Identification of a Cys Motif in the Common β Chain of the Interleukin 3, Granulocyte-Macrophage Colony-stimulating Factor, and Interleukin 5 Receptors Essential for Disulfide-linked Receptor Heterodimerization and Activation of All Three Receptors*

The human interleukin 3 (IL-3) and granulocyte-macrophage colony-stimulating factor (GM-CSF) receptors undergo covalent dimerization of the respective specific α chains with the common β subunit (βc) in the presence of the cognate ligand. We have now performed alanine substitutions of individual Cys residues in βc to identify the Cys residues involved and their contribution to activation of the IL-3, GM-CSF, and IL-5 receptors. We found that substitution of Cys-86, Cys-91, and Cys-96 in βc but not of Cys-100 or Cys-234 abrogated disulfide-linked IL-3 receptor dimerization. However, although Cys-86 and Cys-91 βc mutants retained their ability to form non-disulfide-linked dimers with IL-3Rα, substitution of Cys-96 eliminated this interaction. Binding studies demonstrated that all βc mutants with the exception of C96A supported high affinity binding of IL-3 and GM-CSF. In receptor activation experiments, we found that βc mutants C86A, C91A, and C96A but not C100A or C234A abolished phosphorylation of βc in response to IL-3, GM-CSF, or IL-5. These data show that although Cys-96 is important for the structural integrity of βc, Cys-86 and Cys-91 participate in disulfide-linked receptor heterodimerization and that this linkage is essential for tyrosine phosphorylation of βc. Sequence alignment of βc with other cytokine receptor signaling subunits in light of these data shows that Cys-86 and Cys-91 represent a motif restricted to human and mouse β chains, suggesting a unique mechanism of activation utilized by the IL-3, GM-CSF, and IL-5 receptors.

Cytokine receptor dimerization is a common theme in receptor activation (1). Following the binding of the cognate ligand to cytokine receptors, a sequential process takes place whereby receptor subunits associate and recruit cytoplasmic signaling molecules leading to receptor activation and cellular signaling (2). The general process of receptor dimerization exhibits variations among the cytokine receptor superfamily and may in-volve homodimerization or heterodimerization events depending on receptor subunit composition (3,4). In the case of the growth hormone receptor, growth hormone binds initially to one receptor subunit and induces its homodimerization with a second, identical subunit (5). A similar process probably takes place with erythropoietin and granulocyte colony-stimulating factor, leading, in both cases, to receptor homodimerization and activation (6,7).
With cytokine receptors that comprise multiple subunits, receptor activation is accompanied by homodimerization or heterodimerization of the signaling subunits. For example, in the IL-6 1 receptor system, IL-6 induces dimerization of IL-6R␣ with gp130 (8), homodimerization of gp130, and receptor activation (9). On the other hand, the binding of CNTF to CNTFR␣ induces its association with gp130 and the LIF receptor, and the heterodimerization of gp130 and the LIF receptor is accompanied by receptor activation (10). Similarly, heterodimerization of IL-2R␤ and IL-2R␥ subunits is necessary for IL-2 receptor activation (11,12). Interestingly, in these cases, each receptor ␣ chain constitutes the major binding subunit but does not seem to form part of the signaling receptor complex.
