Molecular Determinants of the Granulocyte-Macrophage Colony-stimulating Factor Receptor Complex Assembly*

The granulocyte-macrophage colony-stimulating factor (GM-CSF) receptor (GMR) is composed of two chains that belong to the superfamily of cytokine receptors typified by the growth hormone receptor. A common structural element found in cytokine receptors is a module of two fibronectin-like domains, each characterized by seven β-strands denoted A–G and A′–G′, respectively. The α-chain (GMRα) confers low affinity GM-CSF binding (K d = 1–5 nm), whereas the β-chain (βc) does not bind GM-CSF by itself but confers high affinity binding when associated with α (K d = 40–100 pm). In the present study, we define the molecular determinants required for ligand recognition and for stabilization of the complex through a convergence of several approaches, including the construction of chimeric receptors, the molecular dynamics of our three-dimensional model of the GM·GMR complex, and site-directed mutagenesis. The functional importance of individual residues was then investigated through ligand binding studies at equilibrium and through determination of the kinetic constants of the GM·GMR complex. Critical to this tripartite complex is the establishment of four noncovalent bonds, three that determine the nature of the ligand recognition process involving residues Arg280 and Tyr226 of the α-chain and residue Tyr365 of the β-chain, since mutations of either one of these residues resulted in a significant decrease in the association rate. Finally, residue Tyr365 of βc serves a dual function in that it cooperates with another residue of βc, Tyr421 to stabilize the complex since mutation of Tyr365 and Tyr421 result in a drastic increase in the dissociation rate (Koff). Interestingly, these four residues are located at the B′-C′ and F′-G′ loops of GMRα and of βc, thus establishing a functional symmetry within an apparently asymmetrical heterodimeric structure.

The granulocyte-macrophage colony-stimulating factor (GM-CSF) receptor (GMR) is composed of two chains that belong to the superfamily of cytokine receptors typified by the growth hormone receptor. A common structural element found in cytokine receptors is a module of two fibronectin-like domains, each characterized by seven ␤-strands denoted A-G and A-G, respectively. The ␣-chain (GMR␣) confers low affinity GM-CSF binding (K d ‫؍‬ 1-5 nM), whereas the ␤-chain (␤ c ) does not bind GM-CSF by itself but confers high affinity binding when associated with ␣ (K d ‫؍‬ 40 -100 pM). In the present study, we define the molecular determinants required for ligand recognition and for stabilization of the complex through a convergence of several approaches, including the construction of chimeric receptors, the molecular dynamics of our three-dimensional model of the GM⅐GMR complex, and site-directed mutagenesis. The functional importance of individual residues was then investigated through ligand binding studies at equilibrium and through determination of the kinetic constants of the GM⅐GMR complex. Critical to this tripartite complex is the establishment of four noncovalent bonds, three that determine the nature of the ligand recognition process involving residues Arg 280 and Tyr 226 of the ␣-chain and residue Tyr 365 of the ␤-chain, since mutations of either one of these residues resulted in a significant decrease in the association rate. Finally, residue Tyr 365 of ␤ c serves a dual function in that it cooperates with another residue of ␤ c , Tyr 421 to stabilize the complex since mutation of Tyr 365 and Tyr 421 result in a drastic increase in the dissociation rate (Koff). Interestingly, these four residues are located at the B-C and F-G loops of GMR␣ and of ␤ c , thus establishing a functional symmetry within an apparently asymmetrical heterodimeric structure.
GM-CSF 1 is a multifunctional growth factor (reviewed in Ref. 1) that stimulates the proliferation of hemopoietic and vascular endothelial cells. Moreover, GM-CSF suppresses apoptosis in hemopoietic precursors (2)(3)(4) while enhancing the response of neutrophils to bacterial antigens and the phagocytic activity of macrophages/monocytes (reviewed in Ref. 5). The GM-CSF receptor is composed of two chains that belong to the superfamily of cytokine receptors typified by the growth hormone receptor (GHR) (6 -8). The ␣-chain confers low affinity binding only (K d ϭ 1-5 nM) (9,10), whereas the ␤-chain (␤ c ) (11) does not bind GM-CSF by itself but confers high affinity binding when associated with the ␣-chain (K d ϭ 40 -100 pM) (12)(13)(14)(15). We have previously shown that the high affinity GM-CSF binding site results from the stoichiometric association of GMR␣ with ␤ c , resulting in a stabilization of the GM⅐GMR complex by 3 orders of magnitude (16). Previous studies also suggest that the low affinity binding of GMR␣ to the ligand may be attributed to a major difference in the off rate when compared with that of the GMR␣⅐␤ c complex, whereas the on rates are not significantly different (17,18).
