The Heparan Sulfate Binding Sequence of Interferon-γ Increased the On Rate of the Interferon-γ-Interferon-γ Receptor Complex Formation*

Interferon-γ (IFNγ), in common with a number of growth factors, binds both to heparan sulfate or heparin-related molecules and to a specific high affinity receptor (IFNγR). Using surface plasmon resonance technology, kinetic analysis of the IFNγ·IFNγR complex formation was performed with the extracellular part of IFNγR immobilized on a sensor chip. At the sensor chip surface, IFNγ was bound by two IFNγR molecules with an affinity in the nanomolar range (0.68 nm). This binding was characterized by an important on rate, k on = 7.3 × 106 m −1·s−1, and an off rate,k off = 5 × 10−3·s−1. This binding assay was used to investigate a possible role of heparin in the IFNγ·IFNγR complex formation. In contrast to growth factors for which binding to heparin is usually required for high affinity receptor interaction, we found in this study that IFNγ bound to heparin displayed a strongly reduced affinity for its receptor. This is consistent with the fact that a cluster of basic amino acids (KTGKRKR, called the C1 domain) in the carboxyl-terminal sequence of the cytokine was involved both in heparin and receptor recognition. To understand how a single domain of IFNγ could be implicated in two discrete functions (i.e. binding to heparin and to IFNγR), we also analyzed in a detailed manner the role of the IFNγ carboxyl-terminal sequence in receptor binding. Using forms of IFNγ, with carboxyl terminus truncations of defined regions of the heparin binding sequence, we found that the C1 domain functioned by increasing the on rate of the IFNγ·IFNγR binding reaction but was not otherwise required for the stability of the complex. Interactions between the IFNγ carboxyl-terminal domain and IFNγR could increased the association rate of the reaction either by increasing the number of encounters between the two molecules or by favoring productive collisions. The mechanisms by which heparan sulfate regulates IFNγ activity may thus include both control of selective protease cleavage events, which directly affect the cytokine activity, and also an ability to modulate the interaction of IFNγ with the IFNγR via competitive binding to the C1 domain.

Interferon-␥ (IFN␥) 1 is a highly pleiotropic protein secreted by activated T-lymphocytes and natural killer cells (1). The active form of this cytokine is a homodimer consisting of two intertwined 143-amino acid polypeptides. Each monomer consists of six helices (denoted A-F), linked by loops, and an unfolded sequence at the carboxyl-terminal side (downstream to the F helix, amino acids 124 -143), which extends away from the molecule (2)(3)(4). IFN␥ mediates its pleiotropic activities through a specific transmembrane receptor (IFN␥R) expressed at the surface of almost all cells (5). This receptor is composed of a ligand-binding subunit, or ␣-chain (IFN␥R␣) and an accessory factor, or ␤-chain which is required for signal transduction (6,7). The IFN␥⅐IFN␥R␣ complex consists of two receptors bound to an IFN␥ dimer, a stoichiometry consistent with the symmetry of the ligand (8,9). IFN␥ residues involved in receptor binding are situated in two areas of the molecule (10,11): the loop connecting the A and B helices (residues 18 -26) and the helix F (residues 108 -124). The carboxyl-terminal sequence (residues 124 -143) is highly flexible and/or adopts multiple conformations (3,4). It contains two small clusters of basic amino acids (C1, residues 125-131, KTGKRKR; C2, residues 137-140, RGRR), which confer on the molecule a important sensitivity to a variety of proteases (12)(13)(14)(15). As a consequence, native IFN␥ is usually a mixture of carboxyl-truncated molecules lacking up to 16 amino acids and therefore may end anywhere from Gln 143 (full-length molecule) to Gly 127 . Since deletion of less than 10 amino acid increases the bioactivity of IFN␥, while more extended cleavages of the carboxyl-terminal domain have the opposite effect, it has been speculated that the cytokine activity can be modulated by limited proteolysis. The portion of the carboxyl terminus that appeared to be the most important in this process is the basic sequence KRKR of the C1 domain, since the beginning of its removal correlated with loss of activity, while cleavages downstream of this sequence resulted in the observed increased activity (13). However, as revealed by the crystal structure of the IFN␥⅐IFN␥R␣ complex, specific interactions between this basic sequence of IFN␥ and IFN␥R have not been identified, and the carboxyl-terminal tail of the cytokine did not appear to be involved in receptor binding (11). Furthermore, while both antibodies directed against amino or carboxyl terminus block biological activity, only synthetic peptides comprising the amino-terminal domain of the cytokine block binding to its receptor in a competitive manner (15). In addition, it has been shown that if a positive charge localized in the carboxyl-terminal part of the cytokine is impor-tant for activity, it can be contributed by a number of different combinations of positions. Therefore, this part of the molecule did not appear to function in a specific manner, and its function remains unclear (16). The carboxyl-terminal domain of IFN␥ also confers on the molecule a high affinity (K d ϭ 1.5 nM) for heparan sulfate (17) or heparin-related molecules (18). A fragment of heparan sulfate that displays high affinity for IFN␥ has been isolated. It consists of two small sulfated heparin-like sequences linked together by an extended internal N-acetylrich domain (19). In the IFN␥⅐heparan sulfate complex, the two carboxyl termini of an IFN␥ dimer interacted with the two heparin-like sequences of the heparan sulfate fragment, mainly through the basic C1 and C2 domains (20).
