Two Different Contact Sites Are Recruited by Cardiotrophin-like Cytokine (CLC) to Generate the CLC/CLF and CLC/sCNTFRα Composite Cytokines*

The cytokines of the interleukin-6 family are multifunctional proteins that regulate cell growth, differentiation, and other cell functions in a variety of biological systems including the immune, inflammatory, hematopoietic, and nervous systems. One member of this family, ciliary neurotrophic factor (CNTF), displays biological functions more restricted to the neuromuscular axis. We have recently identified two additional ligands for the CNTF receptor complex. Both are composite cytokines formed by cardiotrophin-like cytokine (CLC) associated to either the soluble type I cytokine receptor CLF or the soluble form of CNTF receptor α (CNTFRα). The present study was aimed at analyzing the interactions between the cytokine CLC and its different receptor chains. For this purpose, we modeled CLC/receptor interactions to define the residues potentially involved in the contact sites. We then performed site-directed mutagenesis on these residues and analyzed the biological interactions between mutants and receptor chains. Importantly, we found that CLC interacts with the soluble forms of CNTFRα and CLF via sites 1 and 3, respectively. For site 1, the most crucial residues involved in the interaction are Trp67, Arg170, and Asp174, which interact with CNTFRα. Surprisingly, the residues that are important for the interaction of CLC with CLF are part of the conserved FXXK motif of site 3 known to be the interaction site of LIFRβ. Obtained results show that the Phe151 and Lys154 residues are effectively involved in the interaction of CLC with LIFRβ. This study establishes the molecular details of the interaction of CLC with CLF, CNTFRα, and LIFRβ and helps to define the precise role of each protein in this functional receptor complex.

One member of this family, CNTF, displays a narrow spectrum of activity restricted principally to the neuromuscular axis. It promotes the differentiation and survival of a wide range of cell types in the nervous system (18). In particular, CNTF maintains motor neuron viability in vitro, prevents the degeneration of axotomized motor neurons, and attenuates motor deficits in different strains of mice with neuromuscular deficiencies (19 -23). In addition to its activities in the nervous system, CNTF has trophic effects on denervated skeletal muscle and is a regulator of muscular strength in aging (24,25). To exert its biological functions, CNTF binds first to the CNTF receptor ␣ chain (CNTFR␣) and subsequently recruits the transducing subunits gp130 and LIFR␤.
The existence of additional ligand(s) for the CNTF receptor (CNTFR␣) was first suggested by a study comparing the phenotypic consequences of disrupting CNTF versus CNTFR␣ genes. Unlike mice lacking CNTF, those lacking CNTFR␣ die perinatally and display severe motor neuron deficits. Thus, the CNTFR␣ subunit plays a critical role during development by serving as a receptor for a second developmentally important ligand. Moreover, a null mutation in the human CNTF gene does not lead to neurological disease (26).
We have recently identified additional ligands for the CNTF receptor complex as neuropoietin (8) and as the composite cytokine formed by CLC (also known as novel neurotrophin-1/B cell stimulatory factor-3), associated to the soluble type I cytokine receptor CLF (13). We initially observed that CLC, although containing a signal peptide, was inefficiently secreted when expressed in mammalian cells (13). Its secretion could only be induced upon co-expression with CLF, leading to the formation of the CLC/CLF composite cytokine (13). Like CNTF, CLC/CLF recruits cells expressing the tripartite CNTF receptor on their surface; induces the tyrosine phosphorylation of gp130, LIFR␤, and STAT3 in neuroblastoma cells; and acts as a survival factor for motor neurons cultured in vitro (27).
We subsequently observed that CLC could also form a second secreted composite cytokine when associated with the soluble form of CNTFR␣ (sCNTFR␣). Similarly to LIF, CLC/sCNTFR␣ displays activities on cells only expressing gp130 and LIFR␤ on their surface (14). For both CLC/CLF and CLC/sCNTFR␣, there is no covalent link between CLC and the soluble receptor portions. We have also demonstrated that a fusion protein comprising CLC covalently tethered to sCNTFR␣ via a glycine/ serine linker shows enhanced activities compared with the native heterodimeric cytokine (28).
