Identification of ligand-binding site III on the immunoglobulin-like domain of the granulocyte colony-stimulating factor receptor.

The granulocyte colony-stimulating factor receptor (G-CSF-R) forms a tetrameric complex with G-CSF containing two ligand and two receptor molecules. The N-terminal Ig-like domain of the G-CSF-R is required for receptor dimerization, but it is not known whether it binds G-CSF or interacts elsewhere in the complex. Alanine scanning mutagenesis was used to show that residues in the Ig-like domain of the G-CSF-R (Phe(75), Gln(87), and Gln(91)) interact with G-CSF. This binding site for G-CSF overlapped with the binding site of a neutralizing anti-G-CSF-R antibody. A model of the Ig-like domain showed that the binding site is very similar to the viral interleukin-6 binding site (site III) on the Ig-like domain of gp130, a related receptor. To further characterize the G-CSF-R complex, exposed and inaccessible regions of monomeric and dimeric ligand-receptor complexes were mapped with monoclonal antibodies. The results showed that the E helix of G-CSF was inaccessible in the dimeric but exposed in the monomeric complex, suggesting that this region binds to the Ig-like domain of the G-CSF-R. In addition, the N terminus of G-CSF was exposed to antibody binding in both complexes. These data establish that the dimerization interface of the complete receptor complex is different from that in the x-ray structure of a partial complex. A model of the tetrameric G-CSF.G-CSF-R complex was prepared, based on the viral interleukin-6.gp130 complex, which explains these and previously published data.

The granulocyte colony-stimulating factor receptor (G-CSF-R) 1 is a transmembrane protein that is expressed predominantly on cells of the neutrophil lineage and is important in transmitting signals for their proliferation, differentiation, and function (1,2). The extracellular region comprises six structural domains: an N-terminal Ig-like domain followed by five fibronectin type III (FNIII) domains (3) (Fig. 1A). The first two of the FNIII domains (called D2 and D3) form the cytokine receptor homology (CRH) module, which contains four cysteine residues in D2 and a WSXWS motif in D3. These features of the CRH module are conserved in members of the class 1 cytokine receptor family (4). This domain structure is also found in the closest homologue of the G-CSF-R, gp130, which is the shared signal transducing receptor chain of the interleukin (IL)-6 family of cytokines (5). The G-CSF-R and gp130 share 46% sequence similarity in the extracellular region (6). The ligands for the class 1 cytokine receptors have a conserved 4-␣-helical bundle structure but little sequence similarity (7).
In the gp130 family of ligands and receptors, three binding interfaces have been identified (reviewed in Ref. 8). For example, IL-6 interacts with the IL-6 receptor at site I and gp130 at site II. A second gp130 receptor interacts at site III, resulting in a hexameric complex of two components each of IL-6, IL-6 receptor, and gp130 (9). On the IL-6 family of ligands, site I is located at the end of the D helix, site II on the A and C helices, and site III at the N-terminal end of the D helix. Binding of G-CSF to the G-CSF-R results in the formation of a 2:2 complex, but the details of interactions in this complex are yet to be determined (10,11). We have proposed that the G-CSF-R complex is similar to the IL-6 receptor complex but without the site I interaction, because there is no equivalent protein to the IL-6 receptor in the G-CSF-R complex (12) (Fig. 1B). The main receptor-binding site on G-CSF has been shown by mutagenesis to include residues on the A and C helices (13,14) and is equivalent to site II of IL-6 ( Fig. 1D). This site interacts with the "elbow" formed by D2 and D3 of the CRH module (G-CSF-R site II), shown in the crystal structure of G-CSF in complex with the CRH module (15). Mutagenesis studies also identified site II residues on G-CSF-R (16) and showed that the interaction of Glu 19 of G-CSF with Arg 288 of the receptor was particularly important in the site II interface (12).
There is some evidence for an additional binding site in the G-CSF-R complex that is equivalent to site III in gp130 complexes. On G-CSF, mutagenesis has identified some residues that appear to form a second binding site near the N-terminal end of the D helix (13), and we previously showed that Glu 46 of G-CSF in the E helix (Fig. 1D) probably interacted with the Ig-like domain (12). The Ig-like domain of G-CSF-R has been shown by several studies to be important for complex formation. Deletion of most of the Ig-like domain from the G-CSF-R resulted in a 20-fold loss of G-CSF binding affinity and a very low proliferation response in cells expressing the deletion mutant (5). Similarly, substitution of the Ig-like domain with that of gp130 (forming the (gp130-Ig)G-CSF-R chimera; Fig. 1A) resulted in reduced binding affinity and no detectable activity (12). The Ig-like domain is also required for receptor homodimerization with G-CSF in solution (17), and recently, a soluble Ig-like domain was shown to bind G-CSF (18). Taken together, these studies strongly suggest that the Ig-like domain interacts with G-CSF at site III, resulting in receptor dimerization as shown schematically in Fig. 1B. The recently described crystal structure of a 2:2 complex of the CRH module of G-CSF-R (with no Ig-like domain) and G-CSF (shown schematically in Fig. 1C) shows the site II interaction but provides no evidence for a site III interaction (15). The putative site III of G-CSF (Fig. 1D) is exposed in the crystal complex, suggesting that the observed dimerization interface is different from that formed by the complete receptor and may represent an intermediate in complex formation (15).
