PROTEINS: Structure, Function, and Genetics 29:528–544 (1997) Theoretical Investigation of IL-6 Multiprotein Receptor Assembly M.C. Menziani, F. Fanelli, and P.G. De Benedetti* Dipartimento di Chimica, Università di Modena, Modena, Italy ABSTRACT Interleukin-6 (IL-6) is a multifunctional cytokine that regulates cell growth, differentiation, and cellular functions in many cell lineages. Recently, evidences for the formation of an active hexameric complex with an IL-6:IL-6Ra:gp130 stoichiometry of 2:2:2 have been obtained by different experimental approaches. Analysis of the electrostatic potential complementarity between IL-6 and its receptors has been used, in this study, to guide the assembly of homology-based 3D models of the components. The results strongly support a mechanism whereby the active cytokine (IL-6: IL-6Ra) associates with the signal transducing gp130 protein, and the trimeric complex formed further dimerizes to form the hexameric species. Furthermore, computational simulations of the multiprotein complexes provide a rationalization of data from mutation experiments and highlight some key protein–protein interactions which have not yet been the subject of mutagenesis studies. Proteins 29:528–544, 1997. r 1997 Wiley-Liss, Inc. Key words: cytokines; IL-6; IL-6 receptor complexes; electrostatic potential; homology modeling; molecular dynamics INTRODUCTION Interleukin-6 (IL-6) is a multifunctional cytokine that regulates cell growth, differentiation, and cellular functions in many cell lineages.1 The biological effects of IL-6 are initiated by its binding to the extracellular domain of a specific cell surface receptor (IL-6Ra). This low-affinity (1029 M) complex2 constitutes the active cytokine,3 which associates with the extracellular domain of the signal trasducin gp130 protein forming the high-affinity (10210 to 10211 M) ternary complex.4 Signal trusduction requires homodimerization of gp130,5 which takes place via disulfide linkage in the extracellular portion of the receptor, and consists of the activation of tyrosine kinases of the Jak/tyk family.6 The information is next relayed by the phosphorylation of STAT family transcription factors.6 Recently, evidences for the formation of an active hexameric complex with an IL-6:IL-6Ra:gp130 stoichiometry of 2:2:2 have been obtained by different r 1997 WILEY-LISS, INC. approaches7,8 The nucleation of hexameric complexes could be a common biological strategy for cytokines that share structural or functional similarities with IL-6, using the common signaling chain gp130, for example, ciliary neutrophic factor (CNTF), oncostatin M (OSM), leukemia inhibitory factor (LIF), cardiotropin, and IL-11.9 On the basis of structural comparisons, these ligands have been predicted to belong to the longchain helical cytokine superfamily,10 which is characterized by four antiparallel a helices of ,25 residues and the presence of a short helical region in the long AB loop. The prediction has been confirmed, either by x-ray crystallography or NMR, for several members of this family, including granulocyte colonystimulating factor (G-CSF), growth factor (GH), and very recently, LIF, CNTF. The similarities observed in the structure of the cytokine ligands is mirrored in the extracellular portion of their cell-surface receptors.11 Both IL-6Ra and gp130 belong to the hemopoietin receptor superfamily (class I). The most striking feature of these receptors is the conservation of two similar cytokine binding domains containing a pattern of conserved disulfide bonds and a WSXWs motif located just outside the TM domain.12 The crystal structure of some members of this family, such as growth factor (GH), prolactin (PRL), and erythropoietin (EPO) receptor complexes,13–15 shows that both receptor binding domains are related in the overall topology to fibronectin type III domains, folding in b barrels of seven antiparallel b strands. The extracellular region of IL-6Ra consists of one Ig-like domain and two cytokine-receptor domains, while the extracellular part of gp130 is composed of six fibronectin type III modules, but only the second and the third modules form a cytokine receptor domain.16 Recently, molecular modeling-guided mutagenesis studies, based both on the 3D structure of the G-CSF cytokine and on the GH complex paradigm, allowed the identification of several regions of protein– Contract grant support: CNR; Contract grant support: Ministero dell’Università e della Ricerca Scientifica. *Correspondence to: Dr. P. G. De Benedetti, Dipartimento di Chimica, Università di Modena, V. Campi 183, 41100 Modena, Italy. E-mail: firstname.lastname@example.org Received 17 January 1997; Accepted 15 May 1997 THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES protein interactions. According to these findings, three sites are involved in the interaction between IL-6 and its receptors: 1. Site 1 is formed by the C-terminal portion of helix D and by part of the AB loop and interacts with the IL-6Ra receptor. 17–20 2. Site 2 is formed by a limited number of exposed residues on helices A and C and interacts with a gp130 chain.21,22 The location of these sites on the IL-6 surface corresponds to the one of the functionally equivalent sites 1 and 2 on GH.13 3. Site 3 is a composite binding site for a second gp130 chain and encompasses residues at the beginning of helix D and the initial part of the AB loop.23–28 Furthermore, the IL-6Ra has been proposed to interact with the signaling gp130 chain by using residues in the E2 strand and AB2 loop.29 Additional protein–protein contacts have been identified by random30 and rational31 mutagenesis of the hIL-6Ra receptor, while no experimental information is yet available on the sites of gp130 involved in the multiprotein receptors assembly. From these studies, details on the topological location of each component in the putative multimeric complex emerged, and the hypothesis that two ‘‘GH-like’’ IL-6:IL-6Ra:gp130 trimers flank each other with inverted orientations, was reported.8 Starting from this hypothesis we constructed a three-dimensional model of the IL-6 hexameric complex. The assembly of the homology based 3D models of the different components was guided by a detailed analysis of the electrostatic potential complementarity. Moreover, molecular simulations of the multiprotein complexes were carried out in order to highlight molecular determinants for ligand binding and receptor association on the basis of the topological, conformational, and dynamics features of the critical contact residues. METHODS Homology Modeling Sequences of the hormones and receptors were extracted from the EMBL protein sequence database. The crystallographic structure of the human (h), bovine (b), and canine (c) G-CSF (PDB entries 1rhg, 1bgc, and 1bgd, respectively), of the mouse (m) LIF (PDB entry 1lki) and of the hGH receptor complex (PDB entry ghhr) were obtained from the Brookhaven Data Bank.32 Secondary structure predictions were achieved with the Rost and Sander method33 at the Internet address Predict Protein@Embl-Heidelberg.de. Homology modeling was performed by using the program MODELLER.34 The sequence alignment facility implemented in the program THREADER35 was used to fit the hIL-6 sequence into the 3D 529 structure of b, and cG-CSF. The alignments obtained, together with the structural alignment prepared by multiple least-squares superposition implemented in the MALIGN3D command of the MODELLER program34 were used to calculate 3D models for the hIL-6 sequence containing all mainchain and side-chain heavy atoms. Manual modifications of previously reported alignments29 were used for the IL-6Ra and gp130 receptors. Briefly, the program MODELLER derives many distance and dihedral angle restraints on the target sequence from its alignment with the template 3D structures. Then, the spatial restraints and the energy terms enforcing proper stereochemistry are combined into an objective function, and the final models are obtained by optimizing the objective function in cartesian space employing methods of conjugate gradients and molecular dynamics with simulated annealing. The generation of several models for the different components of the IL-6 hexameric complex were obtained by randomizing the cartesian coordinates, allowing a deviation of 64 Å. Refinement and Analysis of the 3D Structures Energy minimization and molecular dynamics studies were carried out using the software package QUANTA version 4.1 (Molecular Simulations, Inc, 200 Fifth Avenue, Waltham, MA 02154) implemented on a Hewlett-Packard 720 workstation. Molecular mechanics and molecular dynamics calculations were performed using the program CHARMM (version 24).36 The minimization procedure consisted of 50 steps of steepest descent, followed by a conjugate gradient minimization until the root-mean-square (rms) gradient of the potential energy was ,0.001 kcal mol21 Å21. The united atom force-field parameters, a 12-Å nonbonded cutoff and a dielectric constant of 80 (see Results and Discussion section) were used. The minimized coordinates of the hormones, receptors, and binary and ternary complexes were used as starting point for dynamics. During dynamics the lengths of the bonds involving hydrogen atoms were constrained according to the SHAKE algorithm,37 allowing an integration time step of 0.001 ps. The a helices conformation