The association of CLC with sCNTFR␣ is similar to that reported previously for CNTF/sCNTFR␣ (29), IL-6/sIL-6R␣ (30), and IL-11/sIL-11R␣ ( (15), where composite cytokines implicating a soluble receptor ␣ component in their structure display functional activities mediated through the appropriate signaling subunits (11,15,(32)(33)(34)(35)(36). A closely related situation also exists for the IL-12 and IL-23 heterodimeric cytokines, composed of an ␣-receptor-like chain (p40) respectively associated to p35 and p19 (37,38). The association between CLC and CLF is also closely related to the recently described heterodimeric cytokine IL-27 where p28 binds the soluble receptor EBI-3 (7,39). It is noteworthy to underline that for CLC, p35, p19, and p28, an association to the soluble receptor counterpart is an absolute requirement to allow the cytokine moiety secretion. These studies reveal an interesting degree of mechanistic promiscuity between the IL-6 and IL-12-type ligands and their multichain receptor complexes.
The IL-6 family cytokines belong to the long chain four ␣-helix bundle class first defined in 1991 (40). Site-directed mutagenesis studies have shown that these cytokines interact successively with their receptor chains through three binding sites, numbered from 1 to 3 by analogy with growth hormone (GH) (41). Cytokines requiring a co-receptor ␣ chain bind to this receptor by site 1 formed by the association of the C -terminal parts of the AB loop and of the ␣D helix. gp130 interacts through binding site 2 located on the solvent exposed faces of the ␣A and ␣C helices of the cytokine. These sites are similar to sites 1 and 2 of GH (41). An additional site of interaction (site 3) corresponds to a second gp130-binding site for IL-6 and IL-11 and to a LIFR␤ binding site for LIF, OSM, CT-1, and CNTF. The LIFR␤-binding site is characterized by an FXXK motif located at the N terminus of ␣D helix of the cytokine and constitutes the signature of this interaction. Based on the structural features of CLF and CNTFR␣, these receptors were expected to interact with CLC via site 1.
The present study was aimed at analyzing the interactions between the cytokine CLC and its different receptor chains. For this purpose, we modeled CLC to define different residues potentially implied in the interactions. We then performed site-directed mutagenesis on these residues and analyzed the biological interactions between mutants and receptor chains. Importantly, the results obtained reveal that interactions of CLC with either sCNTFR␣ or CLF involved different recognition patterns depending on the composite cytokine studied.
All cytokines that have a resolved structure (IL-6, vIL-6, CNTF, LIF, OSM, erythropoietin, and GH) and CLC were analyzed with the Network Prediction Server @nalysis server at the Pôle Bioinformatique Lyonnais (Lyon, France) (44) to build a consensus structure prediction on each sequence.
Molecular Modeling-CLC and CNTFR␣ were homology modeled by satisfaction of spatial restraints using the program Modeler (45) implemented on a SGI Octane work station (SGI, San Diego, CA). Human CNTF (Protein Data Bank code 1CNT) (46) and LIF (Protein Data Bank code 1PVH) (47) were used as templates for the molecular modeling of CLC. In a first approach, the atomic coordinates of CNTF were used to model the hydrophobic core of the protein. This modeling step was required to fix side chain orientations of main residues potentially involved in the interaction of CLC with its receptor subunits. The GH loop fragment between Asn 63 and Leu 76 was subsequently used to model the corresponding part on CLC (Asp 63 -Leu 76 ). In the third step of the analysis, a 12-residue ␣-helix was designed using the Biopolymer module of InsightII (Accelrys, San Diego, CA). This helix has been fitted to the short helix present on the GH-like modeled loop of CLC. This step increased the size of the AB loop ␣AЈ-helix in accordance with the secondary structure prediction (see Fig. 1). In the final steps of the molecular modeling process, the N-terminal part of the AB loop and the ␣D helix of LIF were used to model the corresponding fragment on CLC. CNTFR␣ was modeled based on the structure of IL-6R␣ (Protein Data Bank code 1P9M) (48) and of p40 for the DE loop (Protein Data Bank code 1F45) (49). All of the models were energy-minimized and quality-verified using the Profiles3D (50) and Procheck programs (51).
CLC Site-directed Mutagenesis-The pcDNA3 vector containing the cDNA encoding the human CLC tagged at the C terminus with c-Myc and protein C epitopes was subjected to site-directed mutagenesis using the QuikChange TM site-directed mutagenesis kit (Stratagene, La Jolla, CA) following the manufacturer's instructions. Mutations were performed on the putative site 1 and site 3 residues of CLC as described under "Results." Cell Transfection and Protein Purification-cDNAs encoding wild type CLC, CLC mutants, CLF, or sCNTFR␣ were transfected in Cos-7 cells using the ExGen reagent (Euromedex, Souffelweyersheim, France) according to the manufacturer's instructions. After a 72-h culture period, the supernatants were harvested and subsequently purified by affinity chromatography using an anti-protein C affinity column (Roche Applied Science). Protein purity and concentrations were determined by SDS-PAGE and silver staining of the gels using a bovine serum albumin standard.
