexes by calculating the changes in the accessible surface areas upon complex formation (Fig. 5A). Both cytokines IL-15 and IL-2 utilize residues located in almost identical structural regions, namely the AB-loop, the B-helix, and the CD-loop. In contrast, a remarkable difference is clearly observable between the two ␣ receptors. Although in both receptors, residues involved in the interaction with the ligand are located in the E-loop and D-and C-strands, the interaction of IL-2 with IL-2R␣ also involves residues located in the G*-strand. Because of the absence of strand exchange, the corresponding region in the IL-15R␣ (B-loop) is not involved in the formation of the complex (see Fig. 3). The absence of the participation of the B-loop in the interaction is apparently because of the partial masking of this structural element behind the C-strand (Fig. 4A). Consequently, the interaction area is    Where the binding site potential is displayed, negatively charged is pictured in red, and positively charged is pictured in blue.

IL-15 Receptor Structure
larger for the IL-2⅐IL-2R␣ complex, yet the experimentally derived binding constant is a thousand-fold lower compared with the IL-15/IL-15R␣ complex. This discrepancy must be explained by the different nature of the interaction between the respective cytokines and their receptors. The interaction area of the IL-2⅐IL-2R␣ complex has been described to consist of a hydrophobic core region, surrounded by hydrophilic residues that shield this hydrophobic patch from water molecules. This mode of interaction has been observed in many cytokine⅐cytokine receptor complexes, such as the growth hormone⅐growth hormone receptor and the IL-6⅐IL-6R␣ complex (29,34,35). In contrast, the IL-15/IL-15R␣ interaction area consists almost exclusively of charged residues. This remarkable difference between the IL-15⅐IL-5R␣ and IL-2⅐IL-2R␣ complexes is also reflected by the calculated electrostatic surface potentials of the involved molecules (see Fig. 5B). The highly positive charged surface of the sushi domain of the IL-15R␣ fits nicely against the highly negative charged IL-15 binding region. It is evident, that the interaction between IL-2 and IL-2R␣ involves a much lower number of charged residues. Comparing the kinetic constants of complex formation (k on , k off ) of IL-15⅐IL-5R␣ and IL-2⅐IL-2R␣, the most striking difference is observed in the k off rate constants, which differ by about five orders of magnitude. The kinetic constant k off is the probability that a complex will dissociate and is therefore a direct measure of the heights of the activation barrier between the complex and the free molecules. A complex such as IL-15⅐IL-15R, with such a low k off value is kinetically trapped. This is a typical phenomenon of a polyanion⅐polycation complex. With respect to this, the IL-15/IL-15R␣ complex resembles such a polyanion/polycation interaction, which is responsible for the high affinity of IL-15 to the IL-15R␣ receptor. Specific antibodies against the IL-2R␤ chain are already used in clinical trials to inhibit signaling of both cytokines IL-2 and IL-15 (18), but the soluble IL-15R␣ serves as a potential specific antagonist for IL-15 signaling (24). Because the presented IL-15⅐IL-15R␣ complex is in good agreement with data from site-directed mutagenesis data and the results might therefore serve as a basis for the rational design of more potent IL-15 antagonists, such antagonists can be of valuable therapeutic interest for the treatment of inflammatory diseases.