Mapping of Binding Site III in the Leptin Receptor and Modeling of a Hexameric Leptin·Leptin Receptor Complex*

The leptin·leptin receptor (LR) system shows strong similarities to the long chain cytokine interleukin-6 (IL-6) and granulocyte colony-stimulating factor (G-CSF) cytokine·cytokine receptor systems. The IL-6 family cytokines interact with their receptors through three different binding sites (I–III). We demonstrated previously that leptin has similar binding sites I–III and mapped the interactions between binding site II and cytokine receptor homology domain II (CRH2) (Peelman, F., Van Beneden, K., Zabeau, L., Iserentant, H., Ulrichts, P., Defeau, D., Verhee, A., Catteeuw, D., Elewaut, D., and Tavernier, J. (2004) J. Biol. Chem. 279, 41038–41046). In this study, we built homology models for the CRH1 and Ig-like domains of the LR. The Ig-like domain shows a large conserved surface patch in the β-sheet formed by β-strands 3, 6, and 7. Mutations in this patch almost completely abolished the leptin-induced STAT3-dependent reporter activity. We propose that a conserved cluster of residues Leu370, Ala407, Tyr409, His417, and His418 forms the center of binding site III of the LR. We built a hexameric leptin·LR complex model based on the hexameric IL-6 complex. In this model, a conserved hydrophobic protuberance of Val36, Thr37, Phe41, and Phe43 in the A–B loop of leptin fits perfectly in the CRH2 domain, corresponding to the IL-6 α-receptor, and forms the center of binding site I. The 2:4 hexameric leptin·LR complex offers a rational explanation for mutagenesis studies and residue conservation.

hypothalamic neurocircuitry activates a negative feedback loop resulting in reduced food intake and increased energy expenditure. Decreased leptin concentrations lead to opposite effects. However, leptin is very pleiotropic and also influences, for instance, angiogenesis, blood pressure, hematopoiesis, bone formation, and immune and inflammatory responses (3,4). Many of these effects are mediated by direct peripheral action of leptin, and expression of the LR is now well documented in many peripheral tissues (including ovary, liver, and adipose), hemopoietic precursors, and immune cells (5).
Mature human leptin is secreted as a 146-amino acid protein, and its crystal structure (6) resembles the structures of four-␣-helix bundle cytokines (7). Leptin shows the highest structural similarity to granulocyte colony-stimulating factor (G-CSF) and cytokines of the interleukin-6 (IL-6) family, including IL-6, ciliary neurotrophic factor, and leukemia inhibitory factor (LIF) (8,9). The LR is a member of the class I cytokine receptor family and has no intrinsic kinase activity, depending on cytoplasmic-associated Janus kinase (JAK)-2 for signaling. Extracellularly, different structural domains can be discerned. Two cytokine receptor homology (CRH) domains are separated by an Ig-like domain, followed by two membrane-proximal fibronectin type III domains. The membrane-proximal CRH domain (CRH2) was identified as the main high affinity binding site of the LR (10 -12), whereas the other CRH domain (CRH1) seems to be not essential for leptin binding and receptor activation (10,12). Although the Ig-like domain and the two fibronectin type III modules are not prerequisites for high affinity leptin binding (10), they are essential for LR activation (12,13), but the underlying mechanism has not been fully resolved.
Cytokines of the gp130 family typically interact with their receptor complex through three different binding sites (I-III) (14). The similar binding sites II and III have been found in G-CSF (15,16). The stoichiometry of the signaling complex differs between the G-CSF and IL-6 receptor systems. G-CSF is believed to form a 2:2 tetrameric complex with its receptor, mediated by binding of site II to the CRH domain of a first receptor and by binding of site III to the Ig-like domains of a second receptor (15). The IL-6 receptor complex consists of a hexamer of two gp130 chains, two IL-6 molecules, and two IL-6 ␣-receptor chains. IL-6 binds first to the CRH domain of the IL-6 ␣-receptor through its binding site I in helix D, followed by interaction with the CRH domain of gp130 through its binding site II. These trimers subsequently form a hexamer by interaction of two trimers, in which binding site III in IL-6 interacts with the Ig-like domain.
We showed previously that leptin has binding sites at similar positions as in G-CSF and IL-6 (9). Binding site I is found in helix D and also contains Phe 41 in the A-B loop. Binding site II is found at the surface of helices A and C, and binding site III is found around the N terminus of helix D. Binding site II interacts with the CRH2 subdomain of the LR, whereas residues in binding site III presumably bind to the Ig-like domain of the LR (9,12).