The mechanism of activation of the GM-CSF/IL-3/IL-5 receptor system exhibits features similar to the mechanism employed by the above receptors, although some unique features are becoming evident. One of the most important differences is the contribution that each receptor ␣ chain makes to signaling. This is manifested in two ways: first, unlike IL-6R␣, CNTFR␣, and the IL-2R␣, the cytoplasmic domains of GM-CSFR␣, IL-3R␣, and IL-5R␣ are all required for full receptor activation and signaling (13)(14)(15)(16). Second, IL-3R␣ and GM-CSFR␣ form disulfide-linked dimers with the common ␤ chain (␤ c ) of their receptor (4,17). The disulfide-mediated dimerization of IL-3R␣ with ␤ c and of GM-CSFR␣ with ␤ c is accompanied by tyrosine phosphorylation of ␤ c (4). In all of these cases, however, tyrosine phosphorylation is observed in the disulfide-linked dimers as well as in the monomeric molecules, and hence it is not clear which is the critical species for receptor activation. Furthermore, the location of the cysteines involved in disulfide linkage is not known, nor is it apparent whether they constitute a functionally conserved motif in the cytokine receptor superfamily. We have now performed single alanine substitutions of candidate cysteine residues in the N-terminal cytokine receptor module (CRM) of the IL-3, GM-CSF, and IL-5 receptor ␤ c and examined their contribution to disulfide-linked receptor dimerization, high affinity ligand binding, and receptor activation.

Mutagenesis of Human ␤ c and Expression Plasmid Constructs-Cys-
teine residues were substituted with alanines in the human ␤ chain cDNA using oligonucleotide-directed mutagenesis (Altered-sites, Promega, Sydney, New South Wales, Australia) as described previously (18). The mutations were confirmed by nucleotide sequencing, and the mutant ␤ c cDNAs were subcloned into the eukaryotic expression vector pcDNA1 (Invitrogen, San Diego, CA). The IL-3R, GM-CSFR, and IL-5R ␣ chain cDNAs were cloned into the eukaryotic expression vector pCDM8 (Invitrogen) for transfection (18).
Cell Culture and DNA Transfection-COS cells were maintained in RPMI 1640 medium supplemented with 10% v/v fetal calf serum and transfected by electroporation. Routinely, 2 ϫ 10 7 COS cells were cotransfected in 0.8 ml of PBS at 0°C with 25 g of wild type or mutated ␤ c cDNA together with 10 g of GM-CSFR and 10 g of IL-3R ␣ chain cDNA at 500 microfarads with 300 V. After electroporation, cells were centrifuged through a 1-ml cushion of fetal calf serum, and cells were plated in either 25 ml of medium/150-cm 2 flask or 24-well plates for binding analysis. Transfectants were incubated for 2 days prior to cytokine treatment (18).
The HEK293T cell line, derived from the adenovirus type transformed human embryonic kidney 293 cell line and containing the simian virus 40 large tumor antigen (19), was maintained in RPMI 1640 medium supplemented with 10% v/v fetal calf serum. On the day before transfection, 1.4 ϫ 10 6 cells were plated into 6-cm tissue culture dishes to adhere overnight. Four hours after a medium change, 6 g of wild type or mutated ␤ c cDNA together with 4 g of GM-CSFR␣, 4 g of IL-3R␣, or 4 g of IL-5R␣ cDNA and 0.5 g of JAK-2 cDNA were added to cells in the form of a calcium phosphate precipitate (20), and the cells were placed in an incubator for 4 h to permit the uptake of the DNAcalcium phosphate precipitate. The cells were then washed, replated in 4 plates/150 cm 2 , and placed in the incubator for 48 h prior to cytokine treatment.
GM-CSF, IL-3, and IL-5-Recombinant human IL-3, GM-CSF, and IL-5 were produced in Escherichia coli essentially as described before (21,22). Cytokine purity and quantitation was determined by high performance liquid chromatography analysis. The unit activity of the cytokines based on the ED 50 values in a proliferation assay (23) was 0.03 ng/ml GM-CSF, 0.1 ng/ml IL-3, and 0.3 ng/ml IL-5; each value represents 1 unit of that substance.
Radiolabeling Cytokines-Recombinant IL-3 and GM-CSF were radioiodinated by the iodine monochloride method (24) to a specific activity of about 36 mCi/mg. Routinely, 4 g of protein was iodinated and separated from iodide ions on a Sephadex G-25 column (Pharmacia Biotech Inc.), eluted with PBS containing 0.02% v/v Tween 20, and the iodinated proteins were stored at 4°C for up to 4 weeks.
