Human stem cell factor dimer forms a complex with two molecules of the extracellular domain of its receptor, Kit.

Stem cell factor (SCF) is a cytokine that is active toward hematopoietic progenitor cells and other cell types, including germ cells, melanocytes, and mast cells, which express its receptor, the tyrosine kinase, Kit. SCF exists as noncovalently associated dimer at concentrations where it has been possible to study its quaternary structure; it stimulates dimerization and autophosphorylation of Kit at the cell surface. We have used recombinant versions of human SCF and human Kit extracellular domain (sKit) to study SCF-Kit interactions. By size exclusion chromatography, plus various physical chemical methods including light scattering, sedimentation equilibrium, and titration calorimetry, we demonstrate the formation of complexes containing a dimer of SCF (unglycosylated SCF1-165) plus two molecules of sKit. The concentrations of SCF and sKit in these studies were in the range of 0.35-16.2 microM. The data are analyzed and discussed in the context of several possible models for complex formation. In particular, the sedimentation data are not consistent with a model involving cooperative binding. The Kd estimate for SCF-sKit interaction, obtained by sedimentation equilibrium, is about 17 nm at 25 degrees C. With glycosylated SCF1-165, the Kd is considerably higher.

Stem cell factor (SCF) 1 is encoded by the steel (Sl) locus in the mouse (1)(2)(3), and is the ligand for the tyrosine kinase type receptor, Kit (1,3,4). Kit in turn is encoded by the dominant white spotting (W) locus in the mouse (5,6). From the phenotypes of Sl and W mice, roles for SCF in hematopoiesis, gametogenesis, and melanogenesis could be inferred (7,8). These inferences have been supported by many in vitro and in vivo studies in the last few years; it is clear that SCF influences hematopoietic progenitors, gametes, and melanocytes both developmentally and in the adult animal as well (reviewed in Ref. 8). SCF influences other cell types also, including mast cells (reviewed in Ref. 8), and possibly certain intestinal cell (9,10) and liver cell (11) subtypes. SCF is expressed in different forms, determined by alternate splicing. A longer form, including extracellular domain, transmembrane domain, and cytoplasmic domain, is initially expressed at the cell surface; a presumed proteolytic cleavage can then lead to release of soluble SCF corresponding to amino acids 1-165, derived from the extracellular domain (12)(13)(14).
The alternate splicing leads to a shorter form which lacks the exon 6 amino acids including the proteolytic cleavage site and therefore tends to remain membrane bound (12,13). From several different approaches, there have been indications of functional differences between soluble and membrane-bound SCFs (12,13,(15)(16)(17). Soluble SCF is found at substantial levels in human serum (18).
Soluble SCF exists as noncovalently associated homodimer at concentrations where it has previously been possible to study its quaternary structure (19), although monomer dissociation/reassociation is rapid (20). Whether or not membranebound SCF is dimeric has not been determined.
Kit is a member of the type III tyrosine kinase type receptor family, which also includes the platelet-derived growth factor receptor, the macrophage colony-stimulating factor receptor, and FLK2 (21)(22)(23). It should be noted that platelet-derived growth factor and macrophage colony-stimulating factor are both covalently associated dimers (24 -28). The human Kit cDNA encodes a signal sequence (amino acids 1-23) followed by extracellular domain (amino acids 24 -520; five Ig-like domains), hydrophobic transmembrane region (amino acids 521-543), and cytoplasmic domain (residues 544 -976; split kinase domains) (29). The major expression product at the cell surface is a glycoprotein with a molecular weight of about 145,000 (29). In comparison, the calculated molecular weight of the polypeptide without signal sequence and without glycosylation is 109,740. As with other tyrosine kinase type receptor families, members of the type III family undergo obligatory dimerization and transphosphorylation as the initial event in ligand-induced signal transduction (21). Subsequent signaling events initiated by SCF/Kit have been extensively studied in various cell types (see introduction to Ref. 30). Varying somewhat with the system studied, K d values reported for SCF-Kit interaction have generally been in the range of 0.5-65 nM, although additional higher affinity (16 -310 pM) sites of SCF binding have been described for some cell types (31)(32)(33)(34)(35)(36)(37)(38). Most of these affinity determinations have been arrived at using recombinant Escherichia coli-derived soluble SCF (nonglycosylated). The affinity of CHO cell-derived soluble SCF (glycosylated) has been shown to be relatively lower by a factor of 5-10, and its mitogenic potency is lower also (14).