To further characterize the binding of leptin to its receptor, we built molecular models for the CRH1 and Ig-like domains of the LR. Using site-directed mutagenesis and LR activity measurements, we show that Leu 370 , Ala 407 , Tyr 409 , His 417 , and His 418 form a cluster in the Ig-like domain that is crucial for LR activation. This resembles binding site III of the G-CSF and LIF receptors. We show that the CRH1 domain is less conserved than other domains of the LR and that it can be deleted without any major effect on LR activation. We modeled the leptin⅐LR complex as a hexamer using the coordinates of the IL-6 receptor as a template. In this model, each leptin molecule interacts with three different LR chains.

MATERIALS AND METHODS
Molecular Modeling of the CRH1 Domain-LR sequences in UniProt (17) were detected using PSI-BLAST (18), and all homologs were aligned using T-Coffee (19). For modeling the CRH1 domain, we used the CRH domain structural superpositions and alignments used for modeling the CRH2 domain (20). In this alignment, the CRH domains of the growth hormone receptor (Protein Data Bank code 1AXI), the prolactin receptor (code 1F6F), the erythropoietin receptor (code 1EER), gp130 (codes 1I1R and 1BQU), and the G-CSF receptor (code 1CD9) were superposed and aligned with their homologs from other animal species (19) in the MOE sequence alignment editor (Chemical Computing Group). This alignment and the T-Coffee alignment of the LR CRH1 domains were used to create two profile libraries, which were aligned with each other using T-Coffee. Based on this alignment, 10 molecular models were built with MOE for the mouse LR CRH1 domain using the gp130 receptor CRH domain (code 1I1R) as a template.
Molecular Modeling of the Ig-like Domain-The structure of the gp130 Ig-like domain (Protein Data Bank code 1I1R) was used as a template for modeling. Different structures of the Ig-like domain of gp130 (codes 1I1R and 1N26) were superposed in MOE with the structure of the Ig-like domain of the IL-6 ␣-receptor (1N26) and with 20 Ig-like domain structures that appeared as similar according to the FSSP fold classification (21). This superposition allowed us to detect structurally conserved regions. For alignment of the LR Ig-like domain with its gp130 template, the alignments of several methods were compared. The sequences of members of the G-CSF receptor, gp130, and LR families were aligned using T-Coffee. This alignment was used as the main guidance for modeling. The alignment was compared with the Pfam (pfam06328.2) (22) and Conserved Domain Database (23) alignments for the Lep_receptor_Ig domain.
In the MOE sequence alignment editor, the alignment of the LR Ig-like domain was manually edited; a large shift of the C-terminal sequence of the LR domain was needed to allow a disulfide bridge between Cys 411 and Cys 416 . Smaller shifts were made to ideally position the loops. Using the alignment and the gp130 structure (Protein Data Bank code 1I1R) as a template, we built 150 models in MODELLER Version 8.1 (24) using automodel with introduction of a Cys 411 -Cys 416 disulfide bridge. 10 models with the best DOPE and molpdf scores were selected and evaluated.
Homology Modeling of a Hexameric Leptin⅐LR Complex-A hexameric leptin⅐LR complex was created using the structure of the IL-6⅐IL-6 ␣-receptor⅐gp130 complex (Protein Data Bank code 1P9M) (25) as a template. The leptin⅐CRH2 domain complex was modeled as described (20), and the Ig-like domain was joined to the N terminus of the CRH2 domain model. The model for the CRH2 domain was superposed on the CRH domain of gp130, bringing leptin into a position similar to that of IL-6. The CRH2 domain model without leptin was superposed on the CRH domain of the IL-6 ␣-receptor. Rigid trimers of one leptin molecule interacting with two LR models via their CRH domains were then created by mutually fixing the coordinates of the three molecules. As leptin helices are shorter than IL-6 helices, the rigid trimers were brought closer to each other, followed by a rotation of the Ig-like domain around LR residues 423-424 to fit the critical cluster in the Ig-like domain to binding site III of leptin, bringing oppositely charged residues in both molecules in close contact.