was preserved by means of the nuclear Overhauser effect (NOE) utility provided by the CHARMM program, which allows the imposition of constraints between the backbone oxygen atoms of residue i and the backbone nitrogen atoms of residue i 1 4, whereas the b-sheet structure was maintained by applying NOE constraints between opposite strands. The structures were thermalized to 300 K with 5°C rise per 6000 steps by randomly assigning individual velocities from the gaussian distribution. After heating, the systems were allowed to equilibrate until the potential energy versus time was approximately stable (34 ps). Velocities were scaled 530 M.C. MENZIANI ET AL. by a single factor. An additional 10 ps period of equilibration with no external perturbation was run. Time-averaged structures were then determined over the last 100 ps of each simulation. Data were collected every 0.5 ps. Interaction energies at the monomer–monomer interfaces and at the receptor binding site were computed by using a dielectric constant e 5 4r. Solvent accessible surface area and buried surfaces were determined with a probe radius of 1.4 Å. The fraction of the side chain accessible to the solvent is defined with respect to the maximum accessibility of each amino acid side chain calculated for a fragment GLY-X-GLY. The Protein Health utility and the 3D Profile Analysis program38 implemented in QUANTA, WHAT IF39 and PROCHECK40 programs were used to verify the models obtained. Electrostatic Calculations The electrostatic potentials were calculated by numerically solving the finite difference linearized Poisson-Boltzmann equation using an incomplete Cholesky preconditioned conjugate gradient method as implemented in the University of Houston Brownian Dynamics (UHBD) program, version 4.0.41 The charges of the ionized groups were assigned assuming standard amino acid protonation state at pH 7. His166 of hIL-6Ra was considered in a protonate state. The OPLS parameter set42 with the radii of hydrogen atoms set to 1.2 Å, was used. The ionic strength of 145mM follows a Boltzmann distribution at 300K, and the dielectric constant for the protein interior and surrounding solvent were 2 and 78, respectively. A rolling probe with radius of 1.4 Å was used to determined the solvent-solute dielectric boundary and the dielectric boundary smoothing43 was implemented. Electrostatic boundary conditions were set using the single Debye-Hückel sphere approximation for the hormones receptors.41 A 1003 grid with a 1 Å spacing was used for the calculation of the molecular electrostatic potentials (MEP) of the hormone and receptors. Details of the MEP at critical sites were calculated by the focusing method. According to this method, the potentials from a coarse (1003 grid, 1 Å spacing, in this work) are used to set the boundary potentials for a second more finely spaced grid centered on the region of interest. Grid spacing was set to 0.25 Å and the grids were centered at the atomic coordinates of the Ca atom of residue hIL-6 158 and hIL-6Ra 135 for site 1, hIL-6 100 and hgp130 63 for site 2, hIL-6(A) 33 and hgp130(B) 47 for site 3, hIL-6(A) 24 and hIL-6(B) 91 for site 4, hIL-6Ra 119 and hgp130 114 for site A and, finally, hIL-6Ra(A) 114 and hgp130(B) 167 for site B. RESULTS AND DISCUSSION Three months after the submission of our manuscript to this Journal the 1.9 Å crystal structure of hIL-6 was published.44 Although the coordinates are not yet publicly available, the structure description given in the paper gave us the opportunity of comparing the results of the homology modeling procedure with the experimental structure. Therefore, in the revision of the manuscript we added to the original sections on modeling, refining and assessment of the hIL-6 homology model a new section dedicates to the comparison with the newly determined x-ray structure. Modeling of hIL-6: Template Selection Among the long-chain cytokine subgroup the 3D structure of hGH, bG-CSF, cG-CSF and hG-CSF, mLIF and hCNTF have been solved by crystallographic or NMR techniques (the coordinates of the last one are not publicly available). The structural homology observed between these cytokines is not echoed in any strict conservation of key residues or disulfide bonds. However, previously reported21 multiple sequence alignment between IL-6 and G-CSF from different species shows that the cysteine residues implicates in the disulfide bridges in G-CSF are conserved in IL-6 as well as the overall pattern of hydrophobic and hydrophilic amino acids within the a-helical regions. Moreover, a very similar alignment is obtained by fitting the sequence of IL-6 onto the 3D structure of G-CSF by means of the program THREADER.35 The program, principally designed to solve the inverse folding problem, allows the alignment of a sequence onto a single structure for modeling purposes by using a set of statistically determined pairwaise potentials. The alignment of human IL-6 onto bovine and canine G-CSF significantly differs in the zone of the AB loop, as shown in Figure 1. The manual alignment of the human G-CSF 3D structure onto bovine G-CSF and of the mouse IL-6 sequence onto the human IL-6 sequence is also reported in Figure 1. The structural restraints derived by the G-CSF proteins were exploited for the generation of different 3D models by the program MODELLER.34 Thus, the sets of models A and B were constructed by taking the structure of bG-CSF and cG-CSF, respectively, as a template and the alignment of the sequences reported in Figure 1 as input. The sets of models C-E were based on a multiple structural alignment of bG-CSF, cG-CSF and mLIF which was prepared by multiple least-squares superposition, as implemented in the MALIGN3D command.34 The spatial features are transferred from the bG-CSF and mLIF to the hIL-6 target, in model C and from the cG-CSF and mLIF, in model D. The number of restraints on the structure represented by model E is derived by using bG-CSF cG-CSF and mLIF contemporaneously. The structural alignment obtained for model C is reported in Figure 2. The crystal structure of hG-CSF was not considered in the modeling because large parts of the AB THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES 531 Fig. 1. Threader-based fitting of the sequence of human (h) IL-6 onto the 3D structure of canine (c) and bovine (b) G-CSF. The human G-CSF 3D structure is manually aligned onto bovine G-CSF, and of the mouse IL-6 sequence is aligned onto the human IL-6 sequence. Coordinate sets of G-CSFs from the PDB were used, and the entry name are 1rhg, 1bgc, and 1bgd for human, bovine, and canine, respectively. Helical secondary structure elements as determined by x-ray are shown as boxes enclosing the appropiate residues. Helical secondary structure assignments for IL-6 obtained according to the Rost and Sander methods33 are shown as shadowed residues. Stretches of residues not resolved by x-ray are in lowercase. Fig. 2. Structural alignment of hIL-6 onto bG-CSF and mLIF, derived by MODELLER multiple least-squares superposition.34 X represents stretches of residues not resolved by x-ray. Helical secondary structure elements as determined by x-ray are shown as boxes enclosing the appropiate residues. Helical secondary structure assignments for IL-6 obtained according to the Rost and Sander methods33 are shown as shadowed residues. and CD loops are yet unresolved. mLIF was added to the single target model in the attempt of improving the modeling of the CD loop. In fact, the G-CSF CD loop is four residues shorter and, being a flexible region, a stretch of seven amino acids is unresolved in X-ray structures from different species. Therefore, a long insertion was necessary in hIL-6 model A, with negative consequences on the accuracy of the model in this region. On the contrary, the CD loop in mLIF has a well-defined single main chain conformation throughout its length.45 Fifteen conformations of hIL-6 were calculated for each A-E set of models by randomizing the Cartesian coordinates. The representativity of each randomized conformation is usually judged within the MODELLER framework by the value of its objective function. Because of the low percentage of sequence similarity between the template structures and the hIL-6 sequence, and because of the presence of unresolved stretch of structure in the templates this criterium was not strictly followed for the choice of the best randomized conformation within each A-E 532 M.C. MENZIANI ET AL. TABLE I. Quality Scores for the hIL-6, hIL-6Ra and hgp130 Models and for the Structures Used as Templates LIGANDS 3D-Profile Models hIL-6 model A hIL-6 model B hIL-6 model C hIL-6 model D hIL-6 model E Templates bG-CSF cG-CSF mLIF Scores WHAT IF RECEPTORS Scores 3D-Profile WHAT IF PROCHECK 34.52 32.40 34.40 19.54 27.11 20.57 20.62 20.73 20.66 20.73 20.05 0.05 20.01 20.02 20.08 41.36 40.63 38.19 20.36 20.56 20.50 0.10 20.08 0.10 PROCHECK Models hIL-6Ra hgp130 26.92 22.38 21.41 21.20 20.21 20.17 Templates hGHR 25.97 21.05 20.40 Notes. The minimum 3D-Profile score for a correctly aligned sequence within a given within a given structure is seven. According to WHAT IF models are classified as follows: perfect . 