Construction and binding kinetics of a soluble granulocyte-macrophage colony-stimulating factor receptor alpha-chain-Fc fusion protein.

Granulocyte-macrophage colony-stimulating factor (GM-CSF) activity is mediated by a cellular receptor (GM-CSFR) that is comprised of an alpha-chain (GM-CSFRalpha), which specifically binds GM-CSF, and a beta-chain (betac), shared with the interleukin-3 and interleukin-5 receptors. GM-CSFRalpha exists in both a transmembrane (tmGM-CSFRalpha) and a soluble form (sGM-CSFRalpha). We designed an sGM-CSFRalpha-Fc fusion protein to study GM-CSF interactions with the GM-CSFRalpha. The construct was prepared by fusing the coding region of the sGM-CSFRalpha with the CH2-CH3 regions of murine IgG2a. Purified sGM-CSFRalpha-Fc ran as a monomer of 60 kDa on reducing SDS-polyacrylamide gel electrophoresis but formed a trimer of 160-200 kDa under nonreducing conditions. The sGM-CSFRalpha-Fc bound specifically to GM-CSF as demonstrated by standard and competitive immunoassays, as well as by radioligand assay with 125I-GM-CSF. The sGM-CSFRalpha-Fc also inhibited GM-CSF-dependent cell growth and therein is a functional antagonist. Kinetics of sGM-CSFRalpha-Fc binding to GM-CSF were evaluated using an IAsys biosensor (Affinity Sensors, Paramus, NJ) with two assay systems. In the first, the sGM-CSFRalpha-Fc was bound to immobilized staphylococcal protein A on the biosensor surface, and binding kinetics of GM-CSF in solution were determined. This revealed a rapid koff of 2.43 x 10(-2)/s. A second set of experiments was performed with GM-CSF immobilized to the sensor surface and the sGM-CSFRalpha-Fc in solution. The dissociation rate constant (koff) for the sGM-CSFRalpha-Fc trimer from GM-CSF was 1.57 x 10(-3)/s, attributable to the higher avidity of binding in this assay. These data indicate rapid dissociation of GM-CSF from the sGM-CSFRalpha-Fc and suggest that in vivo, sGM-CSFRalpha may need to be present in the local environment of a responsive cell to exert its antagonist activity.

for binding studies have been problematic because production by cell lines is limited, and simple purification methods are not available. Therefore, we prepared a DNA-construct fusing the sGM-CSFR␣ to the Fc region of mouse IgG2a. Here we describe the cloning, production, purification, binding, and biological activity of this sGM-CSFR␣-Fc fusion protein, with preliminary analyses of binding kinetics.
These primers were used with cDNAs and Taq polymerase as described previously (14,15) to amplify the sGM-CSFR␣ and Fc. The program of amplification was 94°C for 1.5 min followed by 25 cycles at 94°C for 1 min, 60°C for 1 min, 72°C for 1 min, with a final cycle at 72°C for 5 min. The PCR products were stored at 4°C until used. A 2% analytical agarose-gel electrophoresis revealed a band of 1,100 bp and a band of 650 bp, respectively, for sGM-CSFR␣-and CH2-CH3-amplified products, which were therefore purified on a 1% preparative low melting agarose gel (NuSieve, FMC BioProducts, Rockland, ME), extracted with phenol/chloroform (16), and used in the SOEing step.
The SOEing step program consisted of two phases. In the first phase (94°C for 1 min followed by five cycles at 94°C for 1 min, 60°C for 1 min, 72°C for 1 min) the two amplified cDNA products worked as primers for each other to obtain the construct. In the second phase (94°C for 1 min followed by 25 cycles at 94°C for 1 min, 60°C for 1 min, 72°C for 1 min) the primers sGM-CSFR␣ 5Ј and Fc 3Ј were added and the construct amplified. A 2% analytical agarose gel indicated a SOEing product of the desired size. After treatment of SOEing mixture with proteinase K to destroy Taq polymerase, the construct was extracted following the phenol/chloroform protocol and precipitated with ethanol. The construct and the pBabe vector (17,18) were digested with restriction enzymes BamHI and SalI. After alkaline phosphatase treatment of pBabe purified on a 1% preparative low melting temperature agarose gel, ligation was performed following standard procedures (16,19).
The product was transformed into Escherichia coli DH5␣ cells that were plated into LB plates; and minipreparations were made from colonies. Plasmid DNA from minipreps were digested with BamHI/SalI to identify the presence of the desired insert. After amplification of cells with the right insert, plasmid DNA was purified on a Qiagen column (Qiagen Inc., Chatsworth, CA). The purified DNA was checked using two different pairs of restriction enzymes (BamHI/SalI and BamHI/ XbaI), and the expected size of the inserts was confirmed on a 2% analytical agarose gel (ϳ350 bp and ϳ4,750 bp for the BamHI/SalI digestion, and ϳ1,200 -1,300 bp, ϳ2,500 bp, and ϳ4,300 bp for the BamHI/XbaI digestion). The sequence of the insert was also confirmed by sequencing according to previously published protocols (20).