Vectors and Construction of Mutants-The pMET7-mLRlo vector allows expression of the mouse LR with an additional C-terminal Myc tag. Mutations in this vector were generated using the QuikChange site-directed mutagenesis procedure (Stratagene). N-terminal deletion mutants were constructed in this vector by changing the two codons upstream of the codon for the new N terminus to ACCGGT with the introduction of a unique AgeI restriction site. The pMET7-SIgK-HAleptin vector contains the SIgK signal peptide (9). A pMET7-SIgK-HA-leptin_AgeI vector was created by introduction of an ACCGGT AgeI restriction site at the codons for the two C-terminal amino acids Thr and Gly of the SIgK signal peptide. Both pMET7-mLRlo and pMET7-SIgK-HA-leptin_AgeI contain an identical unique SacI site downstream of the coding sequence. The LR AgeI/SacI fragment was ligated into the AgeI/SacI-opened pMET7-SIgK-HA-leptin_AgeI vector, creating a vector in which the SIgK signal peptide is followed by the new N terminus of the LR.
All constructs in this study were verified by DNA sequence analysis of the complete LR sequence insert. The sequences of oligonucleotide primers used for mutagenesis are included in the supplemental table.
Transfection of HEK293T Cells-HEK293T cells were kept in culture at 37°C in an 8% CO 2 humidified atmosphere using Dulbecco's modified Eagle's medium with 4500 mg/liter glucose (Invitrogen) and 10% fetal bovine serum (Cambrex Corp.). For a typical experiment, 4 ϫ 10 5 HEK293T cells were seeded in a 6-well plate a day before transfection and grown using Dulbecco's modified Eagle's medium with 10% serum and 50 g/ml gentamycin. In brief, HEK293T cells were cotransfected using a standard calcium phosphate transfection procedure with the pXP2d2-rPAP1 plasmid, the pUT651 plasmid, and the pMET7-mLRlo plasmid containing the wild-type or mutant LR. The pXP2d2-rPAP1luciferase plasmid contains the luciferase gene under the control of the STAT3-inducible rat Pap1 (pancreatitis-associated protein-1) promoter. The pUT651 plasmid (Eurogentec) contains the Escherichia coli lacZ gene under the control of the constitutively active cytomegalovirus promoter. The day after transfection, cells were resuspended in cell dissociation buffer (Invitrogen). A portion of the cells were used for a luciferase assay; these were seeded in a black 96-well plate (Nunc) and stimulated overnight with different concentrations of mouse leptin (R&D Systems) for STAT3-dependent reporter assays. A portion of the cells were seeded in a black 96-well plate (Nunc) without leptin stimulation for ␤-galactosidase reporter assays. The remainder of the cells were cultured overnight in a 6-well plate, resuspended in cell dissociation buffer, and used for fluorescence-activated cell sorter (FACS) analysis (see below).
STAT3-dependent Reporter Assays-Luciferase assays were performed in duplicate as described (26). After overnight stimulation with leptin, induced luciferase activity was measured by chemiluminescence. mM ATP (final pH 7.8)) was added. Luciferase activity was measured in a Packard TopCount chemiluminescence counter (PerkinElmer Life Sciences). Luciferase data were normalized according to ␤-galactosidase activity. The normalized data were fitted to a one-site binding curve (hyperbola) using GraphPad Prism 2.0 software to determine EC 50 values.
␤-Galactosidase Normalization-␤-Galactosidase activity was measured in triplicate using the Galacto-Star TM chemiluminescence detection kit (Tropix) and the TopCount chemiluminescence counter.
FACS Analysis-After resuspension, cells were incubated with a mixture of two rat anti-mouse LR monoclonal antibodies (27) as primary antibody at a 1:1500 dilution in FACS buffer (phosphate-buffered saline with 1% fetal bovine serum, 50 g/ml gentamycin, and 0.5 mM EDTA). These antibodies specifically recognize the fibronectin type III domain of the LR. As secondary antibody, goat anti-rat IgG conjugated to Alexa Fluor 488 (Molecular Probes) was used at a 1:500 dilution in FACS buffer. Cells were resuspended in 200 l of FACS buffer and measured in a FACSCalibur apparatus (BD Biosciences).

RESULTS
Modeling of the LR CRH1, CRH2, and Ig-like Domains-Our model for the LR CRH2 domain bound to leptin has been described (20). Homology models were built for the mouse LR CRH1 domain using the crystal structure of the gp130 CRH domain as a template (28). The CRH1 domain contains a long insertion at residues 165-183 that could not be reliably modeled.
