Kinetic Analysis of Adenovirus Fiber Binding to Its Receptor Reveals an Avidity Mechanism for Trimeric Receptor-Ligand Interactions*

Most adenoviruses bind to the N-terminal immunoglobulin domain D1 of the coxsackievirus and adenovirus receptor via the head part of their fiber proteins. Three receptor molecules can bind per fiber head. We expressed the D1 domain and the adenovirus type 2 fiber head in bacteria and studied binding interactions by surface plasmon resonance measurements. When receptor domains bind adenovirus fiber independently of each other, the dissociation constant is 20–25 nm. However, when adenovirus fiber binds to receptors immobilized on the sensor chip, a situation better mimicking adenovirus binding to receptors on the cell surface, the dissociation constant was around 1 nm. Kinetic analysis shows that this happens via an avidity mechanism; three identical interactions with high on and off rate constants lead to tight binding of one fiber head to three receptor molecules with a very low overall off rate. The avidity mechanism could be used by other viruses that have multimeric adhesion proteins to attach to target cells. It could also be more general to trimeric receptor-ligand interactions, including those involved in intracellular signaling.

Adenoviruses are responsible for a variety of respiratory, gastroenteric, and ocular infections (1). They are DNA viruses that predominantly infect mammals and birds. In humans 51 types have been identified and grouped into six distinct classes, A to F. Adenoviruses form icosahedral particles with 240 copies of the trimeric hexon protein arranged on the planes and a penton complex at each of the 12 vertices. The penton complex consists of a pentameric base, implicated in virus internalization (2), and an externally protruding trimeric fiber. The fiber protein is responsible for the initial attachment to the host cell, through its C-terminal head domain (3). The atomic structures of type 2 and type 5 fiber heads are known (4,5); furthermore, the fiber shaft has been shown to contain a novel triple ␤-spiral fold (6).
The receptor of most adenovirus types has been identified to be a cell surface protein named coxsackievirus and adenovirus receptor (CAR) 1 (7,8). CAR has been shown to be the receptor for subgroup A, C, D, E, and F virus fibers (9) but not for subgroup B (to which, for example, adenovirus type 3 and 7 belong). Short fibers from group F also do not bind CAR. The receptor-binding site on the adenovirus fiber head has recently been described based on comparative sequence analysis of CAR-binding and non-CAR-binding fiber head domains (4,10) and mutagenesis studies (10,11). The structure of adenovirus type 12 fiber head complexed with the CAR N-terminal Ig D1 domain, which is necessary and sufficient for adenovirus attachment (12), is also known (11). Together, the data show that the primary determinant for CAR binding on the fiber head is the AB loop (for a definition of ␤-strands and loops of the fiber head, see Ref. 5). The AB loops are located on the sides of the head trimer close to an adjacent monomer. Binding of CAR to the fiber does not appear to lead directly to intracellular signaling events important for adenovirus infection because the intracellular domain plus transmembrane helix of CAR can be replaced by another membrane anchor with retention of CARmediated adenovirus infection (13). CAR has been proposed to function as a homophilic cell adhesion molecule (14), a proposition supported by the observation that the CAR D1 domain forms homodimers in solution and in the crystalline state (15).
Here we describe surface plasmon resonance (SPR) binding studies of adenovirus fiber to the D1 domain of CAR, using bacterially expressed proteins. We have determined the relevant association and dissociation constants of this interaction and report an avidity-based mechanism by which the adenovirus fiber binds to three receptor molecules immobilized on a surface. We propose that adenovirus fiber and other multimeric ligands bind their receptors on the cell surface in an analogous way.

EXPERIMENTAL PROCEDURES
Construction of Expression Vectors-For the expression of the adenovirus type 2 fiber head, the plasmid pT7.Ad2fib388 -582 was constructed. A DNA fragment coding for residues 388 -582 was obtained by polymerase chain reaction using pTaq (3) as a template and cloned into the vector pPMD (16), from which the NcoI-BamHI insert was removed (pPMD is a variant of pT7-7, a low copy number vector employing the T7 RNA polymerase system). For the construction of pAB3.CAR15-140H, a DNA fragment encoding residues 15-140 was obtained by polymerase chain reaction using the plasmid pcDNA1-CAR (7) as a template and cloned into the expression vector pAB3 (17), which permits expression under control of the lac promotor-operator and targeting to the bacterial periplasm by means of a pectate lyase leader peptide. The resulting protein contains a C-terminal metal-binding tag.