The purpose of this study was to identify the residues involved in the proposed site III on the Ig-like domain of the G-CSF-R. Site-directed mutagenesis was used to identify residues on the Ig-like domain that interact with G-CSF or are otherwise important for G-CSF-R complex formation. Secondly, we used anti-G-CSF monoclonal antibodies (mAbs) and anti-G-CSF-R mAbs to determine the exposed and buried regions of the complex. Both of these approaches support the proposed site III interaction between G-CSF and the Ig-like domain of the receptor.
After this work was completed, the crystal structure of domains 1-3 of gp130 in complex with viral IL-6 was published (19). This complex is a tetramer of two viral IL-6 and two gp130 subunits. The tetramer is formed by site II binding and the interaction of site III of viral IL-6 with the Ig-like domain of gp130, as we have proposed for the complex of G-CSF with its receptor. These two complexes are likely to be very similar because of the related receptor structures and the ability of viral IL-6 to activate gp130 and G-CSF to activate the G-CSF-R without the requirement for an ␣-receptor (20). We have used the structure of the viral IL-6⅐gp130 complex to make a model of the G-CSF⅐G-CSF-R complex that has allowed us to more clearly interpret our results.

EXPERIMENTAL PROCEDURES
Mutagenesis of the G-CSF-R-Site-directed mutagenesis of the G-CSF-R cDNA in pBluescript (Stratagene, La Jolla, CA) was performed using either the method of Kunkel (21) or a mutagenesis kit (Quickchange site-directed mutagenesis kit; Stratagene). Mutations were con-firmed by nucleotide sequencing. Wild type (WT) and mutant G-CSF-R cDNAs were subcloned into pEFBOS expression vector (22), and the WT G-CSF-R cDNA was also subcloned into pCDNA3 (Invitrogen, Groningen, The Netherlands).
Expression of G-CSF-R Mutants-The WT and mutant G-CSF-R constructs were transfected into Ba/F3 cells by electroporation as described previously (16). Usually six independent cell lines were selected for each mutant based on apparent monoclonality and, where possible, similar levels of receptor expression to WT as determined by flow cytometry screening.
Flow Cytometry-To assess receptor expression in transfected cell lines, the cells were stained with a mAb to the human G-CSF-R, LMM741, as described previously (16). To determine whether the mutant receptors were correctly folded, binding of five conformation-dependent mAbs was measured in a semi-quantitative assay, as described previously (16). Binding of each mAb (median fluorescence channel (MFC) Ϫ background) was compared with that of LMM741 (non-conformation-dependent) and with the binding of the same pair to the WT receptor by calculating the ratio (MFC mAb/MFC LMM741) mutant G-CSF-R ϫ (MFC LMM741/MFC mAb)WT G-CSF-R. The MFC of LMM741 binding to WT G-CSF-R was 49 Ϯ 12 (mean Ϯ S.D. of 12 assays). To determine data variability, the means and standard deviations of the MFC ratios for two other conformation-independent mAbs were calculated. For LMM852, this was 0.995 Ϯ 0.072, and for LMM847, this was 0.995 Ϯ 0.091 (n ϭ 48), compared with the expected value of 1.0. On the basis of these data, a ratio of Ͻ0.7 was considered to be a significant reduction in binding.
Proliferation Assay-Proliferation of transfected Ba/F3 cells in response to G-CSF was determined in a 48-h tritiated thymidine incorporation assay, as described previously (16). The assays were performed twice on at least three independent cell lines expressing each mutant G-CSF-R. The mean EC 50 of the two assays was calculated for each cell line, and then the means Ϯ S.E. of the EC 50 values for the group of cell lines were calculated; thus errors represent variation between cell lines. This was greater than variation between assays.
G-CSF Binding Assay-Tyr 1,3-G-CSF (gift from A. Shimosaka, Kirin, Japan) was iodinated using IODO-GEN (Pierce) either as described (23) or with the modification described by Chizzonite et al. (24). The modification resulted in a slightly higher K d for WT G-CSF-R binding of G-CSF; therefore the K d values for mutant receptors were compared with WT G-CSF-R assayed with G-CSF iodinated by the same method. Binding assays were performed and analyzed as described previously (16). In all cases, the Scatchard transformation of data gave an approximately linear plot with a best fit of single affinity binding.
G-CSF Enzyme-linked Immunosorbent Assay-Binding of 37 anti-G-CSF mAbs to G-CSF mutants was tested in an enzyme-linked immunosorbent assay as described previously (25). All the mAbs were tested at 10 and 1 g/ml using plates coated with WT or mutant G-CSF at 10 ng/well. Those mAbs that showed reduced binding to the mutant G-CSFs were then titrated to determine quantitatively the reduction in Characterization of G-CSF Ligand-Receptor Complexes-Ba/F3 cells expressing WT G-CSF-R or (gp130-Ig)G-CSF-R (12) were incubated for 1 h at room temperature with 100 or 1000 g/ml G-CSF (respectively) or no G-CSF and 10 mM NaN 3 to prevent receptor endocytosis and then washed. The cells at 4°C were then incubated with mAbs at 10 g/ml, washed, incubated with fluorescein isothiocyanate-conjugated antimouse Ig (Silenus, Amrad, Melbourne, Australia; 1:100), and then analyzed by flow cytometry.