The Ig-like domain was modeled using the Ig-like domain of gp130 (Protein Data Bank code 1I1R) as a template (28). The alignment of these domains is shown in Fig. 1. The disulfide and N-glycosylation patterns have been determined for the human LR by Haniu et al. (29) by mass spectrometry. The positions of these glycosylated residues and the disulfide pattern are in complete agreement with the models. These mass spectrometry data indicate that the Ig-like domain of the LR has a disulfide bridge between Cys 350 and Cys 410 and a unique extra disulfide bridge between Cys 411 and Cys 416 that is not found in the Ig-like domains of gp130; the LIF, oncostatin M, or G-CSF receptor; or the IL-6 ␣-receptor. Cys 411 and Cys 416 are completely conserved in all 13 LR sequences. The disulfide-linked loop between ␤-sheets 6 and 7 is shorter in the LR. Modeling Cys 411 and Cys 416 as a disulfide bridge brings Cys 416 into the position of Tyr 116 in gp130. This suggests that, although not conserved in the LR, the linker sequence between the Ig-like and CRH2 domains is of exactly the same length as in the G-CSF receptor and the gp130 family members.
An alignment of 13 LR sequences was built using T-Coffee. The conservation of residues in the alignment was assessed by calculating the ClustalQ scores of the alignment (30). Fig. 2 shows the models for the Ig-like, CRH2, and CRH1 domains, in which the residues are colored according to the calculated residue conservation. Residues at the surface of the CRH2 domain are very well conserved, and residues 501-504 of the CRH2 domain ␤5-␤6 loop, previously shown to be involved in leptin binding (20,31), are completely conserved, as are the residues of the WSXWS motif, which is probably essential for the proper folding of the CRH domains (32,33). The surface residues of the CRH1 domain are clearly less well conserved, with the exception of the WSXWS motif and a small patch formed by Trp 127 , Cys 131 , Pro 224 , Met 225 , and Ser 227 . The Ig-like domain is very well conserved, with a large surface patch of residues in the antiparallel ␤-sheet formed by ␤-strands 3, 6, and 7 and residues in the connecting turns and loops (Fig. 2, A and B).
Mutagenesis of the Ig-like Domain-The conservation of the surface residues in the Ig-like domain suggests a possible role in interaction with leptin. We therefore mutated the conserved residues in this patch and tested the effects of these mutations on LR signaling. HEK293T cells were transfected with the LR mutants and stimulated with increasing concentrations of leptin. Leptin-induced STAT3 activation was measured using the rat Pap1 promoter-luciferase reporter assay, and luciferase activity was normalized using a ␤-galactosidase reporter. The leptin/luciferase activity dose-response curve for these mutants was fitted to a one-site binding curve (hyperbola) using GraphPad Prism 2.0. Cell-surface expression of the mutant LR was assessed by FACS. (A summary of the experiments is provided in the supplemental table.) The A407E and L370A mutations and the N369L/L370S double mutation led to an almost complete loss of signaling capacity (Fig. 3). The other mutations did not significantly alter the EC 50 of the dose-response curve, but some seemed to lower the plateau value of the hyperbolic dose-response curve, corresponding to lower maximal luciferase activity (MLA) in the experiments. This is shown in Fig. 3 for the Y409A, H417A, H418S, H417A/H418S, and Y333A mutations, which showed decreased MLA.
Looking for LR Mutants with Significantly Lowered MLA-The LR mutants were properly expressed at the cell surface according to FACS analysis. The LR surface expression level was determined as the product of (percentage of cells showing LR surface expression) ϫ (geographic  (44). Disulfide bridges in gp130 and in our model are indicated by brackets above and below the alignment, respectively. The ␤-strands are indicated by arrows. The CRH1 and CRH2 domain boundaries are shown below the alignment. LR residues that caused a significantly lower LR activity in this work are colored green. G-CSF and LIF receptor residues that have been shown to be involved in binding site III in previous mutagenesis studies (15,44,45) are colored similarly. gp130 residues involved in the IL-6/gp130 interaction in the crystal structure of the hexameric IL-6 receptor complex are colored yellow (25).

mean of their FACS fluorescence intensity). These values varied between experiments and between different LR mutants.