Protein Expression and Purification-The adenovirus type 2 fiber head was expressed in Escherichia coli strain JM109(DE3) (Promega, Charbonnières, France). Bacteria transformed with pT7.Ad2fib388 -* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
¶ To whom correspondence should be addressed. Present address: Gorlaeus Laboratories, Leiden University, P.O. Box 9502, Einsteinweg 55, 2300 RA Leiden, Netherlands. Fax: 31-71-527-4357; E-mail: m.vanraaij@chem.leidenuniv.nl. 1 The abbreviations used are: CAR, coxsackievirus and adenovirus receptor; RU, resonance units; PAGE, polyacrylamide gel electrophore-582 were grown in LB-Ap medium (10 g/liter Bacto tryptone, 5 g/liter yeast extract, 7.5 g/liter sodium chloride, 100 mg/liter ampicillin) at 37°C to an optical density of 0.5 at 600 nm. They were then cooled to 22°C, 0.5 mM isopropyl-␤-D-thiogalactopyranoside was added, and growth was continued for 16 h at 22°C. Cells from 5 liters of culture were resuspended in 100 ml of 50 mM Tris-HCl, pH 8.0, 25 mM sodium chloride, 1 mM dithiothreitol containing protease inhibitors (Complete TM ; Roche Molecular Biochemicals) and lyzed using a French press. Insoluble material was removed by centrifugation, nucleic acids were precipitated by adding 1% (w/v) streptomycin sulfate, and 70 ml of a saturated solution of ammonium sulfate was added to the supernatant. After centrifugation, the supernatant was loaded onto 75 ml of phenyl-Sepharose FF high subcolumn (Amersham Pharmacia Biotech) equilibrated with PE buffer (25 mM sodium dihydrogen phosphate, 25 mM disodium hydrogen phosphate, 1 mM EDTA, pH 6.8) containing 1.5 M ammonium sulfate. Elution was with a gradient of 1.5-0 M ammonium sulfate; the protein eluted at about 0.6 M. Fractions containing the desired protein were dialyzed against 10 mM Tris-HCl, pH 8.5, 1 mM EDTA and loaded onto a Q10 anion exchange column (Bio-Rad) equilibrated in the same buffer. This column was eluted with a gradient of 0 -200 mM sodium chloride; the protein eluted at about 30 mM. Fractions containing the desired protein were pooled, brought to 50 mM phosphate, pH 6.8, and 1.5 M ammonium sulfate by adding concentrated stock solutions, and loaded onto a phenyl superose 10/10 column (Amersham Pharmacia Biotech) equilibrated in PE buffer containing 1.5 M ammonium sulfate. Elution was with a gradient of 1.5-0 M ammonium sulfate; the protein eluted at about 1.3 M. Fractions containing pure protein as judged by SDS-polyacrylamide gel electrophoresis (PAGE) electrophoresis were pooled and transferred to HN buffer (10 mM HEPES, 150 mM sodium chloride, pH 7.4, with sodium hydroxide) by repeated dilution and concentration using Centricon Plus-20 devices with a nominal molecular mass cut-off of 10 kDa (Millipore, St. Quentin en Yveline, France). An adenovirus fiber protein containing part of the shaft region (residues 319 -592; Ref. 6) was similarly expressed and purified using the plasmid pT7.Ad2fib319 -582.
The D1 domain of CAR was expressed in E. coli with a 12-residue C-terminal extension (GAPAAAHHHHHH) to enable purification by metal-chelating chromatography. The resulting protein will be referred to as CAR D1. Cultures of E. coli strain XL1Blue (Stratagene, La Jolla, CA) transformed with pAB3.CAR15-140H were grown in LB-Ap medium at 25°C to an optical density of 0.5 at 600 nm. Expression was induced with 0.2 mM isopropyl-␤-D-thiogalactopyranoside, and growth continued for 10 h at 25°C. Cells from 5 liters of culture were washed with 150 ml of phosphate-buffered saline solution (pH 7.4). Washed cells were resuspended in 150 ml of 200 mM Tris-HCl, pH 8.0, 0.5 mM EDTA, 0.5 M sucrose, centrifuged as before, and resuspended in 150 ml of distilled water to release the periplasmic fraction. After a 20 min of incubation, cell debris were removed by centrifugation, and protease inhibitors (Complete TM EDTA-free; Roche Molecular Biochemicals) were added. The protein was bound to 2 ml of nickel-nitrilotriacetic acid slurry (Qiagen, Courtaboeuf, France) following the instructions supplied and eluted stepwise with imidazole solutions of increasing concentrations at pH 7.5; the protein eluted between 30 and 200 mM imidazole. Further purification was using a phenyl superose 10/10 column as described for the fiber head protein. The protein eluted at about 1.1 M ammonium sulfate. Fractions containing pure protein as judged by SDS-PAGE electrophoresis were transferred to HN buffer as described above. A protein comprising residues 15-140 without a His tag was also expressed in E. coli and purified (15).