Through sequence alignment, Bazan (6) first predicted that cytokine receptors are made up of two fibronectin folds, each containing seven antiparallel ␤-strands, similar to the immunoglobulin fold. These strands are labeled A-G for the first fibronectin fold and AЈ-GЈ for the membrane proximal fibronectin fold. Crystallization of the GH⅐GHR complex (19), of erythropoietin receptor (EpoR) complexed with an Epo agonist (20), as well as the solution structure of the granulocyte colonystimulating factor receptor (21) confirmed the predicted structure. Interestingly, the RGD integrin binding site of human fibronectin is located at the F-G loop of the four-domain segment (22). In addition, the major ligand binding site for the granulocyte colony-stimulating factor receptor was also identified at the FЈ-GЈ loop just upstream of the WSXWS box. The three-dimensional structure of GMR is not available, but that of the ligand was solved by x-ray crystallography and NMR analysis (23). We have previously constructed a three-dimensional model of the GMR complex on the basis of the GH⅐GHR complex in which the GH crystal was replaced by that of GM-CSF, while the ␣and ␤-chains of GMR were modeled according to the R1 and R2 chains of GHR (24). Through site-directed mutagenesis guided by homology modeling with the growth hormone receptor complex, we identified a single residue, Arg 280 , located at the tip of a ␤-turn of the FЈ-GЈ loop of GMR␣ that drives the ligand recognition process (24). In addition, alanine scanning mutagenesis of the ␤-chain identified two tyrosine residues, Tyr 365 at the BЈ-CЈ loop and Tyr 421 at the FЈ-GЈ loop as crucial for high affinity GM-CSF binding (25)(26)(27). It is not clear, however, whether these two tyrosine residues are involved in ligand recognition or in stabilization of the complex. Finally, two phenylalanine residues of the EpoR, Phe 93 and Phe 205 , located at the E-F and FЈ-GЈ loops, respec-tively, are critical determinants of Epo binding (28), consistent with their locating at the hydrophobic interface between the Epo mimetic peptide and EpoR in the structure of the crystal (20). Together, these studies identified three loops within the two fibronectin folds of the superfamily of cytokine receptors that are potentially involved in ligand binding, the E-F, BЈ-CЈ, and FЈ-GЈ loops. The present study is designed to define the functional determinants of the GM-CSF receptor assembly, through a convergence of binding studies performed at equilibrium and kinetic studies of ligand-receptor interaction. In addition, molecular dynamics simulations were performed on a model of the GM-CSF receptor assembly in order to determine amino acids potentially located at contact points between the ligand and receptor and to guide our mutagenesis aimed at defining the functional interface of the complex.

MATERIALS AND METHODS
Cells and Plasmids-Jurkat cells (ATCC) were passaged three times weekly, at 5 ϫ 10 4 /ml in Iscove's medium (Life Technologies, Inc.) supplemented with 10% fetal calf serum (Life Technologies). The human GMR␣ and ␤ c open reading frame, cloned in pME18S, were a generous gift of Dr. Toshio Kitamura (Tokyo, Japan) (11). Site-directed mutagenesis was performed through polymerase chain reaction overlap, as described previously (24). All polymerase chain reaction-generated fragments were completely sequenced in both orientations after cloning. In addition, the reactivity of all GMR␣ and ␤ c mutants to two monoclonal antibodies (29) was tested by flow cytometry in order to confirm that the mutation did not cause a major disruption of the three-dimensional structure of the molecule (24). Briefly, cells were labeled with the monoclonal antibody against human GMR␣ (Upstate Biotechnology, Inc., Lake Placid, NY) or the monoclonal antibody against human ␤ c (Pharmingen, Mississauga, Ontario, Canada) at 1 g/10 6 cells, followed by a second incubation with a goat anti-mouse IgG coupled to phycoerythrin (Pharmingen). Mutant GM-CSFs were produced as described previously (30).
Electroporation of Jurkat Cells for Transient Expression of GMR␣ and ␤ c -Exponentially growing Jurkat cells were resuspended at 2.5 ϫ 10 7 cells in a volume of 600 l of medium and electroporated at 960 microfarads, 320 V with 5 g of GMR␣ and 5 g of ␤ c cDNA, together with 15 g of pGEM as carrier DNA. The efficiency of gene transfer was found to be significantly enhanced in the presence of carrier DNA. Cells were harvested after 24 h for binding studies and were more than 80% viable. Kinetic studies indicated that binding was optimal at 24 h postelectroporation. This time point was therefore chosen for all studies.
