A dominant negative granulocyte-macrophage colony-stimulating factor receptor alpha chain reveals the multimeric structure of the receptor complex.

The receptor for the hemopoietic growth factor granulocyte-macrophage colony-stimulating factor (GM-CSF) is composed of two chains, both of which belong to the superfamily of cytokine receptors. The α chain confers low affinity binding only, whereas the β chain (βc) confers high affinity binding when associated with α. Ectopic expression of both chains of the receptor in murine NIH-3T3 fibroblasts results in signal transduction, mitogenesis, and morphologic transformation. The cytoplasmic domain of the GM-CSF receptor α subunit (GMR-α) comprises 54 amino acids that have been shown to be important for signal transduction through the β chain. The present study was designed to address the possibility of receptor oligomerization and its functional implication. Cross-linking studies with 125I-GM-CSF on NIH-3T3 transfectants is consistent with the presence of α and βc dimers and of receptor oligomers. We have, therefore, generated an inert α chain through polymerase chain reaction-mediated truncation of 47 amino acids of the COOH-terminal domain of α (αt), and coexpressed αt, α, and βc in NIH-3T3. In cells in which αt and α are present in stoichiometric proportion within the GM-CSF-binding complex, we provide evidence that αt is dominant negative over wild type α on the basis of two different functional assays: cell proliferation and foci formation. Hence, our results suggest the requirement for at least two functional α chains for signal transduction. Together with the cross-linking studies, our data indicate that the functional GMR is an oligomer that contains at least two α chains.

Granulocyte-macrophage colony stimulating factor (GM-CSF) 1 is a multifunctional growth factor (for review, see Ref. 1) that stimulates the proliferation of hemopoietic cells and also of 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 (for review, see Ref. 5).
GM-CSF binds to a receptor that is composed of at least two different subunits, an ␣ chain (6) and a ␤ chain (7,8), both of which are members of the superfamily of cytokine receptors. This family is characterized by conserved structural features in the extracellular domain, i.e. four conserved cysteine residues, and a typical WSXWS motif in the juxtamembrane region (9,10). Furthermore, the cytoplasmic domain lacks intrinsic enzymatic activities (10 -12). The human GM-CSF receptor ␣ subunit (GMR-␣) is 378 amino acids in length (6), most of which constitutes the extracellular domain, whereas the cytoplasmic tail has only 54 amino acids. GMR-␣ confers low affinity binding and has been shown to be species-specific for its ligand (6,7,13). The GMR ␤ c subunit comprises 881 amino acids with a 432-amino acid cytoplasmic tail (7) and has no affinity for GM-CSF by itself (8). The association of GMR-␣ and GMR-␤ confers medium (14) to high affinity binding (13) and biological activity (8,15,16). Previous works have established the importance of the cytoplasmic domain of GMR-␣ in signal transduction and biological activity (17)(18)(19), but this domain does not seem absolutely necessary under very high concentrations of GM-CSF (20). Furthermore, truncation of the COOH-terminal domain of GMR-␣ does not affect high affinity GM-CSF binding (19). Ectopic expression of both chains of human GMR has been shown to confer GM-CSF responsiveness to the murine fibroblast cell line NIH-3T3 and the pro-B cell line BaF3 (16,21), indicating a conservation of signal transduction pathways.
Interestingly, the ␤ chain, referred to as ␤ common or ␤ c , is shared with interleukin-3 and -5 (IL-3 and IL-5), two cytokines that exhibit significant overlap in biological activity with GM-CSF (for review, see Ref. 22). The cytoplasmic domains of IL-3R, GMR, and IL-5R ␣ chains also share a highly conserved stretch of amino acids just after the transmembrane domain (23). In parallel, IL-2, IL-4, IL-7, IL-9, and IL-15 have also been shown to associate with their cognate receptors with similar heteromeric dynamics (24 -26), each one binding a specific ␣ or ␤ chain and sharing a common ␥ chain. Finally, another family of cytokines, IL-6, IL-11, oncostatin M, ciliary neutrotrophic factor and leukemia inhibitory factor, also bind to receptors that share common subunits, gp130 or leukemia inhibitory factor receptor ␣ (27; for review see Ref. 28). Thus, receptor permutation and matching appears to be a recurring theme for cytokines with overlapping biological activities.