Homology Modeling-A model of the Ig-like domain of human G-CSF-R has been constructed using comparative modeling techniques with the program Modeller (26). The Ig-like domain of gp130 (Protein Data Bank entry 1I1R) (19) has been used as the structural template, with the sequence alignment shown in Fig. 2. Secondary structure elements of gp130 are determined from the x-ray coordinates using STRIDE (27). The arrangement of these ␤ strands gives this domain an H-type Ig-like fold (28). The structure-based sequence alignment has been derived manually, by utilizing secondary structure elements and conserved motifs and residues. The four cysteine residues are modeled to form the same disulfide linkages as in gp130. Models were created with various sequence alignments in the CЈ and D strands, but the best were created with the alignment shown in Fig. 2. For the final alignment 100 models were calculated from which the best was chosen in accordance with the Modeller energy function, ProsaII profiles (29), and Procheck assessment (30). The ProsaII trace indicated no misfolding with a z score of Ϫ5.74, comparing favorably with that of gp130 (Ϫ6.23).
To create a quaternary 2:2 G-CSF⅐G-CSF-R complex, a model of the first three domains of G-CSF-R was made. Human CRH module (D2 and D3) of G-CSF-R in complex with G-CSF via site II was modeled on the mouse G-CSF-R bound to human G-CSF (Protein Data Bank entry 1CD9) (15) using Modeller. The sequence alignment of the human and mouse receptors (data not shown) reveals 65% identity and no insertions or deletions in the CRH module. The A-B loop was missing in the x-ray structure, so for this loop the NMR structure of D3 of the human G-CSF-R (Protein Data Bank entry 1CTO (31)) was used as the template. The best structure was chosen in accordance with the criteria mentioned above. The G-CSF-R Ig-like domain was linked to the CRH module using the same interdomain angle observed in the gp130 structure. The quaternary G-CSF-R complex was created by superimposing the two receptors onto the two gp130 molecules (using domains 1 and 2). This superimposition also revealed very similar site II and III binding modes of G-CSF to G-CSF-R and viral IL-6 to gp130. To avoid steric clashes at site III, the distance between the G-CSF receptors was marginally increased.

Mutation of the Predicted Loop Regions of the Ig-like Domain-
The available data about sites of interaction of cytokine receptors with their specific ligands indicated that the loop regions of FNIII domains rather than ␤ strands were the most likely ligand-binding sites (32)(33)(34)(35)(36). An alignment of Ig-like domains was used to predict the ␤ strands and loop regions of the G-CSF-R Ig-like domain. Residues in the predicted loops of the Ig-like domain that have charged, aromatic, or large hydrophobic side chains were chosen for alanine substitution by site-directed mutagenesis. The final alignment (Fig. 2) that was used for modeling (see later) differed slightly in the assignment of some ␤ strand and loop regions. This final alignment has been used to indicate the positions of the mutations in the data (Table I and Fig. 3). The mutated full-length receptor cDNAs were expressed in the murine pro-B cell line, Ba/F3, to enable measurement of the proliferative response to G-CSF and binding affinity for G-CSF, in comparison with the WT G-CSF-R. A panel of five conformation-dependent mAbs was used to determine whether or not the mutant receptors were correctly folded. Binding was measured by flow cytometry in a semi-quantitative assay and compared with three conformation-independent mAbs (see "Experimental Procedures"). Loss of binding of conformation-dependent mAbs could indicate complete unfolding of the domain or a more subtle alteration of ␤ strands or loop regions. The data (Table I) were normalized with respect to level of receptor expression and level of binding of each mAb to WT G-CSF-R. The epitopes recognized by the conformationdependent mAbs are on the Ig-like domain (groups 1 and 3) and D2 of the CRH module (group 2), whereas the conformationindependent mAb shown (group 4) recognizes the additional FNIII domains (23). A change in conformation of the receptor would be likely to affect binding of several mAb groups, but a specific mutation of the epitope recognized by a mAb would affect only one mAb or group. An increase in binding was not seen with any of the mutants.
Two of the multiple mutants, in the AЈ strand and E-F loop, showed total or almost total loss of all conformation-dependent mAb binding, which correlated with low expression, shown by the low median fluorescence channel of the positive control mAb, LMM741 (Table I). Thus these mutants were probably not properly folded. The mutant in the ␣ helix of the B-C loop showed loss of only group 3 mAb binding, indicating that this region includes the binding site of group 3 mAbs. The CЈ strand mutant showed borderline loss of binding with all mAbs and low level expression, suggesting some alteration in conformation. The D-E loop mutant showed no loss of binding. The G strand mutant showed partial loss of binding of all mAbs but good expression, suggesting that alteration of conformation was not severe.