Two forms of Kit have been reported. They differ in that one (designated Kit A) has a 4-amino acid insert just prior to the transmembrane domain (39). The two forms arise as a result of alternative mRNA splicing (39). In addition, soluble Kit (sKit; i.e. Kit extracellular domain) has been found in the conditioned medium of various cells that express Kit (37,38,40) and, at high levels, in human serum (30).
For the growth hormone receptor, it has been shown that a monomeric ligand can mediate receptor dimerization (41,42). Because the ligands that associate with the type III tyrosine kinase type receptors appear to be dimeric, it is reasonable to suggest that this property may be important for mediating receptor dimerization within this family. However, this has not been definitively shown. Lev et al. (31) have proposed and discussed a model for SCF-mediated Kit dimerization in which it is not essential that SCF be dimeric when bound to Kit, and the same group has shown that Kit-Kit interactions play a role in Kit dimerization (43).
In the present work, we have used recombinant human soluble SCFs and recombinant human sKit (Kit extracellular domain) to study the stoichiometries and affinities of SCF-Kit interactions by a variety of physical chemical methods. We demonstrate the formation of a complex containing two molecules of SCF monomer plus two molecules of Kit extracellular domain.

EXPERIMENTAL PROCEDURES
Materials-E. coli-derived recombinant human SCF 1-165 (unglycosylated) and CHO cell-derived recombinant human SCF 1-165 (glycosylated) were prepared as described previously (43); their concentrations were determined using ⑀ 280 nm 0.1% of 0.536 and 0.540 liters/g⅐cm, respectively; these values were calculated from Trp, Tyr, and cystine composition (45). For the CHO cell-derived human SCF 1-165 , the ⑀ 280 nm 0.1% and any mass/volume concentrations given are representative of the polypeptide portion only (i.e. excluding attached carbohydrate). Molecular weight values of 18,656 and 18,525 (for monomers) or 37,313 and 37,051 (for dimers) (formula weights, calculated from amino acid sequence; note that the E. coli-derived SCF has a methionine at position Ϫ1 (44)) were used along with mass/volume values to arrive at molar concentrations of E. coli-derived or CHO cell-derived (respectively) SCF. Such concentrations therefore refer to SCF monomer in some cases and to SCF dimer in other cases, as indicated. 125 I-labeled E. coli-derived human SCF of specific radioactivity 61 mCi/mg was prepared by the lactoperoxidase method (46). 12.5 Ci (about 16 million cpm) of the labeled SCF was mixed with 2.4 mg of unlabeled E. coli-derived human SCF such that the volume of the mixture was 0.4 ml. This solution was applied to a Superose 6 size exclusion column (1 ϫ 29 cm; Pharmacia Biotech Inc.) equilibrated with PBS. Eluting fractions were collected, and those corresponding to the single peak of A 280 -absorbing material, representing SCF, were pooled to give a solution of SCF at 1.71 mg/ml and about 6 million cpm/ml. CHO cell-derived recombinant human sKit was expressed and purified as described (38). Based on the expressed sequence and the expected processing of the signal sequence, the material represents amino acids 24 -520 according to the numbering of Yarden et al. (29), i.e. it corresponds to the entire extracellular domain of Kit. Note that the 4-amino acid insert (residues 510 -513) characteristic of the Kit A isoform is present. The material is heavily glycosylated, with mostly N-linked carbohydrate but some O-linked carbohydrate as well (30). Mass/volume concentrations for the sKit were determined using an ⑀ 280 nm 0.1% of 1.185 liters/g.cm, which is a value calculated from Trp, Tyr, and cystine composition (45); thus these concentrations represent the sKit polypeptide only, exclusive of carbohydrate. Molar concentrations of sKit were determined from mass/volume values using a molecular weight value of 55,815 (calculated from amino acid sequence of the polypeptide).
Size Exclusion Chromatography-Size exclusion chromatography of SCFs, sKit, and incubated mixtures of SCFs and sKit was carried out using a Superose 6 column (1 ϫ 29 cm; Pharmacia) equilibrated with PBS at 4°C at a flow rate of 0.25 ml/min. A 280 of eluting material was monitored with a Pharmacia UV-M monitor.