Transfection of PA317 Cells, Transduction into SP2/0 Cells
PA317 cells (a retroviral packaging line (17,18)) were transfected with the purified DNA using LipofectAMINE reagent (Life Technologies, Inc.) and the described protocol (21). Briefly, 2 ϫ 10 5 cells in 2 ml of DMEM (DMEM containing L-Gln, oxalate/pyruvate/insulin, penicillin/ streptomycin, and 10% fetal calf serum) were seeded in each well of a six-well culture plate and grown until 80% confluence. 1.5 g of DNA in 100 l of serum-free DMEM was added to 12 l of LipofectAMINE reagent diluted to 100 l with serum-free DMEM. After gentle mixing and incubation at room temperature for 45 min to allow the DNAliposome complex to form, the complex solution was diluted to 1 ml with serum-free DMEM and added to the rinsed adherent cells. Cells were kept in a 37°C 5% CO 2 incubator for 5 h and then rinsed, and serum-containing DMEM was added. 72 h after transfection, the cells were passed 1:10 into DMEM containing 5 g/ml puromycin. Cells were grown until confluent, split twice, and grown for 1 week in medium without puromycin. The supernatant was then used to transduce SP2/0 cells.
3 ϫ 10 6 SP2/0 cells, grown in RPMI medium, containing L-Gln, sodium pyruvate, penicillin/streptomycin, 25 mM Hepes, and 15% fetal calf serum, was resuspended in 2 ml of medium with 0.5 ml of PA317 supernatant and 10 l of Polybrene (0.4 mg/ml). After 3 h of incubation, 8 ml of RPMI medium was added; after 2 days of incubation, cells were grown in medium containing puromycin. SP2/0 cells were then cloned and the presence of insert confirmed by PCR. An enzyme-linked immunosorbent assay (ELISA) (see below) was performed on PCR-positive clone supernatants to identify the clones producing the fusion protein.

Purification and Characterization of sGM-CSFR␣-Fc
Clones that presented the highest binding to GM-CSF in two ELISAs were expanded, and the fusion protein from the supernatants was purified on a staphylococcal protein A (SPA) column. To 140 ml of filtered supernatant glycine and NaCl were added to reach, respectively, 1.5 and 3 M final concentrations; the pH was adjusted to 8.0. After loading the sample, the column was washed with 10 mM boric acid, 3 M NaCl, pH 8.9, and eluted with 0.1 M sodium citrate, pH 3.5 (fractions collected into tubes containing 1 M Tris, pH 9.5). Elution fractions containing proteins were concentrated using Centricon-30 (30 kDa cutoff) filtration systems (Amicon, Beverly, MA).
The sGM-CSFR␣-Fc fusion protein was analyzed by SDS-polyacrylamide gel electrophoresis. An 8% acrylamide gel, 1.5 mm thick, was used in an electrophoresis system to analyze both oxidized and reduced forms of the fusion protein. 45 g of purified sGM-CSFR␣-Fc fusion protein purified from various cell clones was loaded into gels, run at 100 volts for ϳ1.5 h using 25 mM Tris, 250 mM glycine, 0.1% SDS, pH 7.5, and then stained with Coomassie Blue.