The molar extinction coefficients of the expressed proteins were determined by UV spectroscopy combined with amino acid analysis. Molar extinction coefficients at 280 nm of 35.5 ϫ 10 3 M Ϫ1 cm Ϫ2 for the adenovirus type 2 fiber head (residues 388 -581), 38.2 ϫ 10 3 M Ϫ1 cm Ϫ2 for adenovirus type 2 head plus part of shaft fiber protein (residues 319 -582), 18.1 ϫ 10 3 M Ϫ1 cm Ϫ2 for His-tagged CAR D1, and 18.5 ϫ 10 3 M Ϫ1 cm Ϫ2 for non-His-tagged CAR D1 were obtained. Protein concentrations were then determined by UV spectroscopy alone. Dynamic light scattering measurements were performed with DynaPro-MS800 dynamic light scattering/molecular sizing instrument (Protein Solutions Ltd., Charlottesville, VA).
SPR Measurements and Analysis-Surface activation, protein immobilization, and binding assays were carried out using an upgraded Biacore 1000 SPR measurement apparatus (Biacore, St. Quentin en Yveline, France). Flow cells of a Biacore B1 sensor chip were activated with a cross-linking mixture consisting of 50 l of 0.2 M N-ethyl-NЈ-(3dimethyl-aminopropyl)-carbodiimide hydrochloride and 0.05 M N-hydroxysuccinimide, after which adenovirus type 2 fiber head (25 g/ml in 10 mM sodium acetate, pH 5) or CAR D1 (20 g/ml in 10 mM sodium acetate, pH 5.0) was injected. Different amounts of immobilized protein were obtained by varying the injected volume. Remaining activated groups were blocked with 50 l of 1 M ethanolamine, pH 8.5. A third flow cell was activated with the cross-linking mixture and immediately blocked with ethanolamine to serve as a negative control. No significant binding of fiber head or CAR D1 to the negative control flow cell was observed (not shown).
All binding assays were carried out at 25°C, and HBS buffer (10 mM HEPES, 150 mM sodium chloride, 3 mM EDTA, 0.005% P20 detergent, pH 7.4) was used as running buffer. CAR D1 was diluted in running buffer, and different concentrations were injected over the fiber head surface for 5 min each to study the association and equilibrium phase. The surface was then washed with HBS buffer for 15 min to study the dissociation phase. The fiber head-CAR D1 complex fully dissociated during that period of time, and regeneration steps were not necessary. In a second series of experiments, fiber head was injected over a CAR D1 surface. After each injection, the surface was regenerated with a 2-min pulse of 10 mM hydrochloric acid.
Maximum stoichiometries of binding were determined by comparing the increase in resonance units (RU) during the immobilization procedure with the maximal increase in RU during the binding phase. The increase in RU depends linearly on the increase in bulk refractive index, and the specific refractive index increment is closely similar for a wide range of proteins independent of amino acid composition. The increase in RU can thus be regarded as depending linearly on the mass of the molecules or complexes attached to the surface, within the limits of measurement. The stoichiometry can then be calculated by dividing the fractional increase in RU by the fractional molecular mass of the concerned species.