In this study, we investigated whether or not IFN␥ bound to heparin could also interact with its high affinity cell surface receptor. Using the Biacore system, we analyzed kinetic aspects of the IFN␥⅐IFN␥R complex formation and, in particular, the importance of the heparin binding sequence of IFN␥ for this interaction. We found that IFN␥ binding to heparin and to IFN␥R are mutually exclusive and that the C1 sequence increased the association rate of the IFN␥⅐IFN␥R binding reaction.

MATERIALS AND METHODS
Equipment and Reagents-An upgraded Biacore system, certified CM5 sensor chips, and HBS buffer (10 mM HEPES, 150 mM NaCl, 3.4 mM EDTA, 0.05% surfactant P20, pH 7.4) were from Biacore AB. Biotin-LC-hydrazide was from Pierce, and streptavidin was from Sigma. Proteolytic enzymes (activated factor X, carboxypeptidase Y, endoproteinase Arg-C) and enzyme inhibitor (pefablock) were from Boehringer. Recombinant IFN␥ (IFN␥ 143 , batch number L 405), IFN␥ lacking the last 19 amino acids on the carboxyl-terminal side (IFN␥ 124 ), and the two monoclonal antibodies (293-4-45 and 13-16-2) used in this study were kind gifts of Roussel Uclaf company. The extracellular part of the human IFN␥ receptor (sIFN␥R) ␣-chain, produced in Sf9 cells, was a kind gift of Dr. L. Ozmen (Hoffmann La Roche), and fractionated heparin was from Sanofi Recherche.
Preparation of IFN␥ Lacking Carboxyl-terminal Amino Acids and the IFN␥ Assay-Full-length IFN␥ (IFN␥ 143 ) was digested with either activated factor X (1 unit/ml in 50 mM Tris, 150 mM NaCl, pH 8, buffer for 1 h at 25°C), carboxypeptidase Y (20), or endoproteinase Arg-C (2.5 units/ml in 50 mM Tris, 150 mM NaCl, pH 8, buffer for 4 h at 25°C). Enzymatic reactions were stopped with 1 mM pefablock and analyzed by 15% SDS-polyacrylamide gel electrophoresis and by electrospray ionization mass spectrometry. For that purpose, digested samples were made 0.4 mg/ml in 10 mM ammonium acetate buffer and analyzed using a Perkin-Elmer Sciex API III ϩ triple quadrupole mass spectrometer equipped with a nebulizer-assisted electrospray (ionspray) source. Samples were directly infused into the source using a syringe pump at a flow rate of 5 l/min and an injector equipped with a 1-l internal loop or were desalted using a reversed phase high pressure liquid chromatography column coupled to the electrospray source. The ionspray probe tip was held at 5000 V, and spectra were recorded in the 900 -1800 range of mass-to-charge (m/z) ratios in steps of 0.4 m/z, with a 1.7-ms dwell time. The signal was averaged over four scans. Specific antiviral activity of the IFN␥ samples were determined with a microtiter inhibition of cytopathic effect assay using Wish cells against vesicular stomatitis virus (21).