Iodinated GM-CSF Binding Assays-Purified GM-CSF was iodinated using the Bolton-Hunter reagent, as described previously (16,24). The specific activity of the ligand was determined through radioimmunoassay. Binding assays were carried out essentially as described previously (16). Briefly, 10 6 cells were incubated with the radioligand for 4 h at 4°C in a total volume of 100 l of bicarbonate-free Iscove's medium. Free GM-CSF was separated from cell-bound GM-CSF by centrifugation over a cushion of silicon/paraffin oil. Nonspecific binding was determined in the presence of a 100-fold excess of cold ligand. GM-CSF saturation curves and competition curves were analyzed by curve fitting using the ALLFIT program, based on a nonlinear regression curve fitting routine (31). The loss in free energy of binding introduced by the respective mutations were calculated as follows (32).
Kinetics of GM-CSF Binding-The association rate constant (K on ) was determined at three concentrations of radioligand, as described previously (24). For the dissociation rate constant (K off ), GM-CSF binding was performed at equilibrium at 1-4 nM radioligand for the various GMR mutants and at 150 pM for the wild type GMR complex. Dissociation was initiated through the addition of a 100-fold excess of cold ligand and was allowed to proceed at 4°C for 1-30 min.
Homology Modeling and Molecular Dynamics of the GM-CSF Receptor Complex-All calculations were performed using the software package SYBYL (Tripos Associates, St. Louis, MO) on a Silicon Graphics Indigo 2 Extreme workstation. The Tripos force field was used for energy calculations, and a dielectric constant of 1 was used to approximate a protein receptor environment. A model of the GM-CSF receptor complex has been constructed and reported elsewhere (24). Briefly, the x-ray crystal structure of human growth hormone bound to its receptor (19) was obtained from the Brookhaven protein data base. This complex was subjected to 1000 steps of conjugate gradient minimization. Using the sequence alignment reported by Goodall et al. (33), the extracellular domains of the two subunits of human growth hormone were transformed one amino acid at a time into the GM-CSF receptor ␣-chain and the membrane proximal cytokine receptor motif ␤ c . After the transformation was completed, the side chains of the individual amino acids were packed using the scan subroutine in SYBYL. The backbone dihedryl angles were held fixed, while the amino acid side chain dihedryl angles were rotated one at a time until a sterically acceptable conformation was obtained. The x-ray crystal structure of GM-CSF has been reported (23). The human GH ligand was excised from the hormone receptor complex, and GM-CSF was positioned into the new receptor complex visually in approximately the same orientation and position as was human GH. When the new GM-CSF receptor complex displayed reasonable steric complementarity, the entire complex was minimized as above, allowing all atoms to relax. Three separate molecular dynamics simulations of 20 ps each were performed on this complex using different randomized starting velocities for each simulation. Changes in secondary structural elements and distances between selected amino acids were monitored for each trajectory. A model of the ␣-homodimer complexed to GM-CSF was constructed in a similar manner and subjected to two 20-ps molecular dynamics simulations.

Role of the BЈ-CЈ and FЈ-GЈ Loops of the ␣-Chain of the GM-CSF Receptor in Ligand
Binding-In our three-dimensional model of the GM-CSF receptor complex, several loops appear to be oriented toward the ligand: the E-F loop of the first cytokine receptor domain as well as the BЈ-CЈ and FЈ-GЈ loops of the second cytokine domain. In order to directly address the importance of these loops in ligand binding, we chose to generate chimeras in which the three loops of GMR␣ were substituted by the corresponding loops of IL-3R␣ (15). Our rationale for choosing IL-3R␣ is based on the evidence that it is the closest to GMR␣ in primary sequence, in predicted three-dimensional structure, as well as in functional and biological properties (18,33,34), yet IL-3R␣ demonstrates exquisite binding specificity for its ligand. The primary sequences of the three loops as well as their flanking ␤-strands are illustrated in Fig.  1. The alignment shows that the primary sequences of the loops are highly variable, except for the arginine at the FЈ-GЈ loop identified in our previous study (24) and for the WS box as shown by Bazan (6). In addition, the FЈ-GЈ loops are conserved in length, whereas the E-F and BЈ-CЈ loops are more variable. In order to preserve the overall structure of the chimeric molecule, we have chosen to maintain the length of the loops as in GMR␣ such that amino acid residues of the interleukin-3 receptor were removed (E-F chimera) or amino acid residues of GMR␣ were added (BЈ-CЈ chimera) to those of the interleukin-3 receptor loops as shown in Fig. 1. NIH 3T3 cells were transiently transfected with the different GMR␣ chimeras alone and submitted to GM-CSF binding assays (Fig. 2). Substitution of the BЈ-CЈ loop as well as the FЈ-GЈ loop of GMR␣ with those of IL-3R␣ completely abrogated GM-CSF binding, while the E-F loop of the first fibronectin domain showed only a 2-fold decrease in ligand binding.
