Covalent dimerization of CD28/CTLA-4 and oligomerization of CD80/CD86 regulate T cell costimulatory interactions.

T lymphocyte receptors CD28 and CTLA-4 bind costimulatory molecules CD80 (B7-1) and CD86 (B7-2) on antigen-presenting cells and regulate T cell activation. While distinct functional roles have been ascribed to each of these molecules, little is known about how they interact. To better characterize these interactions, we have used surface plasmon resonance to perform equilibrium and kinetic binding analyses of extracellular fragments of CD28/CTLA-4/CD80/CD86. We show that CTLA-4 and CD28 binding are both characterized by rapid kinetic on-rates and rapid dissociation rates. Native disulfide-linked homodimers of CD28 and CTLA-4 bound with two kinetically distinct binding sites, one of high avidity and slow dissociation and one of low avidity and more rapid dissociation. Monomeric CTLA-4 bound only with low affinity and rapid dissociation. Therefore, covalent dimerization of CTLA-4 is required for its high avidity binding. Oligomerization of CD80/CD86 is also required for high avidity CTLA-4 binding since CTLA-4 bound with low avidity to monomeric CD86. This contrasts with the ability of CD80/CD86 on antigen-presenting cells to bind CTLA4Ig with high avidity and predicts their organization as oligomers or clusters that permit multivalent binding. Thus, covalent receptor dimerization and ligand oligomerization are two key features of the CD28/CTLA-4/CD80/CD86 receptor system that control ligand binding and may regulate signal transduction by controlling the duration of receptor occupancy.

Despite the structural similarities and common binding partners of CD28/CTLA-4 receptors and CD80/CD86 ligands, different functional properties have been attributed to each of these molecules. Engagement of CD28 by monoclonal antibodies (mAbs) or B7 ligands greatly enhances T cell activation (3)(4)(5)(6), whereas engagement of CTLA-4 by specific mAbs may suppress T cell proliferation (14 -16). Additional evidence that CTLA-4 negatively regulates the immune system comes from studies of CTLA-4-deficient mice, which have a severe lethal lymphoproliferative phenotype (18,19). Thus, CD28 and CTLA-4 apparently have different functions during an immune response (17). CD80 and CD86 have also been reported to have different functions, with CD86 preferentially stimulating the production of interleukin-4, but not interleukin-2 (20,21).
The different biological properties of CD28 and CTLA-4 receptors and CD80 and CD86 ligands may emanate from differing binding properties or from different patterns of expression. Relatively little is known about how these molecules interact. Studies using recombinant soluble immunoglobulin fusion proteins have shown that CTLA-4 binds with higher avidity to CD80 and CD86 than does CD28 (9 -12). Amino acid residues responsible for the increased binding activity of recombinant CTLA4Ig were localized to the CDR1 and extended CDR3-like regions of the Ig variable-like domain (8). CD80 and CD86 had similar equilibrium properties for binding to CTLA-4 and CD28, although different kinetic properties were noted for binding of these ligands to .
Despite these studies, many fundamental properties of CD28/CTLA-4/CD80/CD86 interactions remain unknown. For instance, it is not known how the subunit structures of CD28 and CTLA-4 affect their binding properties. While these molecules have two binding sites per homodimer (13), previous studies have shown only single equilibrium constants for CD80/CD86 binding to CD28 and CTLA-4 (9 -12). It is also unclear how previous binding determinations were affected by multivalent Ig fusion proteins. Moreover, very little information is available on the binding properties of CD28, as studies on this molecule have been hampered by its weak avidity.
It was therefore of interest to better characterize the binding interactions of CD28/CTLA-4/CD80/CD86. In this study, we have used surface plasmon resonance (SPR) to measure the precise equilibrium and kinetics of binding of defined fragments of CD28/CTLA-4 and CD80/CD86. SPR provides a sensitive means to measure, in real time, binding interactions between biological molecules without problems associated with other techniques (22). Binding of soluble proteins to immobilized ligands is measured by the accumulation of mass on the surface of a sensor chip. SPR can be used to directly measure equilibrium binding constants and also to estimate kinetic association (k on ) and dissociation (k off ) rate constants, which can then be used to calculate equilibrium binding constants.
The results of our analysis demonstrate previously unknown roles for CD28/CTLA-4 dimerization and CD80/CD86 oligomerization in regulating binding interactions between these molecules.