Biotinylation of the sIFN␥R-The purified sIFN␥R (22) was prepared at 0.27 mg/ml (8 M) in 20 mM phosphate buffer, pH 6, and reacted for 20 min in the dark and at 4°C with 10 mM sodium periodate to oxidize the glycan parts of the molecule. The reaction was quenched with 15 mM glycerol, and the sample was dialyzed against the 20 mM phosphate buffer. Biotin-hydrazide (including a 2.5-nm spacer arm between the biotin and hydrazide groups) was then added to a concentration of 5 mM, and the mixture was incubated for 6 h at 4°C. The sample was made to 100 mM ethanolamine, and then extensively dialyzed against phosphate-buffered saline, pH 7.2. Biotinylation was checked by Western blot analysis, and samples were aliquoted and stored at Ϫ80°C.
Preparation of the Biacore Binding Surface-The flow rate of the running buffer (HBS) was maintained at 5 l/min, and the temperature was maintained at 25°C. Two flow cells of a CM5 sensor chip were activated with 50 l of a mixture of 0.2 M 1-ethyl-3-(3 dimethylaminopropyl)carbodiimine 0.05 M N-hydroxy-sulfosuccinimide, after which 50 l of streptavidin (0.2 mg/ml in 10 mM acetate buffer, pH 4.2) was injected. Unreacted groups were blocked with a 50-l injection of 1 M ethanolamine, pH 8.5. Approximately 8000 resonance units (RU) of streptavidin were fixed on the surface by this procedure. Biotinylated sIFN␥R (10 l/ml in HBS) was then injected for 1 min on one of the two streptavidin surfaces (the other one being a negative control). Both flow cells were then conditioned with 10 2-min pulses of 10 mM HCl. This resulted in the attachment of 800 RU of IFN␥R. The conversion of RU to surface concentration of proteins was performed using a conversion factor of 1000 RU ϭ 1 ng/mm 2 .
Binding Assay-Test samples were diluted in HBS maintained at 25°C and injected over the IFN␥R surface at a flow rate of 50 l/min. This high flow rate was necessary to reduce mass transport effect due to the high association rate of the proteins being studied. Using the Kinject command, usually 200 l of IFN␥ or IFN␥-derived molecules were injected across the IFN␥R surface, after which the formed complexes were washed at 50 l/min with HBS to study the dissociation phase. The IFN␥R surface was regenerated with a 2-min pulse of 10 mM HCl. For kinetic analysis, a complete set of sensorgrams were recorded with eight different IFN␥ concentrations in the range 0 -0.5 g/ml.
Kinetic Analysis-Sets of sensorgrams were analyzed with the Biaevaluation 2.1 software, provided with the machine, using both linear transformation of the primary data and nonlinear fitting of the sensorgrams. Briefly, the equation for the measured binding rate (dR/dt) as a function of the binding response (R) is dR/dt ϭ k on CR max Ϫ (k on C ϩ k off )Rt n , where k on and k off are the association and dissociation rate constant, C is the concentration of the analyte (IFN␥), R max is the binding capacity of the immobilized ligand (IFN␥R), and Rt n is the amount of analyte bound to ligand at time t n . Kinetic constants (k on and k off ) can be obtained by linear transformation of a set of sensorgrams using a plot of ln(dR/dt) versus time for each analyte concentration. These plots give a line of which the slope is k s . A secondary plot of these slopes (k s ) versus C is then used to determine k on and k off from the linear relationship k s ϭ k on C ϩ k off . Association and dissociation rate constants can also be extracted from a single sensorgram (i.e. a single analyte injection). In that case, the integrated forms of the rate equa- ) for association, where R eq is the steady-state binding response and t a and t d are the start times for association and dissociation) were fitted to the experimental data by nonlinear regression (23). In some cases to derive kinetic constants, numerical integration, which allows fitting of data to complex interaction models, was performed (Biaevaluation 3.0 software). Data were analyzed by global fitting of both association and dissociation phases for several concentrations simultaneously. Affinity (dissociation equilibrium constants, K d ) were calculated from the ratio of dissociation and association rate constants (K d ϭ k off /k on ).