Since our previous results indicate that a mutant ␣, which is unable to bind GM-CSF, can still associate with wild type ␤ c to favor ligand binding (24), we co-transfected the different chimeric GMR␣ with wild type ␤ c . Results shown in Fig. 2C and Table I indicate that the substitution of interleukin-3 receptor sequences at the BЈ-CЈ loop of GMR␣ converted ligand binding to a single low affinity binding site of 1-4 nM, whereas the E-F chimera can direct both high and low affinity GM-CSF binding when co-expressed with wild type ␤ c in NIH 3T3 cells (Fig. 2B), much in the same way as wild type GMR␣ (data not shown). These results confirm the overall importance of the FЈ-GЈ loop in ligand binding as observed previously (24) and reveal the importance of the BЈ-CЈ loop in this process. In contrast, the contribution of the E-F loop of the first fibronectin domain appears to be minimal.
Contribution of the BЈ-CЈ Loop of GMR␣: Molecular Dynamics of the GM⅐GMR Complex-We have previously shown that a single residue at the FЈ-GЈ loop, Arg 280 , drives the ligand recognition process on the ␣-chain and establishes a salt bridge with residue Asp 112 of the ligand (24). No other residue of this loop appears to be important except for residue Asp 278 , which, in our three-dimensional model, points toward residue Lys 191 of the linker domain of GMR␣ and could thus establish an intrachain salt bridging (24). We therefore focused on the BЈ-CЈ loop because of its functional importance in ligand binding and evaluated the three-dimensional model of the GM-CSF receptor complex in order to identify residues within this loop region that may interact with the ligand. In the x-ray crystal structure of the original human growth hormone receptor complex, the BЈ-CЈ loop was found to be distant (ϳ20 Å) from the ligand. Because of this, the BЈ-CЈ loop region of the constucted model of the GM-CSF receptor complex was also distant from the ligand. However, during several molecular dynamics simulations of this complex we observed considerable flexibility in this loop region of the complex (Fig. 3). Although the extent of motion of the BЈ-CЈ loop varied for each simulation, in each case this loop tended to move in close proximity to the ligand. By monitoring the movement of individual amino acids within this loop, we identified Tyr 226 moving in close contact with the Leu 115 side chain of the ligand (Fig. 4A) in all of the simulations, indicating a possible hydrophobic interaction involved in ligand binding.
In addition, we also constructed a model of a low affinity GM-CSF receptor complex (Fig. 5) in order to identify interactions that are established between GMR␣ and the ligand, in the absence of ␤ c . Previous work from our laboratory and elsewhere suggested that GMR␣ binds GM-CSF as a dimer (35,36). Since GM-CSF is a monomer, we modeled the ␣ homodimer complexed to its ligand on the basis of the structure of the growth hormone-receptor complex, i.e. with asymmetric binding on either side. As in the ␣⅐␤ configuration, the FЈ-GЈ loop of the first ␣-chain establishes a contact point with the ligand through a salt bridge between ␣ 1 Arg 280 and (GM)Asp 112 . Again, after molecular dynamics simulations, we observed that the BЈ-CЈ loop of the second GMR ␣-chain moved into the vicinity of the ligand (Fig. 5). Specifically, we identified Tyr 226 in the vicinity of Arg 23 on the ligand, suggesting a potential contact point determined by -charge interaction (Fig. 4B). Hence, according to our three-dimensional models of the high and low affinity binding complexes, the nature of Tyr 226 interactions may vary from a hydrophobic interaction with Leu 115 on site 1 (␣⅐␤ heterodimer and ␣⅐␣ homodimer) to a -charge interaction with Arg 23 on site 2 (␣⅐␣ homodimer). It is thus anticipated that Tyr 226 may be more important for ligand binding in the ␣⅐␣ configuration, since it establishes two contact points with the ligand as opposed to one contact point, respectively.  (B and C), 125 I-GM-CSF was added at the indicated concentrations. The upper curve represents total binding, whereas the lower curve shows nonspecific binding determined in the presence of a 100-fold excess of cold GM-CSF. The curves passing through the data were calculated with the program ALLFIT (24,31). For wild type GMR␣ (not shown) or the ␣E-F chimera (B) associated with ␤ c , the data were better described by a two-binding site model with high and low affinity binding, whereas the data observed with the ␣(BЈ-CЈ) chimera and wild type ␤ c were better described by a one-binding site model (low affinity only). K d values are shown in Table I.