Immunoprecipitation and Western Blotting-72 h after transfection with the appropriate cDNAs, the cell supernatants were harvested, and the cells were lysed in 10 mM Tris-HCl, pH 7.6, 5 mM EDTA, 50 mM NaCl, 30 mM sodium pyrophosphate, 50 mM sodium fluoride, 1 mM sodium orthovanadate, protease inhibitors (1 g/ml pepstatin, 2 g/ml leupeptin, 5 g/ml aprotinin, 1 mM phenylmethylsulfonyl fluoride), and 1% Brij 96. The cell lysates and supernatants were then immunoprecipitated with either the anti-c-Myc, the anti-CLF (AN-F-C6), or the anti-CNTFR␣ (AN-C2) mAb (10 g/ml) overnight at 4°C. The complexes were then isolated using beads coupled to protein A (Amersham Biosciences), submitted to SDS-PAGE, and transferred to Immobilon membranes (Millipore). The membranes were successively incubated with the biotinylated HPC4 mAb and a polystreptavidin-peroxidase solution. The reaction was visualized on x-ray film using ECL reagent (Amersham Biosciences) according to the manufacturer's instructions.
Tyrosine Phosphorylation Analysis-After a 24-h incubation period in serum-free medium, SK-N-GP cells were stimulated for 10 min with the appropriate cytokines (52). The cells were then lysed as described under "Immunoprecipitation and Western Blotting" except that the detergent used was Nonidet P-40. The lysates were subjected to SDS-PAGE and immunoblot analysis with a mAb specific for the tyrosine phosphorylated form of STAT3. The membranes were stripped in 0.1 M glycine, pH 2.5, for 15-24 h and neutralized in 1 M Tris-HCl, pH 7.6, before reblotting with an antibody recognizing all of the STAT3 isoforms.
Flow Cytometry Analysis-BAF GLC cells were successively incubated for 30 min at 4°C with transfected Cos-7 cells supernatants as indicated, the appropriate primary antibody or an isotype control antibody (10 g/ml) and a phycoerythrin-conjugated anti-mouse antibody. Fluorescence was subsequently analyzed on a FACScalibur flow cytometer from Beckton Dickinson (Mountain View, CA).
Biological Assays-For proliferation assays, BAF GLC cells were seeded in 96-well plates at a concentration of 5 ϫ 10 3 cells/well in RPMI 1640 medium containing 5% fetal calf serum. Serial dilutions of cytokines tested were performed in triplicate. After a 72-h incubation period, 0.5 Ci of [ 3 H]thymidine was added to each well for the last 4 h of the culture, and the incorporated radioactivity was determined by scintillation counting. For IL-6 production assays, KB cells were plated in 96-well plates at a concentration of 5 ϫ 10 3 cells/well in culture medium containing 220 pg/ml of CNTF, LIF, WT purified CLC, or mutant purified CLC as indicated. After 48 h, the supernatants were harvested, and their IL-6 content was determined by enzyme-linked immunosorbent assay as described previously (13).

Molecular Modeling of CLC-
To build a precise CLC model, we first performed a multiple sequence alignment of the cytokine with orthologs and paralogs of the IL-6 family. A subgroup of this family of molecules was retained based on the involvement of LIFR␤ in their receptor complex. In addition to CLC, the subgroup encompasses CNTF, CT-1, LIF, and OSM. The analysis was carried out using the T-Coffee program ( Fig. 1) (42). A number of important residues defining putative contact sites between CLC and its three-receptor subunits were identified. Equivalent residues were previously reported as crucial for the interactions of the defined subgroup of cytokines with LIFR␤ (site 3) or CNTFR␣ (site 1). Candidate residues important for the site 1 interaction with CNTFR␣ are well conserved between CLC and CNTF ( Fig. 1). This was demonstrated with Trp 67 in the CLC sequence, which is conserved with Trp 64 in human CNTF, as well as for CLC Arg 170 and Asp 174 corresponding to Arg 171 and Asp 175 in CNTF. Site 3 identifies a region of interaction between LIFR␤ and its cognate ligands. This is defined by a conserved FXXK motif, which is present in CLC between residues Phe 151 and Lys 154 conserved in human CNTF (Phe 152 and Lys 155 ) (53). Site 2 identifies a region of interaction between the cytokine ligand and gp130. This motif is less prominent and consists more of exposed surfaces of ␣A and ␣C helices with multiple discrete and weak contact points for gp130. A number of candidate residues in CLC were identified but not further analyzed (data not shown).