20.5, very good or good 20.5 4 21.0, normal 21.0 4 22.0, very poor 22.0 4 23.0, bad ,23.0. PROCHECK score for a structure solved at 1.5Å resolution is within 20.5 and 0.3. set of models. Independent checks were preferred such as: a) evaluation of the overall fold and sidechain packing of the models provided by the 3D profile program38 in Quanta, b) the quality factor furnished by the WHAT IF program,39 which assesses how normal or abnormal each side chains environment is with respect to the average packing environment for all the residues of the same type in highly resolved PDB structures, and c) several assessments of the structure’s overall quality offered by PROCHECK40 and synthesized in the quality geometry factor indicator, which measures how atypical each residue is as far as its covalent geometry and dihedral angles are concerned, with respect to the ‘‘typical’’ distribution defined by the PDB structures. The results for the best randomized conformation of each set of models are summarized in Table I. A comparison with the scores obtained for the template structures, which are also listed in Table I, shows that pretty good models are, in general, obtained in this modeling exercises, although the percentage of similarity between the sequences is very low. According to this analysis, and to the inspection of the 3D-Profile plots, model A is the one with the lowest chances of occurrence of misfolded regions, furthermore, it achieves the best agreement with the experimental data, as shown in the following section. Refinement and Evaluation of the hIL-6 Model All hydrogen atoms were added to the heavy atom of model A and optimization of the side chain packing was carried out by CHARMM minimization keeping the backbone atoms fixed in their original position. Then the entire molecule was allowed to relax and the resulting structure was subjected to a 150 ps dynamic run. The average structure obtained was further minimized. The refinement procedure followed was previously optimized on the crystal structure of the GM-CSF.46 A substantial improvement in the side-chain packing and hydrogen bonding network is obtained after refinement. Moreover, in the final model 96% of the residues resides in the most favoured area of the Ramachandran plots. Only Arg86 is in a generously allowed region. This residue contributes to the packing of the BC loop against helix A and helix D by forming a charge reinforced hydrogen bond with Ser29 and a salt bridge with Asp142. Analysis of the Model Compatibility of the hIL-6 model with the experimental structural information available prior to structure determination Table II summarizes the most significant results obtained by site-directed mutagenesis studies on hIL-6. The structural location of the mutated residues, which is also reported in the Table, allows to distinguish between the probable candidates for intermolecular interactions and for ligand architecture. Mutated residues, listed in Table II are reported in bold in the text. Inspection of the fractional solvent accessibility (SAS) and of the average atomic fluctuations (rms) of the residues during dynamics gives an estimation of site exposure, and consequently, of reactivity propensity. Human IL-6 presents two potential glycosilation sites,49 Asn27 and Asn126. They are exposed residues located, in the 3D model, at the end of helix A and in the middle of the CD loop, respectively; the fraction of side chain exposed to the solvent is greater than 50%, and the average rms fluctuations of the side chains are of the order of 2 Å. Furthermore, in chemical modification studies51 the reactivity of Metionines for the oxidation reaction was found to decrease in the order Met31 Met99 Met166 Met143 Met49. These residues are located in the AB loop (Met31 and Met49) in helix C (Met99) and in helix D (Met166 and Met143) and, a part for THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES 533 TABLE II. Binding Properties of hIL-6 Mutants and Structural Location of Mutated Residues. Residues Listed in This Table are in Bold in the Text Residues Arg12A Glu33A Lys36A Leu133V Leu140V Met143A Leu147V Leu149V Arg150M Phe152L Phe155A Leu156V Gln157D/N/A Ser158R Ser159R/K/P/I Leu160R/Q/Y Arg161K/N/G/A/P/S/W Ala162T/R/P/F/L Leu163P/V Arg164G/L/W/A Gln165H/E Gln157W,Gln165H Gln157I,Ser158R,Gln165A (IRA) Met166K Residues Tyr13D,Gly17F,Ser100R,Val103D (DFDR) Gln141E, Thr144P Leu39A Leu39D,Glu41F,Asn42W,IRA DFRD,Leu39D,Glu41F,Asn42W, Gln57Y,Ser58K,IRA Trp139R,Asp142R Thr144D IL-6Ra binding Ref. < < < 5 5 < 5 < < > < < < > < < < < < < < < > < 24 24 24 19 19, 47 18, 47 19, 47 19 18 48 49 19 17 17 48, 17 48, 17 48, 18, 17, 49 48, 17 17, 19 48, 49 20 20 20 18 sgp130 sgp130 binding dimerization activity activity Met31, the amount of SAS decreases in a parallel degree with that of the oxidation reactivity. The reactivity features of Met31 can be explained by the high mobility (average rms fluctuations: 2.26 Å) of its side chain during dynamics which compensates for the small amount of SAS found in the minimized average structure of hIL-6. Moreover, 1H-NMR studies provided information on the mutual spatial positions of Trp139, Met143, His146 and Met166.47 Trp139 is positioned, in our model, at the beginning of helix D, it packs against the sulfurs of Met31 and Met143 and the hydrophobic microenvironment is completed by Leu44 and His146. Recently, a comparative NMR study on folding topologies of hIL-6 and its mutants allowed the identification of residues involved in hydrophobic packing of the four helix bundle on the basis of the exposed exposed exposed vs. T144 and I148 (HD) vs. L44 (AB loop) exposed vs. AB loop vs. AB loop core exposed core inter-helix core exposed exposed core inter-helix exposed inter-helix inter-helix exposed exposed C-terminal Ref. Structural position < < < 21, 22 50, 26 24 27 HA, HC HD vs. CD and BC loop AB loop vs. L147 (HD) AB loop < < < 28 8 8 AB loop HD HD vs. L133 (CD loop) < < Structural position HA AB loop AB loop CDloop HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD HD structural changes observed by conservative amino acid replacement of Leu residues with Val residues.19 Consistently, our 3D model predicts that Ile11 and Ile18 in helix A, Leu73, Phe76, Tyr79, and Tyr82 in helix B, Val97 in helix C, and Ile148, Leu149, Phe152, Phe155, and Leu156 in helix D contribute to the hydrophobic core of the cytokine, most of them being completely buried and displaying no or very low mobility during dynamics. Moreover, no significant structural changes in the NMR data are observed for conservative amino acid replacement of Leu140, Leu147 and Leu163 with a Val residue. These residues are buried in our model, but show a significant mobility during the time evolution of the trajectories (average rms fluctuations of 1.39 Å, 1.29 Å and 1.12 Å, respectively). Finally, an important role in packing the AB loop against the four helix bundle is assigned to the 534 M.C. MENZIANI ET AL. Fig. 3. hIL-6 x-ray structure44 versus hIL-6 homology model. Secondary structure elements and other relevant features revealed by the crystal structure (first row) and by the homology model (second row) of hIL-6 are highlighted on the sequence as boxes enclosing the appropriate residues. hydrophobic residues Leu44, Leu46, Pro47 and Met49 on the basis of the low tolerance to mutation of these residues highlighted in a recent study where phagedisplayed libraries of mutants were generated by fully randomizing four different amino acids in the predicted AB loop.27 The role of Leu44 has been described above, Pro47 faces Leu39 and Leu147 in helix D, while Leu46 and Met49 are part of an extended hydrophobic cluster together with Ile148 in helix D and Leu 80 and Thr79 in helix B. Compatibility of the model with the crystal structure A quantitative comparison of the hIL-6 model obtained (hIL-6 model A) with the crystal structure newly determined can not be presented here since the IL-6 coordinates are not yet available. However, a qualitative evaluation of consistencies and inconsistencies can be attempted. In Figure 3, the secondary structure elements and other relevant features revealed by the crystal structure of hIL-6 are highlighted on its sequence (first row) as boxes enclosing the appropriate residues. In the second row the secondary structure features of the final IL-6 3D selected model (hIL-6 model A) are reported. Again, boxes enclosing appropriate residues highlight important secondary structure elements. A good agreement in the position and length of the four principal helices is observed. The stretch of residues at the beginning of loop B, predicted to form a minihelix on the basis of the structure of G-CSF from different species and of secondary structure prediction methods,33 shows no interpretable electron density. A similar situation is observed in the crystal structure of CNTF.52 However, in this case the Authors forward the hypothesis that the stretch of unresolved residues might fold in helical conformation on the basis of the comparison with the 3D structure of LIF45 and of the amino acids conservation in their sequences. The structural elements of the rest of the AB loop appears to be very similar in the model and crystal structure, although the first b turn is misplaced by one residue. The difficulty of predicting insertions or region with little sequence similarity with know structure is evident in loop CD. In fact, the crystal structure presents in the C-terminal portion of this loop a short helix which, being an unique feature of IL-6, is missing