Laser Light Scattering/Size Exclusion Chromatography-The online light scattering/chromatography system used three detectors in series: an absorbance monitor at 280 nm (Knauer A293), a laser light scattering detector (Wyatt Mini-Dawn), and a refractive index detector (Polymer Laboratories PL-RI). A Superdex 200 column (1 ϫ 30 cm; Pharmacia) equilibrated with PBS was used at room temperature with a flow rate of 0.5 ml/min. Because sKit is a glycoprotein, its refractive increment is unknown, and its total molecular weight cannot be determined from the light scattering data. However, as we have shown previously, polypeptide molecular weights (without the carbohydrate contribution) can be determined by combining the signals from all three detectors and using the extinction coefficient of the polypeptide alone (47,48). Calibration was accomplished using ribonuclease, ovalbumin, and bovine serum albumin monomer (Sigma) as molecular weight standards.
Sedimentation Equilibrium-Sedimentation equilibrium experiments were carried out at 25°C using 6-channel charcoal-Epon cells in a Beckman Optima XL-A analytical ultracentrifuge, with concentration distributions measured by absorbance at 280, 230, or 215 nm. The experimental and data analysis procedures followed those described previously (49). Samples of sKit, E. coli-derived SCF, and CHO cellderived SCF were run separately. Samples containing sKit and SCF were made up at ratios of 0.5, 1, or 2 sKit per SCF dimer and a range of sKit concentrations from 3 M to 300 nM. The data were recorded at three different rotor speeds. Solvent densities were measured with a Mettler-Paar DMA-02 density meter.
The sedimentation data on the mixtures were fitted to various heterogeneous association models. The fitting procedure accounts for the different molar extinction coefficients of the various species at the measurement wavelength. During the fitting, the reduced molecular weights of sKit and CHO cell-derived SCF were held fixed at the values determined when they were run alone; the molecular weight calculated from sequence for E. coli-derived SCF was used because its sedimentation data were consistent with that value. The data for up to 18 samples were globally fitted to determine the dissociation constants. During fitting, the fitted parameters were constrained to values that correctly accounted for the total concentrations of sKit and SCF placed into the cell.
Titration Calorimetry-SCF and sKit solutions were dialyzed against the same PBS to minimize heat due to buffer differences. The data were obtained at 25°C in a Microcal Omega calorimeter and were analyzed using both the Microcal Origin analysis package and the program TITRATE, 2 which provides a more robust analysis of the error in the fitted parameters as well as the ability to include a constant baseline heat per injection in the fitting.

RESULTS
Size Exclusion Chromatography-The stoichiometry of complex formation between SCF and sKit was investigated by mixing SCF and sKit in different molar ratios, incubating, and analyzing complex formation by size exclusion chromatography. The results are given in Fig. 1 for E. coli-derived SCF and sKit. It can be seen that the presence of SCF along with sKit results in the appearance of an early eluting species that must represent the sKit⅐SCF complex. (SDS-polyacrylamide gel electrophoresis confirmed that both sKit and SCF were part of this species (data not shown).) When sKit and SCF dimer are present at a 4 to 1 molar ratio during the incubation, integration of the chromatogram peaks shows half of the sKit remaining present in free form and the other half as complex ( Fig. 1, chromatogram 6); note in chromatogram 6 that the absorbance of the peak representing complex is somewhat greater than that for the free sKit by an amount representing the SCF associated with the sKit. No free SCF is apparent. When sKit and SCF dimer are present at molar ratios of 2 to 1, 1 to 1, and 0.5 to 1, the size exclusion chromatography results show that all sKit is present as complex. Again in each case the absorbance of the peak representing complex is greater than that of the free sKit ( Fig. 1, chromatogram 1) by an amount representing the associated SCF. In the case of the sKit to SCF dimer molar ratio of 2 to 1 (Fig. 1, chromatogram 5), no free SCF remains; in the case of the sKit to SCF dimer molar ratio of 1 to 1 (Fig. 1, chromatogram 4), half of the total SCF remains free; and in the case of the sKit to SCF dimer molar ratio of 0.5 to 1 (Fig. 1, chromatogram 3), three-quarters of the total SCF remains free. Taken together, these results strongly indicate that at the concentrations of sKit and SCF used, there is the formation of complex containing equimolar amounts of sKit and SCF monomer.