Functional Importance of the Aromatic Ring at Position 226 of GMR␣-Since Tyr 226 of GMR␣, located at the BЈ-CЈ loop, moved into a strategic position after the simulations, we addressed its importance as well as that of another aromatic residue of the same loop, Tyr 221 . Two mutations were introduced at these positions, a tyrosine to phenylalanine mutation aimed at preserving hydrophobic interactions conferred by the aromatic ring as well as a tyrosine to glutamine mutation, which removes the possibility of hydrophobic interaction. When binding was performed in the absence of ␤ c (Fig. 6A and Table  I), Y226F was able to associate with GM-CSF to the same extent as wild type GMR␣. In contrast, the Y226Q mutation showed a complete loss of GM-CSF binding as illustrated in Fig. 6A and determined through saturation analyses (data not shown). In comparison, mutation of Tyr 221 , which is also located at the BЈ-CЈ loop to either glutamine or phenylalanine did not affect GM-CSF binding under the same conditions ( Fig. 6A and Table I). The loss of low affinity binding caused by the Tyr 226 mutation was not due to structural alterations of the molecule, as assessed by reactivity with a monoclonal anti-GMR␣ (Fig. 7). Our observations are therefore consistent with an important function for the aromatic ring at position 226 in the ␣⅐␣ configuration.
The impact of the Y226Q mutation was further evaluated in the presence of ␤ c . Data shown in Fig. 6C indicate that the Y226Q mutant exhibits a single GM-CSF binding site of high to intermediate affinity, whereas the low affinity binding site was lost. Thus, the Y226Q mutation resulted in a loss of binding free energy of 0.5-0.8 kcal/mol when co-transfected with wild type ␤ c , which accounts for almost half of the loss observed with the BЈ-CЈ chimera (Table I). Together, our results indicate an important role for the aromatic ring in establishing hydrophobic interactions or -charge interaction at position 226 of GMR␣ in the absence of ␤ c and, to a lesser extent, in the presence of ␤ c .
Binding studies performed at equilibrium allow for an estimation of the energetic contribution of individual residues but do not discriminate between a role in the association process itself (K on ) or in the dissociation rate (K off ). The contribution of Tyr 226 to the GMR complex was therefore investigated through determination of the kinetic constants for the GMR␣ Y226Q mutant and wild type GM-CSF (Fig. 8).
In the absence of ␤ c , there was a 3.5-fold increase in the dissociation rate for the Y226Q mutant, consistent with the observed shift in the equilibrium constant determined by saturation analysis (data not shown). In the presence of ␤ c , there was a 5-fold increase in the off rate when the aromatic residue at position 226 of the ␣-chain was mutated to a hydrophilic residue, comparable with that observed with the ␣-BЈ-CЈ chimera or with the ␣R280M mutant. This mild increase in K off is not comparable with those produced by mutations of tyrosine residues of ␤ c , as discussed underneath. There was a significant decrease in the on rate, resulting in an overall 4 -8-fold increase in the dissociation constant. In comparison, mutation of Arg 280 produced a 10 -12-fold decrease in K on , as reported previously (24). Together these decreases in K on suggest that ␣-Tyr 226 contributes to the ligand recognition process, albeit not to the same extent as ␣-Arg 280 .
Comparative Contribution of Tyrosine Residues on the ␤-Chain to the Stability of the GM⅐GMR Complex-We first compared the kinetics of binding of the ␣-chain alone (low affinity) to that of the ␣⅐␤ c complex (high affinity). Data shown in Table I indicate a 100-fold difference in the dissociation rate between the ␣ and ␣⅐␤ complex, whereas the difference in the association rate was 10-fold resulting in a 1000-fold difference in affinity for the ligand. These results are in agreement with the difference in K d observed between ␣ and the ␣⅐␤ complex as determined through saturation analysis. Our observations are therefore consistent with the view that the recruitment of ␤ c results in stabilization of the GM⅐GMR complex. The difference in the on rates between ␣ and the ␣⅐␤ complex is, however, higher than anticipated (18) and suggests that ␤ c can contribute to ligand binding as described elsewhere (24,25,27) albeit to a lesser extent when compared with the ␣-chain.