Western Blot Analysis
Western blot analysis was performed using mAbs against GM-CSFR. 75 g/well of both oxidized and reduced SPA-purified sGM-CSFR␣-Fc fusion protein were loaded into three wells of each of two 8% polyacrylamide gels. After the electrophoretic run as reported previously, the gels were cut in thirds, each strip containing two wells, one with the sample and one with the high molecular mass markers, and transferred to Immobilon-P membranes (Millipore, Bedford, MA) for 1 h at room temperature, under stirring conditions, using 50 mM Tris, 150 mM NaCl, 20% methanol, pH 7.5, as the transfer buffer. The membranes were then blocked overnight at 4°C with 5% non-fat dry milk in 50 mM Tris, 150 mM glycine, 0.05% Tween, pH 7.5 (5% NFDM in TTBS). The next morning they were washed three times for 10 min with TTBS, and the three pairs of oxidized and reduced membrane strips were incubated for 2 h at room temperature with one of the following monoclonal antibodies diluted in TTBS: (a) 7 g/ml mouse IgG2a anti-human GM-CSFR (Pharmingen, San Diego) (this antibody binds GM-CSFR on the same site recognized by GM-CSF); (b) 7 g/ml mouse IgM anti-human GM-CSFR (Pharmingen) (this antibody binds GM-CSFR on a site different from that recognized by GM-CSF); (c) no first antibody. The membranes were then washed three times for 2 min with 5% NFDM in TTBS and twice for 2 min with TTBS. Samples from a and b were then incubated for 2 h at room temperature, under shaking conditions, with a 1:40,000 dilution of biotin-goat anti-mouse IgG (Fab-specific) (Sigma) in TTBS. After washing the membranes three times for 2 min with 5% NFDM in TTBS and twice for 2 min with 1 ϫ TTBS, they were incubated for 2 h at room temperature, under shaking conditions with 0.5 g/ml avidin-horseradish peroxidase (HRP; Pierce) in 1 ϫ TTBS; sample c was incubated under same conditions with goat anti-mouse IgG (HϩL) HRP (Life Technologies, Inc.). The membranes were washed six times with 1 ϫ TTBS, and the TMB substrate solution was added (Kirkegaard and Perry Laboratories, Gaithersburg, MD) until color development. The color reaction was then stopped by distilled water rinsing. coated with 50 l of 10 g/ml GM-CSF (Sargamostin Leukine, Immunex Corporation, Seattle) in 0.1 M NaHCO 3 , and after overnight incubation at 4°C they were washed five times with 1 ϫ PBST and blocked with 200 l of 2% bovine serum albumin in 1 ϫ PBST at 37°C for 1 h. The plates were washed five times with 1 ϫ PBST, and 50 l of clone supernatants was added, incubated overnight at 4°C, and washed seven times with cold 1 ϫ PBST. 100 l of 1/3,000 cold dilution in 1 ϫ PBST of goat anti-mouse IgG (HϩL) HRP was added per well and the plate kept overnight at 4°C. After seven washes with cold 1 ϫ PBST, 100 l of 0.1 mg/ml substrate TMB dihydrochloride (Sigma) in 0.05 M phosphate citrate, 0.03% sodium perborate buffer, pH 5.0, was added per well. After color development at room temperature for 10 min, the enzymatic reaction was stopped with 20 l of 2 N H 2 SO 4 per well and the plate read at 450 nm.
In the second ELISA the wells were coated with 50 l of 5 g/ml SPA (Sigma) in 0.1 M NaHCO 3 for 1 h at 37°C, washed five times with 1 ϫ phosphate-buffered saline with 0.1% Tween-20 (PBST), and blocked with 200 l of 2% bovine serum albumin in 1 ϫ PBST at 37°C for 1 h. The wells were washed five times with 1 ϫ PBST, 150 l of clone supernatants was added, and the wells were incubated overnight at 4°C and washed seven times with cold 1 ϫ PBST. 50 l of 10 ng/ml biotinylated GM-CSF (developed by standard methods (22)) was added per well, and the plate was incubated overnight at 4°C, washed seven times with cold 1 ϫ PBST, and 100 l of 0.1 g/ml avidin-HRP conjugate added per well. After incubation overnight at 4°C and seven washes with cold 1 ϫ PBST, 100 l of substrate was added to each well, and after color development at room temperature for 10 min, the enzymatic reaction was stopped with 20 l of 2 N H 2 SO 4 per well and the plate read at 450 nm.
The third was a competitive ELISA, used to test highly concentrated fusion protein preparations after purification on an SPA column. An ELISA plate was coated with SPA and blocked as reported previously. 50 l of 1 mg/ml sGM-CSFR␣-Fc was added per well and incubated overnight at 4°C. After three washes with 1 ϫ PBST, 50 l of unmodified ("cold") GM-CSF was added at various dilutions and incubated for 1 h at 37°C. 50 l of biotinylated-GM-CSF at 10, 1, and 0.1 ng/ml was added per well, and the plate was incubated for another hour at 37°C. After washing, binding was detected by avidin-HRP as reported above.

Radioligand Assay
Binding of purified sGM-CSFR␣ to 125 I-GM-CSF was also tested as follows. The wells of a radioimmunoassay plate (Wallac, Gaithersburg, MD) were coated with 50 l of 5 g/ml SPA overnight at 4°C, washed three times with PBST, and blocked with 200 l of 2% bovine serum albumin in 1 ϫ PBST at 37°C for 2 h. The plate was washed three times with PBST and then 50 l of purified fusion protein per well was added at 500, 250, 125, 62.5, 31, 15.5, and 0 g/ml in 2% bovine serum albumin and PBST; as positive control anti-GM-CSF mAb 126.213 was used at same dilutions. After a 3-h incubation at 37°C, the plate was washed three times with cold PBST and incubated with 25 l of 388 pM 125 I-GM-CSF (1 ϫ 10 5 to 3 ϫ 10 5 cpm) per well, with or without a saturating amount (100 nM) of cold GM-CSF at room temperature for 1 h. The plate was quenched on ice, washed three times with ice-cold PBST, and the wells were cut out from the plate, transferred into tubes, and the cpm counted.

Proliferation Assay
Inhibition of GM-CSF-dependent cells line MO7E (24 h) (from R. Zollner, Genetics Institute, Cambridge, MA) by sGM-CSFR␣ was tested as follows. 10 4 cells/well were incubated for 1 day with or without 250 pM GM-CSF in the presence of different concentrations of sGM-CSFR␣ (250, 50, 10, 2, and 0 g/ml). 20 l of 5 mg/ml MTT (Sigma) in 1 ϫ PBS was added per well, and after 5 h 100 l of 10% SDS in 0.1 N HCl was used to solubilize the precipitated crystals. After overnight incubation at 37°C, A 570 nm was detected. A parallel assay was set as control, using IL-2-dependent CTLL cells (5 ϫ 10 3 cells/well) in the presence of 20% concanavalin A supernatant as the stimulus.

Biosensor Assay
An IAsys biosensor was used to characterize binding kinetics of sGM-CSFR␣-Fc to GM-CSF. All experiments were done at 25°C.