Our previous work indicates that GMR-␣, -␤ c , and GM-CSF associate in stoichiometric proportion to form a high affinity slowly dissociating ternary complex (13). More recently, IL-6 has been shown to form hexamers in solution with IL-6R and gp130 in the proportion 2:2:2 (27). Furthermore, ectopic expression of ␤ c with a point mutation in the transmembrane domain (V449E) has been shown to confer ligand-independent growth to the hemopoietic cell line FDC-P1 (29). By analogy with a similar mutation in neu, it is suggested that the V449E mutation triggers constitutive ␤ c homodimerization and, by extrapolation, that wild type (w/t) ␤ c may also be able to form homodimers. There is, however, no direct evidence for a higher order of receptor association.
In the present study, affinity cross-linking indicates the presence of both ␣ and ␤ c dimers within the GMR complex. Our data also indicate that ␣ can homodimerize in the presence of ligand, even when ␤ c is absent. Using functional assays, we provide evidence that a GMR-␣ truncated in its cytoplasmic domain acts as a dominant negative mutant over w/t GMR-␣, suggesting higher order association and a functional role for the oligomerization of GMR-␣.

Construction of the COOH-terminal Truncated GMR-␣-
The cytoplasmic domain of GMR-␣ was deleted by PCR as follows. Wild type GMR-␣ cDNA, cloned in the plasmid vector pGEM7 (Promega, Madison, WI), served as a template for the PCR using two oligonucleotides, primer F (ACCAGCCGAGAAATTGG) (position 737-754, located in the extracellular domain of the receptor) and GMR-␣ t primer (CGCTCTA-GACTACTGTATCCTAAGGAACCTTTT) (position 1208 -1188, covering the first 21 nucleotides of the intracellular domain, to which 12 mismatched nucleotides were added to create a stop codon and one XbaI site. The PCR fragment and vector with w/t insert were digested with PstI and XbaI and ligated. The truncated insert (GMR-␣ t ) was then excised with XbaI and EcoRI and recloned into the expression vector pME18 (graciously provided by Dr. Toshio Kitamura, DNAX, Palo Alto, CA; 8, 16). The insert was sequenced to confirm its identity.
Plasmids, Cells, and Transfection-NIH-3T3 cells were graciously provided by Dr. A. Veillette (Cancer Center, McGill University, Montreal, Québec) and were maintained in Iscove's modified Dulbecco's medium (IMDM, Life Technologies, Inc.) supplemented with 10% fetal calf serum (FCS, Life Technologies, Inc.) at 37°C in a fully humidified incubator containing 5% CO 2 . Wild type ␤ c cDNA (KH97) was cloned in the same expression vector as GMR-␣ cDNA but in the absence of the cassette conferring neomycin resistance. The plasmid conferring hygromycin resistance pCEP4 was from Invitrogen (San Diego).
The transfection strategy is shown in Fig. 1. In a first step, stable transfectants expressing either w/t GMR-␣ or -␤ c only or both w/t GMR-␣ and -␤ c contained within the pME18-neo expression vector were generated using G418 selection (500 g/ml) for ␣/␤ c and hygromycin selection (500 g/ml) for ␤ c alone. The transfections were done by calcium phosphate precipitation at a molar ratio of 5:1 of KH97/ pME18-␣ or of KH97/pCEP4. Briefly, 6 -15 g of plasmid DNA was diluted in 100 l of sterile H 2 O and added to an equal volume of 4 ϫ CaCl 2 , pH 7.9 (2 mM Tris (BDH, Poole, UK), pH 8, 0.2 mM EDTA (BDH), pH 8, 500 mM CaCl 2 (BDH)). The mixture was then added slowly with air bubbles to 200 l of 2 ϫ HEPES-buffered saline, pH 7.1 (50 mM HEPES (Fluka Biochemica, Ronkonkoma, NY), 280 mM NaCl (BDH), 1.5 mM Na 2 HPO 4 ⅐7H 2 O (BDH)). After 30 min at room temperature, 400 l was distributed evenly over a 60-mm culture dish of NIH-3T3 cells seeded at 200,000 cells/dish the day prior to transfection. The cells were incubated overnight at 37°C, the precipitate was removed, and the cells were fed with culture medium (IMDM ϩ 10% FCS). The selection was applied 2 days after the transfection. Several independent clones were chosen after 14 days in selective medium on the basis of their binding properties and response to GM-CSF in a proliferative assay (16). The cells were expanded in the presence of 200 g/ml G418 or 250 g/ml hygromycin, respectively. Clones 13 (␣/␤ c ) and 9.2 (␤ c only) were retained for further transfections.