Previous studies identified two residues of the ␤-chain that are important for high affinity GM-CSF binding, Tyr 365 at the BЈ-CЈ loop and Tyr 421 at the FЈ-GЈ loop. These studies, however, did not discriminate between a direct contribution to the ligand recognition process or a structural function involved in complex stability. We therefore mutated these two residues into Gln and verified that the mutations did not induce gross structural changes in the molecule through staining with a monoclonal antibody against ␤ c (Fig. 7). The binding kinetics of mutant GMR ␤Y365Q and ␤Y421Q were then compared with those of wild type ␤ c , in association with wild type ␣-chains. The association rate (K on ) for ␤Y365Q was 4-fold lower than that of wild type ␤ c , whereas the K on of ␤Y421Q was unaffected by the mutation (Fig. 8), suggesting that Tyr 365 but not Tyr 421 is important for ligand association. The dissociation rate constant (K off ) for ␣⅐␤Y365Q and ␣⅐␤Y421Q were, however, 20-and 30fold higher than that of wild type ␣⅐␤ c . These results indicate that the difference in binding affinities observed for the GMR ␤Y365Q and ␤Y421Q mutants compared with wild type GMR is TABLE I Binding characteristics of the GM-CSF receptor complex: contribution of individual residues of GMR␣ and of ␤ c Where indicated, the curves passing through the data were better described with models for either one or two binding sites. The binding free energy of the ␣ ⅐ ␤ complex is 13 kcal/mol, whereas that of ␣ alone is in the range of 9.6 -10.2 kcal/mol.
a Kinetic analyses were performed as described in the legend to Fig. 7. b Binding constants were determined at equilibrium through analysis of saturation curves. c ND, not detected.
mainly due to an increase in the dissociation rate constant. The extrapolated cumulative effect of the two mutations on the dissociation rate constant was 50-fold, which recapitulated the difference between the ␣⅐␤ complex and ␣ alone. Consistent with these kinetic properties, our three-dimensional model also positions ␤-Tyr 365 in the vicinity of the ligand, whereas ␤-Tyr 421 was remote from the ligand, either in our original model (24) or after molecular dynamics (Fig. 4). Together, our observations indicate that the aromatic rings at positions 365 and 421 of ␤ c are critical to the stability of the liganded complex, consistent with the role assigned to ␤ c , and that Tyr 365 also fulfills a ligand association function for ␤ c . Despite the locating of Tyr 365 in the vicinity of helix A, the closest residue on the ligand was Arg 24 , which was still at a distance of 9 Å. In contrast, we found on closer examination of this interface that the neighboring residue Arg 23 was located within 5 Å of ␤-Asp 369 , suggesting the possibility of electrostatic interaction at this position. We therefore determined the importance of the charge at position 23 of the ligand, through analysis of the capacity of mutant GM-CSF to compete for the binding of wild type radioiodinated GM-CSF. Binding characteristics were determined at equilibrium. As illustrated in Fig.  9, the dissociation constant for wild type GM-CSF is 58 pM, which is comparable with values estimated through saturation analysis (Table I) and comparable with published results (12,13,17,37,38). Removal of the charge (R23N) results in a 2-fold shift in the dissociation constant. In contrast, reversal of the charge at position 23 (R23D) induces a 5-fold increase in K d (Fig. 7), indicating that Arg 23 has an apparent contribution of 0.9 kcal/mol. Furthermore, mutations of Asp 369 of ␤ c affected the binding equilibrium to an extent that was comparable with that observed with Arg 23 mutations on the ligand (data not shown). Together, our observations indicate the importance of the charge at position 23 of the ligand, which is required for proper association between the ligand and ␤ c . Since Glu 21 is also critical for high affinity binding (39), with an apparent contribution of 2.5 kcal/mol, the observations are consistent with a hot spot of binding energy on helix D (Glu 21 and Arg 23 ) of GM-CSF, which establishes a functional interface with ␤ c FIG. 3. Three-dimensional model of the GM⅐GMR complex after molecular dynamics. The structure of the complex with hydrophobic interaction and short and long range electrostatic components of the recognition site(s) for GM-CSF on the ␣-chain (site 1) or ␤ c (site 2) are illustrated separately at the top and bottom, respectively. The BЈ-CЈ loops are illustrated in green, and the FЈ-GЈ loops are shown in orange. Basic residues are shown in cyan, acidic residues in red, and aromatic rings in yellow.