Ig Fusion Proteins and Thrombin
Fragments-Chimeric mAb L6 was obtained from Bristol-Myers Squibb Pharmaceutical Research Institute (Seattle, WA). Anti-CTLA-4 mAbs 10A8 and 11D4; anti-CD28 mAb 9.3; and Ig fusion constructs CTLA4Ig, CD28Ig, CD80Ig, and CD86Ig have also been described (11,23). For preparation of recombinant forms of the extracellular domains of CD28 and CTLA-4, Ig fusion proteins were prepared that contained a synthetic thrombin cleavage site between the extracellular domains and the Fc regions. Polymerase chain reaction products encoding the extracellular domains of CD28 and CTLA4Ig (9,10) were digested with HindIII and BclI restriction enzymes and ligated into a HindIII/BamHI-digested Ig expression vector encoding a synthetic thrombin cleavage site 5Ј to the hinge region of a human IgG1 Fc domain (24). The resulting thrombin cleavage site-containing fusion constructs were designated CTLA4tIg and CD28tIg, respectively. Ig fusion proteins were produced by transiently transfected COS cells or stably transfected Chinese hamster ovary cells and purified by affinity chromatography on immobilized protein A-Sepharose (Repligen, Cambridge, MA) (9). The purified recombinant Ig fusion proteins had properties indistinguishable from those of CD28Ig and CTLA4Ig (9,10). For preparation of the dimeric extracellular domains of these proteins, the fusion proteins were digested with purified bovine thrombin (Armour Pharmaceutical Corp., Kankakee, IL) at a final concentration of 5 units/mg of protein at 37°C for 40 -60 min. The cleaved extracellular domains were purified from other digestion products (13) and are designated CTLA-4tp, and CD28tp, respectively. The predicted amino acid sequence of CTLA-4tp contains residues 1-127 (numbered as in Ref. 8) of CTLA-4 fused to a thrombin-cleaved linker peptide (-PDSDpgggggrlv-, where lower-case letters denote linker sequences). CD28tp contains residues 1-134 of CD28 fused to a thrombin-cleaved linker (-PSKPdpgggggrlv-). Concentrations of CTLA-4tp and CD28tp were estimated using extinction coefficients of 2.2 and 1.7 ml/mg, respectively; these were experimentally determined by measuring the A 280 of solutions whose protein concentration was determined by amino acid analysis. Molecular weights were calculated using monomeric molecular weights. CTLA-4tp and CD28tp preparations were analyzed by gel permeation chromatography to ensure their homogeneity; all preparations used in this paper contained Ͻ2% of high molecular weight aggregate(s). Some preparations of CTLA-4tp contained excessive amounts of high molecular weight aggregate(s), so these were subjected to further purification by size fractionation of the main dimeric species by gel permeation chromatography. CTLA-4tp and the previously described thrombin fragment, CTLA-4t (13), were used interchangeably in these experiments since they had identical binding properties. Preparation of CTLA4X C120S (a monomeric form of CTLA-4 with a cysteine to serine mutation in the position of the interchain disulfide of CTLA-4) and CD86t (a monomeric form of the extracellular domain of CD86) was described previously (13).
Analytical Techniques-Two columns were used for gel permeation chromatography. In most experiments, protein fragments were analyzed on a TSK-GEL G3000 SWXL column (Tosohaas, Montgomeryville, PA) equilibrated in phosphate-buffered saline containing 0.02% NaN 3 at a flow rate of 1 ml/min. In some experiments, protein fragments were analyzed on a Waters SW 991 column equilibrated with phosphatebuffered saline containing 10 mM Tris-HCl, pH 7.4, and 0.01% NaN 3 at a flow rate of 0.35 ml/min. Equivalent results were obtained with both columns. SDS-PAGE was performed on Tris/glycine gels (Novex, San Diego, CA). Gels were stained with Coomassie Blue, and images of wet gels were obtained by digital scanning.
BIAcore Analysis-All experiments were run on BIAcore TM or BIAcore 2000 biosensors (Pharmacia Biotech AB, Uppsala) at 25°C. Ligands were immobilized on research-grade CM5 sensor chips (Pharmacia) using standard N-ethyl-NЈ-(dimethylaminopropyl)carbodiimid/Nhydroxysuccinimide coupling (22,25). Proteins were diluted in 10 mM sodium formate buffer, pH 4.0, and incubated with activated sensor chips for times varied to optimize the degree of derivatization needed for particular experiments. After coupling, excess N-hydroxysuccinimide groups were inactivated with ethanolamine HCl (22,25). Control chips were either derivatized with chimeric mAb L6 or mock-derivatized by activating and then inactivating with ethanolamine without the addition of protein. Identical results were obtained with both types of control chips. After binding of CTLA-4 or CD28, CD80Ig-and CD86Ig-coated chips were regenerated with a 60-s pulse of 50 mM sodium citrate and 500 mM NaCl, pH 4.0 or 5.0, respectively. Control experiments showed that CD80Ig-and CD86Ig-derivatized surfaces were able to withstand Ͼ30 rounds of regeneration without losing binding capacity. Before use, all sensor surfaces were subjected to several rounds of analyte binding, followed by regeneration to ensure a stable level of derivatization. Mobile phase buffer for immobilization was phosphatebuffered saline, pH 7.4, containing 0.005% P20 surfactant (Pharmacia). For binding assays, 200 g/ml bovine serum albumin was added to the mobile phase buffer.