In a second step, the truncated GMR-␣ t cDNA and the pCEP4 plasmid (Invitrogen) conferring hygromycin resistance were cotransfected into clone 13 at a molar ratio of 2.5:1, i.e. 12 g of GMR-␣ t cDNA and 3 g of pCEP4. The cells were selected using 500 g/ml hygromycin (clone 13). In parallel, GMR-␣ t cDNA was also transfected into clone 9.2 (expressing w/t ␤ c ), and G418-resistant clones were selected in the presence of 500 g/ml G418. Twenty-five to 30 independent clones from each transfection were screened for both GMR-␣ and ␣ t expression by reverse transcriptase (RT)-PCR for GM-CSF binding and GM-CSF-de-pendent cell proliferation. Five to six stable clones were retained from each transfection and characterized further.
RT-PCR-To select clones with variable levels of expression of w/t and truncated GMR-␣, total RNA was extracted, and 5 g was subjected to an RT reaction using an oligonucleotide that covers the SV40 polyadenylation signal (PAD1) (GCTTTATTTGTGAAATTTGTGATG) contained in the vector pME18, the murine ribosomal S16 antisense primer as an internal control, and the Moloney murine leukemia virus RT (18 units, Life Technologies, Inc.). The PCR (25 cycles; Ref. 30) was performed using 1 l of RT product (from a total volume of 20 l), Vent buffer (10 ϫ, New England BioLabs, Beverly, MA), 5 mM dNTP, 0.5 g each of GMR-␣ t and F primers, 100 mM MgSO 4 (New England BioLabs), and 1 unit of Vent enzyme (New England BioLabs), with annealing temperatures of 58°C.
Binding Assay and Saturation Analysis-Purified recombinant GM-CSF was iodinated with the Bolton-Hunter reagent (DuPont NEN). Specific activity was determined by radioimmunoassay (13,31) and confirmed independently by enzyme-linked immunosorbent assay (32). To cover concentrations up to 20 nM, iodination conditions were chosen to yield a moderately low specific activity (300 -500 cpm/fmol). Cells were distributed in 24-well plates (Linbro, ICN Biomedicals, Costa Mesa, CA) at a concentration of 86,000 cells/well (for saturation analysis and binding assays on transiently transfected NIH-3T3) or confluent monolayers in 35-mm tissue culture dishes (Falcon, Becton Dickinson, Lincoln Park, NJ) for screening. After overnight adherence, the cultures were washed once, and the binding reaction was initiated with the indicated concentrations of 125 I-GM-CSF in a total volume of 100 l/well or 350 l/35-mm tissue culture dish of bicarbonate-free IMDM supplemented with 1% bovine serum albumin. Where indicated, 100fold excess cold GM-CSF was added to the binding reaction to determine the nonspecific binding. The reaction was allowed to proceed for 3 h at 4°C with rocking and was stopped by three rapid washes with ice-cold phosphate-buffered saline. Cells were collected by the addition of 100 l of trypsin. Binding assays were done in duplicate for screening (in 35-mm tissue culture dishes) and in triplicate for all subsequent experiments, which were performed in 24-well plates. Saturation curves were analyzed with the program SCAFIT using a nonlinear curve fitting routine (13,15,31).