The secondary structure prediction of CLC indicated the presence of the four long ␣-helices and an additional short ␣-helix in the C-terminal part of the AB loop we named AЈ (Fig.  1). The length of the interhelical loops are compatible with the characteristic up-up-down-down topology of the structural fold of the cytokines of the IL-6 family (40). We have modeled CLC on the basis of this multiple sequence alignment and the secondary structure prediction of CLC. We used CNTF as molecular modeling template for the helical core of CLC and LIF for the N-terminal parts of the AB loop and of the ␣D helix. The ␣AЈ helix in the AB loop was modeled de novo using the Biopolymer module of InsightII (Accelrys) and integrated in the structure during the molecular modeling process. Fig. 2 displays a ribbon representation of the energy-minimized CLC model. The putative interaction sites (sites 1 and 3) with CNTFR␣, CLF, and LIFR␤ are boxed and detailed in boxed colored zoomed views (Fig. 2, green for site 1 and pink for site 3). The site 1 of CLC is centered by the Trp 67 , a residue also conserved in CNTF. The ␣AЈ helix in the C-terminal part of the AB loop is a crucial element of the general shape of the site 1. Additional residues in the N-terminal part of the ␣AЈ helix (Ala 54 , Ala 60 , and Thr 61 ) are involved in the formation of hydrogen bonds with Trp 67 and residues of the ␣D helix (especially Arg 170 and Asp 174 ). This hydrogen bond network participates to the stability of the site 1 interface by packing the ␣AЈ helix against the four helix bundle and the Trp 67 side chain in the correct orientation. The site 3-binding epitope is classically composed of a sequential conserved FXXK motif (Fig. 2, pink  box). The phenylalanine residue, Phe 151 , is involved in a network of hydrogen bonds allowing the right orientation of its side chain. Moreover, Phe 151 contributes to the definition of a hydrophobic and aromatic cluster. The lysine residue in position 154 protrudes out of the surface of the protein, allowing an interaction with the receptor chains.
Molecular Docking of CLC with CNTFR␣, CBD, and LIFR␤ Ig-like Domain-To have a better understanding of CLC interactions with CNTFR␣ (site 1) and LIFR␤ (site 3) molecular docking was carried out. The complex between IL-6R␣ and human IL-6 was used to build the docked CLC/CNTFR␣ complex (Protein Data Bank code 1P9M) (48). The energy-minimized structure of the complex is presented in Fig. 3A. Trp 67 was shown to be a hot spot for the interaction. The side chain orientation of this residue is favorable for contacting the CNTFR␣ side chains. Trp 67 , Arg 170 , and Asp 174 are involved in an hydrogen bond network stabilizing the side chain orientation of the Trp 67 residue. The Trp 67 aromatic ring is also implicated in an aromatic cluster comprising the Phe 172 , Phe 199 , and Phe 238 residues of CNTFR␣. The Thr 61 side chain is exposed to solvent and is in an unfavorable orientation to be able to interact with CNTFR␣.
We performed the same approach to elucidate the interaction between LIFR␤ and CLC. The IL-6/gp130 complex (Protein Data Bank code 1P9M) (48) was used to model the interaction between the Ig-like domain of LIFR␤ and CLC site 3. As we have previously shown (54), the rotameric orientations of Asp 214 and Phe 284 of LIFR␤ Ig-like domain are favorable for interacting with the Lys 154 and Phe 151 , respectively. The molecular docking between both molecules has been carried out based on these observations. The structure of the docked complex was energy-minimized, and the resulting model is presented in Fig. 3B. To allow for an easier purification and identification of the proteins, wild type and mutant CLC were tagged with a c-Myc epitope followed by a protein C tag. Corresponding cDNAs were subcloned in a mammalian expression vector and transfected in Cos-7 cells. In accordance with the work we previously published (13,14), a strong expression of wild type CLC was detected in cell lysates, whereas the protein was not secreted into the culture media (Fig. 4A). A similar behavior was observed with CLC-generated mutants. Very slight and erratic signals were detected in culture supernatants for mutants T61A and K154A, very likely reflecting release of CLC into the supernatant by necrotic cells.