To assess the effect of the LR surface expression level on the leptin/ luciferase dose-response curves, HEK293T cells were transfected with increasing amounts of wild-type LR expression vector ranging from 0.05 to 2 g of the pMET7-mLRlo plasmid. The leptin/luciferase dose-response curve data were fitted to a one-site hyperbolic binding curve using GraphPad Prism 2.0. The LR expression levels did not seem to affect the EC 50 significantly, but had a drastic effect on MLA. MLA increased with increasing LR surface expression levels (Fig. 4A). When the percentage of cells showing LR surface expression was between 2.5 and 11%, the MLA/LR surface expression level did not differ by Ͼ56% (Fig. 4B). We therefore looked at the data sets for mutants with expression levels between 2.5 and 11% transfected cells and determined their relative MLA versus the wild-type control. A detailed analysis of these data is provided in Table 1. This identified the Y333A, Y409A, H417A, H418S, and H417A/H418S mutants as having significantly lower MLA.

Surface Compatibility of Binding Site III in Leptin and the Proposed Binding
Site III in the Ig-like Domain-Mapping of these mutants in our molecular model for the Ig-like domain showed that Leu 370 , Ala 407 , Tyr 409 , His 417 , and His 418 cluster together on the surface of the antiparallel ␤-sheet formed by ␤-strands 3, 6, and 7 (Fig. 5). This cluster forms a possible binding site III of the LR, interacting with binding site III in leptin. Comparison with residues involved in ligand binding in the Ig-like domains of gp130 and the G-CSF and LIF receptors (Fig. 1) shows that the cluster indeed overlaps with binding sites in these related receptor systems. We looked at the surface compatibility of binding site III in leptin and the proposed binding site III in the Ig-like domain. Leptin residues that affect LR signaling without affecting binding to CRH2 and that are thought to possibly affect binding site III are shown in Fig. 6A. A side view of the leptin molecule in Fig. 6C shows that binding site III around the N-terminal end of helix D forms a flat surface. Such flat surfaces are also found in binding sites III of the LIF and G-CSF receptors. A side view of the Ig-like domain shows that the ␤-sheet formed by ␤-strands 3, 6, and 7 and containing the conserved cluster critical for LR activation also forms a flat surface. This  HEK293T cells were transiently cotransfected with pMET7-mLRlo plasmids encoding different LR mutants, the pXP2d2-rPAP1-luciferase reporter construct, and the pUT651 plasmid for normalization. The transfected cells were either stimulated with different concentrations of leptin for 24 h or left unstimulated. Luciferase measurements were performed in duplicate and normalized using the ␤-galactosidase reporter. Data are representative of at least four separate transfection experiments. As positive control (PC ), wild-type pMET7-mLR was used; an empty vector was transfected as negative control (NC). Percentages of cells with LR cell-surface expression as determined by FACS are shown. The concentrations of leptin used (bars from left to right) were 0, 3.7, 11, 33, 100, 300, and 900 ng/ml. cps, counts/s. might offer a binding site for binding site III of leptin because both binding sites display complementary electrostatic surface potentials (Figs. 5B and 6D). The Ig-like domain has a positive electrostatic surface potential due to the presence of His 417 , His 418 , Arg 419 , and Lys 337 , whereas binding site III in leptin has a negative surface potential due to the presence of Glu 108 , Asp 111 , Glu 115 , and Glu 122 . Manual docking in a hexameric leptin⅐LR complex model (see below) showed that these residues possibly interact and that a hydrophobic interac-tion is possible between LR Leu 370 and leptin Val 30 . With the exception of Asp 111 , all these residues are conserved.
Effect of Partial or Complete Deletion of the CRH1 Domain-Niv-Spector et al. (34) demonstrated that mutations in 39 LDFI 42 of leptin can lead to a leptin mutant that still binds to the receptor, but loses its receptor activation capacity. The authors suggest that 39 LDFI 42 is part of binding site III and that it interacts with 325 VFTT 328 in the LR. Mutagenesis of 325 VFTT 328 to alanine in the mouse LR leads to a drastic  reduction of receptor activity (34). This model differs from our hexameric model, in which Phe 41 is part of binding site I. 325 VFTT 328 is the C-terminal ␤-strand of the CRH1 domain model (Fig. 1). Previous stud-ies demonstrated that large parts of the CRH1 domain can be removed without large effects on LR signaling (10,12). However, in these studies, the last strand of the CRH1 domain was not deleted. We therefore made deletion mutants of the LR lacking the entire CRH1 domain (amino acids 1-328), residues 1-321, or the N-terminal fibronectin type III subdomain of CRH1 (amino acids 1-233). When tested in rat Pap1 promoter-luciferase assays, these deletions did not significantly alter the EC 50 of the leptin/luciferase dose-response curve and had only a very limited effect on MLA (Table 1). These data argue against a critical role for the CRH1 domain in LR activation and against a role for 325 VFTT 328 in binding site III interactions.