The function of the mutant receptors was assessed by determining proliferation in response to G-CSF and binding of G-CSF (Fig. 3). Because there was variation in the level of mutant G-CSF-R expression, we tested the response of cells expressing less than 1000 WT G-CSF-R (Fig. 3, WT-low). These cell lines were, if anything, more sensitive to G-CSF than those with higher receptor numbers; therefore there is no evidence that low receptor number alone (down to 700 receptors/cell) can account for reduced activity. The most interesting mutation was of the G strand. Although there was some loss of mAb binding to this mutant, there was a 23-fold reduction in proliferation and 2.3-fold reduction in binding affinity. With the other mutants, the loss of biological activity correlated with loss of mAb binding and therefore probably reflected indirect effects on receptor function. The E-F loop mutant gave an anomalous result with complete loss of mAb binding and no proliferation activity but unexpectedly only slightly reduced G-CSF binding affinity. To further investigate this result, one cell line (300 receptors/cell) was tested for proliferation at higher G-CSF concentration. It did respond, but with an EC 50 of only 6400 pM (a 320-fold reduction in activity). This is the only mutant tested that has lost ability to bind conformationdependent mAbs but could still bind G-CSF and give a (low) proliferation response.
The AЈ strand, G strand and E-F loop multiple mutation sites were selected for introduction of single residue alanine substitutions. The point mutations did not affect mAb binding except for three mutants shown in Table I (L15A, Q73A, and L86A). The small reduction of only group 1 mAb binding to the Q73A mutant indicates that Gln 73 contributes to the LMM711 binding site. Proliferation and binding data for all the mutants are given in Fig. 3. The mutant receptors were all expressed at over 2000 receptors/cell (Fig. 3). The only mutant showing a reduction in proliferative response with no loss of mAb binding was Q87A, showing that Gln 87 is important for receptor function and may interact with G-CSF. The effects of the mutations on binding affinity were not significant (Fig. 3).
Mutation of Residues in the F and G ␤ Strands and Other Charged Residues-Because mutations of single residues had only small effects, it was possible that there was a relatively large interaction surface between the Ig-like domain and G-CSF, with each contact residue contributing a relatively small amount of binding energy to the complex. We chose to mutate predicted solvent-accessible residues in the F and G strands (because of the role of Gln 87 ) and other exposed charged residues. All of these mutants (listed in Fig. 3) except H5A were well expressed. Interestingly, three of the mutants (F75A, E93A, and R95A) in the F and G strands showed total loss of binding of LMM711, a strongly neutralizing mAb, thus identifying the mAb binding site (Table I). This loss of binding was seen even with 100 g/ml LMM711 (data not shown). Two other mutants, H5A and D57A, also showed some general loss of mAb binding. The selective reduction of LMM793 binding to the L89A mutant suggests that Leu 89 affects binding of this mAb. Mutations F75A, L89A, and Q91A in the F and G strands and H5A in the A strand of G-CSF-R gave greater than 4-fold reduction in EC 50 for proliferation; therefore these residues as well as Gln 87 are important for receptor function (Fig. 3). Although the H5A mutant was not well expressed (550 receptors/ cell), it appeared to be folded correctly, and so it was not excluded. Some additional mutants gave smaller but statistically significant reductions in proliferation response to G-CSF (Leu 15 , Lys 26 , Arg 41 , Trp 82 , and Glu 93 ). Of these residues, Leu 15 appears to be structurally important because of loss of mAb binding (Table I), but the other residues could possibly contribute to G-CSF-R function. There were no detectable effects on binding affinity except for mutant C79A, which, although it had a small loss of binding affinity, had no effect on proliferation (Fig. 3).
The Effect of Combinations of Mutations-Most of the mutations in the F and G ␤ strands reduced the proliferation response, but the effects were quite small. To provide further evidence about the importance of these residues for receptor function, we tested the effect of combining mutations. Firstly, we combined the Q87A mutation with each of the other residues that had significantly altered receptor function. Secondly, we tested the F and G strand single residue mutants with a G-CSF mutant that had strongly reduced binding through ligand site II, (E19A)G-CSF (see Fig. 1D for location of Glu 19 ). Although (E19A)G-CSF bound almost normally to the WT G-CSF-R, it did not bind to a chimeric receptor, (gp130-Ig)G-CSF-R, in which the Ig-like domain was replaced with that of gp130 (Fig. 1A) (12). We expected that the effect of mutations in the Ig-like domain (receptor site III) might be greater in the absence of a strong site II interaction.
The double F and G strand mutants had similar mAb binding to that expected from the single mutants, with the exception of the Q87A,E93A mutant, which had a slightly greater loss of mAb binding than expected (Table I). The double mutations had a greater effect on proliferation (10 -30-fold) than the single mutations (4 -8-fold), but the EC 50 ratios for proliferation were less than would be predicted by multiplying the EC 50 ratios for the individual mutants, with the exception of mutant Q87A,E93A (Table II). However, the increased effect observed

Function of the Ig-like Domain of the G-CSF Receptor
with this mutant may have been partly caused by an increased conformational change that was suggested by the mAb binding data. The double mutants all showed significantly reduced binding affinity for G-CSF that was similar to or greater than that expected from the single mutants (Table II). When the response of the single residue F and G strand receptor mutants was tested with (E19A)G-CSF in comparison with WT G-CSF, the results fell into two groups (Table III). In the first group (F75A, C79A, Q91A, and R95A mutants), the EC 50 ratios of mutant/WT receptor were similar with the two G-CSFs, implying an independent effect of the receptor and G-CSF mutations. In the second group (W82A, Q87A, L89A, and E93A mutants), the receptor mutation had a much greater effect (5-10-fold) on the response to (E19A)G-CSF than on the response to WT G-CSF. This apparently synergistic effect was the most pronounced with the Q87A mutant that had the greatest effect as a single mutation. The second group of mutants also gave some indication of synergy in their binding affinities (Table III), particularly Q87A, which was reduced 9-fold in its binding affinity for (E19A)G-CSF. These two sets of data from double mutants provide stronger evidence for the importance of residues in the F and G strands in a binding interface on the Ig-like domain.