Where possible, equilibrium data were extracted from the sensorgram at the end of each injection and used to calculate the equilibrium dissociation constant independently of the kinetic analysis. Kinetic constants (k on and k off ) were derived from the association and dissociation curves of the sensorgrams either by linear transformation of the primary data or nonlinear fitting of the sensorgrams and by numerical integration (global fitting) of the data to different interaction models, using the Biacore BIAevaluation 3.1 software supplied with the apparatus. Binding curves obtained when CAR D1 was allowed to bind to immobilized fiber head were fitted to a simple A ϩ B 7 AB model provided with the software. When fiber head bound to immobilized CAR D1, we used the rate Equations 1-3 to fit our data, which describe "trivalent binding" (one trimeric fiber head can bind up to three monomeric CAR D1 molecules), according to Equations 4 -7, where A is the trivalent fiber head, B is the CAR D1 molecule, k on1 , k on2 , and k on3 are the association rate constants of Equations 1, 2, and 3, respectively, and k off1 , k off2 , and k off3 are the dissociation rate constants of Equations 1, 2, and 3, respectively.

Expression in Bacteria and Purification of Recombinant
Proteins-The adenovirus type 2 fiber head protein containing residues 388 -582 of the coding sequence, henceforth to be referred to as "fiber head," was expressed in E. coli and purified by a combination of hydrophobic interaction and anion exchange chromatography. The preparation yielded 3-5 mg of purified protein/liter of bacterial culture. N-terminal sequence analysis gave the correct sequence (AITIGNKNDD) and showed that the introduced methionine was post-translationally removed. The same protein was previously expressed in insect cells (3). We also expressed a construct containing the head domain plus part of the shaft domain containing residues 319 -582 (6).
Based on comparison with the extracellular domain of CD4, of which the structure is known (18), we defined the N-terminal Ig domain of CAR (D1) as containing residues 15-140 of the CAR coding sequence. Residue 15 in this numbering was thought to be the N-terminal residue of the mature CAR protein; residues 1-14 were thought to be a signal sequence that is removed after targeting to the cell membrane (7). It was later shown that the mature CAR protein starts at residue 20 (19). We expressed the CAR D1 domain with a C-terminal His tag in E. coli using an expression system in which the protein is exported to the periplasm. Purification was by metal-chelating and hydrophobic interaction chromatography (see "Experimental Procedures") to homogeneity as judged by SDS-PAGE. The preparation of CAR D1 yielded 2-3 mg of pure protein/liter of bacterial culture. N-terminal sequence analysis gave the sequence DFARSLSITT, which confirmed the identity of the expressed protein and the removal of the leader sequence. We also expressed and purified a nontagged version of CAR D1 (15). CAR D1 has also been successfully expressed in the cytoplasm of E. coli by Freimuth et al. (12).
Aggregation State of the Bacterially Expressed Proteins-The natural adenovirus type 2 fiber is a trimer of exceptional stability. It is resistant to incubation with SDS at moderate temperatures and only dissociates into monomers if incubated with SDS at higher temperatures (20,21). Type 2 fiber head containing residues 388 -582 expressed in baculovirus-infected insect cells has been shown to be trimeric by covalent crosslinking (3), crystallographic analysis (4), and dynamic light scattering. Our bacterially expressed fiber head protein (also containing residues 388 -582) shows the same resistance to SDS as fiber protein isolated from adenovirus and as the fiber head expressed in baculovirus-infected insect cells. It is also trimeric when analyzed by gel filtration of SDS-PAGE without prior boiling of the sample and monomeric when the sample is heated before applying it to the gel (Fig. 1, B and C). Furthermore, it forms crystals identical to the protein produced in baculovirus-infected insect cells. 2 The bacterially expressed head plus shaft construct is also trimeric in SDS at moderate temperature or when analyzed by gel filtration or dynamic light scattering. The mass obtained by dynamic light scattering for the head plus shaft construct was 89.0 kDa, with no discernible polydispersity, closely corresponding to a value of 85.7 kDa expected for a trimer. We can therefore safely assume that both proteins are trimeric under the nondenaturing conditions of our surface plasmon resonance assays.
The aggregation state of non-His-tagged CAR D1 has been studied previously (15). It was found to dimerize in the crystal and also in solution. In solution, a dissociation constant of 16 M could be measured by equilibrium centrifugation. We also performed equilibrium centrifugation experiments with Histagged CAR D1 and obtained a dissociation constant of 25 Ϯ 10 M. Within error, this value overlaps the one obtained for non-His-tagged CAR D1. Because our surface plasmon resonance assays contained CAR D1 at nanomolar concentrations, the protein can be regarded as monomeric under these conditions.