Binding of IFN␥ to Sensor Chip Immobilized IFN␥R-Using
the Biacore technology, we set up a binding assay to analyze the IFN␥/IFN␥R interaction, in particular the kinetic aspects. The Biacore instrument uses surface plasmon resonance to measure changes in refractive index when a soluble analyte (here IFN␥) binds to an immobilized ligand (IFN␥R). Biotinylated sIFN␥R was immobilized on a streptavidin-activated sensor chip, at a density of 0.8 ng/mm 2 (800 RU). Biotin groups had been attached to the glycan part of the receptor (a part not required for ligand binding (24)) to ensure that 100% of the molecules remain active upon immobilization. When IFN␥ was injected over the IFN␥R surface, a typical sensorgram was obtained (Fig. 1a), with an association phase (A), equilibrium (E), and, when IFN␥ was replaced by running buffer alone, a dissociation phase (D). Preincubation of the cytokine with increasing concentrations of its soluble receptor completely inhibited the interaction with the immobilized IFN␥R, demonstrating the specificity of the binding (Fig. 1a). In addition, upon injection of IFN␥ over a control surface (containing streptavidin only) no binding was observed (Fig. 1b).
Maximum binding was approximately 400 RU. Since we had 800 RU of immobilized IFN␥R, this suggests that each IFN␥ dimer (34 kDa) has been bound by two IFN␥R molecules (32 kDa each). Dimerization of IFN␥R by its ligand also occurs both in solution and at the cell surface (9). To further confirmed this stoichiometry at the sensor chip surface, we also injected sIFN␥R on a preformed IFN␥⅐IFN␥R complex. As shown on Fig. 1c, the complex was unable to bind additional IFN␥R, demonstrating that it already contained two IFN␥R molecules per IFN␥ dimer. Nonlinear least squares regression analysis of the sensorgram (Fig. 1a, upper curve), using a simple A ϩ B ϭ AB model, gave a k off ϭ 1.2 ϫ 10 Ϫ3 ⅐s Ϫ1 and a k on ϭ 4.5 ϫ 10 6 M Ϫ1 ⅐s Ϫ1 and therefore an equilibrium dissociation constant K d ϭ 0.26 nM. This is higher than the affinity of IFN␥ for its soluble receptor (1.5 nM) reported with other methods (9). However, it should be noted that we observed deviation of the data from the simple model we used. This may be caused by factors that derive from the biosensor method itself (25), such as ratelimiting mass transport effect or rebinding of the ligand during the dissociation phase (see below).
Binding of IFN␥ to Heparin and to IFN␥R Are Mutually Exclusive-One of our objectives was to determine whether IFN␥ bound to heparin was still able to interact with its recep-tor. For that purpose, IFN␥ was preincubated with increasing concentrations of different heparin molecules and then injected over the IFN␥R surface (Fig. 2). The first set of sensorgrams show that a 12-kDa heparin fragment bound to IFN␥ strongly decreased the on rate of the reaction (Fig. 2a). As a consequence, the amount of IFN␥ bound to its receptor was strongly reduced. This heparin fragment (12 kDa, approximately 20 disaccharides in length) represented the minimum length that binds efficiently to IFN␥ and functions by bridging two IFN␥ monomers (19). Smaller heparin fragments (4.5 kDa, approximately 8 disaccharides in length) displayed a strongly reduced ability to inhibit the IFN␥⅐IFN␥R complex formation (Fig. 2b), while a hexasaccharide (1.8 kDa) was completely inactive, even at a 50-fold molar excess (Fig. 2c).
The C1, but Not the C2, Domain of IFN␥ Is Involved in Receptor Recognition-To investigate which of the two basic clusters, C1 and C2, involved in heparin binding could be implicated in receptor interaction, IFN␥ preincubated with two different monoclonal antibodies (mAb), was injected over the IFN␥R surface. These two mAbs (293-4-45 and 13-16-2) defined two overlapping sequences of the carboxyl-terminal part of IFN␥ (amino acids 125-134, which encompass the first basic domain C1 of the carboxyl-terminal part of the cytokine for mAb 293-4-45, and amino acids 132-138, which overlap the second basic domain C2 for mAb 13-16-2) and are described elsewhere (26). Coincubation of IFN␥ with mAb 293-4-45 dra-FIG. 1. IFN␥ binding to immobilized IFN␥R. a, IFN ␥ (0.5 g/ml), either alone or in combination with 2.5, 5, 10, or 20 g/ml soluble IFN␥R, was injected for 4 min (from 130 to 370 s) to an IFN␥R activated sensor chip, after which the samples were replaced by running buffer. The binding response in RU was recorded as a function of time and showed the association phase (A), the equilibrium (E), and the dissociation phase (D). When coincubated with soluble IFN␥R, the binding of IFN␥ was strongly reduced. b, IFN␥ (eight concentrations from 0 to 0.5 g/ml) was injected on a sreptavidin-activated sensor chip, which showed that the binding in a was specific to the immobilized IFN␥R. c, IFN␥ (0.5 g/ml) was injected on the IFN␥R-activated sensor chip until the equilibrium was reached. After a quick wash with running buffer, soluble IFN␥R (1, 2, 10, 20 g/ml, from top to bottom) was injected on the bound IFN␥. The soluble IFN␥R could not bind to a preformed IFN␥⅐IFN␥R complex, demonstrating that at the surface of the sensor chip the cytokine was already bound by two IFN␥R molecules. matically reduced the binding of the cytokine to its receptor (Fig. 3a). Coincubation of IFN␥ with mAb 13-16-2 also reduced the binding but to a smaller extent (Fig. 3b). Both mAbs were also used to probe a preformed IFN␥⅐IFN␥R complex. For that purpose, a saturating dose of IFN␥ was injected over the IFN␥R surface until equilibrium was reached. The complex was then quickly washed with running buffer, after which mAbs were injected (Fig. 3c). These data showed that mAb 13-16-2 could bind to the IFN␥⅐IFN␥R complex, whereas mAb 293-4-45 only displayed a weak interaction. Together these results suggest that the 125-134 sequence of IFN␥ becomes buried in the cytokine-receptor complex but that the 132-138 sequence remains freely accessible.