in the homology model. Since the target structure is characterized by a long stretch of unresolved residues in this region (see Figure 1) the loop has been constructed from scratch in hIL-6 model A, whereas the structural restraints of mLIF were exploited in hIL-6 model C (obtained by multiple least-square superposition of bG-CSF and mLIF, see previous section). Retrospectively, the second choice was not correct because the CD loop of mLIF is tightly packed against the four-helix bundle and presents an extended conformation.45 On the contrary, a better agreement with the experimental structure would have been obtained by considering the secondary structure prediction results,33 shown in Figure 1. The structural comparison reported by Somers et al.44 between the newly determined hIL-6 structure and G-CSF revealed a good agreement in the packing arrangements and interhelix angles. However, IL-6 presents a kink in the middle of helix B due to a break in the hydrogen bonding which is not conserved in G-CSF and, as a consequence, in our homology model. Thus, notwithstanding differences between the model and the experimental structure might be THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES observed in several regions once the coordinates will be available, a careful inspection of the side chain contacts as described in the Somers’s paper shows a gratifying agreement. In particular consistency with experiment is achieved in the following features: a) the hydrophobic core is formed by residues I7, I11, I14, I18, L21, T25, I69, L73, L80, L83, F87, A94, V97, T101, L104, L108, T145, L149, F152, L156, S159, and L163 (see Figure 3, for numbering system comparison). b) Hydrogen bonds are observed between the following amino acid side chains: K111 and S4, S90 and E24, E77 and K102, R86 and D142. c) The side chains of T101, T145 and S159 are at hydrogen bonding distance with the main-chain oxygen atoms of V97, Q141 and F155, respectively. d) Hydrophobic interactions are observed between L66 and M166 and the side chains of L44, L46, P47 and M49 point into a cleft between helix B and D (see previous section for an extensive description). As expected, inconsistency with the experiment is mainly observed for the cluster of hydrophobic side chains stabilizing the position of the E minihelix, which corresponds to a misfolded region in the IL-6 homology model. In summary, the following conclusions can be drawn on the basis of a qualitative comparison between the homology model and the newly determined 3D structure: a) the sequence alignment obtained by means of the program Threader gives, in this case, a satisfactory result. Although a recent analysis of the Thread algorithm performance suggests that sequence alignments implicit in the ‘‘correct’’ threading of a sequence through a structure are frequently incorrect,53 confidence on the alignment result can be gained by the agreement with previously published21,45,51 multiple alignment which included sequences of other members of this family of cytokines and by the capability of the model derived to fit the available experimental data. b) According to the site-directed mutagenesis information available at present the misfolded region in loop CD is not involved in binding or dimerization regions, on the contrary, region of the AB loop putatively involved in the interaction with IL-6Ra and in complex dimerization seems to be satisfactorily predicted. Modeling of the hIL-6Ra and hgp130 Receptors The hGH receptor was used as reference structure for modeling the portions of the hIL-6Ra and hgp130 receptors putatively involved in the IL-6 binding and signal transduction. The presence of four cysteine residues in the N-terminal domains and a structural PXPP motif in the interdomain linker, conserved in all the three receptor chains facilitated the alignment of the first cytokine binding domain. The alignment of the second domain can be based only on 535 motives such as VRXR and WSXWS12 not strictly conserved among different receptors, and on secondary structure predictions. Several 3D models of the hIL-6Ra and hg130 receptors were obtained by randomizing the initial cartesian coordinates. Basically, the alignment recently presented by Salvati et al.29 was used as input. Manual modifications of this alignment were necessary in order to improve the quality of the models in the EF loop regions of the receptors. This was done by a trial-and-error procedure till no stretches of negative scores were found in the Profile 3D plots. The sequence alignment which furnishes the best models is reported in Figure 4. The overall quality of the models obtained is comparable to the resolution of the template hGH receptor as shown by the 3D profiles, WHAT IF and PROCHECK scores, reported in Table I. The initial model of the IL-6Ra and gp130 receptors were subjected to the same refinement procedure used for the ligands. All backbone angles of the final models obtained after energy refinement have trans-planar v angles and fall into allowed region of f_ c space. The structure consistency with the available experimental data was checked. The side chains of Asn107 and Asn131 in IL-6Ra54 and Asn4, Asn30 and Asn100 in gp1304 are putative N-linked glycosylation sites and are exposed to the solvent in the models. Several hIL-6Ra residues involved in ligand binding and signal transduction have been identified by means of random30 and rational29,31 mutagenesis studies and are listed in Table III. Mutated residues listed in Table III are in bold in the text. A rationalization of the role of some amino acids might be attempted on the basis of the 3D model presented. In fact, in virtue of the 3D model obtained, a structural role might be hypothesized for several mutants which show impaired ligand binding. Thus, for example, mutation of Lys10 to Gln or Asp103 to Val prevents the establishment of an electrostatic interaction between the two residues which may participate in the stabilization of the relative orientation between the two cytokine binding domains of IL6Ra. A similar interaction is also present in the x-ray structure of the GHR, the PRLR and EPOR complexes.13-15 Substitution of Val76 to glycine was reported to result in decreased IL-6 binding and in loss of biological activity. In our model Val76, located in the F strand, is part of an hydrophobic zone composed of Phe39, and three residues of the EF loop Phe73, Tyr74 and Ile75. The interaction of Ser72 with the backbone of Ser77 helps to maintain the EF loop local architecture, in agreement with the results obtained from point mutations of the Ser71, Ser72, Phe73, Tyr74 motif, very recently reported.31 A large hydrophobic patch forms the core of the second domain of IL-6Ra. To this cluster contribute: 536 M.C. MENZIANI ET AL. Fig. 4. Sequence alignment of the cytokine binding domains of hIL-6Ra and hgp130 onto hGHR. b-Strand regions, as determined by x-ray are boxed in hGHR, and secondary structure assignments obtained according to the Rost and Sander method33 are shown as shadowed residues. Trp130, Pro127, Pro102, Leu137, Ala180, Pro105, Phe139, Trp124, Tyr143, Leu178, Ile108, Leu141, Val122, Val176, Ile165, Val110, Leu120, Pro200 and Trp119. Ten of these residues have been mutated and, according to the 3D model, structural disturbance might be responsible for the decreased ligand binding capability showed by the mutants. A low tolerance to mutation of the WSXWS motif is common to several members of the cytokine receptor family. Very recently the crystal structure of the EPO 15 receptor complex provided detailed insight into the role of these residues in structure stabilization. The b bulge of the WSXWS sequence interacts with the F2 b-sheet of the VRXR consensus sequence, and a stack of the aromatic side chains interleaved by positive charged residues is formed. These p cation system interactions are observed also in the GHR and PRLR complexes.13,14 In our 3D model WSXWS takes part in an extended p cation system interactions consisting of three aromatic groups (Phe151, Trp189 and Trp192) that stack between five charged residues (Glu140, Arg142, Arg144, Arg179 and Glu191). Two glutamine residues, Gln181 and Gln177, hydrogen bond and orient the arginines, while Ser190 and Ser193 form hydrogen bonds with the main chain of residues Pro104 and Pro105. Most of these residues have been mutated and the mutants showed no IL-6 binding capability, furthermore, some of them presented a very low expression level. Structure–function relationships of the gp130 are, as yet, poorly characterized. The hydrophobic residues participating in the core of the second domain of IL-6Ra and those responsible for the p cation system interactions are largely conserved in the hgp130 receptor. Thus, a very similar packing is observed in this portion of the two receptor models. On the contrary, significant differences are observed in the domain-domain interface. In particular, the absence of the electrostatic interaction between residue 10 in module I and residue 103 in module II, which restrict the domain association in IL-6Ra, is only partially compensated by a number of hydrogen bonding and hydrophobic intermodule interactions involving Asn10, Lys13, Tyr63, Val90, Tyr91, Val93, Ser124, Val125, Lys176 and Lys180. Moreover, Lys92 is completely buried at the interface and makes hydrogen bonds to