CAR D1 Binding to Immobilized Adenovirus Type 2 Fiber Head-To determine conditions suitable for kinetic analysis, we prepared surfaces with different amounts of immobilized adenovirus type 2 fiber head. CAR D1 was injected onto these surfaces, using different flow rates. Because of the fast dissociation, it was not necessary to include a regeneration step after each CAR injection, and a 15-min injection of running buffer was enough to completely remove CAR D1 from the immobilized fiber head. The primary data were analyzed by linear transformation ("linearization," see Ref. 22) using a simple A ϩ B 7 AB model to give the kinetic parameters reported in Table I. Increasing the flow rate from 10 to 50 l/min over a surface containing 1200 RU of fiber head did not change the kinetic parameters (Table I), indicating that the binding reaction was not limited by mass transport effects. We also investigated surfaces on which 600 or 1500 RU of fiber head were immobilized. There was no significant dependence on the surface density of the kinetic parameters of CAR D1 binding to immobilized fiber head (Table I). The binding of CAR D1 molecule to immobilized fiber head can thus be characterized by an on rate constant k on ϭ 2.7 Ϯ 0.3 ϫ 10 5 M Ϫ1 s Ϫ1 and a dissociation rate constant k off ϭ 6.7 Ϯ 0.6 ϫ 10 Ϫ3 s Ϫ1 (errors are standard deviations from the five different experiments in Table I), leading to an average dissociation constant of 24.9 Ϯ 1.2 nM. Equilibrium data were also extracted from the sensorgrams to calculate the equilibrium affinity constant using a Scatchard plot (23). A average value of 23.6 Ϯ 1.7 nM was returned for the dissociation constant (Table I). The fact that this value is almost identical to the value yielded by the kinetic analysis supports the model used and the kinetic analysis itself. Fig. 2A shows the binding curves when CAR D1 was injected over a surface containing 600 RU of immobilized fiber head. This set of data was also analyzed by numerical integration (also called "global fitting"), which provides a stringent test of the assumed model and returns better parameter estimates (22). As shown in Fig. 2, the fitted curves and the experimental data were virtually indistinguishable, using the A ϩ B 7 AB model. The returned kinetic values were k on ϭ 3.1 ϫ 10 5 M Ϫ1 s Ϫ1 and k off ϭ 6.6 ϫ 10 Ϫ3 s Ϫ1 , giving a K d ϭ k off /k on ϭ 21 nM, in reasonable agreement with the above analysis (Table I). Equilibrium data were extracted from the sensorgrams (Fig. 2B) and plotted according to the Scatchard (23) representation (Fig.  2C). The straight line obtained shows that CAR D1 recognizes a single class of noninteracting binding sites, characterized by an affinity constant of Kd eq ϭ 26 nM, a value consistent with the result given by the kinetic analysis of the data.
The stoichiometry of the complex indicated that, on average, each immobilized fiber head can bind a maximum of two CAR D1 molecules. Because of the trimeric organization of the fiber head and from the published structure of the adenovirus type 12 fiber head-CAR N-terminal domain complex (11), a 3:1 stoichiometry would have been expected. This discrepancy is presumably due to some of the binding sites of the fiber head becoming inaccessible during the immobilization procedure (on average one per fiber head trimer; see also Fig. 4).
To further check the validity of our results, we also immobilized an adenovirus type 2 fiber protein containing the head domain plus part of the shaft region (residues 319 -582; Ref. 6) to a surface density of 700 RU. CAR D1 was injected at 20 l/min. A dissociation constant of 30 nM and a maximum binding of 2.3 CAR D1 molecules/trimer were observed. These values correlate very well with those found for the fiber head alone. The similar dissociation constant indicates that the shaft region is not involved in receptor binding, while the fact that the average number of available binding sites is now somewhat larger is consistent with the fact that some of these trimers will have been attached by the shaft domain and thus have all three receptor binding sites available.