BIAcore Data Analysis-Sensorgram base lines were normalized to zero response units (RU) prior to analysis. Samples were run over mock-derivatized flow cells to determine background RU values due to bulk refractive index differences between solutions. Equilibrium dissociation constants (K d ) were calculated from plots of R eq versus C, where R eq is the steady-state response minus the response on a mock-derivatized chip, and C is the molar concentration of analyte. Binding curves were analyzed using commercial nonlinear curve-fitting software (Prism, GraphPAD Software, San Diego, CA). Experimental data were first fit to a model for a single ligand binding to a single receptor (one-site model, i.e. a simple Langmuir system, A ϩ B 7 AB), and equilibrium dissociation constants Subsequently, data were fit to the simplest two-site model of ligand binding (i.e. to a receptor having two non-interacting independent binding sites), as described by the equation The goodness-offits of these two models were analyzed visually by comparison with the experimental data and statistically by an F test of the sums-of-squares. The simpler one-site model was chosen as the best fit unless the twosite model fit significantly better (p Ͻ 0.1).
Association and dissociation analyses were performed using BIAevaluation 2.1 software (Pharmacia). Association rate constants (k on ) were calculated in two ways, assuming both homogeneous single-site interactions and parallel two-site interactions. For single-site interactions, k on values were calculated according to the equation R t ϭ R eq (1 Ϫ exp Ϫks(tϪto )), where R t is the response at a given time, t; R eq is the steady-state response; t o is the time at the start of the injection; and k s ϭ dR/dt ϭ k on ⅐C ϩ k off , where C is the concentration of analyte, calculated in terms of monomeric binding sites. For two-site interactions, k on values were calculated according to the equation ). For each model, the values of k on were determined from the calculated slopes (to ϳ70% maximal association) of plots of k s versus C.
Dissociation data were analyzed according to one-site (AB ϭ A ϩ B) or two-site (AiBj ϭ Ai ϩ Bj) models, and rate constants (k off ) were calculated from best fit curves. The one-binding site model was used except when the residuals were greater than machine background (2-10 RU, according to machine), in which case the two-binding site model was employed. Half-times of receptor occupancy were calculated using the relationship t1 ⁄2 ϭ 0.693/k off .

Preparation of Defined Fragments of the Extracellular Domains of CTLA-4 and CD28
-Accurate measurement of CD28/ CTLA-4/CD80/CD86 interactions required preparation of fragments of these molecules with defined valencies. In a previous study (13), we described the preparation and characterization of dimeric and monomeric forms of the extracellular domain of CTLA-4 and a monomeric form of CD86 (CD86t). Preparation of these fragments was accomplished by thrombin cleavage of (presumed) dimeric CTLA4Ig and CD86Ig at a nonclassical cleavage site in the Fc hinge region of these molecules (13). While these fragments were useful, they were difficult to produce in sufficient yield because of inefficient thrombin cleavage. To improve the yield of these extracellular fragments, we re-engineered the CTLA4Ig and CD28Ig fusion proteins such that a synthetic thrombin recognition site was introduced between the extracellular domains of CTLA-4 or CD28 and the Fc domain (see "Materials and Methods"). The resulting CTLA4tIg and CD28tIg fusion proteins had binding properties indistinguishable from those of our original fusion proteins, but could be more efficiently cleaved by thrombin. 2 From these new fusion proteins, we prepared fragments of the extracellular domains of CTLA-4 and CD28, termed CTLA-4tp and CD28tp, as described under "Materials and Methods." CTLA-4tp and CD28tp retained binding activities essentially equivalent to those of the corresponding Ig fusion proteins as indicated by monitoring binding activities during purification. These proteins also retained the ability to bind anti-CTLA-4 and anti-CD28 mAbs (data not shown), indicating that they had native conformations. These fragments were characterized by SDS-PAGE and gel permeation chromatography in the experiments shown in Fig. 1. Under nonreducing conditions, CTLA-4tp migrated at M r ϭ 55,000 -60,000 when analyzed by either technique; upon reduction, it migrated at M r ϭ 25,000 -35,000. CTLA-4tp therefore was a disulfide-linked dimer. CD28tp migrated at M r ϳ 65,000 and 45,000 by SDS-PAGE under nonreducing and reducing conditions, respectively. In contrast, during gel permeation chromatography, CD28tp eluted at M r ϳ 180,000 and 65,000 under nonreducing and reducing conditions. The elution volume of CD28tp during gel permeation chromatography was therefore greater than would be predicted by SDS-PAGE. This difference may be due to the predicted extensive glycosylation of CD28tp, or it may indicate that CD28tp in solution existed as a tetramer. At present, we cannot distinguish between these possibilities, but to simplify subse-quent calculations, we have assumed that CD28tp was dimeric. Neither CTLA-4tp nor CD28tp preparations used for binding measurements had significant amounts of aggregated material present. Monomeric CD86 (CD86t) and monomeric CTLA-4 (CTLA4X C120S , prepared by site-directed mutagenesis of the cysteine residue that mediates the intrachain disulfide bond) were prepared as described (13).
