Structural Analysis of the Mechanism of Adenovirus Binding to Its Human Cellular Receptor, CAR

See allHide authors and affiliations

Science  19 Nov 1999:
Vol. 286, Issue 5444, pp. 1579-1583
DOI: 10.1126/science.286.5444.1579


Binding of virus particles to specific host cell surface receptors is known to be an obligatory step in infection even though the molecular basis for these interactions is not well characterized. The crystal structure of the adenovirus fiber knob domain in complex with domain I of its human cellular receptor, coxsackie and adenovirus receptor (CAR), is presented here. Surface-exposed loops on knob contact one face of CAR, forming a high-affinity complex. Topology mismatches between interacting surfaces create interfacial solvent-filled cavities and channels that may be targets for antiviral drug therapy. The structure identifies key determinants of binding specificity, which may suggest ways to modify the tropism of adenovirus-based gene therapy vectors.

Many viral infections are initiated by the specific binding of specialized proteins or attachment factors on the virion's surface to glycoprotein receptors on the surface of host cells. Enveloped viruses, such as human immunodeficiency virus (HIV), attach to host cells by means of spike-like membrane glycoproteins, whereas most nonenveloped viruses, such as poliovirus, attach by means of specialized domains integral to their capsids. Adenoviruses (Ad) are hybrids: They are nonenveloped but have trimeric fibers (320 to 587 residues) emanating from the vertices of their icosahedral capsid, which terminate in globular knob domains (∼175 residues). Infection is initiated by the formation of a high-affinity complex between these knob domains and a host cell surface protein. For the majority of the >50 known Ad serotypes, this receptor is CAR, which is also the cellular receptor for group B coxsackieviruses (1). CAR is an integral membrane protein of unknown cellular function that is expressed in a wide range of human and murine cell types. Its two extracellular immunoglobulin superfamily (Ig) domains mediate Ad and coxsackievirus B infections, possibly with different parts of the molecule. After CAR binding, Ad replicates within the nuclei of cells, causing frequent and generally mild infections of the upper respiratory and gastrointestinal tracts. Their efficiency of infection and their ability to transport large amounts of DNA are two of the properties of Ad that have prompted their development as gene delivery vectors for experimental gene therapy (2). Understanding the molecular details of the knob-CAR complex is one step toward developing better gene-delivery systems with altered tropism or increased receptor binding affinity, or both.

The knob domain from Ad serotype 12 (Ad12) and the NH2-terminal domain of human CAR (CAR D1) were overexpressed in Escherichia coli as soluble protein fragments. To determine the molecular basis for the specificity and tight affinity of this interaction (3), we crystallized and determined the structure of the Ad12 knob alone and in complex with human CAR D1 to 2.6 Å resolution (4). The data collection and refinement statistics are shown in Table 1. Adopting the nomenclature of Xia, used for the structure of the Ad5 knob (5), Ad12 knob momomers also have an eight-stranded antiparallel β-sandwich fold. The packing of strands J, C, B, and A (called a sheet) from adjacent monomers form the tightly packed trimer interface and create a large cavity located along the central threefold axis of symmetry that was proposed as a potential receptor binding site (5). Strands G, H, D, and I (R sheet) form a solvent-exposed β sheet that was also identified as a putative receptor-binding site (5). The major difference between the structures of the two knobs, which both bind CAR, is in the HI loop, which in Ad12 (residues 548 to 556) is well ordered but in Ad5 (residues 536 to 549) is a five-residue extended and disordered loop (6).

Table 1

Summary of data collection statistics. Numbers in parentheses refer to data in the outermost resolution shell. AU, asymmetric unit. I, intensity. R free is based on ∼3% of the data.

View this table:

To locate the actual CAR-binding site on Ad12 knob, we solved the structure of the Ad12 knob–human CAR D1 complex (Table 1). It has a triskelion shape with three CAR D1 monomers bound per knob trimer (Fig. 1A). This multiplicity for CAR binding coupled with the high affinity of the interaction may contribute to the high efficiency of Ad infection. In the complex, CAR D1 (residues 1 to 123 of the mature protein) adopts a β-sandwich fold that is characteristic of Ig variable (Igv) domains (Fig. 1B). CAR D1 does not bind at either of the predicted sites on knob. Instead, each CAR D1 molecule binds at the interface between two adjacent Ad12 knob monomers, consistent with the observation that most neutralizing antibodies to knob are directed against the trimer, rather than the monomer (7). In contrast to some viruses, which undergo a conformational change to expose the binding site, we find that complex formation in Ad occurs without any notable rearrangement in the knob structure. This may also be true of CAR D1, because in the complex it is structurally homologous to the unliganded forms of CD4 and the myelin adhesion molecule (8). Compatible with the role of CAR D1 in attaching adenovirus to the cell membrane, its COOH-terminus lies 180° away from the NH2-terminal end of the knob trimer and hence the viral fiber axis. Although the current model does not provide direct information concerning the orientation of CAR D2 (residues 126 to 222) relative to D1, comparisons of CAR D1 with structures of homologous proteins, solved with both D1 and D2 domains, suggest that CAR D2 does not make extensive contacts with knob (9). This is consistent with the observation that soluble CAR D1 alone is sufficient for knob binding and is an inhibitor of viral infection for cells in culture (3).