Immunoperoxidase Assay-NIH-3T3 cells were seeded at 5 ϫ 10 5 cells/60-mm tissue culture dishes and left to adhere overnight. They were transfected using the CaPO 4 method with 5 g of pME18-neo vector containing the cDNA for GMR-␣ t or GMR-␣ w/t. Total DNA concentration was brought up to 10 g with pGEM4 as a carrier. Sixteen hours later, cells were washed and distributed into six wells FIG. 1. Production of stable NIH-3T3 transfectants expressing ␤ c (clone 9.2) and w/t ␣ (clone 13) and/or ␣ t . NIH-3T3 cells were transfected with w/t GMR-␣ and ␤ c cDNA by calcium phosphate precipitation. Independent clones were selected using 500 g/ml G418. Clone 13 was retained after binding and proliferation assays with GM-CSF. Clone 13 was then transfected with GMR-␣ t cDNA and pCEP4 plasmid conferring hygromycin resistance. Six independent clones were retained according to their varying expression ratio of ␣ t to ␣ w/t and binding assays. In parallel, NIH-3T3 cells were also transfected with GMR-␤ c w/t chain cDNA and pCEP4 plasmid. Independent transfectants were selected with 500 g/ml of hygromycin. Clone 9.2 was retained and subjected to GMR-␣ t cDNA transfection. Stable transfectants were selected using 500 g/ml G418. Five independent clones were retained using binding assays. each in 24-well plates to adhere overnight. Unless otherwise stated, all of the following steps were carried at 4°C. After a 10-min blocking step with 500 ml of IMDM supplemented with 10% normal goat serum (NGS), cells were labeled with 200 l of mouse anti-human GMR-␣ antibody (Upstate Biotechnology Inc.) at a concentration of 2 g/ml IMDM, 1% NGS for 30 min. Following this, cells were further blocked with 300 l of IMDM, 10% NGS for 10 min and washed once with 400 l of IMDM, 1% NGS. The peroxidase-coupled goat anti-mouse IgG (Sigma Chemical Co.) was added in a total volume of 200 l of IMDM, 1% NGS, 10% FCS at a final dilution of 1:75 for 30 min. Cells were washed twice with 400 l of IMDM, 1% NGS. The peroxidase was revealed with 200 l of o-phenylenediamine from Sigma at 1 mg/ml in water with 0.7 l/ml hydrogen peroxide (30%, Sigma) and incubated at 37°C for 3 min. The reaction was stopped with 50 l of 4 N sulfuric acid. The supernatants were harvested and centrifuged at 13,000 rpm for 3 min, and 200 l was used to read the optical density at 490 nm.

125
I-GM-CSF Cross-linking-All steps were done on ice or at 4°C, using two 60-mm confluent tissue culture dishes/clone. After washing, bound GM-CSF was covalently cross-linked with 1 mM BS 3 (Pierce) in ice-cold phosphate-buffered saline for 30 min with rocking (33). The reaction was stopped with 125 l of 10 ϫ quenching buffer (10 mM Tris, 1 mM EDTA, 150 mM NaCl) (13). Cells were lysed in 200 l of 50 mM HEPES, pH 8, containing 1% Triton X-100 and 1 mM phenylmethylsulfonyl fluoride for 15-20 min. Cells debris was removed by a 15-min centrifugation at 13,000 rpm, and the cleared lysates were kept at 4°C for up to 3 weeks. Protein concentration was determined according to Bradford (34) (Bio-Rad). Samples were resolved by electrophoresis on 7.5% or 4 -8% gradient denaturing SDS-polyacrylamide gel as indicated. The gel was then dried on a Nytran membrane (Schleicher & Schuell) and exposed for 3-10 days to a PhosphorImager screen before scanning.
Proliferation Assay-Cells were seeded at a concentration of 25,000/ 35-mm dish in duplicate and allowed to grow for 30 h. The cells were then serum starved in 1.5 ml of Opti-MEM (Life Technologies, Inc.) for 18 h following which they were stimulated with GM-CSF at the indicated concentrations for 2 days. [ 3 H]Thymidine (specific activity of 20 Ci/mmol, DuPont NEN) was added (3 Ci/ml) for the last 20 h of incubation. Cells were then collected and retained on fiberglass filters (Schleicher & Schuell) that were sequentially washed with 4 ml of phosphate-buffered saline, 4 ml of 10% trichloroacetic acid, 4 ml of H 2 0, and 4 ml of methanol. Filters were air dried prior to liquid scintillation (EcoLume, ICN, Costa Mesa, CA) counting. Data were analyzed and EC 50 determined with the program ALLFIT (15,31).