CLC Associates with sCNTFR␣ to Form a Secreted Composite Cytokine via Site 1-We previously reported that to be secreted CLC needed a preassociation with either CLF or sCNTFR␣ (13,14). To assess the secretion capacities of CLC mutants, we co-expressed them with sCNTFR␣ (Fig. 4B). After a 72-h culture period, culture supernatants and lysates were analyzed for their CLC content using an anti-tag antibody for detection. When mutants were expressed together with sCNTFR␣, eight of the nine mutations introduced in the putative site 1 of CLC abrogated the secretion of the protein (Fig. 4B). Only the T61A mutant was significantly secreted. In contrast, the mutations performed on the putative site 3 did not affect the secretion of the CLC/sCNTFR␣ composite cytokine, leading to a secretion level similar to that observed with the wild type cytokine. These results demonstrate that, to be secreted, CLC associates with the sCNTFR␣ through binding site 1. In contrast, mutations introduced in site 3 did not affect the secretion of CLC, indicating that the interaction between sCNTFR␣ and CLC was not site 3-dependent.
CLC Associates with CLF to Form a Secreted Composite Cytokine via Site 3-We then co-expressed the same set of CLC mutants with its alternate counterpart CLF. Similarly, after a 72-h culture period the samples were harvested and tested by Western blot for the presence of CLC in cell lysates or supernatants (Fig. 4C). In contrast to the observations in Fig. 4B, none of the site 1 mutations affected the interaction between CLC and CLF, and CLC was readily detectable in the supernatants of the cultures (Fig. 4C). More surprisingly, some of the mutations introduced in site 3 altered the secretion of CLC. Only a faint secretion of CLC K154A mutant was observed compared with the wild type form of the cytokine, and secretion of the double mutant F151A/K154A was entirely abrogated. These results clearly indicate that CLF interacts with CLC via site 3 or through a motif overlapping site 3.
Secreted Mutants Remain Associated with Their Co-receptors-We further assessed whether the secreted mutants were still able to complex with the CLF or sCNTFR␣ co-receptors. After transfection of cDNA coding for CLC mutants with either sCNTFR␣ or CLF in Cos-7 cells, the supernatants were immunoprecipitated with an anti-CNTFR␣ or an anti-CLF antibody and Western blotted with an anti-protein C antibody to reveal associated CLC. When immunoprecipitation was performed with an anti-CLF antibody (Fig. 5A), site 1 CLC mutants were still able to interact with CLF, indicating that site 1 is not required for a cognate interaction between CLC and CLF. In addition, as observed in Fig. 4C, when introducing mutations in site 3, we could not detect the CLC/CLF composite cytokine after the immunoprecipitation step. These results sustain the idea that CLC physically recognizes CLF via a site 3 interaction.
A mirror image was obtained after co-expression and immunoprecipitation of sCNTFR␣ (Fig. 5B), which correlates with the secretion pattern observed for the mutated CLC in Fig. 4B. Thus, site 1 CLC mutants could not be immunoprecipitated with sCNTFR␣, indicating that CLC interacts with sCNTFR␣ via site 1. In contrast, purified fractions obtained using site 3-mutated proteins still contain the CLC/sCNTFR␣ composite cytokine, indicating that CLC does not interact with sCNTFR␣ via site 3.
These opposing results obtained using either CLF or sCNTFR␣ also demonstrate that the mutations introduced in CLC do not disturb the three-dimensional structure of the protein, because a nonreactive mutant with a given counterpart becomes reactive when associated to its other counterpart. Furthermore, all forms of CLC generated above were still immunodetected by a panel of 5 different anti-CLC monoclonal antibodies (data not shown).
Taken together, these results lead to the conclusion that CLC interacts with sCNTFR␣ through site 1 and with CLF through site 3. To the best of our knowledge, this is the first example of an interaction between a cytokine site 3 and a soluble ␣-chain receptor, which is typically involved in site 1 interactions.