Adenovirus Type 2 Fiber Head Binding to Immobilized CAR D1-In a second series of experiments, CAR D1 was immobilized on a sensor chip and allowed to bind fiber head injected over the surface. This system mimics the CAR presentation at a putative cell surface, and thus may represent a more physiologically relevant assay. In preliminary experiments we prepared different surfaces with different amounts of immobilized CAR D1, over which fiber head was injected. Whatever the conditions were, the sensorgrams now showed much slower dissociation and could not be fitted to a simple A ϩ B 7 AB binding model (data not shown). Visual inspection of the sensorgrams showed that equilibrium was now much more difficult to attain (compare the association phases of Figs. 2A and 3A), and the formed complexes are now very stable (compare the dissociation phases of Figs. 2A and 3A). Although the CAR D1 molecule spontaneously dissociated from the immobilized fiber head (binding curves returned to the base line in less than 15 min with running buffer alone), injection of 10 mM hydrochloric acid was required to dissociate the complex when the CAR D1 molecule was immobilized on the surface (data not shown). To minimize all possible artifacts, including mass transport effect and rebinding (see below), we used a surface with a small amount of immobilized CAR D1 (200 RU) for a more detailed analysis. Evaluation of the sensorgrams indicated that the binding was complex and could not be fitted to a simple A ϩ B 7 AB model. Such a model returned a 2 value (which describes the closeness of the fit) of 23.1 (a 2 value below 10 is considered acceptable; BIAevaluation 3.1 software handbook).
Thus, as a first approach, we measured equilibrium data. Apart from the very highest fiber head concentrations, the measurements do not really attain equilibrium; it appears that the CAR D1 surface is able to capture fiber head even from dilute solutions to saturate more and more available binding sites. Technical limitations of the apparatus used prohibited us from using even longer contact times. However, from the six highest fiber head concentrations we could calculate an approximate affinity constant independently of the kinetic process (Fig. 3B). A K d of 1.2 nM was estimated (Table II), which represents an apparent 20-fold increase in affinity compared with the binding of CAR D1 to immobilized fiber head. Interestingly, the stoichiometric analysis indicated that one fiber head bound three CAR D1 molecules at the sensor chip surface   (Fig. 4), a point consistent with the trimeric nature of the adenovirus fiber head. We then evaluated our kinetic data using a trivalent binding model: A ϩ B 7 AB, AB ϩ B 7 AB 2 , and AB 2 ϩ B 7 AB 3 , where A is a trivalent molecule (fiber head in this context) that binds to B, a monomeric component (here CAR D1). Global fitting of the data using this model returned a 2 value of 3, indicating that the fitting procedure has been considerably improved and now describes the kinetic data much better (Fig.  3A). The initial binding step (A ϩ B 7 AB) was characterized by association and dissociation rate constants of 2.4 ϫ 10 5 M Ϫ1 s Ϫ1 and 5.5 ϫ 10 Ϫ3 s Ϫ1 , respectively. Interestingly, these val- Although not all points have reached equilibrium and the data do not represent a complete binding isotherm, this analysis does give an idea about the order of magnitude of the dissociation constant. C, CAR D1 immobilized on a sensor chip surface was saturated with fiber head (injected at 80 nM). Subsequently, CAR D1 in solution (sCAR at 0, 32.5, 65, 130, and 325 nM) was injected at 20 l/min, and the response in RU was recorded. No significant increase in RU was observed, indicating that no fiber binding sites for CAR were available on the fiber head in the preformed complex.  4. Graphical representation of adenovirus fiber binding to its receptor. The CAR D1 domain is shown in dark gray, and adenovirus fiber head is in light gray. A, monovalent binding reaction of the CAR N-terminal domain to immobilized adenovirus fiber head. Shown are CAR D1 binding and dissociating on the left and a front view of the complex on the right. The rate constants of binding k on and dissociation k off and the equilibrium dissociation constant K d of the complex are quoted. In our experiments, about two CAR molecules could bind simultaneously to one fiber head, indicating that on average one of the three CAR binding sites of the fiber head was unavailable, presumably because of the covalent cross-linking (see text). B, trivalent binding reaction of the adenovirus fiber head to the CAR N-terminal domain tethered to a surface (by chemical cross-linking to a surface like in our experiments or by a membrane anchor to a human cell membrane). Shown are the trimeric fiber head binding to CAR D1 on the left, the trivalent complex in the middle, and the fiber head dissociating from CAR on the right. In our experiments, fiber head bound three CAR molecules simultaneously. ues are close to those found when monomeric CAR D1 bound to immobilized fiber head, thus giving an affinity of 23 nM. Because these values did not depend on the molecules which have been immobilized, they were not affected by the immobilization process and thus are likely to be the true values.