The L39A/D40A/F41A/I42A, L39A/D40A/F41A, F41S, and S120A/ T121A leptin mutants completely retained their effect on LR activation with all the CRH1 domain deletion mutants (data not shown), showing that impaired interaction with VFTT in the LR is not the cause of the effects of these leptin mutations. The mutations might, however, drastically affect the binding site I interaction. The mutants of the LDFI motif proposed by Niv-Spector et al. (34) always contained an L39A and/or I42A mutation. Leu 39 and Ile 42 are completely buried in the leptin structure and interact with the buried Val 124 in the N terminus of helix D. Residues 120 -122 in the N terminus of helix D are in the center of binding site III. It is therefore not impossible that mutations of Leu 39 and Ile 42 also indirectly affect binding site III.
Conservation of Three Putative Binding Sites in Leptin-We showed previously that mutations in three different sites on the leptin surface can affect LR activation. These sites correspond to three clusters of conserved surface-accessible residues. The largest conserved cluster is the previously identified binding site II at the surface of helices A and C (9,20). A second conserved area is found around binding site III (9). As shown in Fig. 6B, site III shows a patch of very conserved solvent-ex-   (9), are shown as space-filling spheres and are colored according to the effect on MLA. Yellow, inactive; white, 30% of the wild-type leptin level; cyan, Ͻ60% of the wild-type leptin level; blue, Ͻ80% of the wild-type leptin level; dark blue, no significant effect. B, same view of the molecule with the residues colored according to conservation as described in the legend to Fig. 2. The very conserved Ser 120 and Thr 121 residues (involved in the S120A/T121A mutation, which abrogates signaling) are colored yellow. Binding sites I and III are separated by an unconserved ridge. C, side view of the leptin molecule showing that binding sites I and III are spatially apart. Binding site III forms a flat surface. Binding site II, formed by the parallel helices A and C, is extremely conserved. D, electrostatic surface potential of the leptin molecule shown in the same orientation as in A and B. The upper (10 kcal mol Ϫ1 q Ϫ1 ; blue; where q Ϫ1 is elementary proton charge) and lower (Ϫ10 kcal mol Ϫ1 q Ϫ1 ; red) values of the potential are identical as in Fig. 6 for reason of comparison. posed residues, including the very conserved surface-exposed residues Val 30 , Ser 70 , Ser 120 , Thr 121 , and Glu 122 . A third conserved surface cluster contains the conserved residues Val 36 , Thr 37 , Phe 41 , and Phe 43 (Fig. 6, B  and C). We showed previously (9) that a F41S mutation lowers the STAT3 signaling of the LR and classified this residue as a possible binding site I residue based upon structural superposition with IL-6.
Homology Modeling of a Hexameric Leptin⅐LR Complex-To assess the possibility of a hexameric leptin⅐LR complex with three binding sites/leptin molecule, we built models for the complex of leptin with its receptor using the crystal structure of the IL-6⅐IL-6 ␣-receptor⅐gp130 complex (25). Both the gp130 and IL-6 ␣-receptor chains were replaced with the LR CRH2 domain/Ig chain as described under "Materials and Methods" (Fig. 7). In this hexameric model, each leptin molecule interacts with three LR molecules. 1) The conserved surface patch around Phe 41 interacts with the CRH2 domain of LR2, which takes the place of the ␣-receptor chain. The very conserved residues Val 36 , Thr 37 , Phe 41 , and Phe 43 form a hydrophobic protuberance on the leptin surface, which fits right between the cleft between the two fibronectin type III subdomains of the CRH2 domain of LR2. In our model, these four residues interact with the very conserved LR residues Leu 469 , Leu 504 , Leu 528 , and Leu 617 and the conserved LR residue Ser 530 . The very conserved LR residues Leu 469 and Tyr 470 seem to fit in a conserved hydrophobic pit in the leptin surface formed by residues 36 -39 and 41-43. Q134S/D135S and Q138S/Q139S/V142A mutations in helix D of leptin superposing with binding site I in IL-6 also lower STAT3-dependent LR signaling (9). In the model, these residues are involved in interactions with residues of the ␤10 -␤11 and ␤15-␤16 loops of the CRH2 domain. However, Gln 134 , Asp 135 , Gln 138 , and Val 142 are not conserved in human leptin, and Gln 138 and Val 142 are replaced with the more hydrophobic tryptophan and leucine residues. The interaction via these residues might differ between mouse and human leptins. 2) Binding site II interacts with the CRH2 domain of LR1. 3) Binding site III interacts with the Ig-like domain of LR3. As helix D of leptin is shorter than helix D of IL-6, LR1 and LR2 are much closer to each other and even interact directly.