Equilibrium Binding Analysis of CTLA-4 and CD28 -We initially used SPR to measure equilibrium binding of CTLA4Ig, CD28Ig, CD80Ig, and CD86Ig fusion proteins using the methods of van der Merwe et al. (26). This procedure was devised for determining the equilibrium binding properties of lymphocyte surface receptors with low binding avidity and rapid kinetic off-rates. CD80Ig and CD86Ig were coupled to sensor chips, and solutions of CTLA4Ig and/or CD28Ig were added in the mobile phase. The levels of derivatization of the chips were adjusted to ensure that steady-state (equilibrium) binding was achieved at low concentrations of analyte. Similar results were obtained at different levels of chip derivatization. Equilibrium binding of CTLA4Ig at a given concentration was determined by comparing the steady-state RU on CD80Ig-or CD86Igcoated chips with the value obtained on control chimeric mAb L6 or mock-derivatized chips. From this comparison, levels of specific binding were determined at different concentrations. Equilibrium dissociation constants (K d ) could then be determined from binding isotherms by curve fitting. Similar results were obtained whether we measured binding of CTLA4Ig to immobilized CD80Ig or CD86Ig or binding of CD80Ig or CD86Ig to immobilized CTLA4Ig (data not shown). Thus, the equilibrium constants obtained were not greatly influenced by which protein was immobilized, indicating that protein immobilization to the sensor chip did not significantly affect binding activity (data not shown; also see results presented in Tables I  and II, below). For most subsequent experiments, we chose to measure binding of CTLA-4 and CD28 fragments with defined valency to immobilized CD80Ig and/or CD86Ig.
We measured equilibrium binding of monomeric CTLA-4 (CTLA4X C120S ), dimeric CTLA-4 (CTLA-4tp), and dimeric CD28 (CD28tp). Representative experiments demonstrating equilibrium binding of these fragments to CD80Ig are shown in Figs. 2-4, respectively. A summary of all equilibrium binding experiments is given in Table I. The binding isotherm for CTLA4X C120S was adequately fit by a one-site model as judged by visual examination of the fitted curves (Fig. 2); the more complicated two-site model also did not statistically improve the goodness-of-fit significantly (p Ͼ 0.1). Therefore, single equilibrium dissociation constants (K d ) were chosen for binding of CTLA4X C120S to CD80Ig and CD86Ig (Table I).
In contrast, the binding isotherm for dimeric CTLA-4tp ( Fig.  3) showed more complicated equilibria. Binding saturation was achieved over a wider concentration range than for CTLA4X C120S , and although binding of CTLA-4tp was measurable at much lower analyte concentrations than for CTLA4X C120S , binding equilibrium for CTLA-4tp was not achieved at lower concentrations, even over relatively long periods of time (compare Figs. 2 and 3). For binding of CTLA-4tp, the two-site model clearly increased the goodness-of-fit, both visually (Fig. 3) and statistically (p Ͻ 0.1). Both high and low avidity equilibrium dissociation constants (K d ) were estimated for CTLA-4tp (Table I), with the high avidity sites accounting for ϳ80 and ϳ60% of the total binding to CD80Ig and CD86Ig, respectively. The low avidity K d values of CTLA-4tp binding to CD80Ig or CD86Ig were similar to the K d values for CTLA4X C120S (Table I). The high avidity K d1 values for CTLA-4tp binding to CD80Ig and CD86Ig were ϳ200and ϳ500-fold less, respectively (indicating stronger binding), than the K d values for CTLA4X C120S . Thus, CTLA-4tp bound more strongly (avidly) than did CTLA4X C120S .