Saturation Binding Assays-Binding assays were performed on confluent monolayers in 24-well plates over a concentration range of 10 pM to 10 nM 125 I-labeled GM-CSF or IL-3 in binding medium (RPMI containing 0.5% (w/v) BSA/0.1% (w/v) sodium azide) essentially as described previously (25). After incubation at room temperature for 2 h, radioligand was removed, and the cells were briefly washed twice in binding medium. Specific counts were determined after lysis of the cell monolayer with subsequent transfer and counting on a ␥ counter (Cobra Auto Gamma; Packard Instruments Co., Meridien, CT). Dissociation constants were calculated using the EBDA and LIGAND programs (26) (Biosoft, Cambridge, United Kingdom).

Analysis of Receptor Cell Surface Expression by Flow Cytometry-
Cell surface expression of transfected receptor subunits was confirmed by indirect immunofluorescence staining using anti-receptor ␣ and ␤ chain specific monoclonal antibodies. Staining was performed as described previously (17) and analyzed with a EPICS Profile II flow cytometer (Coulter Electronics, Hialeah, FL). Surface Labeling of Cells and Immunoprecipitation-COS cells were cell surface-labeled with 125 I by the lactoperoxidase method as described previously (28). Approximately 10 8 cells were washed twice in PBS and then labeled with 1 mCi of 125 I (NEN Life Science Products and AMRAD Pharmacia) in PBS. Cell were lysed in lysis buffer consisting of 137 mM NaCl, 10 mM Tris-HCl (pH 7.4), 10% glycerol, 1% Nonidet P-40 with protease inhibitors (10 g/ml leupeptin, 2 mM phenylmethlysulfonyl fluoride, 10 g/ml aprotinin), and 2 mM sodium vanadate for 30 min at 4°C followed by centrifugation of the lysate for 15 min at 12,000 g 4°C. Following a 1-h preclearance with protein A-Sepharose (Pierce) at 4°C, the supernatant was incubated for 18 h with 5 g/ml antibody. Protein-immunoglobulin complexes were captured by incubation for 1 h with protein A-Sepharose followed by six subsequent washes in lysis buffer. Samples were boiled for 10 min in SDS sample load buffer either in the presence or absence of 2-mercaptoethanol (i.e. reducing or nonreducing) before separating immunoprecipitated proteins by SDS-PAGE. Immunoprecipitation from HEK293T cells was carried out similary except the cells were not surface labeled.
SDS-Polyacrylamide Gel Electrophoresis-Immunoprecipitated proteins were analyzed by SDS-PAGE on polyacrylamide gels. Samples were boiled in SDS loading buffer for 5 min prior to loading. Molecular weights were estimated using SeeBlue TM prestained standards (Novex French's Forest, New South Wales, Australia). Radiolabeled proteins were visualized using an ImageQuant PhosphorImager (Molecular Dynamics, Sunnyvale, CA).

Rationale for Mutagenesis of N-terminal Cysteine
Residues of ␤ c -To study the molecular events involved in the activation of the IL-3, GM-CSF, and IL-5 receptors, we replaced several extracellular cysteine residues of ␤ c by alanine residues. So as to target Cys available for intermolecular interactions, we sought to avoid Cys involved in structurally important intramolecular disulfide bonds. By homology with other cytokine receptors, domains one and three are expected to possess two disulfide bonds each. This is clearly the case with domain three, which contains only four Cys residues. However, domain one ␤ c possesses seven Cys, of which only Cys-34, Cys-45, and Cys-75 could be aligned readily with equivalent Cys residues in other receptors (Fig. 1). Of the remaining Cys residues, Cys-86, Cys-91, and Cys-96 are conserved in murine ␤ c . Of these, Cys-96 is followed by an Ile at position 98, which aligns with conserved hydrophobic residues in other cytokine receptors, suggesting that this Cys is part of the second conserved disulfide bond. Cys-96 and Cys-91 are proposed to lie in an extended loop between the D and E beta strands of the first domain. Although Cys-86 and Cys-91 were favored candidates for intermolecular disulfide bond formation, we chose to also mutate the nearby Cys-96 and Cys-100 and the single Cys residue in domain 2 at position 234.