Model of the Ig-like Domain Binding Sites-We have created a model of the Ig-like domain of G-CSF-R based on the template structure of gp130. A sequence alignment of all the known G-CSF-R Ig-like domain sequences, including cow and pig expressed sequence tags, and gp130 sequences has been created and is shown in Fig. 2. The alignment highlights the conservation of four cysteine residues along with predominantly buried hydrophobic and structurally important residues. Secondary structure elements of the homologous gp130 Ig-like domain are identified. The disulfide bond linkages used for G-CSF-R are Cys 23 -Cys 78 and Cys 3 -Cys 29 and follow those that are found in gp130 (19,37). The backbone of the G-CSF-R Ig-like domain model closely follows that of gp130, apart from a deletion after the ␣1 helix and an insertion in the D-E loop region. Modeling details are described under "Experimental Procedures." The positions of residues that had a greater than 4-fold effect on activity when mutated (Phe 75 , Gln 87 , Leu 89 , Gln 91 , and His 5 ) are shown on the model in Fig. 4A. These residues form a cluster that is likely to be a binding interface, with the exception of Leu 89 , which is largely buried. The epitopes recognized by the mAbs are shown in Fig. 4B. The binding site of LMM711 (a strongly neutralizing mAb) partially overlaps the binding interface depicted in Fig. 4A. The ␣1 helix in the B-C loop that is recognized by the group 3 mAbs does not overlap the binding interface, consistent with the partial, indirect effect of these mAbs on receptor function. A molecular surface representation of the Ig-like domain model showing the binding interface has  a At least three independent cell lines were tested for proliferation response to G-CSF. EC 50 is the concentration of G-CSF giving 50% of maximum response. For WT G-CSF-R, EC 50 ϭ 29 Ϯ 6 pM (mean Ϯ S.E.). Mutants were all significantly different from WT (p Ͻ 0.01), determined by Student's t test.
b G-CSF binding data analyzed by Scatchard transformation to determine K d and the number of receptors per cell (mean Ϯ range of two assays). Mutants were all expressed at Ͼ3,000 receptors/cell. G-CSF binding of all mutants was significantly different from WT (p Ͻ 0.01). WT K d ϭ 65 Ϯ 5 pM (12 assays). a A typical cell line was used for each mutant and WT receptor. The data are the means of two assays. For WT G-CSF-R, EC 50 for proliferation with WT G-CSF was 15 pM, and that with (E19A)G-CSF was 170 pM.
b One binding assay was performed. The K d for WT G-CSF-R binding WT G-CSF was 62 pM, and binding (E19A)G-CSF was 77 pM. been compared with the equivalent view of the gp130 Ig-like domain in Fig. 4C. The residues that are the most important for G-CSF-R function (magenta) are remarkably similar in position to the viral IL-6 binding surface of the gp130 Ig-like domain. Two G-CSF-R residues that are also likely to be central for ligand binding (Ile 88 and Asp 90 ) are shown in green. These were not tested because they were predicted to be buried from our original alignment.
Mapping of the Surface of the G-CSF Ligand-Receptor Complex with mAbs-To further analyze the receptor complex, we mapped the exposed and inaccessible regions of the complex by saturating the receptor with G-CSF and then measuring mAb binding to the complex by flow cytometry. Because the binding of G-CSF to its receptor is almost irreversible, there is little displacement of G-CSF by the mAbs in a short term incubation, and binding of a mAb indicates that its epitope is exposed. Both anti-receptor and anti-G-CSF mAbs were used (Table IV). The characteristics and partial epitope mapping of the anti-G-CSF mAbs have been described previously (25); representative mAbs from each epitope group were tested. Neutralizing anti-G-CSF mAbs belong to epitope groups 1 and 2, plus the unique LMM207. Additional information about the epitopes of the neutralizing mAbs was obtained by measuring their binding to G-CSF mutants in an enzyme-linked immunosorbent assay (data summarized in Table IV). The epitopes described may not be complete because every possible exposed residue in the G-CSF-R and in G-CSF has not been tested. Binding of LMM741, which binds to the FNIII domains of the receptor, was used as a positive control for level of receptor expression. In addition to cells expressing WT G-CSF-R, we tested cells expressing the chimeric receptor, (gp130-Ig)G-CSF-R (Fig. 1A). This receptor appears unable to dimerize in the presence of G-CSF, thus forming a 1:1 complex, (12) and may therefore reveal epitopes that are buried only in the dimerization interface of the WT G-CSF-R complex.