This equimolar stoichiometry was confirmed with the use of 125 I-labeled SCF. 125 I-labeled SCF, prepared as described under "Experimental Procedures," was mixed with unlabeled SCF (E. coli-derived) to give a preparation having specific radioac-tivity of 163 cpm/g. A mixture of sKit (at 40 g/ml) and the SCF preparation having molar ratio of sKit to SCF dimer of 1 was prepared, incubated, and subjected to size exclusion chromatography as in Fig. 1. Fractions containing complex were collected and pooled. Values of 2916 cpm/ml for radioactivity and 0.071 for absorbance at 280 nm were obtained for the pool. From the specific radioactivity of the SCF (163 cpm/g), it could be calculated that SCF was present in the pool at 17.9 g/ml or 0.96 M (for SCF monomer). Because ⑀ 280 nm 0.1% for the SCF is 0.536, it could also be calculated that absorbance due to SCF was 0.010. Thus absorbance due to Kit was 0.061 and, using ⑀ 280 nm 0.1% of 1.185, sKit was present in the pool at 51.3 g/ml or 0.92 M, and the molar ratio of sKit to SCF monomer determined in this way was 0.96 for the complex.
A priori, the equimolar amounts of sKit to SCF monomer in the complex could represent 1:1 or 2:2 complex. (Note that all references to stoichiometries for the complex, e.g. 1:1, 1:2, 2:2, will indicate the ratio of sKit to SCF monomer). The elution positions of marker proteins of known molecular weights are indicated by arrows in Fig. 1. However, because molecular shape as well as molecular size influences elution position, molecular weight values for SCF, sKit, and SCF⅐sKit complex cannot be reliably determined relative to marker proteins, and we did not attempt to determine stoichiometry of the complex on the basis on molecular weights estimated from elution positions.
Size Exclusion Chromatography with Light Scattering Detection-To distinguish between 1:1 and 2:2 stoichiometries for the complex, we accurately determined molecular weights of eluting species by using light scattering in combination with the size exclusion chromatography. The molecular weights determined by light scattering are independent of molecular shape and elution volumes. The size exclusion chromatography experiments were done using a system with refractive index and absorbance detectors in addition to the light scattering detector ("Experimental Procedures"). This allowed us to calculate the molecular weight of the complex (excluding contributions of carbohydrates) with an accuracy of Ϯ5%. Experiments with sKit alone and E. coli-derived SCF alone gave molecular weights of 55,600 and 38,600, respectively, consistent with the formula weights of 55,815 for sKit monomer and 37,313 for SCF dimer. The value for SCF is also consistent with that determined previously (and below) by sedimentation equilibrium (19). We then applied a sample containing 0.60 mg/ml (10.75 M) sKit plus 0.18 mg/ml SCF (2.2 mol of sKit/mol of SCF dimer), which produced a large peak corresponding to the sKit⅐SCF complex and a small peak of free sKit.
As we have described previously (49), to calculate the polypeptide molecular weight of the complex we need its extinction coefficient, which depends on the (assumed) stoichiometry. We therefore used an approach in which the calculated molecular weight is deemed valid only if it is consistent with the stoichiometry assumed. If we assume the complex contains one sKit plus two SCF monomers (one SCF dimer), the data imply M r ϭ 163,000, which is inconsistent with the theoretical value of 93,128. With an assumed stoichiometry of two sKit plus four SCF monomers (two SCF dimers) for the complex, the light scattering data give M r ϭ 163,000, again inconsistent with the theoretical value of 186,256. In contrast, when we assume a stoichiometry of two sKit plus two SCF monomers (one SCF dimer), the calculated M r of 149,000 agrees very well with the theoretical value of 148,943. We therefore conclude that the complex indeed contains two sKit and two SCF monomers (one SCF dimer).
Size Exclusion Chromatography of sKit/CHO Cell-derived SCF Complexes- Fig. 2 summarizes size exclusion chromatography experiments identical to those of Fig. 1 except that CHO cell-derived SCF (glycosylated) was used instead of E. coliderived SCF (nonglycosylated). Note that CHO cell-derived SCF alone elutes earlier than E. coli-derived SCF, but the apparent complex elutes later than that observed with E. coliderived SCF (Fig. 1) and elutes increasingly later as the initial amount of SCF becomes less. As noted in the Introduction, prior data have indicated that the mitogenic potency and receptor binding affinity of CHO cell-derived SCF are somewhat lower than those of E. coli-derived SCF. The likely explanation for the data of Fig. 2 is that the affinity of CHO cell-derived SCF for sKit is sufficiently low that during the running of the size exclusion columns, dissociated SCF is constantly being separated away from the sKit. In other words, there is a shifting equilibrium during the running of the columns such that the absorption peaks do not represent stable complexes held together with high affinity (as is the case with sKit and E. coli-derived SCF).