Immobilization and Regeneration Conditions
GM-CSF and SPA immobilization on a carboxymethyl-dextran cuvette (IAsys) was performed in the instrument using reagents supplied in the coupling kit as recommended by the manufacturer. Briefly, the cuvette was first washed with 200-l washes of alternatively deionized water and 10 mM HCl to swell the cuvette matrix and finally equilibrated in PBST. The carboxymethyl-dextran layer on the sensor chip was then activated by a 10-min incubation with a mixture of 100 l of 13 mg/ml N-hydroxysuccinimide and 100 l of 76 mg/ml EDC to form succinimidyl esters that are reactive with amino groups. After washes with PBST, 200 l of 10 g/ml GM-CSF or SPA in 10 mM sodium acetate, pH 5, was added to the cuvette, and after a response plateau was reached signifying completion of the reaction, the cuvette was washed out with PBST. The remaining activated groups were blocked by injection of 200 l of 1 M ethanolamine, pH 8.5, for 2 min. After reequilibration with PBST, the cuvette was ready for the binding experiments. The cuvette matrix was regenerated to remove bound ligand using 200 l of either 0.5 M sodium carbonate, pH 11.0, or 10 mM HCl and 0.5 M NaCl and reequilibrated with PBST.

Binding Assays
Immobilized GM-CSF-200 l of sGM-CSFR␣-Fc at 75, 300, 600, and 900 nM in PBST was added to the cuvette. After a binding plateau was achieved it was incubated with PBST and the cuvette regenerated for the next experiment.
Immobilized SPA-A 200-l solution containing 20 g of sGM-CSFR␣-Fc in PBST was added to the cuvette and allowed to bind for 6 min (phase 1). PBST was added for 5 min (phase 2). GM-CSF at various concentrations in PBS was then added and allowed to bind for 5 min (association phase, phase 3) for 5 min. PBST was then added for an additional 5 min (dissociation phase, phase 4). No regeneration was performed between runs in these assays.

Data Analysis
Data were collected automatically and analyzed subsequently using the Microsoft Excel program (Microsoft, Redmond, WA). Plots were fitted using Cricket Graph (Cricket Software, Malvern, PA). The sGM-CSFR␣-Fc surface prepared by capturing sGM-CSFR␣-Fc on immobilized SPA was corrected for a downward drift as dissociation of sGM-CSFR␣-Fc from the SPA occurred along with GM-CSF binding to sGM-CSFR␣-Fc. Accordingly, control experiments were run in which sGM-CSFR␣-Fc was allowed to bind to SPA for 6 min (phase 1), the receptorassociated surface was washed for 5 min (phase 2), and then phosphatebuffered saline was added for an additional 5 min (to simulate the association phase of GM-CSF, phase 3) followed by a final 5-min wash (simulating the dissociation phase of GM-CSF from sGM-CSFR␣-Fc, phase 4). In these experiments, the negative slope of the dissociation of GM-CSFR␣-Fc from SPA decreased slightly after each of the washes (in phases 2, 3, and 4). Therefore, correction for sGM-CSFR␣-Fc dissociation from SPA was performed separately for each phase. The slope of dissociation of sGM-CSFR␣-Fc from SPA in the final 200 s of phase 2 was calculated for each experiment and used to normalize this portion of the sensorgrams. Similarly, the slope of the final 200 s of phase 4 was calculated and used to correct the dissociation phase of GM-CSF from sGM-CSFR␣-Fc. A weighted average of the slopes from phases 2 and 4 was used to predict a slope for phase 3, and this was used to correct the association phase of GM-CSF to the sGM-CSFR␣-Fc. Evaluation of these corrections on the mock experiments gave reasonable agreement with the observed data (data not shown). This triphasic correction was used for each experiment with immobilized SPA.
Kinetic analysis was performed as described previously (23). In the case of a bimolecular interaction of two species A and B whose association and dissociation are regulated, respectively, by k on and k off for giving the final product AB, If B is immobilized on the cuvette, the response R is proportional to the amount of product AB, thus, where R max is the maximum response. Therefore, a plot of dR/dt versus R will have a slope of Ϫ(k on [A] ϩ k off ) ϭ Ϫk s , and a plot of k s versus [A] will have a slope of k on .
In the case of the dissociation phase, any free A from dissociation of the product is assumed to be removed by the buffer so that there is no reassociation ([A] ϭ 0). Therefore, Therefore, the slope of a plot of ln(R 0 /R) versus time represents k off . In practice, reassociation may not be eliminated by the wash, and its effect may be detected in interactions characterized by fast-on fast-off kinetics such as this one. This can result in the ln(R 0 /R) versus time plots deviating from a straight line. The equilibrium dissociation constant (K D ) was calculated from the equation and from a plot of R/concentration versus R, analogous to a Scatchard plot of Bound/Free versus Bound, where the slope corresponds to Ϫ1/K D .

Cloning of sGM-CSFR␣-Fc
We had previously cloned a soluble form of the GM-CSFR␣ into a retroviral vector (12). We used this clone, which encodes for sGM-CSFR␣, to prepare a fusion protein of the sGM-GM-CSFR␣ with the Fc portion of murine IgG. The sGM-CSFR␣-Fc construct (shown in Fig. 1) was obtained as follows. sGM-CSFR␣ cDNA from clone 9 and the CH2-CH3 cDNA region of mouse anti-reovirus mAb 9BG5 (IgG2a) were PCR amplified using the primers described under "Materials and Methods." An analytical 2% agarose gel indicated a band of ϳ1,100 bp and a band of ϳ650 bp, respectively, for sGM-CSFR␣ and CH2-CH3. After purification on a 1% low melting temperature agarose gel the two products were fused in a PCR-SOEing step, obtaining a construct of ϳ1,700 bp. The construct was digested and ligated into the BamHI/SalI site of the pBabe-puro vector. After transformation of E. coli DH5␣ cells, a colony with the right sized insert was selected, grown, and DNA purified on a Qiagen column. The construct was verified by restriction enzyme mapping and sequencing as described under "Materials and Methods."