Most mAbs, both anti-receptor and anti-ligand, were unable to bind to the WT G-CSF-R complex, indicating that their binding sites were inaccessible (Table IV). Anti-G-CSF-R mAbs LMM775 and LMM793 bound, but at a reduced level, indicating that the helix in the B-C loop, ␣1, of the receptor Ig-like domain (Ser 30 -Asp 33 ; Fig. 4B) was at least partially exposed. The other anti-G-CSF-R mAbs did not bind, as expected from their ability to block G-CSF binding in a competition assay (23). The only other mAb able to bind was the anti-G-CSF mAb LMM308 showing that the N-terminal peptide of G-CSF was exposed in the complex.
In cells expressing the (gp130-Ig)G-CSF-R chimera, two mAb epitopes were exposed that were inaccessible in the WT G-CSF-R complex, whereas LMM308 binding was not changed. Binding of the anti-G-CSF mAb LMM351 was increased to 46% of the control level. This mAb recognizes peptide 34 -46 of G-CSF (end of A helix and E helix) and is the only mAb that binds to ligand site III and not site II, therefore suggesting that ligand site III is important for interaction with the Ig-like domain. Our previous data (12) and the data of Young et al. (14) suggested that Glu 46 of G-CSF was important for interaction with the receptor, probably with the Ig-like domain (12). Four additional anti-G-CSF mAbs recognized this residue strongly, i.e. binding to (E46A)G-CSF was reduced by more than 100fold. However, none of these mAbs was able to bind to the chimera complex (LMM205 and LMM350 and data not shown), probably because other residues required for binding (e.g. in ligand site II) were inaccessible. The crystal structure of the complex shows that ligand site III is not in contact with the CRH module (15), consistent with the results indicating that it interacts with the Ig-like domain. The neutralizing mAbs binding to D2 of the receptor (group 2) were able to bind to the complex at about 70% of the binding in the absence of G-CSF. This result was not expected because previous data indicated that the epitope was close to receptor site II (23). The binding to the chimeric G-CSF-R complex suggests that these mAbs recognize an epitope that is blocked partly by receptor dimerization and partly by G-CSF binding.
Model of the G-CSF-R Complex with G-CSF-A model of the The key residues in G-CSF signaling are highlighted in space-filling representation. Key exposed residues are magenta, whereas the buried Leu 89 is pink. B, antibody binding sites of G-CSF-R Ig-like domain. Residues recognized by neutralizing mAb LMM711 are yellow. Residues recognized by LMM775 and LMM793 antibodies (Ser 30 -Asp 33 ), which are able to bind to the tetrameric complex, are shown in gray. C, molecular surface representations of G-CSF-R (left) and gp130 (right) Ig-like domains depicted in the same orientation as in A and B. G-CSF-R residues with greater than 4-fold effect on signaling are magenta, whereas those with a 2-4 fold effect are mauve. Residues with an effect of less than 2-fold are in yellow, whereas those not mutated are in gold. Residues in green (Ile 88 and Asp 90 ) were not mutated but seem likely to contact G-CSF via the site III interaction. The gp130 residues in contact with viral IL-6 via the site III interaction (19) are magenta. The coordinates of the model are available from the authors on request.
human G-CSF-R (Ig-like domain plus CRH module) tetrameric complex with G-CSF was prepared to enable the incorporation of all the available structural and biological data. The 2:2 complex was built using the crystal structures of the murine G-CSF-R (15) and gp130 (19) complexes as templates and has the same 2-fold symmetry as seen in the gp130 complex (Fig. 5). G-CSF interacts with one receptor via site II and a second receptor via site III, but there are no G-CSF-G-CSF or receptorreceptor interactions. Previous experiments suggest that the G-CSF-R subunits will make contact through domains 4 -6 (38,39). Residues that are proposed to form the receptor site III interface in this study are shown on the model. In the receptor Ig-like domain, Phe 75 , Gln 87 , and Gln 91 are likely to contact G-CSF. His 5 on the other hand appears to be at the periphery of the contact area of receptor site III with G-CSF, and it is unclear whether or not it is directly involved in G-CSF binding. Reidhaar-Olson et al. (13) have previously suggested that G-CSF residues Lys 40 , Val 48 , Leu 49 , and Phe 144 form ligand site III. In the model, the G-CSF residues Glu 46 , Leu 49 , and Phe 144 and are contact residues with the Ig-like domain, whereas Lys 40 and Val 48 are peripheral and less clearly involved. The critical site II interaction of Glu 19 of G-CSF with Arg 288 of the receptor is also shown. The human G-CSF-human G-CSF-R site II interface is very similar to that determined experimentally for the human G-CSF-mouse G-CSF-R complex (15). DISCUSSION Alanine scanning mutagenesis of the Ig-like domain of the G-CSF-R has shown that residues Phe 75 , Gln 87 , and Gln 91 in the predicted F and G ␤ strands form a binding site for G-CSF. The binding site of mAb LMM711 was also on the F and G strands of the Ig-like domain, overlapping the G-CSF binding site, consistent with the strongly neutralizing effect of this mAb (23). In contrast, anti-receptor mAbs LMM775 and LMM793 that were only partially neutralizing bound to a region outside of the G-CSF binding site. Further information about the ligand-receptor complex was obtained from determining mAb binding to ligand-receptor complexes. These results showed that the N terminus of G-CSF is exposed in the WT G-CSF-R complex (mAb LMM308). In addition, a G-CSF site III epitope (mAb LMM351) is buried in the WT G-CSF-R complex but exposed in a nondimerizing, chimeric G-CSF-R (1:1) complex.