Foci Formation-Cells were seeded at a concentration of 125,000/ 35-mm dish, in IMDM supplemented with 10% FCS, in the presence or absence of GM-CSF at the indicated concentrations. The cultures were refed with fresh medium every 3-4 days (35,36). Foci were visible on day 6 -7 and were scored without fixation on day 10 using an inverted microscope.

RESULTS
The Multimeric Nature of the GMR Complex-GM-CSF cross-linking studies were done with clone 13 expressing w/t ␣ and ␤ c and clone 18 expressing w/t ␣ alone. Previous studies occasionally suggested the presence of high molecular mass complexes, in addition to the two major bands corresponding to ␣ and ␤ c (13,21). To determine the size of these high molecular mass complexes, the cross-linked proteins were resolved by electrophoresis in a denaturing continuous 4 -8% polyacrylamide gel. Under these conditions, there was a linear relationship between R F values and the molecular masses of protein standards ranging between 67 and 669 kDa (Fig. 2B). In clone 18, 125 I-GM-CSF cross-linked to a single ␣ chain, as for clone 13 ( Fig. 2A). Furthermore, two additional bands were observed at 186 kDa, corresponding to an ␣ dimer (␣ 2 ) with a single molecule of ligand, as well as a slowly migrating complex of more than 800 kDa. All three bands were observed in clone 13 as well. However, data observed with clone 13 indicate that the ␣ dimer is not favored in the presence of ␤ c . Several additional bands are observed with clone 13 and not clone 18. The most prominent at 160 kDa corresponds to a single molecule of ␤ c cross-linked to the ligand. The two bands of higher molecular masses correspond potentially to ligand ␤ c dimer cross-linking (290 kDa) and to the ternary complex GM⅐␣⅐␤ c (250 kDa) (Fig.   2). That the bands corresponding to GM⅐␣⅐␤ c and GM 2 ⅐␣ 2 ⅐␤ c2 should be fainter than the neighboring bands may also be attributed to inefficient ␣⅐␤ c cross-linking, as reported elsewhere (7,13,23,36). Cross-linking of 125 I-GM-CSF to all observed bands is competed by 125-fold excess cold GM-CSF, indicating their specificity. Our observations therefore suggest a higher order of association within the GMR complex.
Coexpression of COOH-terminal Truncated GMR-␣ (␣ t ) and Wild Type ␤ c in NIH-3T3 Cells-It has been shown previously that the cytoplasmic tail of the ␣ subunit is important for signal transduction through ␤ c (19). Given the importance of the COOH-terminal domain in signal transduction and our crosslinking studies, which revealed the presence of ␣ dimers, we reasoned that an inert COOH-terminal truncated GMR-␣ could inhibit signal transduction through the w/t receptor, if the biologically active complex consists of at least two ␣ chains. We therefore addressed the question of whether or not a fully truncated COOH-terminal ␣ chain could act as a dominant negative receptor over w/t ␣.
To this end, we used three types of NIH-3T3 stable transfectants: clone 18 expressing only w/t ␣, clone 9.2 expressing only w/t ␤ c , and clone 13 expressing both w/t ␣ and ␤ c . The level of ␤ c which is the limiting element for high affinity binding differs between clones 9.2 (lower level) and clone 13 (higher level) (data not shown). The effect of COOH-terminal truncation on ligand binding and signal transduction was first examined by transfecting GMR-␣ t into clone 9.2. Five subclones (9.2 variants) expressing high levels of both GMR-␣ t and w/t ␤ c were selected. Binding assays performed at 200 pM GM-CSF confirmed that the introduction of ␣ t in 9.2 restored ligand binding (Table I). Selected clones were then tested for their proliferative response to GM-CSF in serum-free medium. The positive control, clone 13 expressing w/t GMR-␣ and -␤ c , was shown to respond well to GM-CSF in this assay (Table I and Ref. 16). In contrast, clone 9.2 and its variants showed no proliferation increase on exposure to saturating concentrations of GM-CSF (800 pM) (Table I), confirming that ␣ t was inert.