Mutation of Cys-86 and Cys-91 Selectively Disrupt Ligandinduced, Disulfide-linked Heterodimer Formation-Expression plasmids encoding IL-3R ␣ and wild type (wt) or C86A, C91A, C96A, C100A, and C234A mutant ␤ c were co-transfected into COS cells. After 48 h, the cells were 125 I surface-labeled and either left unstimulated or stimulated with IL-3. IL-3R␣ and ␤ c were then immunoprecipitated with specific MAbs 9F5 and 8E4, respectively, and the proteins were resolved by 6% SDS-PAGE under either nonreducing or reducing conditions. We found that the ␤ c mutants C100A and C234A behaved very similarly to wt ␤ c . Both mutants allowed the formation of two high molecular weight complexes in response to IL-3 ( Fig. 2A), which, as with wt ␤ c , contain IL-3R␣ and ␤ c (Ref. 17 and data not shown). These two complexes were immunoprecipitated by both anti-IL-3R␣ mAb 9F5 and anti-␤ c mAb 8E4 ( Fig. 2A). In the absence of IL-3, the anti-IL-3R␣ mAb 9F5 immunoprecipitated only monomeric IL-3R␣, whereas the anti-␤ c mAb 8E4 immunoprecipitated monomeric ␤ c as well as the high molecular weight complex corresponding to disulfide-linked ␤ c homodimers (4, 17) ( Fig. 2A). As with the disulfide-linked dimers, the noncovalent IL-3R␣ and ␤ c heterodimers were not affected by mutating Cys-100 or Cys-234 because both the anti-IL-3R␣ mAb and the anti-␤ c mAb co-immunoprecipitated both IL-3R␣ and ␤ c , and they did so only in the presence of IL-3 (Fig. 2B).
In contrast, the mutants C86A, C91A, and C96A had a profound effect on disulfide-linked receptor dimerization. In the presence of IL-3, anti-IL-3R␣ mAb 9F5 and anti-␤ c mAb 8E4 did not immunoprecipitate the high molecular weight complexes corresponding to IL-3R␣ and ␤ c heterodimers ( Fig. 2A). In fact, under nonreducing conditions, very little or no monomeric ␤ c was immunoprecipitated by either mAb; most of the label was observed in the high molecular weight region, probably representing aggregated ␤ c . With the anti-IL-3R␣ mAb 9F5, a nonspecific band migrating slightly faster than ␤ c was seen ( Fig. 2A). Under reducing conditions, however, monomeric ␤ c could be detected (Fig. 2B). An important difference was noted between the C86A and C91A mutants on one hand and the C96A mutant on the other hand. Anti-IL-3R␣ mAb 9F5 co-immunoprecipitated C86A and C91A in the presence of IL-3 but did not co-immunoprecipitate C96A (Fig. 2). Reciprocally, in the presence of IL-3, the anti-␤ c mAb co-immnunoprecipitated IL-3R␣ with C86A and C91A but not with C96A (Fig. 2). This was more clearly seen under reducing conditions (Fig. 2B) than under nonreducing conditions ( Fig. 2A), where an overall lower signal was observed.
To verify that the surface expression levels of the individual ␤ c mutants was similar to ␤ c wt, COS cells transfected with the various constructs were analyzed by flow cytometry. Flow cytometry analysis indicated that the surface expression of wt ␤ c and the Cys 3 Ala ␤ c mutants was very similar both in terms of percentage of transfected cells expressing the different receptor subunits and in absolute levels (Fig. 3), suggesting that the mutations did not affect subunit transport and expression at the cell surface.