These data are consistent with a complex that resembles the gp130-viral IL-6 structure rather than the complex described by Aritomi et al. (15) (Fig. 1C). The site II interface between the ligand and CRH module of the receptor is essentially the same in both complexes. It is the dimerization interface that differs. In the gp130 complex with viral IL-6 (19), the site III interface is formed between the Ig-like domain and viral IL-6, thus forming a tetrameric complex analogous to our proposed G-CSF-R complex. Our mutagenesis data have identified a similar site III on the Ig-like domain of the G-CSF-R. In addition, the E helix of G-CSF (forming part of site III) is exposed in the crystal structure in an orientation that could not interact with an Ig-like domain. This structure is incompatible with our data showing that G-CSF site III is buried in the WT complex. In the crystal structure of the G-CSF-R complex (without the Ig-like domain), the N terminus of G-CSF is buried in the dimerization interface, whereas our mAb binding data (LMM308) show that the N terminus of G-CSF is exposed in the WT receptor complex. It seems likely that in the presence of the receptor Ig-like domain, a quite different G-CSF-R dimerization interface forms from that reported by Aritomi et al. (15).
The model of the G-CSF-R Ig-like domain was prepared using the gp130 Ig-like domain as a structural template. The disulfide bond configuration used was that of gp130, because the four Cys residues involved are conserved in all the known sequences of G-CSF-R and gp130 (Fig. 2). Modeling the experimentally determined disulfide linkage reported in Haniu et al. (40) with Cys 3 -Cys 78 and Cys 23 -Cys 29 proved very difficult. In the crystal structure of gp130 and the model of G-CSF-R, the sulfur atoms of these cysteines are ϳ8 and 9.5 Å apart, respectively, far greater than the 2 Å required for a disulfide bond. To accommodate the Cys 3 -Cys 78 disulfide, the A and AЈ strands would need to shift by three residues. This would increase the size of the AЈ-B loop, which, in gp130, forms significant contact with D2. Formation of the Cys 23 -Cys 29 disulfide bond would necessitate a large conformational change at the N-terminal end of the Ig-like domain. Thus the experimentally determined linkages would prohibit domain 1 folding in a traditional Ig-  (16). The G-CSF peptide is the smallest peptide produced by proteolytic digestion or peptide synthesis that was recognized by the mAb (25). b mAbs at saturating concentration (10 g/ml) were tested for binding to Ba/F3 cells expressing WT G-CSF-R or (gp130-Ig)G-CSF-R saturated with G-CSF by flow cytometry. The median fluorescence channel was compared with that for LMM741 (in the absence of G-CSF), an anti-G-CSF-R mAb that recognizes the FNIII domains. like manner. Given the high degree of structural similarity observed among all experimentally determined members of this family, and the conservation of ligand-receptor binding sites, the experimentally observed disulfide assignment of the Ig-like domain seems unlikely.
Most of the single alanine substitutions reduced receptor activity by 2-3-fold. According to the model, these mutations were distributed widely over the Ig-like domain; therefore, it seems unlikely that all of these small effects indicate direct ligand interaction and more likely that these mutations indirectly affected receptor function. We concluded that a reduction in activity of at least 4-fold was required to indicate a residue that contributed directly to ligand binding. By this criterion, residues His 5 , Phe 75 , Gln 87 , Leu 89 , and Gln 91 were found to be important for receptor function, with Gln 87 being the most important. Leu 89 is probably not directly involved because it is largely buried in the model. Similarly, the effect of the Leu 89 mutation on binding of LMM793 is likely to be indirect. Gln 87 , Leu 89 , and Gln 91 are conserved in the four known G-CSF-R sequences, consistent with their importance in structure and/or function. In addition, Gln 87 is conserved in four of five gp130 sequences and is involved in the receptor site III interface with viral IL-6 (Gln 91 in human gp130). The other residues, His 5 and Phe 75 , are conserved in three and two other G-CSF-R sequences, respectively (Fig. 2). These residues may interact with residues that are not conserved among the G-CSFs, or the alternative residues (Tyr 5 , Leu 75 ) may be able to make similar contacts with G-CSF. Several of the smaller effects on proliferation were statistically significant, and the residues responsible (Lys 26 , Arg 41 , Trp 82 , and Glu 93 ) are found on the periphery of the binding site in the model; therefore they may be part of site III. Although the effect of L86A was not significant, it was only slightly less than a 4-fold and is also on the periphery of site III.
None of the single residue mutations (except Cys 79 , which is unlikely to be important because it does not significantly affect proliferation) had a significant effect on binding affinity. Detection of loss of binding affinity is difficult in the presence of intact site II binding. By comparing the binding affinity of monomeric and dimeric receptor complexes, the increase in binding affinity attributable to the site III interaction appears to be about 20-fold (12,41). Given that several residues contribute to receptor site III, the effect of a single mutation was expected to be small; however, the effect on binding affinity of mutating two residues in the F and G strands was significant. Similarly, the effect of several of the single F and G strand mutations on binding of (E19A)G-CSF, which is the central ligand site II mutant (Fig. 1D), was increased.