The binding isotherm for dimeric CD28tp also showed complicated equilibria (Fig. 4). Binding of CD28tp was measurable only at very high analyte concentrations. This dictated that large quantities of purified CD28tp were required for binding measurements, and with the amounts of material available, it was not possible to achieve saturation binding. Despite the limited concentration range of analyte used, the CD28tp bind-ing isotherm was better fit by a two-site model, although visually the difference between one-and two-site models was less compelling than with CTLA-4tp (compare Figs. 3 and 4). The two-site model also gave statistically better fits of the data (p Ͻ 0.1). That support for the two-site model was weaker for CD28tp than for CTLA-4tp may reflect experimental limitations of the weak binding interactions of CD28tp or the limited concentration range of analyte used; alternatively, it may indicate that CD28tp binding is better explained by even more complicated models. However, since the kinetic binding analysis presented below also supports a two-site rather than a one-site model for CD28tp binding, we have chosen to use the simple two-site model to estimate both high and low avidity equilibrium dissociation constants (K d ) for CD28tp (Table I). The percentage of high avidity binding sites (K d1 ) for CD28tp was less than for CTLA-4tp (ϳ15 and ϳ23% for binding to CD80Ig and CD86Ig, respectively). The reason for this is presently unknown.
Kinetic Binding Analysis of CTLA-4 and CD28 -We also determined binding kinetics for CD28/CTLA-4/CD80/CD86 interactions since these are likely to be critical determinants of the biological effects of dynamic receptor/ligand interactions. We analyzed the kinetics of CTLA-4 and CD28 binding to CD80Ig-and CD86Ig-coated sensor chips. Representative determinations of the association and dissociation kinetics for CTLA4X C120S , CTLA-4tp, and CD28tp binding to CD80Ig are shown in Figs. 5 and 6, respectively. A summary of the association (k on ) and dissociation (k off ) rate constants is presented in Table II.
For determination of association rate constants, sensor chips were derivatized with relatively low amounts of CD80Ig or CD86Ig so that association was not limited by mass transport of analyte to the chip surface. The binding responses over time were determined at several concentrations of analyte for CTLA4X C120S , CTLA-4tp, and CD28tp. Curves were fitted to one-and two-site models, and the slopes (dR/dt ϭ k s ) were calculated as described under "Materials and Methods." When k s values were plotted versus analyte concentration, linear relationships were reliably and reproducibly obtained when k s values were obtained with one-site models. Values of k on could then be calculated from the slopes of plots of k s versus analyte concentration. When association data were analyzed using two-site models, less reliable k on values were obtained for all combinations. In some experiments, it was not possible to calculate two meaningful k on values. In other experiments, when two k s values could be calculated, the difference between the resulting k on values was always relatively small (less than ) from two to five independent experiments. K d values were obtained by curve fitting experimental data to one-and two-site models as described under "Materials and Methods." p values for the statistical improvement in choosing a particular two-site model over a one-site model are given. When a two-site model was chosen (p Ͻ 0.1), both high avidity (K d1 ) and low avidity (K d2 ) dissociation constants were calculated. The proportion of total binding attributable to the high avidity site is indicated as percent K d1 . Calculated K d values were derived from experimentally determined kinetic rate constants (see Table II) using the equation K d ϭ k off /k on . Where two k off values were determined (see Table II), two values for K d are given; the calculated percent K d1 was determined from the percentage of total binding attributable to the more slowly dissociating (higher avidity binding) component (see Table II

CD28 and CTLA-4 Binding to CD80 and CD86
ϳ10-fold). Finally, in no case did use of the more complex two-site model improve the linear relationship (as judged by correlation coefficient (r) and the standard error of measurement) between the k s values and concentration. For these reasons, we therefore concluded that use of two-site models of association was not justified. We determined a single association rate constant (k on ) for each combination of proteins (Table  II). Monomeric CTLA-4 bound with k on values within ϳ2-6-fold those for dimeric CTLA-4 (for CD80Ig and CD86Ig, respectively). The similarity in these k on values argues that the native conformation of monomeric CTLA-4 was maintained. The rapid k on values determined for CTLA-4 were near the limits of instrument detection. The apparent dissociation rate constants (k off ) can be influenced by the rebinding of dissociated analyte. For these determinations, therefore, it was desirable to minimize rebinding by derivatizing sensor chips with moderately low levels of CD80Ig or CD86Ig and by maximizing analyte concentrations. Dissociation of monomeric and dimeric CTLA-4 and dimeric CD28 was measured from CD80Ig-derivatized (Fig. 6) or CD86Ig-derivatized (data not shown) sensor chips. Dissociation curves were fit with one-and two-binding site models (Fig. 6). Dissociation of monomeric CTLA-4 from CD80Ig (Fig. 6, A and B) and CD86Ig (data not shown) was approximated by a one-binding site model such that the residuals between the observed and calculated curves were within experimental variation of BIAcore measurements. Analysis of these dissociation curves gave very rapid k off values (Table II), corresponding to half-times of receptor occupancy of ϳ8 and ϳ2 s for CD80Ig and CD86Ig, respectively. The k off values calculated in this way were in good agreement with those estimated from the y intercept of the k s versus concentration plots used to calculate k on values (Fig. 5). From the determined k off and k on values for monomeric CTLA-4 binding to CD80Ig and CD86Ig, equilibrium binding constants (K d ) could be calculated (Table I), which were in good agreement with the experimentally determined K d values.