Sedimentation Equilibrium of SCF Alone and sKit Alone-Before studying the sedimentation behavior of complexes between sKit and SCF, we characterized each protein by itself. Data (not shown) for E. coli-derived SCF at three loading concentrations (500, 250, 125 g/ml) and three rotor speeds give an excellent fit to a single species. Using a calculated protein partial specific volume of 0.7433 ml/g (50) and the  (19), the SCF elutes unusually early, probably because of an elongated shape.) measured solvent density of 1.00399 g/ml for PBS, these data imply a molecular weight of 37,360 with a 95% confidence interval from 36,700 to 38,020, which includes the formula weight of 37,313 for an SCF dimer. These data confirm the value of 36,080 Ϯ 1670 determined previously (19) and confirm that SCF is dimeric at the concentrations being studied. Similarly, data obtained for CHO cell-derived SCF at three rotor speeds and at a high loading concentration of 650 g/ml were well fitted assuming an ideal single species. By methods described previously (49), we used the measured buoyant weight, together with the formula weight of 37,051 for the polypeptide dimer, the calculated polypeptide partial specific volume (0.7432 Ϯ 0.005 ml/g), and estimates of the carbohydrate partial specific volume to derive a carbohydrate weight for CHO cell-derived SCF of 16,900 [14,900 -18,100], 3 and a total molecular weight of 53,900 [51,900 -55,100], with most of the uncertainty being due to that in the carbohydrate partial specific volume. Again, these molecular weight values are very close to those reported previously for the CHO cell-derived SCF (19).
The sKit behaved as an ideal single monomeric species, with no appreciable self-association at concentrations up to about 1 mg/ml. Using the polypeptide formula weight of 55,815 and partial specific volume of 0.732 ml/g, the data give a carbohydrate weight for sKit of 18,900 [ Sedimentation Equilibrium of sKit Plus SCF Mixtures-Sedimentation equilibrium experiments were performed on mixtures of sKit and E. coli-derived SCF over a range of sKit⅐SCF molar ratios, total concentrations, and rotor speeds in order to verify the binding stoichiometry and to quantitate the strength of the binding interactions. These studies confirmed the light scattering conclusions that complex containing two receptors plus two SCF monomers (one SCF dimer) can be formed. Fig. 3 shows sedimentation data at 8,000 rpm for a sample with 2 mol of sKit/mol of SCF dimer (Fig. 3, diamonds) at relatively high protein concentrations (30 -400 g/ml). For such a plot (ln[A] versus [radius] 2 /2), a single species will give a straight line whose slope is proportional to the molecular weight of the species. The solid line shown in Fig. 3 indicates the slope expected for a 2:2 complex, and the dashed line indicates the slope for a 1:2 complex calculated from sedimentation equilibrium data for the separate components. This comparison clearly indicates that this sample contains primarily the 2:2 complex. Also shown in Fig. 3 are data for a sample with the same sKit loading concentration but having 1 mol of sKit/mol of SCF dimer (Fig. 3, squares). The significantly lower slope indicates that excess ligand leads to a rapid fall off in the fraction of 2:2 complexes in favor of 1:2 complexes. Moreover, these data (as well as additional data at lower sKit to SCF ratios) show no evidence of 2:4 complexes (two sKit plus four SCF monomers (two SCF dimers)).