Characterization of Enzymatically Processed IFN␥-To directly analyze the role of the heparin/heparan sulfate binding sequence in receptor interaction, we prepared various forms of IFN␥ lacking defined domains of the carboxyl terminus, making use of the great enzymatic susceptiblility of this part of the cytokine (13). Cleavage of IFN␥ was performed with activated factor X, carboxypeptidase Y, or endoproteinase Arg-C (Fig. 4). The M r of these truncated forms of IFN␥, as measured by electrospray ionization mass spectrometry analysis (Table I), were 16,254, 15,705, and 15,208 for activated factor X-, carboxypeptidase Y-, and endoproteinase Arg-C-treated IFN␥. This corresponded to molecules from which 6, 10, or 14 carboxyl-terminal amino acids were removed, and this was consistent with the known substrate specificity of the enzymes used. These molecules therefore ended at Arg 137 , Gln 133 , and Arg 129 and will be referred to as IFN␥ 137 , IFN␥ 133 , and IFN␥ 129 (see Fig. 4). As a control, IFN␥ 143 was also analyzed, and the M r obtained (16,908) was exactly the M r expected for the fulllength molecule. In addition, deletion of six (IFN␥ 137 ) or 10 (IFN␥ 133 ) amino acids resulted in the expected increased antiviral activity (13), while deletion of 14 (IFN␥ 129 ) or 19 (IFN␥ 124 ) amino acids resulted in the expected decreased activity for such truncated molecules (data not shown).
Kinetic Analysis of the Interaction between IFN␥ and IFN␥R: Role of the Carboxyl-terminal Sequence-IFN␥ 143 (full-length IFN␥), and the carboxyl termini-deleted IFN␥ were injected over the biacore IFN␥R surface, each with a range of concentrations to produce sets of sensorgrams from which association and dissociation phases can be analyzed. The dissociation phase started at 370 s, when the injection of IFN␥ was changed to a perfusion buffer. To this part of the sensorgrams was applied a simple AB ϭ A ϩ B dissociation model. A good fit was found for IFN␥ 124 only (Fig. 5e), which gave an off rate in the range 4.7-5.2 ϫ 10 Ϫ3 ⅐s Ϫ1 . For the others, the higher the observed on rate (see below) the worse was the fit. Presumably, the fast on rate caused a rapid rebinding of the dissociated molecules, a phenomenon that occurs as increasing numbers of free immobilized ligand (IFN␥R) are regenerated at the surface of the sensor chip during the dissociation phase. Since free IFN␥R, and therefore the rebinding effect, increased with time during the dissociation, only the first 60 s were used for the analysis. In addition, to prevent rebinding of dissociated molecules, soluble IFN␥R was included in the perfusion buffer during dissociation (e.g. see Fig. 1c). Including IFN␥R at 10-20 g/ml in the perfusion buffer, a simple dissociation model can now be assumed, characterized by an off rate of 5-5.6 ϫ 10 Ϫ3 ⅐s Ϫ1 . Interestingly, this k off was similar to the k off of IFN␥ 124 (for which no rebinding occurred). Therefore, the addition of soluble IFN␥R in the perfusion buffer was effective in preventing rebinding and necessary for accurate kinetic analysis. Furthermore, these data showed that cleavages of the carboxyl terminus of IFN␥ did not modify the off rate of the reaction and, thus, that the carboxyl-terminal sequence of  (97,400, 66,200, 45,000, 31,000, 21,500, 17,000, and 14,400). Lane B, IFN␥ 143 . Lane C, activated factor X-treated IFN␥. Lane D, carboxypeptidase Y-treated IFN␥. Lane E, endoproteinase Arg-Ctreated IFN␥. Lane F, IFN␥ 124 . Lane G, a mixture of lanes B-F. The carboxyl-terminal sequence of IFN␥ is also shown. The two clusters of basic amino acids C1 and C2 are in boldface type, and the cleavage sites, as determined by electrospray ionization mass spectrometry (see Table  I), are indicated by arrows.