Transfection and Expression of the Fusion Protein
sGM-CSFR␣-Fc in the pBabe vector was transfected into a retroviral packaging cell line, and the virus-containing medium was used to transduce SP2/0 cells. These SP2/0 cells were selected and subcloned as described under "Materials and Methods." Clones were tested for the presence of the construct by reverse transcriptase PCR, with 9 of 96 clones expressing the sGM-CSFR␣-Fc by this assay. Several subclones were isolated and grown to confluence; the selective media were removed and supernatants collected. These supernatants were screened for production of sGM-CSFR␣-Fc by two different ELISAs as described under "Materials and Methods." An example of one ELISA is shown in Fig. 2, comparing the binding of clone supernatants with that of neutralizing anti-GM-CSF mAb 126.213 (24). In this example, the ELISA plate was coated with SPA to which the sGM-CSFR␣-Fc was bound. Biotinylated GM-CSF was then added, and binding was detected by streptavidin-HRP conjugate. This assay shows significant binding of several clones to GM-CSF. Similar results were obtained in an assay in which the ELISA plate was coated with GM-CSF, then the sGM-CSFR␣-Fc was bound and binding detected by anti-mouse Ig-HRP (data not shown). Clones that consistently displayed binding were chosen for further analysis.

Purification and Characterization of sGM-CSFR␣-Fc
sGM-CSFR␣-Fc was purified from ϳ140 ml of clone supernatant on an SPA column, as described under "Materials and Methods." The partially purified sGM-CSFR␣-Fc was analyzed in an 8% acrylamide gel (Fig. 3). In the gel run under oxidizing conditions there is a high molecular mass band of ϳ160 -200 kDa, representing multimers of the fusion protein, whereas in the gel run under reducing conditions there are two main bands, at ϳ60 kDa and ϳ43 kDa (sGM-CSFR␣-Fc monomer and uncharacterized contaminant, respectively). We scanned the gel using Image 1.41 (Wayne Rasband, NIMH, Bethesda, MD) and performed densitometry, allowing us to estimate that the sGM-CSFR␣-Fc fusion protein represents ϳ50% of the total protein purified by the SPA column. Gel filtration chromatography (using a Sephacryl S-200HR column) was used to evaluate the molecular mass of the sGM-CSFR␣-Fc multimer. The chromatogram indicated that the oxidized sGM-CSFR␣-Fc runs at ϳ160 kDa (data not shown). When compared with the reducing SDS-polyacrylamide gel electrophoresis of the fusion protein, this suggests that the sGM-CSFR␣-Fc forms trimers. We have seen the ϳ60 -200-kDa form from several sGM-CSFR␣-Fc preparations consistently and reduce to ϳ60 kDa upon reduction consistently.
Western blot analysis with anti-human-GM-CSFR mAbs to detect the receptor portion of the fusion protein and antimouse-IgG to detect the Fc portion of the fusion protein indicated a high molecular mass band for the oxidized forms (sGM-CSFR␣-Fc trimer) and a strong band at 60 kDa for the reduced form (sGM-CSFR␣-Fc monomer) (Fig. 4).

Binding Analysis of sGM-CSFR␣-Fc
The SPA-purified sGM-CSFR␣-Fc was evaluated next for binding to GM-CSF and for bioactivity. A competitive ELISA assay (Fig. 5) was carried out to confirm that the sGM-CSFR␣-Fc bound to native GM-CSF. An ELISA plate was coated with SPA followed by the sGM-CSFR␣-Fc. We then added unlabeled GM-CSF at various dilutions, followed by biotinylated GM-CSF, with binding detected by an avidin-HRP conjugate. We saw increasing binding with increasing amounts of biotinylated GM-CSF (compare the A 450 nm without competitor for 7, 70, and 700 pM biotinylated GM-CSF), and this was inhibited competitively by increasing amounts of unlabeled GM-CSF. This indicates that unmodified GM-CSF binds to our fusion protein. A radiolabeled binding assay (Fig. 6) confirmed the binding of the sGM-CSFR␣-Fc fusion protein to 125 I-GM-CSF. In this experiment, increasing the amount of added fusion protein increased the binding 125 I-GM-CSF, and this binding was blocked fully by excess cold GM-CSF. This confirms the binding of GM-CSF to the sGM-CSFR␣-Fc fusion protein.

Bioactivity of sGM-CSFR␣-Fc
We next evaluated the ability of the sGM-CSFR␣-Fc to inhibit the biological activity of GM-CSF using a GM-CSF-dependent cell line. MO7E cells are a myelomonocytic cell line and are dependent on GM-CSF for growth. We evaluated the growth of MO7E cells in the presence of GM-CSF with or without added sGM-CSFR␣-Fc using the MTT assay (see "Materials and Methods"). The presence of increasing amounts of sGM-CSFR␣-Fc resulted in increasing inhibition of prolifera-tion of the MO7E cells (Fig. 7). We found that 250 g/ml sGM-CSFR␣-Fc produced 75% inhibition in cellular proliferation. In contrast, the sGM-CSFR␣-Fc had no effect on the growth of the IL-2-dependent cell line CTLL (data not shown). This indicates that the sGM-CSFR␣-Fc is a specific biological antagonist of GM-CSF.