Further quantitative analysis of these sedimentation equilibrium data requires fitting to an appropriate binding model. Our size exclusion chromatography, light scattering, sedimentation equilibrium, and titration calorimetry (next section) studies all indicate that the fundamental interaction stoichiometry is one sKit per SCF monomer and that the largest complex is the 2:2 complex, so we need only consider models that are consistent with this stoichiometry. Fig. 4 schematically illustrates four such binding models. Because at the protein concentrations used in these studies SCF will be essentially completely dimeric, in the first three of these models SCF dimer is treated as a single species with two binding sites for Kit (these are the models used previously to treat the interactions of neurotrophins with the receptors TrkB and TrkC (49)). In Model 1, these two binding sites per SCF dimer have fixed, 3 The values in square brackets are 95% confidence intervals. Fig. 1, except that CHO cell-derived SCF was used instead of E. coliderived SCF. equivalent affinities. In Model 2, the two binding sites have fixed inequivalent affinities. Model 2 is included for completeness, but because there is no reason to expect that a symmetric SCF dimer will present inequivalent sites, this was not considered an appropriate model for this system. In Model 3, the two binding sites are initially equivalent, but they interact with either positive or negative cooperativity to alter the affinity for binding the second receptor. It is important to note that in this model the actual mechanism of cooperativity is irrelevant; for example, it could arise either through creation of a receptorreceptor interaction site, as has been proposed (31,43), or through changes in SCF conformation or other mechanisms. Lastly, in Model 4 we allow for a finite dissociation of SCF into monomers, while keeping fixed the intrinsic interaction between Kit and an SCF monomer. 4 That is, there is no cooperativity in this model, and the monomer 7 dimer equilibrium of SCF is assumed to be independent of whether Kit is bound. Fig. 5 shows the results of a simultaneous fit of Model 1 to data for sKit plus E. coli-derived SCF mixtures made up at both the two sKit per SCF dimer and one sKit per SCF dimer molar ratios, and at three different rotor speeds. This fit is excellent and returns an intrinsic dissociation constant for each site of 17 [12][13][14][15][16][17][18][19][20][21] nM. As implied by Fig. 3, these data fit poorly (with about 10-fold higher variance (not shown)) to a model having only one sKit binding per two SCF monomers (one SCF dimer). Indeed, the fitted parameters imply that, averaged across the cell, the sample made at a molar ratio of two sKit per SCF dimer shown in Fig. 3 has 83% of its absorbance from the 2:2 complex, 11% from the 1:2 complex, 6% from free sKit, and 0.2% from free SCF, whereas the sample made at a molar ratio of one sKit/SCF dimer has 52% of its absorbance from the 2:2 complex, 43% from the 1:2 complex, 1% from free sKit, and 4% from free SCF. A second, independent set of experiments gave equivalent results, returning an intrinsic dissociation constant of 17 [14 -20] nM.

FIG. 2. Size exclusion chromatography of incubated mixtures of sKit and CHO cell-derived SCF. Experiments were done as for
We have also examined fits to the cooperative binding model, Model 3. Because an examination of model data showed that FIG. 4. Models for receptor-ligand interactions. In all models the shapes of the receptor (R) and ligand (L or L 2 ) are schematic and not intended to imply any specific structural features (the models are completely defined by the association reactions and their dissociation constants). In Models 1-3, the SCF dimer is treated as a nondissociable molecule with two binding sites. In Model 4, SCF dimer is allowed to reversibly dissociate into monomers, and this dissociation is assumed to be independent of receptor binding. The upper panel shows the raw data points and the fitted curves; for the sake of clarity, the data sets at 11,000 and 14,000 rpm have been displaced upward by 0.3 and 0.7 absorbance, respectively. All 2548 data points were globally fitted to Model 1 (Fig. 4), in which each SCF dimer has two equivalent sites to which sKit molecules bind independently. The best fit dissociation constant from this analysis is 17 nM. The bottom panel shows that the residuals from this fit do not exhibit large systematic deviations. the distribution of species is more sensitive to cooperativity when there is a larger excess of SCF in the samples, these fits included samples made at molar ratio of 0.5 sKit/SCF dimer. With these data sets, the cooperative model does not improve the fit over a fit to Model 1, and it can be concluded with 95% confidence that the dissociation constant for the second sKit on an SCF dimer is between 0.87 and 1.26 times that of the first. We therefore conclude that there is no binding cooperativity in this system, and we emphasize that this conclusion is based on data at high protein concentration and thus is free of any influence from possible dissociation of SCF dimers.
Model 4, which incorporates the possibility of appreciable dissociation of SCF dimer to monomers, can also be considered. Although our sedimentation data show no evidence for dissociation of SCF dimers, they also cannot rule out a monomer 7 dimer K d of the order of a few nanomolar or less. We therefore used model 4 to verify that such a K d value would not significantly affect our estimates of the K d for SCF association with sKit.