Our data indicate that the first binding event was the binding of one fiber head molecule to one immobilized CAR D1 molecule. However, the complex (one fiber head bound to one CAR D1 with an affinity of 23 nM) was then further stabilized by the association of two additional CAR D1 molecules, leading to the high overall affinity observed. These two additional binding reactions at the chip surface are characterized by on rates k on2 ϭ 0.15 RU Ϫ1 s Ϫ1 and k on3 ϭ 0.08 RU Ϫ1 s Ϫ1 and off rates k off2 ϭ 0.60 s Ϫ1 and k off3 ϭ 0.41 s Ϫ1 (Table II). The overall off rate of the fiber head-3 CAR D1 complex was 2.2 ϫ 10 Ϫ4 s Ϫ1 , and this represents a 25-fold increase in stability versus the values measured for the fiber head-1 CAR D1 complex. Taking into account the overall off rate (2.2 ϫ 10 Ϫ4 s Ϫ1 ), the complex (fiber head-3 CAR D1) was found to have an overall K d of 2.2 ϫ 10 Ϫ4 s Ϫ1 /2.4 ϫ 10 5 M Ϫ1 s Ϫ1 ϭ 0.9 nM (Table II). This value is lower than but similar to the affinity we determined using quasi-equilibrium data (Kd eq ϭ 1.2 nM), suggesting that our kinetic analysis is correct.
To demonstrate that the high apparent affinity is not due to rebinding of soluble fiber head, we first injected fiber head (at 80 nM) over a surface containing 400 RU of CAR D1 to form a stable complex and then injected soluble CAR D1 over the preformed complex. As shown in Fig. 3C, the preformed complex was unable to bind significant amounts of additional CAR D1, demonstrating that each bound fiber head molecule already bound three CAR D1 molecules (i.e. there were no more binding sites available in the preformed complex). Because the dissociation rate constant was not increased by the presence of soluble CAR molecules in the running buffer during the dissociation phase, these data also showed that a possible rebinding effect did not contribute to the low dissociation rate we measured.

DISCUSSION
Viruses initiate infection by attaching themselves to the surface of a susceptible host cell and have evolved to use a variety of cell surface molecules for this purpose. In this paper, we report an avidity-based mechanism by which adenovirus fiber binds to the CAR D1 N-terminal Ig domain to form a highly stable complex, with a stoichiometry of three CAR D1 molecules to one fiber head trimer. This stoichiometry is consistent with the published crystal structure of the adenovirus type 12 fiber head-CAR D1 complex (11).
We initially designed binding experiments in which CAR D1 molecules were allowed to bind independently of each other to immobilized fiber head. Both kinetic and equilibrium analyses indicated that CAR D1 recognizes its binding site on the fiber head with an affinity of around 24 nM. However, the binding of soluble CAR to fiber head is only a poor mimic of the physiological interaction between a cell surface and the virus. We thus adopted a more physiologically relevant system, in which CAR D1 was coupled to a solid phase, through the flexible dextran matrix of a sensor chip (receptor molecules on a cell surface are also "flexible" because of the liquidity of the cell membrane). Not only does this system better represent CAR expressed at the cell surface, but it also showed a 25-fold increase in affinity by an avidity-based mechanism. This avidity mechanism allows adenovirus to employ binding interactions with high individual off rates in tandem to achieve an overall low off rate (for a graphical explanation see Fig. 4) and give rise to highly stable cell surface attachment (dissociation constant around 1 nM). In the case of the entire adenovirus particle, this avidity could be further enhanced by the fact that more fiber proteins interact simultaneously each with three receptors. The mechanism we propose has been described in qualitative terms by various authors (for a review see Ref. 24). However, we believe we have for the first time analyzed it in kinetic detail for one of the systems, adenovirus attachment to its receptor.
A cratic entropy contribution has been quantified for association processes in solution (25), although the usefulness of the concept is not universally agreed (26,27). It would suggest an increase in affinity of greater than the 25-fold we measured. However, we are considering a process involving multiple interactions at a surface, in our case a Biacore chip, but with analogies to a biological membrane. We suggest that the decrease in degrees of liberty experienced by the trimeric fiber head after binding to the first CAR D1 molecule may account for the lower affinity we measure but feel that a more detailed explanation, which should then also take into account thermodynamic behavior (28), falls outside the scope of our current work.