main-chain oxygen atoms of residue Cys7 and Asp88 and a p cation interaction with Phe87. A similar situation has been recently observed in the crystal structure of the human Tissue Factor.55 IL-6 Ternary Complex The assembly of the trimeric complex 3D model is based on the assumption of a ligand binding mode similar to the one revealed by the crystal structure of the hGH complexed with its receptor. Thus, a least squares fit procedure was used to obtain a superposition of the conserved b strands Ca backbone of the IL-6Ra and gp130 receptors onto the b-stands of the hGH receptor. The Ca-carbon atoms of the IL-6 ahelices were superpositioned onto the aligned residues of the hGH. In this way, a rough overlap of the main axis of the two hormones can be achieved, but adjustments in the orientation of the IL-6 ligand 537 THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES TABLE III. Binding Properties of hIL-6Ra Mutants and Structural Position of Mutated Residues. Residues Listed in This Table are Reported in Bold in the Text Residue Arg9Q,Lys10S Trp20L Pro67L,Glu68Q Ser71Y Ser72I/L/PR/V Phe73H/I/Y Tyr74/A/G/R/V Val76G Asp103V Pro104A,Pro105H Asn116D Val122S, Trp124L Ser133G/I/R/V Phe134S/V Tyr135A/I/R/V Ser133A,Phe134V Arg136S,Leu137I Glu140V,Leu141H Arg142S,Tyr143D Val164G,Ile165M Ile165D His166S,Asp167G/V His166S Asp167V,Ala168D Asp167G His166S,Asp167V,Ala114D,Asn116D Gly171D His174Y,Val175L Arg173A,His174I Val176M,Gln177H Arg179G Arg179G,A180T Glu183A,Phe184I Gly187V,Glu188D Trp189H,Ser190A Ser190A,Ser193A Glu191D, Trp192R Thr199K,P200L with respect to the two receptors were necessary to remove the major steric overlap. Furthermore, the mutual orientations of the three components was adjusted in order to maximize the complementarity of the electrostatic potentials computed on IL-6 and IL-6Ra for the binary complex, and on IL-6:IL-6Ra and gp130 for the ternary complex (Fig. 5). Recently, electrostatic analysis was successfully used to distinguish between homo and hetero-oligomeric receptor binding properties of IL-4 and G-CSF.56 Assuming standard amino acid protonation state at pH 7, the hIL-6 is globally neutral. The net charges of hIL-6Ra and hgp130 are 11.0e and -1.0e, respectively, by considering only His166 of hIL-6Ra in a protonate state and all the other histidine IL-6 binding < < < 5 5 5 5 < < < 5 < 5 < < < < < < < < 5 5 5 < 5 < 5 < < < < < < < < < < IL-6 signaling 5 5< 5 < < < 5 < < < < < < < < < Ref. Structural position 15a 15a 15a 17a 17a 17a 17a 15a 15a 15a 16a 15a 17a 17a 17a 15a 15a 15a 15a 15a 15a 15a, 16a 15a 15a 15a 16a 15a 15a 15a 15a 15a 15a 15a 15a 15a 15a 15a 15a AB loop B sheet EF loop EF loop EF loop EF loop EF loop EF loop G-A2 G-A2 A2-B2 loop B2 sheet B2-C2 loop B2-C2 loop B2-C2 loop B2-C2 loop B2-C2 loop C2 sheet C2 sheet E2 E2 E2 E2 E2 E2 E2 E2F2 loop E2F2 loop E2F2 loop F2 F2 F2 F2-G2 loop F2-G2 loop F2-G2 loop F2-G2 loop F2-G2 loop G2 residues as neutral. The electrostatic potential surrounding the hIL-6 ligand is asymmetric and, a part from a region of negative potential confined to the end of helix A, part of the AB loop and the N and C-terminal of helix B, the rest of the surface is covered by positive potential. The IL-6Ra receptor presents a large region of negative potential surrounding the AB, CC’, C’E loops, the N-terminal part of the EF loop and the hinge region connecting the two cytokine binding domains. The second domain, on the contrary, is mainly surrounded by a positive potential. Intriguing is the electrostatic potential distribution of gp130 which reflects the extreme flexibility of its interaction surface. It presents two different faces with respect to the main plane on 538 M.C. MENZIANI ET AL. Fig. 5. Electrostatic isopotential surfaces for the IL-6 receptor complex and its components. The blue and red contours represent the isopotential surface at 10.5 kcal (e·mol) and 20.5 kcal (e·mol), respectively. Important sites for intermolecular interactions are highlight by arrows on ribbon diagrams of the molecules. which the two domains lie. One face (shown in Fig. 5) is characterized by spots of positive and negative potentials, while on the other side a large spine of positive potential is contoured by zones of negative potential, stretching over the F strand and the CC’ loop. In order to characterize, in a more detailed way, the MEP distribution on the receptor and on ligand putative binding sites, we used the focusing method41 to describe the electrostatics in the important regions. Table IV reports the values of the potentials (EP), the solvent accessible surface area (SAS), the resulting density of the potential (r) and the list of the residues contributing to the generation of the surfaces putatively involved in the ligand–receptor and receptor–receptor recognition. Site 1. Interaction between hIL-6 and hIL-6Ra The site 1 interface of IL-6 and IL-6Ra models shows the most significant complementarity of the electrostatic potentials. This argues for the funda- mental importance of the electrostatic interaction for IL-6 receptor recognition and supports the hypothesis of the sequential assembly of the ternary complex model proposed by Taga et al..3 A description of the interactions established by the charged residues responsible for the complemetarity of the electrostatic potential during the dynamics simulation of the binary and ternary complex is given in Table V. Several intermolecular interactions involve residues (in bold) identified to be important for binding and dimerization activity (Tables II and III, and following text). A good agreement between the experimental evidences and the role assigned to critical contact residues in our model is observed. Moreover, a recent study27 aimed at improving the interaction of the IL-6 AB loop residues with hIL-6Ra proposed, on the basis of the consensus sequence derived from fully randomized phage libraries, Asp53 and Phe56 to be direct contact points to the receptor. In fact, these residues are not tolerant to substitution. On the contrary, a 20-fold improvement in binding of the IL-6Ra is observed for the Lys48V, Ala50R, Glu51M, Lys52E quadruple mutant. Accordingly, in our model, THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES 539 TABLE IV. Electrostatic Potential (EP), Solvent Accessible Surface (SAS) and Density of the Potential (r 5 EP/SAS) at the Surface Generated by the Residues Putatively Involved in Ligand-Receptor and ReceptorReceptor Recognition EP kcal/(e · mol) Site 1 hIL-6 hIL-6Ra Site 2 hIL-6 hgp130 Site A hIL-6Ra hgp130 SAS (Å)2 r kcal/(e · mol)Å2 Residues 3480.88 1350 2.58 23497.29 1607 22.18 12, 15, 36, 48, 52–53, 56–57, 60, 150–151, 154–155, 157–158, 161–162, 164–165 38–39, 67–68, 73–76, 95, 97, 131–136, 158–159, 183–186 21450.58 2565.34 1030 1034 21.41 2.48 7, 10, 13–14, 16–17, 20–21, 91–93, 95–96, 98–100, 102–103, 105–107 11–14, 35–38, 46, 58–60, 62–65, 124 2060.11 2365.62 871 793 2.36 20.46 113–114, 116–119, 154, 163–164, 166–167, 169 104, 106–107, 112, 114, 118, 153, 154, 155–158, 160 Asp53 makes intermolecular charge reinforced hydrogen bonding interactions with the Arg9 and Ser14 receptor residues, while Phe56 packs against the side chains of the Pro12 and Leu13 receptor residues. In addition, the interactions established by Lys48 and Glu51 can easily be compensated and improved by the charged mutants in position 50 and 52. It is worth noting that, although the hydrophilic groups are required to give specificity, the interaction is dominated by the hydrophobic forces, in fact, more than 60% of the total interaction energy between the hormone and the IL-6Ra receptor (-364.20 kcal/mol), computed on the minimized average structure of the binary and ternary complexes, is due to the dispersion forces as represented by the vdW component of the total interaction energy. A salt bridge is observed in the final model between the Arg164 residue of the IL-6 cytokine and the Asp178 residue of the gp130 receptor. The loss of ligand binding capability toward the IL-6Ra, observed for the Arg164A mutant (Table II) seems to be due to structural disturbance,48 in fact, it reduces the a-helical content by 25% as compared to that of the wild-type molecule. In our model, besides contributing to the stabilization of the interaction with the gp130 receptor, Arg164 interacts with Glu5, in helix A, strengthening the packing of the four helix bundle. Site 2. Interaction between hIL-6 and hgp130 Ligand receptor complementarity of the MEP at site 2 is less marked but the dipolar nature of the interface putatively involved in the interaction showed by both ligand and receptor might be exploit to guide the docking. At the 0.5 kcal/(e·) isopotential surface, the gp130 receptor shows to the ligand a positive pole around the CC’ and the second half of the EF loops, while the AB loop, the interdomain linker GA2, and the B2C2 loop constitute a negative pole. On the contrary, the ligand offers to the receptor a negative surface around the middle of helix A and a positive surface surrounds the N-terminal part of helix A and helix C (Fig. 5). Analysis of the final ternary complex, as resulted from the minimized average structure, reveals just a few specific contact points between polar residues and, among these, Tyr13 and Ser100 have been shown to be important for gp130 binding,21 once combined in the quadruple mutant Tyr13D,Gly17F, Ser100R,Val103D (DFRD). However, by taking into account the synergistic effect of the combined mutations (the only single mutant showing a detectable and reproducible decrease in activity is Tyr13D)21 and their nonconservative nature, it seems to be reasonable to impute the loss of activity to structural changes or to global modifications of the electrostatic potential (the r values for the DFRD mutant at this site is 1.82 kcal/(e·mol)), which prevent the correct orientation of the active cytokine IL-6:IL-6Ra complex with respect to gp130. In this way, because of the generous tolerant nature of the gp130 receptor, the binding of the gp130 is not totally impaired,28 but its efficient association with the IL6-Ra receptors is prevented. Site A. The hIL-6Ra:hgp130 dimerization interface Following the hypothesis formulated in the previous paragraph, the bridging role of the IL-6 ligand with respect to the two receptors is a key factor for the correct orientation of the two receptors which results in a productive receptor–receptor interaction, as also confirmed by the complementarity of the r index at this interface. Residues in the A2B2 loop and E2 strand of the IL-6Ra have been identified to be involved in stabilizing the heterodimeric receptor on the base of a rational site-directed mutagenesis study recently published.29 IL-6Ra presents a positive surface at this interface with the only exception of a small spot of negative potential around the Asp167 residue, on the contrary, gp130 presents a negative surface at this site, which surrounds a large zone of positive potential around the Lys114 residue. This scenario is in full agreement with the results obtained by rational 540 M.C. MENZIANI ET AL. TABLE V. Hydrogen Bonding Pattern for the IL-6 Trimeric Complex. Residues Identified to be Critical for Binding by Site Directed Mutagenesis Stdies are Reported in Bold (see text) IL-6 ARG ARG ARG SER ARG LYS LEU ALA ASN LYS LYS GLU GLU GLU GLU GLU ASP ASP CYS GLN GLN GLN GLN SER LYS ARG LYS GLU GLN GLN GLN ARG ARG ARG IL-6Ra 12 12 12 19 22 36 39 40 42 48 48 51 51 51 51 51 53 53 55 57 57 57 57 58 68 150 153 154 157 157 157 161 161 161 aDistances HH11 HH12 HH21 OG O HZ3 O O OD1 H HZ3 H OE2 OE2 OE1 OE2 OD1 O O H HE22 HE22 HE22 H HZ2 HH21 HZ2 OE2 OE1 OE1 HE22 HE NH2 NH2 GLU ASP GLU ARG GLN CYS GLN GLN GLN ASN TYR GLU ARG ARG GLN TYR ARG SER ASN SER PRO HIS SER SER GLU GLY GLU ARG ARG GLU GLU PHE PHE GLU 183 158 183 138 186 97 95 95 95 41 74 68 37 37 40 74 9 14 131 11 127 128 132 11 68 185 183 9 136 183 183 134 134 183 OE1 OD2 OE2 HH21 HE21 O HE21 HE22 HE21 O OH OE2 HH11 HH21 HE21 HH HH21 HG HD22 O O ND1 OG O OE1 O O HH11 H H OE2 O O OE2 DHAa d(D-A)a IL-6 171.50 156.57 167.30 171.45 169.40 174.56 99.51 162.90 169.36 99.44 168.54 172.65 155.07 154.71 158.44 162.02 162.70 164.11 158.66 164.49 174.07 103.43 100.29 170.59 166.42 148.60 169.78 176.75 171.86 156.33 174.02 154.10 157.21 170.86 2.07 2.38 2.05 1.90 2.10 1.93 2.57 2.25 2.06 2.77 1.92 2.14 2.30 2.38 2.57 2.10 2.05 2.19 2.18 2.11 2.05 2.91 2.53 2.46 2.03 2.43 1.92 1.99 1.99 2.11 2.29 2.19 2.09 2.20 GLU ARG LYS GLN TYR TYR ASP GLU MET SER SER SER LYS GLN GLN GLN GLN LYS THR ARG ARG IL-6Ra ASN TRP TRP TRP TRP HIS HIS HIS HIS HIS HIS HIS ASP ASP gp130 5 6 9 10 13 13 16 92 99 100 100 100 102 106 106 106 109 110 124 164 164 O HH12 HZ2 HE21 OH OH OD1 O O OG HG HG HZ1 HE22 OE1 OE1 OE1 HZ3 HG1 HE HH22 LYS PHE GLU TYR TYR LYS LYS LYS LYS LYS ASP TYR HIS SER GLU TRP TRP GLU HIS ASP ASP gp130 116 HD22 GLN 119 H THR 119 HE1 GLU 119 NE1 ARG 119 NE1 ARG 161 H ASP 162 H ASP 162 O ASP 166 HD1 THR 166 NE2 ARG 166 ND1 THR 166 ND1 THR 167 H THR 167 OD2 LYS 180 64 177 63 63 176 13 46 41 13 58 59 40 35 36 37 37 36 40 178 178 HZ3 O O O OH HZ3 H HZ3 HZ2 NZ2 O O O OG H H HE1 OE2 NE2 OD2 OD2 DHAa d(D-A)a 175.1 169.5 159.8 167.4 130.4 159.6 173.5 174.8 172.7 130.5 92.2 177.9 177.3 176.7 162.1 167.4 168.7 171.4 159.0 158.7 160.6 2.02 2.22 1.95 2.33 2.90 2.94 2.16 1.99 1.90 2.90 2.98 2.04 1.98 2.10 2.30 2.08 2.06 1.98 2.55 2.35 2.22 DHAa d(D-A)a 160 O 106.0 2.57 158 O 108.1 2.75 148 OE2 157.1 2.34 154 HH21 90.9 2.74 154 HH22 110.2 2.43 149 OD1 174.7 2.23 149 OD1 165.3 2.12 149 H 167.2 2.05 158 OG1 173.0 2.02 154 HE 133.4 2.80 158 H 169.0 2.42 158 H 169.0 2.42 158 OG1 177.3 2.46 114 HZ3 167.3 2.14 (Å) and angles (deg) are averaged throughout the molecular dynamics analysis period. site-directed mutagenesis study at this site.29 In fact, double mutation of the His166S, Asp167V prevents the binding with specific residues of gp130 (see Table V), while the introduction of negative charged residues at position 114 and 116 (Ala114D, Asn116D) might result in a significant alteration of the local electrostatic potential (the r values for the SVDD mutant at this site is 0.66 kcal/(e·mol)). This suggests that local, rather than global, charge complementarity controls productive interaction at this site. The Hexameric Complex Rules for the assembly of the hexameric complex are dictated by the following experimental evidences: a) contact between the two gp130 molecules in the hexameric complex has been inferred by the observation that tyrosine kinase activity was associated with the disulfide-linked dimeric protein;5 b) In the examer, the IL-6 ligand establishes interactions with two distinct gp130 chains. The first chain interacts with ligand residues in the helix A and C,the second chain interacts with a composite binding site formed by residues at the beginning of helix D and the initial part of the AB loop.22–28 An equivalent region is involved in the binding of LIF and CNTF to the LIFR chain responsible for signal transduction through heterodimezation of gp130;57,58 c) a very recent study on the influence of IL-6 dimerization on formation of the high affinity hexameric complex shows that the increased binding affinity of the IL-6 dimer (IL-6D) is due to its ability to crosslink two IL-6Ra molecules, while its reduced biological potency results from a decreased ability of THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES the IL-6D:(IL-6Ra)2 complex to obtain a productive coupling with the soluble portion of gp130, in fact the residual biological activity of IL-6D seems to be due to the dimer dissociation in active monomers.59 These results strongly supports a mechanism whereby monomeric IL-6 crosslinks IL-6R and gp130 to form a trimeric complex, two such complexes further dimerizing to form the hexameric species.59 The tentative model of the hexameric receptor complex assembly reported in Figure 6 best conciliates most of the experimental information available and the constraints imposed by the complementarity of the electrostatic potentials on the surface of the trimeric complexes (Fig. 5). In this assembly one trimer is rotated by an angle of about 45° with respect to each other and the two IL-6 molecules are faced in a head to head position (opposite the adjacent amino and carboxy termini). Although this bears a superficial resemblance with the dimerization mode of the M-CSF and SCF cytokines10,46 the long helices of the IL-6 are more tightly packed in structure therefore, the interpenetration of the dyad, which stabilized the dimeric structure of the two short-chain cytokines,46 is prevented in this case and only a small amount of each IL-6 surface results buried. This confirms the hypothesis that IL-6 in stable dimers are oriented differently to those purported to occur in the hexameric complex.59 The electrostatic potential surface computed on the minimized average structure of the trimer presents significant differences with respect the ones computed on the single components. The most striking features of the electrostatic potential of the trimer is a more diffuse positive potential over the IL-6Ra with respect to the dimeric complex and a significant increase of the negative potential area associated with the complexed gp130 with respect to the isolated one. This might constitute the driving force for the hexameric receptor assembly and furnishes one more argument to support a dimerization mechanism involving the two preformed trimeric complexes. The electrostatics in the important regions of protein–protein interactions are reported in Table VI and the pairs of residues which are mainly involved in the stabilization of the complex are listed in Table VII. Due to the size of the hexameric complex, molecular dynamics simulations were not carried out on the system, and the hydrogen bonding pattern reported in Table VII refers to the minimized structure. The interaction at site 3 occurs between IL-6 of the first trimer (IL-6(A)) and the gp130 chain of the second trimer (gp130(B)) and concerns, in our model, a very limited area. However, several residues predicted to be responsible of the second gp130 chain recruitment by site-directed mutagenesis experiments are involved in electrostatic interactions with residues of the gp130 receptor mainly located in the 541 Fig. 6. Stereo view of the IL-6 hexameric receptor complex. The picture was prepared by using the MOLSCRIPT program60 according to secondary structure assignment. C’E loop and B sheet. These are the zones of the gp130 receptor which display the greater variability of residues among different species (human, mouse and rat). The double mutant Trp139R, Asp142R significantly reduces the potential density (the r values is -1.9 kcal/(e·mol)) at this site with devastating