Lastly, we have also used sedimentation equilibrium to compare the affinities of sKit for glycosylated and nonglycosylated SCFs. Analysis of 12 data sets for sKit plus CHO cell-derived SCF at molar ratios of two sKit per SCF dimer and at one sKit per SCF dimer using Model 1 gave a K d of 590 [540 -630] nM, which implies a 35-fold weaker binding than that for E. coliderived SCF. The sedimentation data therefore confirm the weaker binding suggested by the size exclusion chromatography studies. Because SCF molecules with different glycosylation have different affinities (14), the 590 nM K d value determined here represents an average value. However, it is likely that the data analysis does not weight molecules of different affinity equally, so this value probably does not represent a true overall average affinity.
Titration Calorimetry-The binding of two sKit molecules per two SCF monomers (one SCF dimer) was also confirmed by titration calorimetry. Fig. 6 (top) shows the results for additions of E. coli-derived SCF to a solution of sKit. The large exothermic response due to binding ends after 7-8 injections (at which point total SCF corresponds to about 0.5 mol of SCF dimer/mol of sKit), followed by a constant small response due to a small mismatch of buffers and/or a small temperature difference between buffers. After subtraction of this constant heat per injection, the heat per injection data could be fitted well to Model 1 (Fig. 6, bottom), returning a stoichiometry of 2.03 [1.94 -2.11] sKit binding sites/two SCF monomers (one SCF dimer) and a ⌬H of Ϫ13.0 [Ϫ11.9 to Ϫ14.2] kcal/mol for the binding of each sKit. The estimated dissociation constant, 60  nM, has a large uncertainty but is consistent with that found by sedimentation equilibrium. DISCUSSION By a variety of physical chemical methods, including size exclusion chromatography, light scattering, sedimentation equilibrium, and titration calorimetry, we have clearly shown that SCF and sKit can form a stable 2:2 complex (i.e. two sKit per two SCF monomers or two sKit per SCF dimer). Our studies have utilized recombinant soluble forms of human SCF and human Kit.
The interactions that we describe and the conclusions that we reach provide insight into the possible nature of SCF-Kit interactions in vivo. The 2:2 complex formation clearly provides a mechanism for Kit dimerization, which is known to occur in vivo and to be required for initiation of signal transduction (21,31,51). However, because all of our data is obtained in vitro, we cannot definitely state that such a complex would transmit or initiate biological signaling in vivo, and, as always, there are differences between in vitro and in vivo situations. As men-tioned in the Introduction, both SCF and Kit are present in membrane-bound as well as soluble (extracellular domain) forms in vivo, and interactions involving the former may not be very well represented by our studies; the kinetics and energetics of interactions at surfaces will differ from those in solution, and our studies obviously would not reflect any interactions that might involve transmembrane or intracellular portions of SCF and Kit. In addition, even for soluble SCF and soluble Kit in vivo, concentrations are much lower than those in our in vitro studies (mean values in human serum, respectively, are 3.3 ng/ml (177 pM, for SCF monomer) and 324 ng/ml (5.8 nM) (18,30)). Clearly, SCF is mostly dimeric at the concentrations used in our in vitro studies, but we cannot exclude the possibility that SCF could be significantly dissociated into monomers at the low concentrations found in vivo.
Nevertheless, our studies suggest that the dimeric nature of SCF is important to the mechanism of Kit dimerization. This would differ from cases of monomeric ligands that can mediate dimerization of their receptors (41,42). Note that macrophage colony-stimulating factor and platelet-derived growth factor, whose receptors are members of the Type III tyrosine kinase family along with Kit, are covalently held dimers (24 -28).
It is important to ask whether our results are in accord with other proposed models for interaction of SCF and Kit and for Kit dimerization. A sequential model for this process was proposed by Lev et al. (31). In this model, the initial binding of Kit to an SCF dimer to form a 1:2 complex is monovalent (i.e. it involves only one of two binding sites/SCF dimer) and has relatively low affinity. This SCF binding triggers a conformational change in Kit that exposes or creates a receptor-receptor interaction site (a "receptor dimerization site"). Consequently a second Kit associates, giving rise to an unstable 2:2 complex in which the receptors are held together only by the dimerization site, and only one of the receptors interacts with SCF. Next, the second Kit molecule binds to the second site on the SCF dimer, giving rise to a stable, high affinity 2:2 complex held together both via the receptor dimerization site and via the SCF dimer. This model also proposes that when SCF is in great excess over Kit, a single Kit dimer may bind two SCF dimers (each held by only a single interaction with a receptor), forming a 2:4 complex. A key feature of this model, i.e. that receptor dimerization can occur independently of the bivalent nature of the SCF ligand, was invoked primarily to account for the observation that human SCF is able to heterodimerize mouse and human Kit on cells that express both receptors, even though the affinity of human SCF for mouse Kit is very low. This group has proposed that a Kit dimerization site is present in immunoglobulin-like domain 4 of the Kit extracellular domain, because recombinant Kit truncations that lack immunoglobulin-like domain 4 are unable to dimerize (43).