Binding Properties of the Mutants-We next examined by flow cytometry the binding properties of the CLC mutants on a target cell. Ba/F3 cells were modified to stably express the functional receptor for CNTF, i.e. the membrane form gp130, LIFR␤, and CNTFR␣ (BAF GLC) (55). This cell line acquires the ability to bind CLC/CLF when the tripartite CNTF receptor is expressed on its surface (13). Because all except one mutant (F151A/K154A) were secreted when co-expressed together with CLF, we chose to study the membrane binding of the CLC/CLF composite cytokine. Concentrations of wild type and mutated forms of CLC were quantified by Western blot. Binding to the cell surface was monitored by flow cytometry using an anti-c-Myc antibody allowing the recognition of the CLC/CLF complex bound to the cell surface as described previously (14) (Fig. 6). Most of the mutations introduced in site 1 affected the interaction of CLC with CNTFR␣ expressed on the surface of BAF GLC (Fig. 6). This was in agreement with the results obtained in Fig. 4B showing that site 1 is essential for a correct association between CLC and sCNTFR␣ to allow for secretion of the composite cytokine. One discrepancy with the secretion pattern was noted for the D174A mutant and to some level for the Q165D/T166H mutants, which, despite a lack of detectable association when co-expressed together with sCNTFR␣ (Fig.  4B), were still able to bind to the membrane form of receptor (Fig. 6). Regarding F151A and K154A single mutants, we failed to detect any association of them to the membrane (Fig. 6). These results show that CLC recognizes surface CNTFR␣ principally via site 1 residues Ala 60 , Trp 67 , and Arg 170 .
Functional Properties of Site 1-or Site 3-mutated Forms of CLC-We then assessed the functional properties of CLC mutants in a proliferation assay using the BAF GLC cell line. Mutated proteins with an inactivated site 1 (W67A and R170A/ D174A) or site 3 (F151A/K154A) were purified from transfected cell lysates and subjected to silver staining for quantification. Similarly, wild type CLC, as well as the T61A mute mutant were immunoseparated and used as controls in BAF GLC proliferation assays (Fig. 7). After a 72-h culture period, the proliferative response of the cells to the mutants was determined.
The results showed a lack of proliferation of BAF GLC cells grown in the presence of site 1-or site 3-mutated cytokines. The same CLC mutants were further studied by analyzing their ability to recruit the STAT3 signaling protein in the SK-N-GP neuroblastoma cell line expressing the gp130, LIFR␤, and CNTFR␣ receptor subunits (Fig. 8). A robust STAT3 tyrosine phosphorylation was observed in response to WT or to T61A CLC, as well as to CNTF and LIF, the control cytokines. In contrast, site 1 (W67A and R170A/D174A) and site 3 (F151A/ K154A) mutants failed to induce an activation of STAT3. We next assessed the ability of the purified proteins to induce an IL-6 production in the KB carcinoma cell line. This cell line, which expresses gp130 and LIFR␤, was stably transfected with CNTFR␣ (KB-CNTFR␣). KB-CNTFR␣ cells were previously shown to produce IL-6 in response to CNTF or to CLC (13). Purified CLC mutants were thus added to KB-CNTFR␣ cell cultures for 72 h followed by IL-6 supernatant determination by enzyme-linked immunosorbent assay (Fig. 9). Similarly to the proliferation and STAT3 activation assays, a complete impairment in the induction of IL-6 was seen when adding the mutated site 1 (W67A and R170A/D174A) or site 3 (F151A/ K154A) proteins to the cultures. Taken together, these functional data are in accordance with the preceding experiments and reinforce the contribution of sites 1 and 3 in CLC receptor recognition and activation.

DISCUSSION
Cytokines of the IL-6 family (CLC, CNTF, LIF, CT-1, NP, OSM, IL-6, IL-11, and IL-27) are characterized by the involvement of the gp130 receptor chain in their multimeric receptor complex. The recent discovery in this family of the three composite cytokines (CLC/CLF, CLC/sCNTFR␣, and IL-27) demonstrates a striking similarity between the IL-6 and IL-12 family cytokines (13,14). Although CLC possesses a signal peptide, this cytokine is not secreted. It must bind noncovalently to one of its soluble ␣ chain receptor components to form a secreted composite cytokine. This situation was first described for IL-12 where the cytokine component p35 must preassociate to its counterpart p40 before being secreted (38). It is noteworthy that p40 has the structure of a soluble ␣ chain receptor because it displays the so-called cytokine-binding domain (CBD) preceded by an Ig-like domain at the N terminus (49). Nevertheless, contrary to CLC/CLF and CLC/sCNTFR␣, p35 and p40 are associated by disulfide bridges that confer a higher stability to the complex. More recently, p40 was also found to bind to another ␣-helical protein called p19 to form the heterodimer IL-23 (37). A p40-related protein termed EBI-3 was identified as the p28 subunit counterpart in the IL-27 heterodimer (7, 56). All of these heterodimeric cytokines are formed in the producer cell by the association between a nonsecreted protein folded in a four ␣-helix bundle characteristic of a classical cytokine (40) and a soluble receptor-like protein. The association between the two counterparts allows the subsequent secretion of the heterocomplex.