the cytokine was not involved in the stability of the complex.
During ligand injection (130 -370 s), three phases can be distinguished. The first phase was diffusion-controlled (the reaction was limited by mass transport of the analyte), and, in our case, this was observed (see below) for the molecules that displayed a high on rate (IFN␥ 143 , IFN␥ 137 , and IFN␥ 133 ) but not for the others (IFN␥ 129 and IFN␥ 124 ). The second phase was determined by the kinetics of the IFN␥/IFN␥R interaction, and the third one was equilibrium. It was immediately apparent (see Fig. 5) that the binding rate of the reaction increased for IFN␥ 137 and IFN␥ 133 , compared with full-length molecule (Fig.  5, a-c) but decreased for IFN␥ 129 and IFN␥ 124 (Fig. 5, d-e). Kinetic analysis of the binding curves was first performed with the Biaevaluation 2.1 software. As indicated above, in some cases (IFN␥ 143 , IFN␥ 137 , and IFN␥ 133 ), the early phase of the association was dominated by mass transport. However, as the reaction proceeds, the free binding sites decrease and the binding rate progressively depends on the intrinsic rate constant. Thus, as a first approach, when analyzing the experiments, the mass transport-limited parts of the sensorgrams were excluded from the data used for evaluation, as already reported in several other cytokine-receptor binding studies (27,28). In this way, the kinetic part of each sensorgram could be confidently fitted on the basis of an A ϩ B ϭ AB model, and deviations of the data points from the fitted curve were quantitatively in the range expected from the background noise. As detailed under "Materials and Methods," analysis was performed both by nonlinear fitting and by linear transformation of the data. Both methods gave similar kinetic constants, and the results are summarized in Table II. However, mass transport limitation has been identified as a potential problem, leading to underestimation of the association rate constant (29). Thus, to better assess the observed changes in on rates, caused by cleavages of the IFN␥ carboxylterminal sequence, we also analyzed our data by numerical integration, using the Biaevaluation 3.0 software. Here, in contrast to the previous analysis, the entire time courses of the reactions were fitted to binding models, including mass transport-limited binding reaction. Global fitting of binding curves, recorded with IFN␥ 129 and IFN␥ 124 , gave respective association rate constants of 2.48 and 1.9 ϫ 10 6 M Ϫ1 ⅐s Ϫ1 . It is noteworthy that these values are in close agreement with those determined either by linear transformation of the data or by nonlinear fitting (Table II). In contrast, global fitting of the binding curves obtained with IFN␥ 143 , IFN␥ 137 , and IFN␥ 133 was only possible when mass transport limitation was introduced in the model. Data are reported in Table II and showed that these three molecules bound IFN␥R with respective on rate constants of 0.73, 1.3, and 1.57 ϫ 10 7 M Ϫ1 ⅐s Ϫ1 . These values are, on average, twice as much as those determined with the first approach and showed that k on values were underestimated when only parts of the sensorgrams were used for the  Full-length IFN␥ (IFN␥ 143 ), activated factor X (aFX)-, carboxypeptidase Y (CY)-, or endoproteinase Arg-C (ArgC)-treated IFN␥ was analyzed by electrospray ionization mass spectrometry. Cleavage points (last amino acid) were determined from the measured relative molecular mass (M r ). (e) were injected over an IFN␥R-activated surface at a flow rate of 50 l/ml during 4 min (from 130 to 370 s), after which running buffer was injected, and the response in RU was recorded as a function of time. Each set of sensorgrams was obtained with IFN␥ molecules at (from top to bottom) 0.5, 0.4, 0.3, 0.25, 0.2, 0.15, 0.12, and 0 g/ml. Association phases where the binding was not limited by mass transport were used for the kinetic analysis either by linear transformation of the primary data and by nonlinear fitting of the sensorgrams (see Table II) with the Biaevaluation 2.1 software. Global fitting of the binding curves, using numerical integration (Biaevaluation 3.0 software) was also performed (see Table II). fitting procedure.