Analysis of dissociation curves for dimeric CTLA-4 and CD28 was more complicated (Fig. 6, C and D, and E and F, respectively). In neither case did a one-binding site model provide a satisfactory fit to the experimental data (Fig. 6, C and E); residuals were greater than the experimental variation, especially in the early phases of the dissociation curves. However, for both dimeric CTLA-4 and CD28, two-binding site models gave reasonable agreement between the observed and expected binding (Fig. 6, D and F), and residuals were within experimental variation. From these models, two k off values could be calculated (Table II), consistent with the calculation of two equilibrium binding constants (K d ) (Table I). From the kinetic constants (k on and k off ) (Table II), equilibrium binding constants (K d ) for CTLA-4tp and CD28tp binding could be calculated (Table I). These, for the most part, agreed well with the experimentally determined values, both in magnitude and in percentage of total binding.
High Avidity Binding of CTLA-4 Is Due to Rebinding-Binding of dimeric CTLA-4tp was characterized by two K d values and two k off values (Tables I and II). The more rapidly dissociating (weaker binding) k off2 values for dimeric CTLA-4tp approximated the k off value for monomeric CTLA4X C120S . Slower dissociation (stronger binding) could indicate rebinding of dimeric CTLA-4 during the dissociation measurements (11). To test this possibility, we measured k off rates of CTLA-4tp from immobilized CD80Ig and CD86Ig in the presence of increasing amounts of soluble CD80Ig (Fig. 7). This would be expected to occupy free CTLA-4tp binding sites and prevent their rebinding to immobilized CD80Ig or CD86Ig.
As shown in Fig. 7, the rate of dissociation of CTLA-4tp increased in the presence of increasing concentrations of soluble CD80Ig. Very high concentrations (Ͼ1 mg/ml) of CD80Ig were required to produce these effects, probably because of the high density of CD80Ig and CD86Ig immobilized on the sensor chips. The apparent dissociation constants (k off ; values determined using a one-binding site model) for CTLA-4tp measured in the presence of increasing competitor increased asymptotically and approached values determined for monomeric CTLA4X C120S (Table II). The more slowly dissociating, higher avidity component of dimeric CTLA-4tp binding was therefore reduced when rebinding was blocked. In other experiments, we measured the effects of increasing concentrations of CD80Ig on the dissociation of CTLA4X C120S and CD28tp from immobilized CD80Ig and CD86Ig. In contrast to its effects on dissociation from CTLA-4tp, soluble CD80Ig did not greatly affect dissociation of these other molecules (Ͻ2-fold effect). Thus, rebinding of monomeric CTLA-4 or dimeric CD28 affected dissociation measurements less than it did for dimeric CTLA-4. It should be emphasized that the k off values determined for these molecules in the absence of competition were near the limit of detection, so further increases may not have been possible to measure.
Effects of CD86 Valency on High Avidity Binding of CTLA-4 -It was also important to determine the effects of valency of CD80/CD86 molecules on CTLA-4 binding avidity. We therefore measured binding of monomeric CD86 to immobilized CTLA4Ig. The use of immobilized CTLA4Ig was justified by other experiments (data not shown) showing that CTLA4Ig gave similar equilibrium and kinetic constants as CTLA-4tp for binding to immobilized CD80Ig and CD86Ig.
We compared equilibrium binding of monomeric CD86t and multimeric CD86Ig to immobilized CTLA4Ig (Table III). For monomeric CD86t, the data were adequately fit by a onebinding site model, yielding a low affinity K d of ϳ1400 nM. This was similar to the K d measured for binding of monomeric CTLA4X C120S to immobilized CD86Ig (ϳ2200 nM) ( Table II). The association constants (k on and k off ) for CD86t binding to CTLA4Ig were estimated as 3.7 ϫ 10 5 M Ϫ1 s Ϫ1 and 0.5 s Ϫ1 , which were also very similar to those measured for binding of monomeric CTLA4X C120S to immobilized CD86Ig (Table II). Thus, both equilibrium and kinetic binding constants for binding of monomeric CD86t to CTLA4Ig reflect low avidity binding. In contrast, binding of multimeric CD86Ig gave both high and low avidity equilibrium binding constants (K d ϭ 7 and 110 nM), which were very similar to those determined for dimeric CTLA-4tp binding to immobilized CD86Ig (Table II). Thus, valency of both CTLA-4 and CD86 affects binding avidity. DISCUSSION Previous studies showed that a covalent CTLA-4 homodimer contained two binding sites for CD80/CD86 molecules, but how these two sites affected binding avidity was not examined (13). Indeed, studies on the binding properties of CTLA-4 and CD28 identified only single avidity classes of binding sites on these molecules (7, 9 -12). Here, using the more sensitive techniques of SPR, we show that dimeric CD28 and CTLA-4 show both high and low avidity binding, consistent with their 2:2 binding stoichiometries.