The ␤ c Mutants C86A and C91A Do Not Disrupt IL-3 and GM-CSF High Affinity Binding-We next examined the ability of Cys mutants of ␤ c to support high affinity IL-3 or GM-CSF binding. COS cells were transfected with IL-3R␣ and GM-CSFR␣ and either wt ␤ c or the Cys 3 Ala ␤ c mutants and subjected to saturation binding studies with 125 I-IL3 and 125 I-GM-CSF. Scatchard transformation of the saturation binding curves were performed and the K d and receptor numbers determined using the Ligand program. We found that ␤ c bearing Cys 3 Ala substitutions at positions 86, 91, 100, and 234 were able to form high affinity binding sites ( Fig. 4 and Table I). The range of affinities for IL-3 high affinity binding of these Cys 3 Ala ␤ c substitution mutants varied from 31 to 280 pM, compared with 330 pM for wt ␤ c , whereas GM-CSF high affinity binding ranged from 27 to 230 pM, compared with 120 pM for wild type ␤ c . In contrast, COS cell transfectants expressing the C96A ␤ c showed no detectable high affinity binding ( Fig. 4 and Table I).
Although ␤ c mutants C86A and C91A were able to support high affinity binding of IL-3 and GM-CSF, a reduction in the number of high affinity receptors was observed compared with wild type ␤ c and the C100A and C234A analogues ( Fig. 4 and Table I). This is probably a reflection of the tendency of C86A and C91A ␤ c analogues to oligomerize as observed in the immunoprecipitations under nonreducing conditions ( Fig. 2A), thereby reducing the amount of free ␤ c available for interaction with ␣ chain.
C86A and C91A Abolish IL-3-, GM-CSF-, and IL-5-dependent Tyrosine Phosphorylation of ␤ c -It has been previously established that stimulation of cells with IL-3 leads to the formation of disulfide-linked heterodimers of IL-3 receptor ␣ and ␤ c chain, which is associated with phosphorylation of ␤ c (17). Similarly, in the case of the GM-CSF receptor, receptor heterodimerization occurs upon stimulation with GM-CSF, and this is accompanied by ␤ c phosphorylation (4). Here we show that in addition to the IL-3 and GM-CSF receptors, the IL-5 receptor forms disulfide-linked complexes that are similarly accompanied by ␤ c phosphorylation (Fig. 5). The relative proportion of phosphorylated ␤ c in the disulfide-linked heterodimer and in monomeric ␤ c varied between the three receptors. This may be due to kinetic differences in receptor assembly or in the stability of each receptor heterodimer. We have now taken advantage of the inability of ␤ c mutants to form disulfide-linked heterodimers to determine whether the formation of these is necessary for receptor activation or whether noncovalent dimerization is sufficient for activation as measured by ␤ c phosphorylation. We transfected wt ␤ c and the different ␤ c mutants in HEK293T cells together with JAK-2 and either IL-3R␣, GM-CSFR␣, or IL-5R␣ chain cDNA. After 48 h, the cells were either not treated or treated with IL-3, GM-CSF, or IL-5, lysed, and immunoprecipitated with mAb 8E4 anti-␤ c. The immunoprecipitates were separated on SDS-PAGE gels under reducing conditions and Western blotted with antiphosphotyrosine antibody. Mutants C100A and C234A and wt ␤ c , which heterodimerize with the receptor ␣ chain in a disulfide-linked manner in response to ligand, showed phosphorylation of ␤ c . In contrast, mutants C86A, C91A, and C96A, which have lost the ability to heterodimerize in a disulfide- FIG. 1. Alignment of domain 1 of the CRM present in the common ␤ chain of the GM-CSF, IL-3, and IL-5 receptors and other signaling subunits of the cytokine receptor superfamily. Four conserved Cys residues form the basis of the alignment, with the second Cys followed by a conserved Trp and the fourth Cys followed by a hydrophobic residue at the iϩ2 position. The sequences of human and mouse ␤ c and ␤ IL-3 are shown; the numbering corresponds to the human sequence, with residue 1 being the initiation Met. Human and mouse ␤ subunits are aligned with human gp130, the IL-2R␤ and IL-2R␥ chains, the erythropoietin receptor (EPOR), and the growth hormone (GHR). The dashes represent spaces introduced to optimize the alignment. linked manner in response to ligand, have lost the potential to be phosphorylated in response to IL-3, GM-CSF, or IL-5 (Fig. 6).