CLC was the first protein described in the IL-6 family to be heterodimeric. Indeed, it must associate with either CLF or sCNTFR␣ to be secreted. The fact that CLC can form two different heterodimeric cytokines makes it unique within the IL-6 family (13,14). To understand the mode of interaction of CLC with its two soluble receptors and with its membrane receptor chains, we performed molecular modeling-guided sitedirected mutagenesis on CLC. We subsequently analyzed the biological interactions between the mutants and the different receptor chains.
It is well established that when an IL-6 family cytokine binds to its tripartite receptor, it recruits first its specific ␣-chain component via a site 1 interaction formed by the C-terminal part of AB loop and of ␣D helix (41). This is the case for IL-6/IL-6R␣, IL-11/IL-11R␣, and CNTF/CNTFR␣. Moreover, for the heterodimeric cytokine IL-12, p35 also binds its p40 soluble receptor counterpart via a site 1 interaction stabilized by two disulfide bridges (49). CLF and sCNTFR␣ are two short receptor ␣ chains composed of only one CBD and one Ig-like domain. Each of them may be co-expressed with CLC to form two different heterodimeric cytokines. Based on their rough structure and their function, we predicted that these receptors would interact with CLC through its site 1 motif as is the case for the other IL-6 family cytokines and for the p35 IL-12 component. Surprisingly, mutagenesis on the CLC site 1 impaired the interaction with CNTFR␣ but not with CLF, clearly indicating that these receptors interact through two different regions with CLC (site 1 for CNTFR␣ and another site for CLF). These results were in accordance with a preceding study showing a simultaneous interaction between CLC, CLF, and CNTFR␣ (13). Nevertheless, in this former work we could not exclude the formation of a CLC dimer allowing the interaction with the two receptors via site 1 only. Now, for the first time, we show that a cytokine may associate with a soluble ␣-chain counterpart via different sites. Additional work on other cytokines, in particular on the heterodimeric cytokines IL-23 and IL-27, could be helpful to determine whether this type of interaction is only restricted to CLC/CLF or whether it is representative of as yet to be determined cytokine/receptor interactions.
Mutagenesis on site 1 not only impaired the interaction between CLC and sCNTFR␣ but also that with the membranebound CNTFR␣ subunit involved in target cell recognition. These results suggest that CLC binds the two forms of CNTFR␣ via the same site 1. Nevertheless, we observed that the CLC/sCNTFR␣ interaction is more sensitive than CLC/ mCNTFR␣ interaction to some mutations (i.e. D174A and Q165D/T166H). It is likely that when CLC binds CNTFR␣ at the membrane of the cell, the site 1 interaction is immediately relayed and greatly stabilized by the binding to the signaling subunits gp130 and LIFR␤, via sites 2 and 3, respectively. This explanation is in accordance with the values of binding affinities usually observed between a cytokine and either its ␣-chain receptor or its multimeric membrane receptor (K d ϭ ϳ10 Ϫ8 and 10 Ϫ10 nM, respectively).
In this work, site-directed mutagenesis on site 3 was aimed at studying interactions between CLC and the LIF receptor ␤ chain. Indeed, previous work on the IL-6 family had shown that all of these cytokines display, in addition to site 1, two sites involved in interactions with the large transducing chains of the receptor complexes. Site 2, defined by large surfaces of the ␣A and ␣C helices, interacts with the CBD of gp130. Site 3 is more restricted and displays a hot spot of two amino acid residues crucial for the interaction with LIFR␤. These residues are part of a conserved FXXK motif. The receptor counterpart involved in this interaction is the Ig-like domain of LIFR␤ (54). In CLC, when residues Phe 151 and Lys 154 were mutated to alanine, the interactions between CLC and LIFR␤ were abrogated, because supernatant containing these mutants or purified mutants were unable to induce the biological signal in target cells. Single mutants (F151A or K154A) were much less potent than the double mutant F151A/K154A, meaning that both residues are essential for the interaction with LIFR␤. Surprisingly, these mutations also abrogated interactions between CLC and CLF, because single mutants co-expressed with CLF were poorly secreted, and no secretion was observed with the double mutant. One could argue that these mutations could alter the global structure of the cytokine thus preventing its correct folding and functionality. This hypothesis was excluded, however, because the double mutant was still able to be secreted upon interaction with sCNTFR␣. Therefore, we can conclude that CLC interacts with CLF by a site corresponding to, or overlapping, the site 3 motif. This result was unexpected because CLF is a receptor structurally related to the soluble ␣-chains of the family composed of only one CBD, an Ig-like domain and no transmembrane domain.