To verify that the truncation of the cytoplasmic domain of GMR-␣ did not affect its capacity to be expressed at the cell surface, we compared the immunoreactivity of a monoclonal anti-GMR-␣ with NIH-3T3 cells transiently transfected with equal amounts of w/t ␣ and ␣ t . There was no significant difference between the two groups (data not shown). We also compared the capacity of w/t ␣ and ␣ t to associate with ␤ c in order to form a high affinity complex. To this end, various concentrations (0.25-10 g) of truncated or w/t GMR-␣ cDNA in pME18neo (Ϸ6.5 kilobases) were transiently transfected in NIH-3T3 together with 5 g of ␤ c cDNA in pME18 (Ϸ6.0 kilobases) (Table II). Binding assays were then performed at 200 pM radioligand, a concentration that would be sufficient for binding to the high affinity complex (␣⅐␤ c ) with minimal occupancy of the low affinity binding site (␣ alone). Increasing the ␣:␤ c ratio up to 1:1 resulted in increased 125 I-GM-CSF binding (Table II). There was no further increase in binding when ␣ was 2-fold higher than ␤ c , suggesting that at 200 pM radioligand, most of the binding may be attributed to occupancy of the high affinity complex. More importantly, there was no significant difference between ␣ t and w/t ␣ in this assay.
Coexpression of ␣ t , w/t ␣ and ␤ c -We then proceeded to induce ␣ t expression in clone 13, which expresses both w/t ␣ and ␤ c . Twenty-five clones were screened by RT-PCR for obvious variations in ratio of expression of ␣ t versus w/t ␣. Six clones were selected with ratio of ␣ t over w/t ␣ which varied from less than 0.001:1 (13-T6) to 1:0.001 (13-T2) (Fig. 3). These clones were also subjected to binding and proliferation assays to determine their response to GM-CSF. Our data indicated that the binding of 125 I-GM-CSF was not affected by ␣ t expression, which was further confirmed by GM-CSF saturation analyses (Table III). For example, 13-T5 and the w/t clone 13 expressed both high and low affinity GM-CSF binding sites with K d that were comparable. Similarly, the numbers of high and low affinity binding sites/cell were not significantly different between the two clones.
Reduced GM-CSF Responsiveness in Cells Coexpressing ␣ t and ␣-Cross-linking studies were performed to determine the ratio of ␣ t over that of w/t ␣ and ␤ c at the protein levels (Fig. 4). There was a good overall concordance between the levels of mutant relative to w/t proteins and their relative mRNA levels determined by RT-PCR. Thus, clone 13-T6 does not exhibit detectable ␣ t , whereas clones 13-T3 and 13-T5 express equal levels of each type. In contrast, clones 13-T1 and 13-T4 express predominantly ␣ t . Furthermore, 13-T2 expresses very low levels of ␤ c . In this experiment, cross-linked proteins were run on a single 7.5% polyacrylamide gel, and, under these conditions, high molecular mass complexes were not as well resolved as in a continuous 4 -8% polyacrylamide gel (compare Fig. 2 with Fig. 4). Nonetheless, GM-CSF is efficiently cross-linked to ␤ c and to both w/t and ␣ t , suggesting that both w/t ␣⅐␤ c and ␣ t ⅐␤ c complexes exist.