consequences on the gp130 dimerization activity (Table II). Thr144 is a buried residue in our model, and as previously hypothesized,61 its substitution with a negative charge residue might induce local distortions which, indirectly, impair the binding with gp130(B). Substitution of the hIL-6 stretch of residues Thr25Glu37 with the corresponding mouse residues give rise to a IL-6 chimera with reduced biological activity. It shows selective impairment of the in vitro binding to gp130 but not to IL-6Ra.61 According to our model, this substitution could reduce binding both at site 3 and site 4. In fact, in the chimera Glu33 is changed to Met, and Lys36 and Glu37 to Asp. Thus, the ionic pairs interactions between the Glu33 and Lys36 IL-6 residues and the Arg16 and Glu18 gp130 residues and the charge reinforced hydrogen bonding between the Glu37 ligand residues and the Lys48 gp130 residue are destroyed. It is worth noticing that the nonconservative mutations which occur in position 33 and 36 of the mouse AB loop of IL-6 are paralleled by nonconservative mutations of residues in position 16 and 18 (Arg16 to Leu, and Glu18 to Gln) in the mouse gp130 receptor, while the conservative mutation in position 37 (Glu37 to Asp) is mirrored by the conservation of residue Lys48 in the mouse receptor. The effect of the IL-6 chimera on the interaction between IL-6(A) and IL-6(B) might be due, according to our model, to the substitution of residues Lys28 and Asn30, which are involved in interactions with Gly and Ser residues of the IL-6 partner BC loop, respectively. Moreover, also the complementarity of the electrostatic potential, observed at this site is markedly reduced. 542 M.C. MENZIANI ET AL. TABLE VI. Electrostatic Potential (EP), Solvent Accessible Surface (SAS) and Density of the Potential (r 5 EP/SAS) at the Surface Generated by the Residues Putatively Involved in Ligand-Receptor and Receptor-Receptor Recognition EP kcal/(e · mol) SAS (Å)2 r kcal/(e · mol)Å2 23901.08 2581.3 558 747 26.99 3.45 30–33, 37, 41, 139, 142–143 14, 16, 47–52 2633.92 21860.52 1153 850 2.28 22.18 19–20, 22–24, 27–30, 86–93 24, 28, 84–85, 87–89, 91–94 2342.17 21951.44 1100 1115 2.13 21.75 142, 144–148, 151, 173, 175, 177, 192–194, 195–198 136–139, 165–167, 169, 189–195 Site 3 hIL-6(A) hgp130(B) Site 4 hIL-6(A) hIL-6(B) Site B hIL-6Ra(A) hgp130(B) Residues TABLE VII. Hydrogen Bonding Pattern for the IL-6 Exameric Complex. Residues Identified to be Critical for Binding by Site Directed Mutagenesis Studies are Reported in Bold IL-6(A) CYS GLU GLU SER LYS LYS GLU GLU TRP ASP ASP TRP IL-6(A) LYS LYS LYS LYS ASN SER SER GLN GLN SER SER SER SER IL-6Ra(A) GLU ARG ARG ARG LYS LYS GLU aDistances DHAa d(D-A)a O O HH12 HH11 OE2 O HZ1 HZ3 HH11 HH11 HH12 HH21 140.81 175.91 162.36 170.2 170.9 170.10 167.48 127.39 156.13 103.81 96.51 148.50 2.73 2.10 2.25 3.00 2.50 2.49 2.07 2.18 1.97 2.69 2.80 2.16 d(D-A)a 1.98 2.13 2.06 2.01 2.05 1.95 2.13 2.42 2.36 2.59 2.23 2.14 2.29 d(D-A)a 2.11 2.03 2.04 2.16 1.92 2.02 2.53 gp130(B) 32 33 33 34 36 36 37 41 139 142 142 139 H H OE1 OG HZ2 HZ1 OE2 OE2 O OD1 OD1 O ASP ASP ARG ARG GLU PRO LYS LYS ARG ARG ARG ARG 50 50 16 16 18 52 48 48 49 49 49 49 23 23 28 28 30 89 89 93 93 89 89 89 89 HZ1 HZ2 HZ2 HZ2 HD22 HG H H HE21 OG HG OG OG IL-6(B) SER GLU GLU GLU GLU PHE SER GLU GLU SER GLU SER GLU 89 24 92 92 92 87 89 88 88 89 24 90 91 OG OE1 OE1 OE1 OE2 O O OE1 OE1 H OE1 H H DHAa 171.88 169.58 173.73 172.98 171.30 177.72 171.66 164.33 118.21 102.99 178.30 175.33 163.93 45 142 142 144 149 149 195 OE2 HH11 HH21 HH21 HZ2 HZ3 OE2 gp130(B) THR GLU GLU GLU VAL TYR LYS 25 188 188 167 103 195 137 HG1 OE2 OE1 OE1 O OH HZ1 DHAa 166.21 177.83 174.86 164.28 173.10 177.05 159.47 (Å) and angles (deg) are computed on the minimized structure. The predicted contact residues of the IL-6Ra of the first trimeric complex IL-6Ra(A) at the interface with the gp130 chain of the second gp130(B) (site B) are located on strand C2, loop C2C2’ and loop F2G2. Some of the IL6Ra(A) residues involved in the interaction have been mutated. Double mutation of residues Arg142S, Tyr143D gives ambiguous results probably due to structural disturbance,30 while a THEORETICAL MODELS OF IL-6 HEXAMERIC COMPLEXES triple mutant Arg147G, Ser148A, Lys149G shows decreased signal trunsduction, ,50% of normal. Moreover, an additional contact point, established between residues Glu45 and Thr25 in the first cytokine binding domains, is observed between the two receptors. It is interesting to note that all the residues of the gp130 chains involved in the receptor–receptor interactions in both site A and B are conserved among species (human, rat and mouse). The assembly does not allow direct contact between the two gp130 chains. Thus, the hypothesized homodimerization of the receptor,5 according to our model should take place in one of the three membral proximal fibronectin type III modules. Indeed three cysteine residues are found in this portion of the sequence: Cys419, Cys458, and Cys466 (the numbering corresponds to the full sequence).54 6. 7. 8. 9. 10. 11. CONCLUSIONS Three-dimensional molecular modeling and computational simulations on hIL-6, hIL-6:hIL-6Ra and hIL-6:hIL-6aR:hgp130 allow the identification and the quantitative analysis of the molecular determinants for structure stabilization and ligand–receptor recognition. Confidence in the models obtained is provided by the concurrence between key residues highlighted by the models and results from mutagenesis studies. Moreover, analysis of the electrostatic potential complementarity of the complex components supports a stepwise mechanism whereby IL-6 recognizes and binds the IL-6Ra to form the active cytokine IL-6:IL-6Ra. This associates with the signal transducing gp130 protein and the trimeric complex formed further dimerizes to form the hexameric species. Furthermore, the model of the hexameric complex provides a rationalization of data from mutation experiments and predicts some key protein–protein interactions which might be the targets of future experimental investigations. 12. 13. 14. 15. 16. 17. 18. 19. ACKNOWLEDGMENT Financial support from CNR and Ministero dell’Università e della Ricerca Scientifica (funds 40%) is acknowledged. REFERENCES 1. Akira S., Taga, T., Kishimoto, T. Interleukin-6 in biology and medicine. Adv. Immunol. 54:1–78, 1993. 2. Yamasaki, K., Taga, T., Hirata, H., Yawata, Y., Kawanishi, B., Seed, T., Taniguchi, T., Hirano, T., Kishimoto, T. Cloning and expression of the human interleukin-6 (BSF-2/ IFNb2) receptor. Science 241:825–828, 1988. 3. Taga, T., Hibi, M., Hirata, Y., Yamasaki, K., Yasukawa, K., Matsuda, T., Hirano, T., Kishimoto, T. Interleukin-6 triggers the association of its receptor with a possible signal transducer, gp130. Cell 58:573–581, 1989. 4. Hibi, M., Murakami, M., Saito, M., Hirano, T., Taga, T., Kishimoto, T., Molecular cloning and expression of an IL-6 signal transducer, gp130. Cell 63:1149–1157, 1990. 5. Murakami, M., Hibi, M., Nakagawa, N., Nakagawa, T., 20. 21. 22. 23. 543 Yasukawa, K., Yamanishi, K., Taga, T., Tadamitsu, K. IL-6 induced homodimerization of gp130 and associated activation of a tyrosine kinase. Science 260:1808–1810, 1993. Hibi, M., Nakajima, K., Hirano, T. IL-6 cytokine family and signal transduction: A model of the cytokine system. J. Mol. Med. 74:1–12, 1996. Ward, L.D., Howlett, G.J., Discolo, G., Yasukawa, K., Hammacer, A., Moritz, R.L., Simpson, R.J. High affinity interleukin-6 receptor ia a hexameric complex consisting of two molecules each of interleukin-6, interleukin-6 receptor and gp130. J. Biol. Chem. 269:2386–23289, 1994. Paonessa, G., Graziani, R., De Serio, A., Savino, R., Ciapponi, L., Laham, A., Salvati, A.L., Toniatti, C., Ciliberto, G. Two distinct and independent sites on IL-6 trigger gp130 dimer formation and signalling. EMBO J. 14:1942–1951, 1995. De Serio, A., Graziani, R., Laufer, R., Ciliberto, G., Paonessa, G. In vitro binding of ciliary neutrophic factor to its receptors: Evidence for the formation of an IL-6-type hexameric complex. J. Mol. Biol. 254:795–800, 1995. Sprang, S.R., Bazan, J.F. Cytokine structural taxonomy and mechanisms of receptor engagement. Curr. Opin. Struct. Biol. 3:815–827, 1993. Mott, H.R., Campbell, I.D. Four-helix bundle growth factors and their receptors: Protein–protein interactions. Curr. Opin. Struct. Biol. 5:114–121, 1995. Bazan J.F. Structural design and molecular evolution of a cytokine receptor superfamily. Proc. Natl. Acad. Sci. U.S.A. 87:6934–6938, 1990. De Vos, A., Ultsch, M., Kossiakoff, A.A. Human growth hormone and extracellular domain of its receptor: Crystal structure of the complex. Science 255:306–312, 1992. Somers, W., Ultsch, M., De Vos, A., Kossiakoff, A.A. The x-ray structure of a growth hormone-prolactin receptor complex. Nature 372:478–481, 1994. Livnah, O., Stura, E.A., Johnson, D.L., Middleton, S.A., Mulcahy, L.S., Wrighton, N.C., Dower, W.J., Jolliffe, L.K., Wilson, I.A. Functional mimicry of a protein hormone by a peptide antagonist: The EPO receptor complex at 2.8Å. Science 273:464–471, 1996. Horsten, U., Van de Leur, H.S., Müllberg, J., Heinrich, P.C., Rose-John, S. The membrane distal half of gp130 is responsible for the formation of a ternary complex with IL-6 and the IL-6 receptor. FEBS Lett. 360:43–46, 1995. Savino, R., Laham, A., Giorgio, M., Cabibbo, A., Tramontano, A., Ciliberto, G. Saturation mutagenesis of the human interleukin-6 receptor-binding site: Implications for its three-dimensional structure. Proc. Natl. Acad. Sci, U.S.A. 90:4067–4071, 1993. Fontaine, V., Ooms, J., Content, J. Mutagenesis of the human interleukin-6 fourth predicted a helix: Involvement of the Arg168 in the binding site. Eur. J. Immunol. 