DISCUSSION
We previously found evidence for the existence of binding sites I-III in leptin (9). The interactions between the CRH2 domain of the LR and binding site II of leptin have been demonstrated and mapped in detail (9,20).
In the gp130 receptor family and the G-CSF⅐G-CSF receptor complexes, binding site III (located around the N terminus of helix D) interacts with the Ig-like domain of a second receptor. To demonstrate the  role of the Ig-like domain as a possible binding site III homolog in the LR, we first built molecular models for this domain. Analysis of residue conservation of surface residues in the models showed that a large surface patch of the Ig-like domain seems to be particularly conserved. Using site-directed mutagenesis and STAT3-dependent luciferase reporter assays, we have demonstrated that mutations in a conserved cluster of Leu 370 , Ala 407 , Tyr 409 , His 417 , and His 418 have significant effects on LR activation. These residues are found around the antiparallel ␤-sheet formed by ␤-strands 3, 6, and 7 (Fig. 5). Comparison with residues involved in ligand binding in the Ig-like domains of gp130 and the G-CSF and LIF receptors (Fig. 1) showed that the cluster indeed overlaps with binding sites in these related receptor systems. We therefore propose that this cluster forms the core of binding site III of the LR, interacting with binding site III in leptin. Binding sites III of leptin and the LR have compatible shapes and electrostatic surface potentials.
Whereas the interaction partners of binding sites II and III are most likely the CRH2 and Ig-like domains of the LR, respectively, the identity of a putative interaction partner for binding site I is less clear, as leptin does not have an ␣-receptor. We have examined the possibility that the LR would take the position of both the IL-6 ␣-receptor and gp130 by homology modeling of a 2:4 leptin⅐LR complex.
This hexameric model seems to give optimal fits for binding sites I-III. Binding site I interaction involves a hydrophobic protrusion around leptin Phe 41 , which fits in the CRH2 domain (Figs. 7 and 8). Binding site II interactions center around Leu 13 in leptin, interacting with Leu 504 in the CRH2 domain (20,31). Binding site III interactions seem to have a large electrostatic component, with four opposite charges in leptin binding site III and the Ig-like domain. The three conserved clusters at the surface of leptin match perfectly the three binding sites (Fig. 8, A-C), whereas the corresponding interaction sites in the LR are also very conserved. The hexameric model also fits nicely with the leptin mutations that were demonstrated previously to affect CRH2 domain binding or STAT3 MLA (Fig. 8D) (9). The region around Leu 504 in the CRH2 domain plays a central role in interaction with binding sites I and II in leptin (20,31), as is the case for the CRH domain of the erythropoietin and growth hormone receptors (35,36).
A hexameric leptin⅐LR complex offers a rational explanation for several previous observations. 1) We showed previously that LRs in which the box-1 motif is deleted (LR⌬box-1) or three intracellular tyrosines are mutated to phenylalanines (LR-F3) are unable to generate a STAT3-dependent signal; however, when coexpressed, LR⌬box-1 and LR-F3 complement each other and are able to generate a STAT3-dependent signal upon leptin stimulation (12). This phenomenon can be explained by higher order clustering, with more than two LR chains/leptin⅐LR complex (12).
A hexameric model can explain this JAK/STAT complementation (12) by assuming that the LR⌬box-1 chain takes the position of the LR2 and LR4 chains and that the LR-F3 chains take the position of the LR1 and LR3 chains. In this scenario, JAK cross-phosphorylation happens via JAK molecules associated with LR-F3 chains LR1 and LR3, which take the place of gp130. The activated JAK molecules can than activate STAT3 molecules associated with LR3 and LR4. This also explains why the Ig-like domain has to be present in LR-F3 for complementation, whereas it is not strictly required in the LR⌬box-1 chain. Hexameric complex formation requires interactions between binding site III and the Ig-like domains in LR1 and LR3, whereas association of LR2 and LR4 in the complex does not require the Ig-like domain (8).