through several residues, including His 367 , Tyr 365 , and, potentially, Asp 369 . DISCUSSION The proper assembly of a high affinity complex involving GM-CSF and the ␣and ␤-chains of the receptor is essential for signal transduction. In the present study, we define the molecular determinants required for ligand recognition and for stabilization of the complex through a convergence of several approaches, the construction of chimeric receptors, the molec-ular dynamics of our original three-dimensional model of the GM⅐GMR complex, and site-directed mutagenesis. Critical to this tripartite complex is the establishment of four non covalent bonds, three that determine the nature of the ligand recognition process involving residues Arg 280 and Tyr 226 of the ␣-chain and residue Tyr 365 of the ␤-chain. Interestingly, the latter serves a dual function in that it cooperates with another residue of ␤ c , Tyr 421 , to stabilize the complex.
Charge Interactions and -Charge Interactions Determine GM-CSF Binding-The specificity of protein-protein interactions is often conferred by charge interactions or -charge interactions, as in enzyme-substrate or major histocompatibility complex-antigen recognition processes. In the GM⅐GMR complex, we have previously identified two salt bridges that are critical for high affinity binding, one between the ␣-chain and the ligand ␣-Arg 280 -(GM)Asp 112 (24) and the other one between ␤ c and the ligand ␤-His 367 -(GM)Glu 21 (25,39). In the present study, we provide evidence for the importance of two aromatic rings in this process, Tyr 226 of GMR␣ and Tyr 365 of ␤ c . Interestingly, both are located at the BЈ-CЈ loops and contribute to the ligand recognition process, as evidenced by a significant decrease in the on rate when these residues were mutated to hydrophilic residues. Molecular dynamic simulations and functional binding studies concur to indicate that ␣-Tyr 226 serves a critical ligand binding function in the absence of ␤ c . Hence, our results suggest that the low affinity complex is established through the interactions of Arg 280 on the first ␣-chain with site 1 on helix D of the ligand and of Tyr 226 on the second ␣-chain with site 2 on helix A of the ligand. In contrast, in the high affinity complex formed by the ␣⅐␤ heterodimer, Our model also predicts that the contact points between the ligand and ␤ c may be multiple and involve at least three residues at the BЈ-CЈ loop of ␤ c , Tyr 365 , His 367 , and Asp 369 , and three residues on helix A of the ligand, Leu 28 or Arg 24 , Glu 21 , and Arg 23 , respectively.
In the present study, we have identified the role of aromatic residues within the BЈ-CЈ loop of both chains of the GM-CSF receptor that are important for ligand binding. Interestingly, such interactions for the residues reported here and elsewhere contribute to more than 60% of the free energy of the whole complex. It is possible that the remaining free energy may be contributed by hydrophobic interactions between residues that are buried at the ligand-receptor interface. Two aromatic residues that are important for interleukin-6 binding are also found at the BЈ-CЈ loop of interleukin-6 receptor ␣, Phe 230 , and Tyr 231 (40). In addition, another aromatic residue, Trp 169 , also located at the BЈ-CЈ loop of GHR is essential to the growth hormone receptor complex (32). Thus, the conservation of one or more aromatic residues within the BЈ-CЈ loop suggests a nonredundant function forinteractions or -charge interactions at this position in ligand binding. Not surprisingly, in the EpoR that lacks an aromatic residue at the BЈ-CЈ loop, this function is fulfilled by another residue at the E-F loop that is important for ligand binding, Phe 93 (28). Interestingly, the E-F loop of the growth hormone receptor also presents Trp 104 at the hormone-receptor interface, which together with Trp 169 contributes to 60% of the free energy of the complex (32). Since both EpoR and GHR are homodimers, one may speculate that they constitute a subfamily in which the ligand binding function is established at the E-F loop. Although the latter appears to be dispensable for the GMR complex, both chains of GMR have retained two functional interfaces, one at the FЈ-GЈ loop, comparable with EpoR and granulocyte colony-stimulating factor receptor, while the other interface is established at the BЈ-CЈ loop, in the same way as the growth hormone receptor. Hence, within the superfamily of cytokine receptors, each receptor chain, be it part of a homodimeric or a heterodimeric complex, appears to establish two major functional interfaces with the ligand through two of the three loops, E-F of the first fibronectin fold and BЈ-CЈ or FЈ-GЈ of the second fibronectin fold.