24:1041– 1045, 1994. Nishimura, C., Watanabe, A., Gouda, H., Schimada, I., Arata, Y., Folding topologies of human interleukin-6 and its mutants as studied by NMR spectroscopy. Biochemistry 35:273–281, 1996. Cabibbo, A., Sporeno, E., Toniatti, C., Altamura, S., Savino, R., Paonessa, G., Ciliberto, G. Monovalent phage display of human interleukin (hIL-6): Selection of superbinder variants from a complex molecular repertoire in the hIL-6 D-helix. Gene 167:41–47, 1995. Savino, R., Laham, A., Salvati, A.L., Ciapponi, L., Sporeno, E., Altamura, S., Paonessa, G., Toniatti, C., Ciliberto, G. Generation of interleukin-6 receptor antagonists by molecular-modeling guided mutagenesis of residues important for gp130 activation. EMBO J. 13:1357–1367, 1994. Savino, R., Ciapponi, L., Laham, Demartis, A., Cabibbo, A., Toniatti, C., Altamura, S., Ciliberto, G. Rational design of a receptor super-antagonist of human interleukin–6. EMBO J. 13:5863–5870, 1994. Ehlers, M, de Hon, F.D., Boss, K. H., Horsten, U., Kurapkat, G., van De Leur, H.S., Grötzinger, J., Wollmer, A., Brakenhoff, J.P.J, Rose-John, S. Combining two mutations of human interleukin-6 that affect gp130 activation results in a potent interleukin-6 receptor antagonist on human myeloma cells. J. Biol. Chem. 270:8158–8163, 1995. 544 M.C. MENZIANI ET AL. 24. de Hon, F.D., Bos, H. K., Ebeling, S.B., Grötzinger, J., Kurapkat, G., Rose-John, S., Aarden, L.A., Brakenhoff, J.P.J. Leucine-58 in the putative 5th helical region of human interleukin (IL)-6 is important for activation of the IL-6 signal transducer, gp130. FEBS Lett. 369:187–191, 1995. 25. Brakenhoff, J.P.J., Bos, H. K., Grötzinger, J.G., Rose-John, S., Aarden, L.A. Identification of residues in the putative 5th helical region of human interleukin-6, important for activation of the IL-6 signal transducer, gp130. FEBS Let. 395:235–240, 1996. 26. Brakenhoff, J.P.J., de Hon, F.D., Fontaine, V., Boekel, E., Schooltink, H., Rose-John, S., Heinrich, P.C., Content, J., Aarden, L.A. Development of human interleukin-6 receptor antagonist. J. Biol. Chem. 269:86–93, 1994. 27. Toniatti, C., Cabibbo, A., Sporeno, E., Salvati, A.L., Cerretani, M., Serafini, S., Lahm, A., Cortese, R., Ciliberto, G. Engineering human interleukin-6 to obtain variants with strongly enhanced bioactivity. EMBO J. 15:2726–2737, 1996. 28. Sporeno, E., Savino, R., Ciapponi, L., Paonessa, G., Cabibbo, A., Laham, A., Pulkki, K., Sun, R.X., Toniatti, C., Klein, B., Ciliberto, G. Human interleukin-6 receptor super-antagonistswith high potency and wide apectrum on multiple myeloma cells. Blood 87:4510–4519, 1996. 29. Salvati, A.L., Laham, A., Paonessa, G., Ciliberto, G., Toniatti, C. Interleukin-6 (IL-6) antagonism by soluble IL-6 receptor a mutated in the predicted gp130 binding interface. J. Biol. Chem. 270:12242–12249, 1995. 30. Yawata, H., Yasukawa, K., Natsuka, S., Murakami, M., Yamasaki, K., Hibi, M., Taga, T., Kishimoto, T. Structure– function analysis of human IL-6 receptor: Dissociation of amino acid residues required for IL-6 binding and for signal transduction through gp130. EMBO J. 12:1705– 1712, 1993. 31. Kalai, M., Montero-Julian, F.A., Grotzinger, J., Wollmer, A., Morelle, D., Brochier, J., Rose-John, S., Heinrich, P. C., Brailly, H., Content, J. Partecipation of two Ser-Ser-PheTyr repeats in interleukin-6 (Il-6) binding sites of the human IL-6 receptor. Eur. J. Biochem. 238:714–723, 1996. 32. Bernstein, F.C., Koetzle, T.F., Williams, G.J.B., Meyer, E.F.Jr, Brice, M.D., Rodgers, J.R., Kennard, O., Shimanouchi, T., Tasumi, M. The protein data bank: A computerbased archival file for macromolecular structures. J. Mol. Biol. 112:535–542, 1977. 33. Rost, B., Sander, C. Combining evolutionary information and neural networks to predict protein secondary structure. Proteins 19:55–72, 1994. 34. Sali, A., Blundell, T.L. Comparative protein modeling by satisfaction of spatial restraints. J. Mol. Biol. 234:779–815, 1993. 35. Jones, D.T., Taylor, W.R., Thornton, J.M. A new approach to protein fold recognition. Nature 358:86–89, 1992. 36. Brooks, B.R., Bruccoleri, R.E., Olafson, B.D., States, D.J., Swaminathan,S., Karplus, M. CHARMM: A program for macromolecular energy, minimization and dynamics calculations. J. Comput. Chem. 4:187–217, 1983. 37. van Gunsteren, W.F., Berendsen, J.C. Algorithms for macromolecular dynamics andconstraintdynamics. Mol. Phys. 34:1311–1327, 1977. 38. Bowie, J.U., Lüthy, R., Eisenberg, D. A method to identify protein sequences that fold into a known three-dimensional structure. Science 253:164–170, 1991. 39. Vriend G. WHAT IF: A molecular modeling and drug design program. J. Mol. Graphics 8:52–56, 1990. 40. Laskowski, R.A., Mac Arthur, M. W., Moss, D.S., Thornton, J. M. PROCHECK: A program to check the stereochemical quality of protein structures. J. Appl. Cryst. 26:283–291, 1993 41. Devis, M.E., Madura, J.D., Luty, B.A.,Mc Cammon, J.A. Electrostaics and diffusion of molecules in solutions: Simulations with the University of Houston Brownian Dynamics Program. J. Comp. Phys. Commun. 62:187–197, 1990. 42. Jorgensen, W.L., Tirado-Rives, J. The OPLS potential functions for proteins. Energy minimizations for crystal of cyclic peptides and crambins. J. Am. Chem. Soc. 110:1657– 1666, 1988. 43. Devis, M.E., Mc Cammon, J.A. Dielectric boundary smoothing in finite differnce solutions of the Poisson equation: An approach to improve accuracy and convergence. J. Comput. Chem. 12:909–912, 1991. 44. Somers, W., Stahi, M., Seehra, J.S. 1.9 Å crystal structure of interleukin 6: Implications for a novel mode of receptor dimerization and signaling. The EMBO J. 16:989–997, 1997. 45. Robinson, R.C., Grey, L.M., Staunton, D., Vankelecom, H., Vernallis, A.B., Moreau, J.-F., Stuart, D.I., Heath, J.K., Jones, E.Y. The crystal structure and biological function of leukemia inhibitory factor: Implication for receptor binding. Cells 77:1101–1116, 1994. 46. Menziani, M.C., Fanelli, F., De Benedetti, P.G. Computational simulations of Stem-Cell factor/c-Kit receptor interaction. Proteins Struct. Func. Genet. 26:42–54, 1996. 47. Nishimura, C., Futatsugi, K., Yasukawa, K., Kishimoto, T., Arata, Y. Site-specific mutagenesis of human interleukin-6 and its biological activity. FEBS 281:167–169, 1991. 48. Leebeek, W.G., Kariya, K., Schwabe, M., Fowlkes, D.M. Identification of a receptor binding site in the carboxyl terminus of human interleukin-6. J. Biol. Chem. 267:14832– 14838, 1992. 49. Li, X., Rock, F., Chong, P., Cockle, S., Keating, A., Ziltener, H., Klein, M. Structure–function analysis of the C-terminal segment of human interleukin-6. J. Biol. Chem. 268:22377– 22384, 1993. 50. Hammacher, A., Simpson, R.J., Nice, E.C. The interleukin (IL-6) partial antagonist (Q159E, T162P)IL-6 interacts with the IL-6 receptor and gp130 but fails to induce a stable hexameric receptor complex. J. Biol. Chem. 271: 5464–5473, 1996. 51. Nishimura, C., Ekida, T., Masuda, S., Futatsugi, K., Itoh, S, Yasukawa, K., Kishimoto, T., Arata, Y., Chemical modification and 1H-NMR studies on the receptor-binding region of human interleukin-6. Eur. J. Biochem. 166:377–384, 1991. 52. McDonald, N.Q., Panayotatos, N., Hendrickson, W.A. Crystal structure of dimeric human ciliary neutrophic factor determined by MAD phasing. The EMBO J. 14:2689–2699, 1995. 53. Defay, T.R., Cohen, F.E. Multiple sequence information for threading algorithms. J. Mol. Biol. 262:314–323, 1996. 54. Zilberstein, A., Ruggeri, R., Korn, J.H., Revel, M. Structure and expression of cDNA and genes for human interferonbeta-2, a distinct species inducible by growth-stimulatory cytokines. EMBO J. 5:2529–2537, 1986. 55. Muller,Y.A., Ultsch, M.H., de Vos, A.M. The crystal structure of the extracellular domain of human tissue factor refined to 1.7 Å resolution. J. Mol.Biol. 256:144–159, 1996. 56. Demuchuk, E., Mueller, T., Oschkinat, W.S., Wade, R.C. Receptor binding properties of four-helix-bundle growth factors deduced from electrostatic analysis. Protein Sci. 3:920–935, 1994. 57. Hudson, K.R., Vernallis, A.B., Heath, J.K. Characterization of the receptor binding sites of human leukemia inhibitory factor and creation of antagonists. J. Biol. Chem. 271:11971–11978, 1996. 58. Di Marco.A., Gloaguen, I., Graziani, R., Paonessa, G., Saggio, I., Hudson, K.R., Laufer, R. Identification of ciliary neutrophic factor (CNTF) residues essential for leukemia inhibitory factor receptor binding and generation of CNTF receptors antagonists. Proc. Natl Acad. Sci. U.S.A. 93:9247– 9252, 1996. 59. Ward, L.D., Hammacher, A., Howlett, G.J., Matthows, J.M., Fabri, L., Moritz, R.L., Nice, E.C., Weinstock, J., Simpson, R.J. Influence of interleukin-6 (IL-6) dimerization on formation of the high affinity hexameric IL-6receptor complex. J. Mol. Biol. 271:20138–20144, 1996. 60. Kraulis, P.J. MOLSCRIPT: A program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallogr. 24:946–950, 1991. 61. Ciapponi, L., Graziani, R., Paonessa, G., Laham, A., Ciliberto, G. Definition of a composite binding site for gp130 in human interleukin-6. J. Biol. Mol. 270:31249–31254, 1995.