The expression of all mutants compared with wt was monitored by flow cytometry and by Western blot analysis with antibodies to ␤ c and indicated that the levels were very similar between all the mutants. (Figs. 3 and 5D). Because the number of high affinity sites for mutants C86A and C91A was decreased, the lack of phosphorylation with these two mutants could have been the result of a decrease in sensitivity due to less mutant ␤ c being heterodimerized compared with the total amount of mutant ␤ c expressed. To address this possibility, we have examined only ␤ c heterodimerized to the IL-3R␣ by immunoprecipitating with IL-3R␣ antibody and Western blotting with antiphosphotyrosine antibody. The results were identical to those seen when the ␤ c was directly immunoprecipitated, indicating that cysteines in position 86 and 91 are essential for receptor tyrosine phosphorylation (Fig. 7). DISCUSSION We show here that disulfide-linked heterodimerization of the GM-CSF, IL-3, and IL-5 receptors is essential for receptor activation by the cognate ligand. Furthermore, we have identified Cys-86 and Cys-91 in the N-terminal domain of ␤ c as the key Cys residues involved in heterodimerization with the ␣ chain of each receptor. Comparison with other cytokine receptors indicates that these Cys residues constitute a conserved motif present only in human and mouse ␤ c and in ␤ IL-3 , suggesting that it subserves a specialized function restricted to the GM-CSF, IL-3, and IL-5 receptor family.
We have previously shown that the human IL-3 and GM-CSF receptors undergo both noncovalent and disulfide-linked dimerization upon ligand binding (4). We have now extended these observations to the IL-5 receptor, demonstrating that disulfide-linked dimerization is a common theme in this recep- tor subfamily. To identify the Cys residues in ␤ c responsible for disulfide linkage with the IL-3, GM-CSF, and IL-5 receptor ␣ chains, we mutagenized five of the eight Cys residues in the N-terminal CRM of ␤ c that, from alignment with other cytokine receptors, represented the best candidates for intermolecular interactions. We found a range of sensitivities to mutation of these Cys residues that could be correlated with their interspecies and interreceptor conservation.
The first class of Cys residue is exemplified by Cys-100 and Cys-234. These residues are not conserved with even the closely related mouse ␤ chains and we found no phenotype on replacing them with alanine residues. The second class is represented by Cys-96, which is apparently a conserved residue in both the mouse ␤ chains and the cytokine receptor family at large (29) and is inferred to be involved in a structurally conserved disulfide bond. This residue is apparently required for the structural integrity of the first domain of h␤ c if not the entire extracellular portion of the molecule. Although the C96A mutation permitted cell-surface expression of h␤ c , it did not support high affinity binding of GM-CSF or IL-3 despite the  Table I. substitution being well removed in sequence, and presumably spatially distant, from the fourth domain of the receptor that encompasses the majority of the ligand-recognition determinants (18,25). The exact molecular basis for this observation is uncertain, but it may be related to sequestration of h␤ c into very large aggregates ( Fig. 2A) that obscure the ligand-contact site.
The third and most interesting class of cysteine mutation is that of C86A and C91A, analogues that have lost their ability to form disulfide-linked heterodimers but still retain the ability to associate noncovalently with the ␣ subunit upon stimulation with ligand (Fig. 2). Although these mutants exhibited some propensity to aggregate in the absence of stimulation, they retain the ability to interact with ligand as judged by their ability to support high affinity binding. Importantly, these analogues are deficient in phosphorylation of tyrosine residues in ␤ c ; however, receptor-mediated functions and downstream signaling remains to be ascertained.