Structural Determinants of GMR That Define the Ligand Binding Function-The F-G loop of fibronectin harbors the RGD sequence that defines the integrin binding site (22). Similarly, crucial binding determinants for both GMR and EpoR are also located at the FЈ-GЈ loop. Thus, the conservation of a  Table I. FIG. 7. Recognition of wild type and mutant GMR␣ and ␤ c by monoclonal antibodies. Jurkat cells were transiently transfected with wild type or mutant GMR␣ and ␤ c . One day after electroporation, transfected cells were labeled with the monoclonal antibody against human GMR␣ or the monoclonal antibody against human ␤ c and with a phycoerythrin-conjugated goat anti-mouse Ig. Control cells were labeled with the second antibody only. Dead cells were eliminated through staining with propidium iodide. Similar staining with the monoclonal anti-GMR␣ was observed in Jurkat cells expressing GMR␣ in the absence of ␤ c (data not shown). functional interface at the FЈ-GЈ loop suggests that structural elements have evolved to maintain this interface. The crystal structure of EpoR revealed an important structural element that holds the FЈ-GЈ loop in a configuration, which is optimal for presentation of the functional residue Phe 205 , i.e. the WS box also found in all members of the superfamily of cytokine receptors (20). In our three-dimensional model, we have previously shown that the two Trp residues of the WS box of GMR␣ are in the proximity of Arg 276 of the FЈ strand, thus establishing a strong -charge interaction that forces a ␤-turn at the location of Arg 280 , required for ligand association (24). On the ␣-chain, another interaction reinforces this structural requirement, the salt bridge between Asp 278 at the FЈ-GЈ loop, just upstream of the WS box, and Lys 191 of the linker domain (Ref. 24 and data not shown). Since the linker domain holds the two fibronectinlike modules together, we anticipate that interactions may be found at this location in other cytokine receptors.
Finally, an important residue was previously found at the FЈ-GЈ loop of ␤ c , Tyr 421 (26), which was thought to be involved in ligand binding. In contrast, our results indicate that Tyr 421 does not serve a ligand binding function. First, kinetic studies performed here indicate that Tyr 421 is primarily involved in stabilizing the complex. Second, in our three-dimensional model of the complex, Tyr 421 is not oriented toward the ligand. Even when the residue is flipped in trans, we did not detect any residue on the ligand that might be within a reasonable distance of Tyr 421 . Consistent with these observations, solving the structure of domain 4 of ␤ c complexed with an antagonistic antibody 2 shows that Tyr 421 indeed points to a different face than the cytokine binding site located on the B-C loop. Since we have previously shown that a functional GMR complex is at least a hexamer of two GMR␣, two ␤ c , and two molecules of GM-CSF (35), it is tempting to speculate that Tyr 421 is important in establishing a higher order of receptor assembly required for signal transduction. Future studies will be required to resolve this issue.
Symmetry within an Asymmetrical Structure-The ␣-chain of the GMR complex comprises 378 amino acid residues, most of these constituting the extracellular domain (9), while ␤ c has 881 amino acid residues, half of which are located in the intracellular domain (11). Despite its short cytoplasmic tail, GMR␣ cooperates with ␤ c in signal transduction (35,41,42), and residues essential for cellular activation have been located within its 40-amino acid tail. Thus, despite a variability in length, both chains of GMR appear to fulfill specific functions in signal transduction. In addition, the present study identifies a symmetrical disposition of the functional loops involved in GMR complex assembly. Evidence presented here and elsewhere indicates that both BЈ-CЈ and FЈ-GЈ loops of GMR␣ and of ␤ c are critical for the assembly of the GM-CSF receptor and its ligand into a high affinity complex. In our original model, the FЈ-GЈ loop of GMR␣ is the mirror image of the BЈ-CЈ loop of ␤ c , both determining the ligand association process, whereas the BЈ-CЈ loop of GMR␣ appears to be the counterpart of the FЈ-GЈ loop of ␤ c , both more remote from the ligand. After molecular dynamics, the BЈ-CЈ loop of GMR␣ moves closer to helix D of the ligand to establish the second bridge with GM-CSF. As a result, GMR␣ has two loops in the vicinity of the ligand, FЈ-GЈ and BЈ-CЈ, both contributing to the ligand recognition process, whereas ␤ c is predicted to have a single loop (albeit with several contact points), consistent with a ligand binding function which is secondary to that of GMR␣. Hence, after molecular dynamics, the two functional loops of GMR␣ are disposed as mirror images of the corresponding loops of ␤ c , establishing a symmetrical structure within the extracellular domains of the GMR complex. In summary, residues located at the turn of the BЈ-CЈ and FЈ-GЈ loops of both GMR␣ and ␤ c specify the nature of the functional interface of the GM⅐GMR complex.