2) Ligand-independent clustering of the LR has been demonstrated in solution (13,27) and at the cell surface (13,(37)(38)(39). Couturier and Jockers (37) used a bioluminescence resonance energy transfer assay to show that 60% of the LR population exists as dimers and that leptin induces conformational reorganization of these preformed LR complexes. At least part of the LR population exists as disulfide-linked dimers at the cell surface, even in the absence of leptin, and both the isolated CRH2 domain and the two fibronectin type III domains can form disulfidelinked dimers (13). In the hexameric model complex, the C termini of the CRH2 domains of LR1 and LR2 approach each other. It is thus very possible that LR1 and LR2, as well as LR3 and LR4, can be disulfidelinked to each other; our hexameric model suggests that Cys 602 in the CRH2 domain and Cys 672 in the membrane-distal fibronectin type III domain are situated close to the C terminus of the CRH2 domain at positions that probably allow disulfide bond formation. Moreover, mutation of these two cysteine residues leads to a drastic reduction of STAT3-dependent signaling (13). 4 3) We have shown that mutations in binding site II of leptin can have drastic effects on the affinity for the isolated CRH2 domain without affecting the EC 50 value for signaling. This was explained by assuming that leptin binds in a preformed complex with multiple binding sites, partially compensating for the lower affinity. The additional binding site might very well be the second CRH2 domain, which interacts with binding site I.
Several findings support differences between the complex formation mechanisms in the leptin and G-CSF receptor systems. Residue conservation analysis of the G-CSF structure and mutagenesis studies did not show a binding site I in G-CSF. Consistent with this, the G-CSF receptor system did not show JAK/STAT complementation. 4 The interaction of leptin binding site III with the LR Ig-like domain is weak and cannot be demonstrated in binding assays. In contrast, the interaction of G-CSF with the Ig-like domain of the G-CSF receptor can be observed in solution, suggesting a higher affinity for the G-CSF binding site III/Ig-like domain interaction (40). Although binding site III of G-CSF shows a central exposed phenylalanine, no such large hydrophobic interactions are predicted for binding site III of the LR. These differences in affinity for binding site III might explain the need for a hexameric leptin⅐LR complex with an extra binding site I and LR2 and LR4 chains, in contrast with a tetrameric G-CSF⅐G-CSF receptor complex.
It is striking that most mutations in the Ig-like domain do not significantly affect the EC 50 of the leptin/luciferase dose-response curve, but affect only MLA. We found previously (9) exactly the same effect for leptin mutants in binding sites I and III. This can be explained by assuming that the EC 50 value is determined by the binding step that leads to formation of an initial leptin⅐LR complex, which is, to a large extent, determined by binding site II. Formation of the actual signaling complex requires interactions between two initial leptin⅐LR complexes, probably via binding site III. Theoretical fitting of such a reaction scheme shows that the EC 50 value is affected only by the K d1 for the initial binding, but not by the K d2 for the interactions between initial complexes. Changes in K d2 affect the amount of the final receptor complex and thus maximal reporter activity, as observed for binding site I and III mutants.
Like the LIF and oncostatin M receptors, the LR contains an N-terminal CRH1 domain. In the LIF receptor, the CRH1 domain interacts directly with the ciliary neurotrophic factor receptor (41) and is essential for ciliary neurotrophic factor binding and signaling, whereas it is not needed for signaling by LIF and oncostatin M (42). Deletion of the LR CRH1 domain seems to have only a limited effect on signaling in our experimental setup, which involves overexpression of the LR. It cannot be excluded that the role of the CRH1 domain is more important at lower LR concentrations. A Q269P mutation in the CRH1 domain causes the obese phenotype of the fa/fa rat (43) and results in defective (partially constitutive) LR signaling (27). In our model, the Q269P mutation leads to severe steric clashes between the introduced proline residue and the first tryptophan of the CRH1 WSXWS motif, probably affecting the stability or correct folding of this domain.
In conclusion, we have shown here that Leu 370 , Ala 407 , Tyr 409 , His 417 , and His 418 at the surface of the LR Ig-like domain form a conserved cluster that is critical for LR activation. We propose that this cluster in the Ig-like domain of the LR interacts with binding site III of leptin. A conserved hydrophobic protuberance around Phe 41 fits perfectly in the CRH2 domain model and forms binding site I. Combined with our previous studies, these new observations lead us to propose a 2:4 hexameric leptin⅐LR complex in which each leptin molecule interacts with three LR molecules via binding sites I-III.