Biosensor Analysis
GM-CSF Binding to sGM-CSFR␣-Fc-The sGM-CSFR␣-Fc was evaluated for the kinetics of binding using a biosensor. Initially, the binding of sGM-CSFR␣-Fc to immobilized SPA was analyzed to assure that the Fc portion of our receptor was functional. A typical sensorgram obtained is shown in Fig. 8A. After the addition of sGM-CSFR␣-Fc, a bulk phase effect is seen initially (increase in the response from bulk phase refractive index) followed by an association phase. With washing, an initial rapid response decrease (likely from bulk phase refractive index decrease) becomes relatively exponential. Fig. 8B shows a typical dR/dt plot for the association phase. The initial rapid response shift (likely bulk phase effect) is shown in open circles, followed by a relatively linear association shown in filled circles. The plot of dR/dt for the linear phase shows a k s of 2.51 ϫ 10 Ϫ3 /s for this concentration of sGM-CSFR␣-Fc.
Calculation of the off-rate (k off ) is shown in Fig. 8C. Departure from linearity is seen, possibly caused by rebinding, which increases with time. The plot of ln(R 1 /R n ) versus time for the initial phase yields a k off of 2.00 ϫ 10 Ϫ3 /s for this interaction. The relative linearity of the dissociation phase of sGM-CSFR␣-Fc from SPA seen in Fig. 8A encouraged us to attempt to analyze a monovalent interaction of GM-CSF to the sGM-CSFR␣-Fc bound to SPA. This was a result of our desire to analyze a monovalent interaction of GM-CSF with the binding sites on the sGM-CSFR␣-Fc, as sGM-CSFR␣ is monovalent. We felt that the linearity of the dissociation of sGM-CSFR␣-Fc from SPA would allow us to correct for this while analyzing the interaction of GM-CSF with the sGM-CSFR␣-Fc.
For these experiments, a standardized protocol was developed. sGM-CSFR␣-Fc was allowed to bind to SPA for 6 min  A and B) and reducing (lanes C and D) conditions. The corrected molecular masses of prestained molecular mass standards are also shown.
(phase 1), followed by a 5-min wash (phase 2). GM-CSF was then added and allowed to bind (association phase, phase 3) for 5 min, followed by a final dissociation phase of 5 min (phase 4). Preliminary experiments performed with phosphate-buffered saline added in place of GM-CSF at the initiation of phase 3 revealed that the downward slopes of phases 2, 3, and 4 increased sequentially. We therefore corrected for the downward slopes of these three phases separately, as noted under "Materials and Methods." The data from the final 200 s of phases 2 and 4 were used to correct these phases. An intermediate slope was calculated for phase 3 and used to correct this phase.
A typical sensorgram, with the predicted "base lines," is shown in Fig. 9A. The values for the predicted base lines were subtracted from the data points in the sensorgram and yielded the corrected sensorgram shown in Fig. 9B. A relatively rapid initial association phase is apparent, which is supplanted by a more gradual association phase. dR/dt plots of these data were nonlinear (data not shown). Departure from the linearity expected for a single bimolecular interaction (Equation 3) could be caused by multiple modes of interaction with different affinities, cooperativity, or other complex models. Analysis of the dissociation phase is shown in Fig. 9, C-F. The relative linearity of the plot of ln(R 1 /R n ) versus time was seen for several concentrations of GM-CSF, including 400, 800, 1,600, and 3,200 nM. This linear relationship was maintained for at least the first 100 s of the dissociation phase. The calculated k off was remarkably stable for these different experiments, ranging from 2.32 ϫ 10 Ϫ2 to 2.54 ϫ 10 Ϫ2 /s (average 2.43 Ϯ 0.12 ϫ 10 Ϫ2 /s).
Immobilized GM-CSF-The binding of various concentrations of the sGM-CSFR␣-Fc trimer to immobilized GM-CSF was measured using the conditions described under "Materials and Methods." Fig. 10A shows an overlay of sensorgrams obtained with 75, 300, 600, and 900 nM sGM-CSFR␣-Fc. After an association phase showing the binding of sGM-CSFR␣-Fc to immobilized GM-CSF, a washing step with buffer alone was used to effect the dissociation phase.
The association phase of each sensorgram was analyzed by plotting the change in response over time (dR/dt) versus the response (R) according to Equation 3. The dR/dt plots, shown in Fig. 10B, were nonlinear, with at least two phases apparent. This departure from linearity was similar to that seen in experiments with sGM-CSFR␣-Fc bound to immobilized SPA, suggesting a complex process irrespective of the assay orientation. The dissociation phase was analyzed by assuming that it represents the irreversible release of sGM-CSFR␣-Fc and so should follow an exponential decay. Data were plotted as the ln(response at time zero of dissociation/response at time n) versus time according to Equation 5 (Fig. 10C). Again, nonlinearity was observed, but the initial ϳ140 s were quite linear, so rates were calculated for these time points as has been reported previously (25). The dissociation rate constant (k off ) was calcu-lated as the slope of these plots. For this calculation, the highest concentration of sGM-CSFR␣-Fc was used because this concentration gave the most accurate dissociation curve. The value of k off was calculated as 1.57 ϫ 10 Ϫ3 /s Ϫ1 . This indicates a slower off-rate for this experimental configuration than that shown in Fig. 9, C-F (see above), probably because of the multivalent nature of sGM-CSFR␣-Fc. DISCUSSION In this paper we report the preparation and characterization of a fusion protein obtained by expression of a construct made from sGM-CSFR␣ and the Fc portion of a mouse antibody. The fusion protein is produced easily by transduced cells, recovered from their medium, and purified on an SPA column via the Fc portion. This protein should provide a useful means for the study of the interaction of GM-CSF or GM-CSF mimics with the GM-CSFR␣.