The observation of identical phenotypes with either mutation C86A or mutation C91A suggests that these residues may cooperate functionally in the native receptor such as via formation of an additional intramolecular disulfide. This is con-sistent with our molecular modelling of ␤ c , which suggests that Cys-86 and Cys-91 are sufficiently close to allow the formation of such a disulfide bond. 2 In the presence of ligand, this bond is proposed to undergo disulfide exchange with a free sulfhydryl group from the ␣ chain that is brought into proximity via ligand-dependent noncovalent association.
Previous experiments have noted a correlation between disulfide-linked receptor dimerization and receptor activation. IL-6 induces covalent dimerization of two molecules of gp130 (9), and CNTF induces covalent dimerization of gp130 with the LIF receptor (10). Similarly, IL-3 (Fig. 5A), GM-CSF (Fig. 5B), and IL-5 (Fig. 5C) induce covalent dimerization of ␤ c with the corresponding ␣ chain. In all of these cases, concomitant phosphorylation of the receptor has been observed; however, a causal relationship has not been established. The use of the C86A and C91A mutants allowed us to demonstrate that noncovalent receptor associations are not sufficient for receptor tyrosine phosphorylation and that this requires disulfide linkage of receptor subunits. The role of covalent dimerization  Cys-91, Cys-96, Cys-100, and Cys-234 of ␤ c on IL-3 and GM-CSF high affinity binding COS cells were transfected with IL-3R␣, GM-CSFR␣, and either wild type ␤ c or mutated ␤ c carrying alanine substitutions of Cys at positions 86, 91, 96, 100, and 234 and subjected to saturation binding studies with 125 I-labeled IL-3 and 125 I-labeled GM-CSF. The radioiodinated ligand concentration for both IL-3 and GM-CSF ranged from 10 pM to 10 M. Nonspecific binding was determined in the presence of 1 M unlabeled ligand. Scatchard transformation of the saturation binding curves were performed, and the K d and receptor numbers determined using the LIGAND program. In the case of IL-3, due to the extremely low affinity of IL-3R␣, the affinity was estimated to be 50 nM based on our own previous studies (25). In the case of GM-CSF binding, a two-site fit was statistically preferred (P Ͻ 0.05) with all the ␤ c constructs except for C96A, in which no high affinity sites were detected. Values from two representative experiments are shown. Experiment 2 is the same as the experiment shown in Fig 4. IL-3 GM-CSF   , or IL-5R␣ (C) together with wild type or mutant ␤ c were incubated either with medium alone (Ϫ) or with medium containing 6.5 nM IL-3 (ϩIL-3), 6.5 nM GM-CSF (ϩGM-CSF), or 6.5 nM IL-5 (ϩIL-5) for 5 min at 4°C. After cell lysis, proteins were immunoprecipitated with anti-␤ c mAb 8E4, and the immunoprecipitates were separated under reducing conditions on an SDS-7.5% polyacrylamide gel transferred to nitrocellulose and probed with anti-phosphotyrosine antibody 3-365-10 (A-C). To control for the amount of ␤ c present, the filters were also probed with anti-␤ c , mAb, 1C1 (D).

FIG. 7.