Binding of both monomeric and dimeric CTLA-4 to CD80Ig or CD86Ig was characterized by very rapid apparent k on values (ϳ0.2-1 ϫ 10 6 M Ϫ1 s Ϫ1 ). It should be mentioned that determination of kinetic on-rates to ligands immobilized in a polymer matrix can lead to underestimation of the true kinetic on-rates (28). The kinetic on-rates for CTLA-4 binding are comparable to those for other T lymphocyte receptor/ligand pairs, such as CD2/CD58 (Ն4 ϫ 10 5 M Ϫ1 s Ϫ1 ) (26) and T cell receptor-major histocompatibility class I peptide complexes (2.1 ϫ 10 5 M Ϫ1 s Ϫ1 ) TABLE II Summary of apparent kinetic constants Association and apparent dissociation rate constants were determined from experiments performed as described for Figs. 5 and 6, respectively. Shown are means Ϯ S.D. of association (k on ) and dissociation (k off ) rate values from multiple experiments; association rate determinations were performed two to four times, and dissociation rate determinations, four to eight times. Where a two-site model best fitted the data, two values for k off are given. Percent k off1 is the percentage of total binding accounted for by the slower dissociating (higher avidity binding) component. k off values Ͼ400 ϫ 10 Ϫ3 s Ϫ1 are less reliable since they approach the limits of the BIAcore.  (27). The k off values for CD28 binding to CD80Ig and CD86Ig were also very rapid, but were similar to other lymphocyte surface receptors (26,27). This would predict a short time of receptor occupancy for CD28/CD80/CD86 interactions (half-times of less than ϳ1 s for CD28/CD80/CD86, compared with less than or equal to ϳ6 s for CD2/CD58 (26) and ϳ17 s for T cell receptor/major histocompatibility complex class I peptide interactions (27)). These values for CD28 were calculated from FIG. 6. Apparent dissociation kinetics of monomeric CTLA-4, dimeric CTLA-4, and dimeric CD28 binding to immobilized CD80Ig. Shown is a comparison of dissociation models for binding of monomeric CTLA-4 (A and B), dimeric CTLA-4 (C and D), and dimeric CD28 (E and F). Representative dissociation sensorgrams (solid lines) are compared with models (dashed lines) for one-site (left panels) and two-site (right panels) dissociation. Residuals between observed and expected values are expressed as a percentage of the analyte bound at the start of the dissociation (dotted lines). Dissociation of CTLA4X C120S , CTLA-4tp, and CD28tp was measured at concentrations of 200, 200, and 10,000 nM, respectively; in each case, the amount of CD80Ig immobilized was 2290, 3640, and 4300 RU, respectively. Analyses were run at 60 l/min, and the resulting sensorgrams were normalized to t ϭ 0 at the start of the dissociation period. bivalent (higher avidity) binding constants; monovalent (lower avidity) binding would lead to much shorter occupancy times. Comparison of occupancy times for dimeric CTLA-4 is more difficult because of the effects of receptor valency on apparent dissociation rate. Dimeric CTLA-4 has slower kinetic off-rates compared with CD28, but its time of receptor occupancy is still relatively short (half-time of less than ϳ180 s). Monomeric CTLA-4 binding would also lead to a short period of receptor occupancy (less than ϳ8 and ϳ2 s for binding to CD80Ig and CD86Ig, respectively), but the physiological relevance of this is unclear since CTLA-4 primarily exists as a disulfide-linked homodimer.
CD80 and CD86 have very similar equilibrium binding properties, despite sharing very little amino acid sequence homology (11). Nevertheless, CD80 and CD86 were reported to deliver qualitatively different costimulatory signals during T cell activation (20,21). It is therefore of interest to examine the binding properties of these molecules in detail. Certain of the measurements made here confirm measurements made in previous studies (11). For example, the K d values for dimeric CTLA-4 binding to CD80Ig and CD86Ig were similar to those reported previously for CTLA4Ig binding. Also, binding of dimeric CTLA-4 to CD80Ig was 3-4-fold higher than to CD86Ig (11). The difference in the kinetic off-rates of monomeric CTLA-4 binding to CD80Ig and CD86Ig (Table II) was similar to a previous estimate made using transfected cells (11). In other instances, however, SPR has permitted comparisons which were not possible in previous studies. Here we provide the first estimates of affinity constants (as opposed to avidity constants) for monomeric CTLA-4 binding to CD80Ig and CD86Ig (ϳ200 and ϳ2200 nM), which differ by ϳ10-fold. It remains to be determined how different CD80 and CD86 binding properties affect signaling properties of these molecules.