The effect of ␣ t expression on GM-CSF responsiveness was assessed in two biological assays: thymidine uptake and morphologic change. GM-CSF induced a dose-dependent proliferation in serum-free medium in clone 13 (Fig. 5) and foci formation (data not shown). 13-T6, which does not express detectable ␣ t , behaved similarly to the parental clone 13 in both assays (Figs. 5 and 6). In contrast, in clones 13-T3 and 13-T5, which express both mutant and w/t proteins in equal proportion, there was a sharp decrease in GM-CSF responsiveness in the thymidine incorporation assay and a significant shift in EC 50 (Fig. 5 and Table III). Similarly, morphologic change induced by GM-CSF was also reduced drastically in 13-T3 and 13-T5 (Fig. 6). Finally, both cell proliferation and foci formation were minimal in 13-T4 (exhibiting mainly ␣ t expression) in response to GM-CSF.
In summary, our observations indicate a drastic decrease in GM-CSF responsiveness in subclones expressing ␣ t . Since 13-T3 and 13-T5 express both w/t ␣ and ␣ t in stoichiometric proportion, a mere sequestration of the ligand or ␤ c could not explain the observed data. Rather, our observations indicate that a functional GM-CSF receptor comprises at least two ␣ subunits, and they support the hypothesis of a higher order of   complexes at different ␣ to ␤ c ratio NIH-3T3 cells were transiently transfected with the indicated amounts of wild type ␣ (␣ w/t ) or truncated ␣ (␣ t ) together with 5 g of ␤ c as detailed under "Materials and Methods." Binding assays were performed at 200 pM 125 I-GM-CSF. Nonspecific binding was determined in the presence of a 100-fold excess of unlabeled GM-CSF and was subtracted from the total binding. Data shown are the mean Ϯ S.D. of triplicate determinations and are typical of three independent experiments. ␣ i ϭ ␣ w/t or ␣ t .  3. RT-PCR of NIH-3T3 transfectants. RNA was extracted from clone 13 and the various subclones coexpressing w/t GMR and ␣ t (clones 13-T1-T6). Reverse transcription was performed using primer PAD1 and murine S16AS (internal control) as described under "Materials and Methods." One l of the reactions was used for PCR, and 10 l of these reactions was separated by electrophoresis on 1% agarose-TAE gel. Varying ratios of ␣ t to ␣ w/t were observed, whereas the internal control ribosomal MS16 was more constant. association for the active GMR complex. Since we showed previously that ␣ and ␤ c associates in a stoichiometric proportion (13), our observations also suggest in extenso that a functional GMR complex contains at least two ␤ c molecules. DISCUSSION A Dominant Negative GMR-␣-The present study provides evidence that a COOH-terminally truncated GMR-␣ can suppress the function of its w/t counterpart, even when expressed in equal amounts. That the COOH-terminal domain of the ␣ chain might be important for signal transduction was inferred previously from the specificity of response triggered by IL-3R, IL-5R, and GMR, despite their sharing a common signal transducing ␤ chain (for review, see Ref. 37). Indeed, COOH-terminal truncation results in a receptor complex that is no longer competent for signal transduction (11,19, 38 and the present study). A soluble GMR was also previously shown to decrease the response of the cells to GM-CSF (39). The mechanism of inhibition was unclear, however, and could be the result of GM-CSF sequestration by the soluble ␣ chain. Through quantitative cross-linking studies and coexpression of ␣ t and w/t ␣ in stoichiometric proportion, we provide direct evidence that ␣ t acts as a dominant negative mutant of w/t ␣. Moreover, we also verified that ␣ t is not expressed preferentially at the cell surface, nor does it show increased association with ␤ c when compared with w/t ␣. Therefore, mere GM-CSF or ␤ c sequestration was ruled out by the fact that 13-T3 and 13-T5 in which equal amounts of mutant and w/t proteins were present in the GM-CSF binding complexes showed a drastically impaired response to GM-CSF compared with w/t clones. A dominant negative suppression can be inferred from the observations that in cells expressing equal levels of ␣ t and w/t ␣, there was more than a 75% decrease in cell proliferation in response to GM-CSF. A possible explanation may be the composition of the various complexes that are formed. Thus, 25% of the complex would be ␣ w/t2 ⅐␤ c2 and fully functional, whereas the remaining 75% would be inactive since 50% would be ␣ w/t ⅐␣ t ⅐␤ c2 and 25% ␣ t2 ⅐␤ c2 . Comparison between 13-T3 and 13-T5 with 13-T6 or parental shows such a drastic decrease in the amplitude of the response together with a significant shift in EC 50 values, which were 8 -20-fold higher.