After amplification and SOEing of the sGM-CSFR␣ cDNA and the CH2-CH3 region of mouse anti-reovirus mAb 9BG5 (IgG2a), a construct of the expected size was obtained and ligated into pBabe-puro vector (Fig. 1). This construct was used to produce recombinant retrovirus, which was used to transduce SP2/0 cells. After subcloning the sGM-CSFR␣-Fc-transduced SP2/0 cells, supernatants were tested for their binding to biotinylated GM-CSF. For all supernatants we obtained low values in the ELISAs ( Fig. 2 and data not shown), probably due in part to the low concentration of fusion protein in the supernatants and in part to the rapid off-rate of GM-CSF (see below). However, several positive clones were obtained which allowed more extensive analysis of purified sGM-CSFR␣-Fc.
Denaturing polyacrylamide gel analysis of SPA-purified sGM-CSFR␣-Fc under reducing and nonreducing conditions indicated a similar pattern of molecular masses for sGM-CSFR␣-Fc purified from two different clones (Fig. 3). Moreover, both polyacrylamide gel electrophoresis under nonreducing conditions and gel filtration chromatography of the purified oxidized protein indicated that the fusion protein was present as a multimer, probably a trimer of ϳ160 -200 kDa. SDSpolyacrylamide gel electrophoresis performed under reducing conditions indicated a ϳ60-kDa protein (Fig. 3), which represented ϳ50% of the total protein in the preparation. Also seen was a ϳ45 kDa band (which was absent in other preparations) and a smaller band of ϳ25 kDa which is seen at the dye front of the gel in Fig. 3. A gel run in the same conditions alongside fetal calf serum as a control shows bands of ϳ60 and ϳ25 kDa in the fetal calf serum (data not shown). Thus, it is likely that there is antibody contamination from the fetal calf serum used to grow our sGM-CSFR␣-Fc-transduced SP2/0 cells. Bovine IgG under oxidizing conditions likely co-migrates with the multimeric sGM-CSFR␣-Fc, with the bovine IgG heavy chain also co-migrating with the sGM-CSFR␣-Fc under reducing conditions. Alternatively, the other bands seen on the reducing gel could represent proteins that covalently attach to the sGM-CSFR␣-Fc. We think this is unlikely as the quantity of contaminating proteins varies in our different preparations. In addition, a two-dimensional gel run under reducing conditions showed a single band with an estimated pI of 6.7 (data not shown).
The nature of the ϳ60 kDa band was investigated by Western blot analysis, using antibodies against the GM-CSFR␣ and murine IgG. The experiment indicated that the ϳ60 kDa band is detected by two different anti-human GM-CSFR monoclonal antibodies, recognizing the receptor portion of the fusion protein, and by anti-mouse IgG, recognizing the Fc portion of the sGM-CSFR␣-Fc fusion protein (Fig. 4). The other contaminating bands were not recognized by either reagent, making it unlikely that these represent breakdown products. strated by the plots of both dR/dt versus R and k s versus concentration. In the dR/dt versus R plot (Fig. 10B) we represented the first linear portion of the curves for 75, 300, and 600 nM concentrations, but we omitted from the plot the 900 nM concentration. This high sGM-CSFR␣-Fc concentration resulted in very rapid binding in the initial association phase (Fig. 10A). It was therefore impossible to obtain enough data points to draw a reliable dR/dt versus R plot for this concentration. Moreover, with such a rapid association rate, mass transfer becomes the rate-limiting factor in the initial stage of the binding. We therefore declined to calculate a formal k on from these data.
For analysis of the dissociation phase, we chose to analyze only the plot for the highest concentration of sGM-CSFR␣-Fc, 900 nM, because this had the highest response/background ratio. This analysis did not fit the expected exponential decays shown by the curvature of the plot in Fig. 10C. Nonlinearity in dissociation phase analysis has also been seen in the interaction of insulin-like growth factor and insulin-like growth factorbinding protein (26). In this case the effect was attributed to either long term maturation of the analyte-ligand complex or functional heterogeneity of the immobilized ligand. Other potential explanations include multiple sites with different affinities, cooperativity in binding, rebinding, and mass effects. Curve-fitting the data in Fig. 10C showed a fit to the sum of exponentials, and this is perhaps the result of differential binding characteristics of sites within the sGM-CSFR␣-Fc trimer, depending on the number of active receptor sites already in-volved in binding to immobilized GM-CSF. In the absence of a definite physical explanation the data were analyzed by taking the slope of the curve in the first 120 s. In support of this method of analysis are its simplicity and the consistency of the calculated k off with the average value from experiments with different sGM-CSFR␣-Fc concentrations. The k off value determined by this method is 1.57 ϫ 10 Ϫ3 /s. This is ϳ1 order of magnitude slower than for the reverse experimental configuration (see Fig. 9, C-F). This indicates that the trimer formation of sGM-CSFR␣-Fc slows the off-rate, likely because of an avidity effect.