Together these analyses showed that progressive cleavages of the IFN␥ carboxyl terminus progressively increased the on rate of the binding to IFN␥R until the C1 domain was affected. The maximum on rate was observed with IFN␥ 133 (k on ϭ 1.57 ϫ 10 7 M Ϫ1 ⅐s Ϫ1 ). Such a high on rate has been observed already for the interleukin-4 binding to its receptor (30). Once the C1 domain was cleaved (IFN␥ 129 ) or removed (IFN␥ 124 ), the k on decreased. Thus, the integrity of the carboxyl-terminal sequence of the cytokine is important for the association phase of the IFN␥⅐IFN␥R binding reaction. DISCUSSION A number of growth factors and cytokines bind to heparin or heparin-related molecules (31,32). This include the fibroblast growth factor family (33), vascular endothelial growth factor (34), hepatocyte growth factor (35), heparin binding-epithelial growth factor (36), and cytokines such as interleukin-1, -2, -3, -4, -6, -7, and -8 (37)(38)(39)(40)(41), chemokines (42), and IFN␥ (18). Considerable data have been reported during the past few years regarding the physiological significance of the binding of heparin to growth factors, the prototype of which is FGF-2. Importantly, FGF does not bind productively to its high affinity receptor, unless it is itself bound to heparin or heparan sulfate. Such information has not been reported yet for any heparinbinding cytokine.
In this study, we set up a binding assay, suitable for kinetic analysis, in which IFN␥ binds to IFN␥R, with high affinity and specificity, and with a stoichiometry identical to what occurs at the cell surface (one IFN␥ for two IFN␥R, Fig. 1). Using this assay, we first showed that, in contrast to growth factors, IFN␥ bound to heparin displayed a strongly reduced binding to its receptor, which was caused by a reduction of the on rate of the reaction. As a consequence, the mechanism by which heparin may affect IFN␥ activity should be completely different from the mechanism by which heparin modulates the activity of other growth factors (see below). Noteworthy is the fact that only extended heparin fragments (at least 20 disaccharides) could inhibit the IFN␥/IFN␥R binding (Fig. 2). This length was found to be necessary to interact simultaneously with the two carboxyl termini of an IFN␥ homodimer, and the present finding further supports the molecular organization of the heparan sulfate binding site we previously described (19).
The observation that IFN␥ binding to heparin and to IFN␥R were mutually exclusive could indicate that the cytokine carboxyl-terminal domain was also engaged in IFN␥R recognition. The involvement of the IFN␥ C1 sequence (one of the two basic clusters that interacts with heparin) in IFN␥R binding has been reported yet, but its possible mechanistic role has not been established (10,43). In particular, the kinetic analysis performed here first indicated that IFN␥ molecules from which C1 and C2 were removed (IFN␥ 124 ) dissociated from its receptor at the same off rate (5 ϫ 10 Ϫ3 ⅐s Ϫ1 ) as the parent molecules (IFN␥ 143 ). Thus, at equilibrium, the carboxyl-terminal sequence of the cytokine, and especially the C1 domain, was not involved in the stability of the IFN␥⅐IFN␥R complex. This was also supported by the fact that mAb 293-4-45, the epitope of which is C1, did not dissociate a preformed complex (Fig. 3c). In contrast, progressive cleavages of the carboxyl-terminal domain of IFN␥ modulated the on rate of the reaction (Fig. 5). Increased on rate was observed for any cleavage downstream of the C1 domain (IFN␥ 137 and IFN␥ 133 ), whereas cleavages within or upstream of the C1 domain decreased the on rate of the reaction (IFN␥ 129 and IFN␥ 124 ). We believed that the functional importance of the C1 sequence, during the association phase only, that we report here, explains the existing and controversial data on the role of the IFN␥ carboxyl terminus in bioactivity. The mechanism by which removal of the C2 domain further increased the on rate of the IFN␥⅐IFN␥R binding reaction is not known, but it is possible that C1 and C2 compete with each other to bind to some sequence on IFN␥R (see below).