Two of the multiple residue mutations (in the AЈ strand and the E-F loop) in G-CSF-R unexpectedly resulted in a loss of mAb binding to both the Ig-like domain and D2, showing that alteration in the conformation of the Ig-like domain can affect the neighboring domain. Of these mutated residues, Ile 12 , His 14 , and Leu 15 are in contact with D2, whereas the E-F loop is in close proximity to D2, according to the model. Interestingly, three single residue mutants in N-terminal loops of D2 (L129A in the B-C loop and N184A and L186A in the F-G loop) had a similar effect on both domains (data not shown). In the model, Asn 184 is structurally important, forming stabilizing hydrogen bonds, and both Leu residues contact the Ig-like domain, suggesting that interactions between the loops of the two domains are important for receptor conformational stability and/or relative domain orientation. Similarly, in gp130 the Ig-like domain forms extensive hydrophobic contacts with D2, orienting the Ig-like domain in a fixed position with respect to D2 (19). The E-F loop mutant was also unusual because it was able to bind G-CSF with only slightly reduced affinity, despite loss of mAb binding; however, the complex with G-CSF was probably abnormal because signal transduction as measured in the proliferation assay was greatly reduced. It may be that the loss of binding of anti-D2 mAbs was caused by steric hindrance by the mutant Ig-like domain rather than unfolding of D2.
Because the effect of single residue mutations was quite small, double mutants in the F and G strands were made to confirm the importance of this region. In these mutants, the loss of receptor activity was increased without further destabilizing the domain (with one exception: Q87A,E93A), and there was detectable loss of binding affinity, providing stronger evidence for the role of this site. However, the double mutations did not have as great an effect as predicted from the single mutations, indicating that the residues did not behave independently and are probably part of the same binding interface. In addition, the activity of the F and G strand single mutants was determined with the G-CSF mutant (E19A)G-CSF. We have previously shown that Glu 19 in site II of G-CSF interacts with Arg 288 in the CRH module of the receptor (12). This interaction was confirmed in the crystal structure of the complex (15), which showed that Glu 19 is a central residue in the interface of G-CSF with the CRH module and is independent of Key residues are highlighted in space-filling representation. Critical site II residues of G-CSF-R(Arg 288 ) and G-CSF(Glu 19 ) are red. Site III residues of G-CSF (Lys 40 , Glu 46 , Leu 49 , and Phe 144 ) are orange, and G-CSF-R (His 5 , Phe 75 , Gln 87 , and Gln 91 ) are magenta. Exposed mAb LMM775 and LMM793 binding site is gray, whereas the exposed N terminus of G-CSF (here shown at Gly 4 ) is represented by a filled circle. The figures were prepared using MolScript (44) and Raster3D (45). The coordinates of the model are available from the authors on request. N-term, N-terminal.
the Ig-like domain. In fact, four of the F and G strand mutants had a similar effect on the response to both WT-G-CSF and (E19A)G-CSF and thus behaved independently of the (E19A)G-CSF mutation, and four gave a greater effect than expected. This synergistic effect of two mutations was also seen previously with some combinations of receptor and G-CSF mutations (12). We believe that the synergy indicates that a secondary effect in addition to the direct effect on binding is occurring, but the mechanism of this is unclear.
We have proposed a quaternary model of the 2:2 complex of G-CSF⅐G-CSF-R based on the structure of gp130 in complex with viral IL-6. The site II interaction between human G-CSF and human G-CSF-R is essentially unchanged from the crystal structure of the human G-CSF-mouse G-CSF-R complex. The critical Glu 19 -Arg 288 site II interaction, shown in Fig. 5, is conserved in the human G-CSF-human G-CSF-R complex. Previously proposed site III residues on G-CSF are Lys 40 , Val 48 , Leu 49 , and Phe 144 (13) and Glu 46 (12,14). Our model of the site III interaction involves all of the above residues except Val 48 , which is directed away from G-CSF-R. The role of Lys 40 in a site III interaction has been disputed. Alanine substitution of this residue significantly reduces binding to mouse but not human G-CSF-R (13,14). The model of the complex has Lys 40 in close proximity to the C-CЈ loop comprising Gly 43 and Ala 44 in human G-CSF-R. The corresponding residues in the mouse receptor ( Fig. 2) are Gln 43 , Asp 44 , and Glu 45 . These two negative charges are likely to interact with Lys 40 in human G-CSF, whereas there can be no such interaction between human receptor and ligand.
The positions of G-CSF and G-CSF-R glycosylation sites have been mapped on the 2:2 complex. The O-linked glycosylation of Thr 133 in G-CSF (42) is located in the middle of the C-D loop and far away from binding sites II and III. The four N-linked glycosylation sites located on the first three domains of G-CSF-R (40) are all exposed in the model of the complex and would not interfere with G-CSF binding.
The identification of a ligand-binding site on the Ig-like domain of the G-CSF-R has provided strong evidence for the existence of a site III interaction between G-CSF and the receptor. The model that we based on the structure of gp130 in complex with viral IL-6 explains the published structure-function data as well as the data presented here, suggesting that this gp130 complex structure will be more generally applicable in this family of receptors. The dimerization interface described in the crystal structure of the partial G-CSF-R complex by Aritomi et al. (15) is not compatible with the data presented here. Therefore we consider that the partial complex represents either an intermediate in receptor complex formation or a crystallization artifact.