Implications for the Structure of the Biologically Active GMR Complex-The negative effect of truncated GMR-␣ on the function of w/t GMR is better explained by the possibility of ␣ homodimerization and the requirement for at least two normal ␣ chains in a GMR complex that is competent for signal transduction. Moreover, recent reports (18,27,40,41) provide evidence for an active ␤ c homodimer complex in BaF3 and FDC-P1 cell lines. Consistent with our observations, IL-6 has been shown to form a hexameric complex with IL-6R and gp130 in the ratio of 2:2:2 in solution phase binding assays (27). Interestingly, two distinct and differently oriented sites on IL-6 have been implicated in gp130 dimer formation, predicting that a similar mechanism may be operating for GM-CSF receptor Binding constants were determined by computer analysis of saturation curves that cover a range of 5 pM to 20 nM 125 I-GM-CSF with the program ALLFIT. Nonspecific binding was determined in the presence of a 100-fold excess of unlabeled ligand. The half-maximal concentration of GM-CSF requested for biologic activity (EC 50 ) is derived by analysis of the GM-CSF dose-response curves for thymidine incorporation shown in Fig. 5, using the program ALLFIT. Despite comparable binding characteristics, there was an 8-fold difference in EC 50 between 13-T5 and parental cells. assembly. Because of structural homologies between ligands and receptors, computer modeling and site-directed mutagenesis of GM-CSF may allow us to address this possibility directly.
Stoichiometry of the GMR Complex-A direct demonstration of the multimeric nature of the GMR complex was provided by our optimized cross-linking and binding studies with three types of NIH-3T3 GMR transfectants. Thus, clone 9.2 expresses a full-length ␤ c but does not bind GM-CSF even at 100 nM radioligand (data not shown). Clone 18 (␣ alone) binds GM-CSF with low affinity only, whereas clone 13 (␣␤ c ) exhibits both high and low affinity binding. On NIH-3T3 cells expressing the ␣ chain alone (clone 18), GM-CSF was shown to cross-link to an ␣ dimer at a 1:2 ratio of GM to GMR-␣. The dimer configuration was not favored by the presence of ␤ c in clone 13. More importantly, comparison of the binding characteristics as well as the cross-linking data between clone 18 (␣ alone) and clone 13 (␣⅐␤ c ) indicate that both the ternary complex GM⅐␣⅐␤ c and possibly an oligomeric complex are present in cells coexpressing ␣ and ␤ c and constitute the high affinity binding site. The dominant negative effect of ␣ t over w/t ␣ indicates that the oligomeric complex is the minimal structure required for delivering a proliferative signal into the cells.
It has been suggested previously that ␤ c could homodimerize because of an activating point mutation analogous to the one observed in neu, known to confer ligand-independent activation. Our observations provide a direct demonstration for the presence of ␣ and ␤ c homodimers and further underscore the possibility of gain of function mutations that cause homodimerization as reported for tyrosine kinase receptors. A truncated EpoR was first thought to be competent for mitogenic signaling but not for suppression of apoptosis. It was later found that the BaF3-EpoR transfectants also expressed low levels of w/t EpoR and that the truncated EpoR had a dominant negative effect over that of w/t EpoR, probably because of dimerization (42).
Although data are not available for Epo, there is recent evidence for IL-6 that the ligand can also form dimer or tetramer, as shown previously for Steel factor, M-CSF, and ligands for other tyrosine kinase receptors (43). Taken together, the results suggest that cytokine receptors may deliver mitogenic signals to the cells through receptor oligomerization. Since associated tyrosine kinases such as Jak-2, Yes, and Lyn have been implicated in signal transduction by GMR (38,44), the observations are consistent with the view that receptor diand/or oligomerization brings together two or more tyrosine kinase molecules, resulting in their activation. Furthermore, our approach provides a more general strategy to designing dominant negative receptors, either as membrane-anchored molecules or as soluble receptors, and to address the functional importance of receptor oligomerization.