Thus, our data indicate that the sGM-CSFR␣-Fc interaction with GM-CSF is characterized by a rapid dissociation phase, implying that the major energy of binding is contributed by a fast on-rate. These kinetics have relevance to the biological activity of the sGM-CSFR␣. The ability of the sGM-CSFR␣ to function as a biological antagonist is likely caused by competition for free GM-CSF with the transmembrane form of the receptor (tmGM-CSFR␣) present on cells. Given the relatively fast off-rate of the sGM-CSFR␣-Fc, the antagonist activity of sGM-CSFR␣ is then dependent on its continued presence, because once the sGM-CSFR␣ diffuses away, the GM-CSF would dissociate and then be available for binding to the tmGM-CSFR␣. As noted above, the t 1/2 for dissociation is on the order of 27-30 s. Thus, for sGM-CSFR␣ to sequester GM-CSF and remove it from the local environment of a responsive cell, diffusion away from the cell would have to be more rapid than this dissociation rate. Slower rates of diffusion would imply FIG. 9. Kinetics of GM-CSF binding to sGM-CSFR␣-Fc bound to SPA. Panel A, typical sensorgram obtained for GM-CSF binding to sGM-CSFR␣-Fc which was prebound to SPA, as in Fig. 8. 20 g of sGM-CSFR␣-Fc was bound to SPA for 6 min (as in Fig. 10A) at which point the sGM-CSFR␣-Fc was washed out. GM-CSF binding was initiated 5 min after the initial wash-out of sGM-CSFR␣-Fc. The 200 s before the addition of GM-CSF is shown. In this experiment, 800 nM GM-CSF was added. The association phase lasted 5 min, at which point the GM-CSF was washed out for an additional 5 min. Each circle indicates a data point. Two processes are evident: dissociation of sGM-CSFR␣-Fc from SPA (the progressive down-slope), and the association/dissociation of GM-CSF with the sGM-CSFR␣-Fc. The predicted base line (in fact, the continuous dissociation of sGM-CSFR␣-Fc from SPA) was determined as noted under "Materials and Methods" and is shown for the three phases of the experiment (before GM-CSF was added, the association phase, and the dissociation phase). Panel B, the same sensorgram shown in Fig. 10A corrected for dissociation of sGM-CSFR␣-Fc from SPA. The values for the predicted base line were subtracted from the data points shown in Fig. 8A to produce this corrected sensorgram. Panels C-F, dissociation rate constant calculations for 400 nM (C), 800 nM (D), 1,600 nM (E), and 3,200 nM GM-CSF (F). The dissociation phase of the corrected sensorgrams (as shown in panel B) were replotted as the ln(R 1 /R n ) versus time. Data points are shown after a steep initial decline in RU (see Fig. 10B), likely resulting from mass effect. A line fitted to the data was calculated, with a slope corresponding to the k off values, which were 2.53 ϫ 10 Ϫ2 /s (C), 2.34 ϫ 10 Ϫ2 /s (D), 2.32 ϫ 10 Ϫ2 /s (E), and 2.54 ϫ 10 Ϫ2 /s (F). that the sGM-CSFR␣ needs to be present in the vicinity of the cell to act as a competitive inhibitor.
In terms of understanding the high affinity sites formed by tmGM-CSFR␣ and ␤ c , these data would suggest that the contribution of ␤ c is mostly to slow the off-rate, as the on-rate is already relatively rapid. This is supported by experiments reported by Gearing et al. (4), where binding of 125 I-GM-CSF to low affinity sites on HL-60 cells was lost after a short (10-min) incubation in the absence of 125 I-GM-CSF, but the high affinity sites remained. We would postulate, based on our data and theirs, that the primary role of the GM-CSFR␣ in the complex is to capture GM-CSF with a rapid association phase and that the ␤ c functions primarily to slow the dissociation phase. It is interesting to speculate on the role of the ␤-subunit (␤ c ) in slowing the dissociation of GM-CSF from the heterodimeric receptor. Studies of GM-CSF mutants support a key role for Glu-21 in binding to the high affinity but not the low affinity receptor (8,(27)(28)(29)(30)(31). These authors would argue for direct binding of the GM-CSF A-helix, centered on residue Glu-21, to the ␤ c . Such direct binding would account for the slower off-rate seen with the heterodimeric receptor. However, direct evidence is lacking for such an interaction, and conformational changes or aggregation of the GM-CSFR␣ imparted by ␤ c could also account for the slower off-rate seen. Clarification of this matter awaits direct binding data with GM-CSFR␣, ␤ c , and GM-CSF.
When compared with data from other cytokine receptor interactions, the kinetics of binding for GM-CSF to the sGM-CSFR␣-Fc are similar to those reported for other cytokines, such as IL-5 (25). The characteristics of rapid on-rates and somewhat slower, but still rapid, off-rates may be a general characteristic of cytokine-receptor interactions for four-helix bundle cytokines.