In addition to the experiments shown in the present paper, we also carried out point mutations in the Ig-like domain of CLF to inactivate the potential contact site 3 with Phe 151 and Lys 154 of CLC. We tried to express D47A and Y130A CLF double mutants in different mammalian systems but failed to detect any corresponding protein, suggesting that the introduced mutations altered the protein structure. This prevented us from proposing a more complete interaction model for the CLC/CLF complex. LIFR␤ on the other hand is a large transducing subunit with an extracellular domain composed of two CBDs, one Ig-like domain, three FnIII domains, a transmembrane domain, and an intracellular domain. The binding of a cytokine to its receptor via sites 2 and 3 was until now associated with the recruitment of the transducing subunits of the receptor, allowing their dimerization and subsequent intracellular signaling. Such an interaction between a soluble ␣ chain and a cytokine via its site 3 has not previously been reported. Together, the results on site 3 mutants underline that this region of the cytokine may have two functions. First, it may be implicated in the interaction between the cytokine and its co-receptor for the cytokine to be secreted from producer cells, and second, it is essential for the interaction with receptor signaling subunits at the surface of the target cell.
When we compare results from several functional assays with either site 1 or site 3 mutants, we observe a global coher- This phenylalanine residue is conserved in EBI-3 and IL-11R␣. It has been shown that this residue is most likely involved in the interaction of IL-11 with IL-11R␣ (31,57,58). Moreover, these aromatic residues are involved in an aromatic cluster that is probably the key of the interaction between CLC and CNTFR␣. The drastic effect of the R170A/D174A double mutation can be explained by the alteration of the orientation of the side chain of the Trp 67 . These results, as summarized in Fig.  10, highlight the fine degree of regulation between cytokine binding on its receptor and the generation of a specific biological response in target cells.
Among the IL-6/IL-12 family of heterodimeric cytokines, all of the soluble receptor subunits share several cytokine counterparts (CNTFR␣: CNTF, CLC and NP; p40: p35 and p19; and EBI-3: p28 and p35). Going by our current knowledge, it seems that the opposite situation is more restricted because only CLC can bind CNTFR␣ or CLF, and p35 can associate with p40 or EBI-3. This therefore raises the question of whether the receptor component CLF could also bind an as yet unidentified cytokine subunit. In line with this idea, although CLF is es-sential for CLC secretion, it is not essential for its activity on the target cell because both CLC alone and CLC/CLF are able to induce biological activity on cells displaying the appropriate membrane receptor (i.e. CNTFR␣, gp130, and LIFR␤) (13,14). Analysis of CLC and CLF mRNA distribution by in situ hybridization or real time PCR in mouse embryos revealed that both molecules were not strictly co-localized. In particular, CLF was highly expressed in lung and bronchial tubes, whereas CLC was weakly detectable in these tissues. 2 These observations are in accordance with the possible involvement of CLF in other biological functions than the CLC/CLF heterodimer. It has also been reported that the p40 subunit of IL-12 may form the biologically active homodimer (p40) 2 . (p40) 2 was shown to antagonize IL-12, mainly in mouse tissues, by competition with the heterodimer being p35-p40 on the receptor, but it has also agonistic properties described to be involved in immune response to infectious disease (6). We tested the possibility for a molar excess of CLF to neutralize the biological responses observed in the presence of CLC alone. In none of the tested situations could we detect a neutralizing effect of the soluble receptor, suggesting that CLF does not possess antagonistic activities in the studied models. 3 Because CLF is also a soluble molecule that is readily secreted, it must not be excluded that as for p40 for IL-12, it may display biological functions independently of CLC or another cytokine.
In summary, we have demonstrated by a molecular modeling-guided site-directed mutagenesis study that the two heterodimeric cytokines of the IL-6 family, CLC/CLF and CLC/ CNTFR␣ are structurally different. Interactions between CLC and CNTFR␣ occur via the site 1 of CLC, whereas CLC/CLF interactions take place via a site near or overlapping site 3. The identification of CLC contact sites 1 and 3 will be helpful in developing antagonist and superagonist forms of the cytokine for further applications.