Ligand-induced heterodimers of IL-3R␣ and ␤ c with C86A and C91A substitution lack tyrosine phosphorylation of ␤ c . HEK293T cells transfected with IL-3R␣ together with either wild type or mutant ␤ c were incubated with medium alone (Ϫ) or with medium containing 6.5 nM IL-3 (ϩIL-3) for 5 min at 4°C. After cell lysis, proteins were immunoprecipitated with mAb 9F5 (IL-3R␣), and the immunoprecipitates were separated under reducing conditions on a SDS-7.5% polyacrylamide gel, transferred onto nitrocellulose, and probed with antiphosphotyrosine antibody 3-365-10 (A). To control for similar amounts of immunoprecipitated ␤ c , the filters were also probed with mAb 1C1 anti-␤ c (B). growth signal (32). Although they are essential for normal activation (14 -16), the role of the cytoplasmic domains of the receptor ␣ chains remains unclear; however, they may serve to orientate to ␤ chains so as to juxtapose correctly the JAK molecules or to participate directly in certain functions (33,34).
The ␤ chain forms intermolecular disulfides specifically with the ␣ chains of the GM-CSF, IL-3, or IL-5 receptors. A common feature of these ␣ chains is the presence of an N-terminal FnIII-like domain of a type restricted to this subfamily of cytokine receptors. Within this N-terminal domain, all three receptor ␣ chains possess an uneven number of Cys residues in the N-terminal domain, suggesting that these Cys residues are the most likely candidates to act as partners for ␤ c . The IL-5R␣ has only a single Cys residue in the N-terminal domain, at position 86, and this residue has been shown to be important for IL-5 binding to the receptor (35).
The stoichiometry of the IL-3, GM-CSF, and IL-5 receptor complexes is not known. The formation of an intermolecular disulfide bond between Cys-86 or Cys-91 of ␤ c and a Cys in the N-terminal domain of a receptor ␣ chain could potentially occur with either the ␣ chain with which it shares ligand or a second ␣ chain, recruited as part of a hexameric complex, as seen with the IL-6 receptor (36). Because the individual FnIII-like domains of the receptor ␣ chains and ␤ c are likely to be fairly rigid units with a length of 3.5-4.5 nm, the ability of ␤ c to contact ␣ chain will depend on the interdomain angles that they can adopt. The receptor ␣ and ␤ c chains are class 1 cytokine receptors, and the angles observed between the two domains of the CRM in the known structures of this family, growth hormone receptor (5) and erythropoietin receptor (37) are approximately 90°. It is therefore reasonable to infer that the angles between domains 1 and 2 and domains 3 and 4 of ␤ c and between domains 2 and 3 of the receptor ␣ chains will also be approximately 90°. The conformation of the linker peptides between domains 2 and 3 of ␤ c or between domains 1 and 2 of the receptor ␣ chains cannot be gauged by reference to homologous structures. Even if the linker peptides permitted the membrane-distal portions to fold back over the cytokine binding portions of the receptors, this would be unlikely to facilitate a sufficiently close approach of Cys residues to allow formation of the observed intermolecular disulfide bonds. Rather, we propose that the intermolecular disulfide bond forms between ␤ c and an ␣ chain from a second receptor heterodimer (Fig. 8) because this can be accommodated readily with respect to both the sizes of the domains and their interdomain angles.
Based on the likely orientation of their N-terminal domains with respect to the CRM of the ␣ chains, we favor a receptor complex arranged clockwise when viewed from outside the cell in the order ␣ chain 1 /ligand 1 /␤ chain 1 S-S␣ chain 2 /ligand 2 /␤ chain 2 . The disulfide linkage of ␤ c in one receptor heterodimer to an ␣ chain in a second receptor heterodimer would facilitate juxtaposing two ␤ c molecules with their associated JAK kinases and induce receptor phosphorylation. This initial association may also facilitate the formation of a second disulfide between ␤ c in receptor 2 and an ␣ chain in receptor 1 (Fig. 8, A  and B). The formation of a 2:2:2 complex is also consistent with the requirement of two ␣ chains in an active ligand-receptor complex (38). On the other hand, a 1:1:1 stoichiometry has been suggested from experiments using chimeras of ␣ and ␤ c with fos and jun leucine zippers, although the formation of higher order complexes was not excluded (34). The direct measurement of ␣-␤ c interactions in solution may ultimately resolve this question.