However, this sequential binding model is inconsistent with the behavior we observe for sKit and SCF in significant ways. The model predicts that binding will be highly cooperative, because the addition of a second sKit molecule to a 1:2 complex to form the stable 2:2 complex adds the receptor-receptor binding energy in addition to a second receptor-ligand bond. As discussed above, we find no evidence for cooperativity. In other words, the sequential model predicts that 2:2 complex will persist when ligand is in great excess, whereas we find that even a 2-fold excess of SCF (in terms of monomer) over sKit substantially reduces the proportion of 2:2 complex in favor of 1:2 complex (Fig. 3). 5 Note that reported bell-shaped SCF doseresponse curves (43,51) can be interpreted as being consistent with a lack of cooperativity, i.e. if the binding of a second Kit is not favored relative to binding of the first, Kit molecules will tend to be associated singly with SCF dimers when SCF is in excess over Kit. Another way in which our data are inconsistent with the sequential model is that we find no evidence by sedimentation equilibrium for formation of the 2:4 complexes predicted by this model when SCF is in great excess.
The binding affinities determined here by sedimentation equilibrium can be compared with reported K d values determined by techniques involving the binding of radiolabeled SCF. We have previously reported a value of 1.5 nM for the same sKit used in these studies (38), whereas others have reported values of 0.7 nM and 0.3 nM, respectively, for recombinant Kit extracellular domain (32) and for a fusion between Kit extracellular domain and an antibody Fc fragment (34). For the full-length receptor recombinantly expressed on cells, values of 0.5-2 nM have been reported (31,33). For various cell types naturally expressing the full-length receptor, high affinity (K d 16 -310 pM) and low affinity (K d 1.7-65 nM) classes of receptor have been reported, with both being present together on some of the cell types (35)(36)(37). However, in making comparisons between values from different laboratories or techniques, it is important to note that the affinity of sKit for SCF must be strongly temperature-dependent, and the data cited were obtained at either 4 (34,38) or 22°C (32). The fairly large ⌬H for this interaction seen by titration calorimetry (which matches that reported for human growth hormone and its receptor (41)), implies a 4.6 -6.2-fold increase in affinity between 25 and 4°C (assuming ⌬H and ⌬S are temperature-independent). For example, our estimate of 17 nM for the K d from sedimentation equilibrium at 25°C implies a K d of about 3 nM at 4°C, in reasonable agreement with the value of 1.5 nM from the binding of radiolabeled SCF (38). Further, in comparing the present results to those obtained by other methods, it is important to ask whether the stoichiometry of the interactions and multivalent nature of SCF were explicitly considered, particularly with regard to binding to cell surfaces, where the dimerization of Kit should produce nonlinear Scatchard plots that overestimate the true intrinsic affinity (52).
By size exclusion chromatography and by sedimentation equilibrium, we find that glycosylated SCF has a lower affinity for Kit than does nonglycosylated SCF. This conclusion is consistent with previous data (14) showing lower in vitro biological response potency and weaker binding in a competition type receptor binding assay for glycosylated SCF in comparison with nonglycosylated SCF; the differences were shown to be caused particularly by glycosylation at Asn 93 and at Asn 65 (14) and presumably represent steric effects of the glycosylation. As mentioned above, we have also shown that soluble SCF and soluble Kit are both present in human serum (mean levels of 177 pM (for SCF monomer) and 5.8 nM, respectively) and, using a K d value of 1 nM, we had calculated that much (about 85%) of the SCF would be associated with soluble Kit (30). Because native SCF is glycosylated (18), the higher K d values would lead to a calculation of considerably less SCF association with soluble Kit in the serum. The temperature dependence of K d noted above would affect such calculations similarly.