The question arises of whether other trimeric receptor-ligand interactions show the same avidity mechanism. The receptor of group B adenoviruses and the possible second receptor of group F adenoviruses are as yet unknown, but it is possible that they use a similar mechanism. Trimeric cell attachment proteins of other viruses such as reovirus sigma 1 protein and the long and short fibers of bacteriophage T4 (1) are in our opinion likely to also employ a similar mechanism.
Many intercellular communication pathways involve trimeric receptor-ligand interactions by members of the tumor necrosis factor (TNF) receptor family with TNF-like trimeric ligands. The interactions of the TNF receptor with TNF␤ (29) and TRAIL (TNF-related apoptosis inducing ligand) with DR5 (death receptor 5) (30) are surprisingly similar to that of the CAR D1 domain with adenovirus type 12 fiber head (11), with the receptor binding in a groove between two monomers and surface loops located toward the bottom of the trimer being functionally most relevant (10,11,31,32). It is therefore likely that the same avidity mechanism we observe for adenovirus fiber head binding to CAR also exists for ligands binding to the TNF receptor family.
Trimeric interactions also exist intracellularly. TNF receptor and many of its homologues (for instance CD40) bind with their intracellular domains to the C-terminal domains of tumor necrosis factor receptor-associated factors (TRAFs; Ref. 33). TRAFs then mediate signaling to various transcription factors (34). The binding of trimeric TRAF2-C to monomeric and artificially trimerized CD40 has been compared, and it was found that the affinity was 12-fold lower in the case of monomeric CD40 (35).
The avidity mechanism can perhaps be extended to other multimeric cell attachment molecules or whole viruses. The internalization of adenovirus is mediated by the pentameric penton base (2), and by using SPR, 4.2 integrin molecules have been measured to bind to each pentamer with an individual affinity of 73 nM (36). In these measurements, adenovirus was the immobilized ligand, and soluble integrin was the analyte. Perhaps if the integrin were to be immobilized and the penton base or the whole virus were used as analyte, the same avidity mechanism as for the fiber head-receptor interaction can be demonstrated. The affinity of rhinovirus (37) and human echovirus 11 (38) for their receptors has been measured to be in the micromolar range, again by immobilizing the virus and using soluble receptor as analyte. We propose that the avidity mechanism will lead to a much higher effective affinity of virus for a cell bearing multiple copies of receptor. Indeed, a chimeric bivalent receptor binds to rhinovirus with a 17-fold enhanced affinity (37). In the case of influenza virus binding to polyvalent sialisides, the affinity increases by 3 orders of magnitude (39) compared with the normal millimolar affinity for monovalent binding (40).
A study by SPR in which saccharide ligands were immobilized on a sensor chip surface was carried out by Mann et al. (41). They then studied binding of concanavilin A to this surface and inhibition of this binding by monovalent or multivalent competitor ligands. They found a dissociation constant of around 1 M for multivalent concanavilin binding to the surface, dissociation constants of 92-290 M for monovalent binding of saccharides to concanavilin A, and dissociation constants down to 2 M for polyvalent binding of polysaccharides to concanavilin A, a similar avidity effect to the one we observe.
In conclusion, we have analyzed in detail an avidity mechanism that may be widespread in receptor-ligand interactions. Experiments on other model systems using the same strategy (immobilizing the receptors and using soluble receptor-binding proteins or whole viruses) may yield more examples of the avidity mechanism we describe. The next step in this research is studying the attachment of whole viruses in kinetic detail by surface plasmon resonance. This may be more difficult in practice because of the size of whole viruses and their relative instability compared with the isolated receptor-binding proteins. The whole virus may not be able to reach all the receptor molecules immobilized in the dextran surface, and their greater size may lead to overloading of the plasmon resonance signal, whereas their instability may lead to signal changes because of dissociation of the virus rather than receptor-binding or receptor-dissociating events. Nevertheless, the development of new SPR surface chemistries and immobilization strategies may make these studies more feasible in the future.
Finally, adenovirus is investigated intensively as a candidate vehicle for gene therapy (42). To target adenovirus to specific cell types, much research is focused on abolishing CAR binding of adenovirus and retargeting to other receptors (Ref. 43 and references therein). To ensure that recognition of these alternative receptors is as efficient as possible, it might be advantageous to mimic the mechanism of binding to the natural receptor. For achieving the same stoichiometry and avidity mechanism, alternative receptor binding sites should be engineered into the trimeric adenovirus fiber head of adenovirus so that each site is independently accessible by a receptor molecule.