In general, random encounters between two proteins occurs in an orientation where the specific contact residues are not aligned, and the two molecules diffuse away from each other after colliding nonproductively. Thus, many collisions are needed before specific association occurs (44). As stated above, the carboxyl terminus of IFN␥ is highly flexible and can adopt multiple conformations. Of interest in this regard was that one of these conformations places the IFN␥ basic C1 domain in front of a cluster of acidic amino acids in the IFN␥R while the two molecules are bound to each other (11). Attracting forces between these two charged domains could place the two interacting molecules into a proper orientation while they are still some distance apart. In such a way, the specific binding sites may collide with each other more frequently, and this would result in the increased on rate observed when C1 was present in the molecule (IFN␥ 143 , IFN␥ 137 , and IFN␥ 133 ). Another mechanism could be proposed whereby interacting IFN␥ and IFN␥R are held together by relatively nonspecific forces, (between C1 and the group of acidic amino acids found in IFN␥R) long enough to increase their chance of finding a mutually reactive configuration. A mechanism of this sort can be viewed as representing reorientation within an encounter complex (44) and may account for the importance of C1 during the association phase only. It is noteworthy that the association rate can be increased for a variety of molecules, simply by attaching weakly interacting, relatively unstructured polymeric domains to the macromolecules involved. Furthermore, charged residues are particularly suitable for such mechanisms, because electric fields that surround such groups spread out in every direction, making collision geometry less important (44). Together, our data and the proposed mechanisms described above suggest that C1 interaction with IFN␥R is the first event of the IFN␥⅐IFN␥R complex formation. The dissociation rate was identical for all of the IFN␥ forms studied here, irrespective of the presence of C1 and C2 domain (k off ϭ 5 ϫ 10 Ϫ3 ⅐s Ϫ1 ). However, for those forms that contained the C1 region, it was necessary to include soluble IFN␥R in the running buffer during the dissociation phase to prevent rebinding of dissociated molecules. The observation that the C1 sequence caused immediate rebinding of newly dissociated molecules further supports the idea that interaction between C1 and IFN␥R represents the first contact between the two molecules.  Fig. 5 were analyzed as described under "Materials and Methods." Dissociation rate constants (k off ) were measured in the absence of rebinding effect. Association rate constants (k on ) were obtained by linear transformation of the primary data (a) or by nonlinear fitting of the sensorgrams, using the Biaevaluation 2.1 software (b). For IFN␥ 143 , IFN␥ 137 , and IFN␥ 133 , the early phase of the association was dominated by mass transport, and the mass transport-limited parts of the sensorgrams were excluded from the analysis. Data were also analyzed by global fitting (c), using the Biaevaluation 3.0 software, to take into account mass transport effect (see "Results"). The on rate obtained by this method was used for the calculation of K d (K d ϭ k off /k on ). For analytical purposes, rebinding has been eliminated in our study, but this may have a physiological importance. IFN␥ is active at extremely low concentration, well below the K d (for example, a few pM only are required to have an antiviral effect in cell culture). Therefore, free IFN␥R are in large excess, and rebinding of dissociated molecules is likely to occur at the cell surface. As a result, the apparent affinity of IFN␥ for its receptor will be increased (see result from Fig. 1a). High affinity, and in particular a high rate of association, is necessary for achieving a high rate of product formation but also for preventing side reactions such as inactivation.
High affinity binding of IFN␥ to heparin also represents a mechanism by which IFN␥ is protected from inactivation (20). This binding also supports the plasma clearance of the cytokine and localizes it in restricted areas within tissue (45). In particular, it has been found that in vivo heparin protected the C1, but not the C2, domain of IFN␥ from degradation, a process that thus increases the cytokine activity by as much as 600% (46). It is noteworthy that these effects of heparin or heparan sulfate on IFN␥ activities did not depend on IFN␥R, and this is consistent with the fact that binding to heparin and IFN␥R are two independent events. However, since IFN␥ cannot bind heparin and its receptor simultaneously, it remains unclear how the heparin-bound cytokine can be released to interact with its receptor, and this point will be the subject of future studies. Finally, the results of the present work together with our previous reports clearly indicate that IFN␥ and growth factors (such as FGF) belong to distinct groups regarding their regulation by heparan sulfate/heparin-like molecules.