A comparison of the binding properties of CTLA-4 and CD28 is important in light of the apparently different biological functions of these molecules (17). Binding of dimeric CTLA-4 to CD80Ig was 2500-and 270-fold stronger (for K d1 and K d2 , respectively) compared with dimeric CD28. This difference is greater than that previously determined (9, 10) because of the more than ϳ10-fold larger (indicating weaker binding) K d values for high avidity binding of dimeric CD28 measured in the present study. This difference probably arises from aggregation and uncertain valencies of the Ig fusion proteins used previously. Despite the weak binding of dimeric CD28 measured in this study, our analysis predicts that monomeric CD28 would bind even weaker. Binding of dimeric CTLA-4 to CD86Ig was

TABLE III
High avidity binding of CTLA-4 requires multivalent CD86 Equilibrium binding of the indicated pairs of proteins was measured as described in the legend to Fig. 2. High avidity (K d1 ) and low avidity (K d2 ) equilibrium dissociation constants were calculated from best fit curves calculated from a two-site model as described under "Materials and Methods." Values presented are the average of two experiments. 570-and 70-fold stronger (for K d1 and K d2 , respectively) compared with CD28.
Rapid kinetic off-rates facilitate the transitory intercellular interactions characteristic of lymphocytes, but present the problem of how to maintain CD80/CD86 ligand occupancy long enough to generate productive signals. We have demonstrated here two characteristics of the CD28/CTLA-4/CD80/CD86 receptor system that would tend to counteract the effect of fast kinetic off-rates and thereby increase the duration of receptor occupancy. The first of these is the influence of the two CD80/ CD86-binding sites per pre-existing CD28/CTLA-4 covalent dimer; the second, oligomerization of CD80/CD86 molecules on APC.
We have shown that bivalent CTLA-4 and probably CD28 have different kinetic and equilibrium binding characteristics compared with the monomeric molecules. Bivalently bound CD28/CTLA-4 receptors have a decreased likelihood of dissociating from CD80/CD86 ligands. This decreased dissociation due to bivalent binding would prolong the effective receptor occupancy time.
Oligomerization of CD80 and/or CD86 molecules would also tend to prolong CD28/CTLA-4 receptor occupancy. CD80/CD86 molecules are not known to dimerize (7). However, while (dimeric) CTLA4Ig binds with high avidity to APC, CTLA-4 dimers do not bind with high avidity to monomeric CD86. High avidity binding of CTLA4Ig to APC may be accounted for by oligomerization or clustering of CD80/CD86 molecules on the cell surface. Clusters of CD80/CD86 could either pre-exist or may be ligand (CTLA4Ig)-induced. Available data support the existence of pre-existing clusters of CD80/CD86 molecules since high avidity binding of CTLA4Ig is observed under conditions that discourage ligand-induced clustering (i.e. at low temperature, in the presence of metabolic inhibitors, or on fixed cells). The existence of pre-existing oligomers of CTLA4Ig-binding molecules on human Langerhans cells was also suggested by earlier immunofluorescence microscopy studies (29). Preexisting oligomers of CD80/CD86 molecules would favor multiple engagements with CD28 and/or CTLA-4 receptors.
Engagement of CD28 and/or CTLA-4 receptors by clustered CD80/CD86 ligands would increase the local receptor concentration and provide for greater receptor occupancy. Despite the concentration-driven increase in receptor occupancy, the rapid dissociation kinetics of CD28/CTLA-4/CD80/CD86 complexes would still result in transiently unoccupied CD80/CD86 molecules. These could then rebind other CD28/CTLA-4 receptors, perhaps allowing engagement of many CD28/CTLA-4 receptors by relatively few CD80/CD86 molecules. A similar process has been suggested for T cell receptor-major histocompatibility class I peptide complexes (30).
Clustering of CD80/CD86 molecules on the APC surface may explain earlier findings that CD28 aggregation by mAbs enhances signaling through this receptor and alters the biochemical nature of signals generated (reviewed in Ref. 31). Clusters of CD80/CD86 molecules could cross-link different CD28/ CTLA-4 receptors because of their bivalent binding. Cross-linking would also be favored by the rapid kinetic on-and off-rates of binding. Clustering of CD80/CD86 molecules on APC may therefore be an important control point for the activity of this receptor/ligand system.