Figure 1

The Ad12 knob–CAR D1 complex and the importance of the AB loop. (A) Ribbon diagram of the Ad12 knob–CAR D1 complex viewed down the viral fiber. The core of each knob monomer comprises an eight-stranded antiparallel β sandwich, although about 65% of the residues are found in ordered surface exposed loops and turns. The V sheet is colored purple and the R sheet and HI loop are colored magenta. The AB loop is highlighted in yellow, and all other regions are gray. About 2050 Å2 of primarily hydrophobic surface per monomer is buried upon trimerization of the knob domains. In Ad12 knob, the interfacial residues are D415, P417, P418, I426 (AB loop), V450, K451 (CD loop), and Q487, Q494, S497, and V498 (E and F strands) from one monomer, and P517, P519, N520, and E523 (FG loop) from the other. (B) Ribbon diagram of the CAR D1 domain from the complex colored in a rainbow from blue to red. Strands in the foreground are D, E, B, and A, and strands in the background are C", C′, C, F, and G from left to right, respectively. In CAR D1, the interfacial residues are P33, E37, L39, V48, D49, V51, L54, S56, Y61, D62, E63, Y64 (strands C, C′, and C"), K102, K104, A106, and P107 (later half of F). This figure was generated in MOLSCRIPT (29). (C) Amino acid sequence alignment of residues in the AB loop for all knob subgroups. Adenovirus subgroups A, C, D, E, and the long fiber of F bind CAR, whereas subgroup B and short fiber knobs of F do not. Residues conserved in CAR-binding serotypes are colored blue, those with 1 amino acid different are colored in red, and all other residues are colored in black. On a wider alignment, with the use of all available adenovirus sequences, the pattern of conservation of proline and aspartate was unchanged (14).

The nature of the specificity of Ad12 knob for CAR is suggested by the structure of the complex, particularly at the interface that is formed by four loop regions of Ad12 knob interacting with a single face of the CAR D1 sandwich. Specifically, the AB loop, the carboxyl ends of the DE loop, and the very short F strand of one knob monomer and the FG loop of the adjacent knob monomer interact with strands C, C′, C", and the second half of strand F in CAR D1. The AB loop (colored yellow in Fig. 1A) contributes over 50% of interfacial protein-protein interactions, including the three hydrogen bonds involving conserved atoms in Ad12 knob [D415 O on knob forms a hydrogen bond with K104 Nζ on CAR D1 (D415 O/K104 Nζ), L426 O/Y64 OH, and K429 Nζ/E37 Oɛ2] (10), and thus may be the key anchor for the complex. This loop spans the width of the CAR D1, held at one end by D415 and at the other end by E425, which changes rotomer conformation upon CAR D1 binding to accommodate the approaching side chain of Y61. In the middle of the AB loop, the conserved residue P418 makes contacts with residues E37, V51, and L54 in CAR D1. Substitution of L54 with alanine in CAR D1 completely abrogates CAR binding, which is consistent with a role for these interactions in complex stability. The importance of the AB loop is further emphasized when comparing the sequence of non–CAR-binding Ad serotypes such as subgroup B (serotypes 3 and 7), which have evolved to bind a different receptor (11), or serotypes 40 and 41, where two types of fiber exist on the same virus but only one fiber type binds CAR (12) (Fig. 1C). The sequences in this region of the non–CAR-binding serotypes diverge widely from each other and from known CAR-binding serotypes. Specifically, the knob domains of non–CAR-binding fibers either have insertions or deletions in this loop relative to the conserved residues, P418 and N419.

To further explore the idea that the AB loop is important in CAR binding, we constructed a number of knob variants that mimic non–CAR-binding knobs in this region. The double substitution P417E and P418A, which converts the Ad12 AB loop to a sequence similar to Ad3, the insertion of TI before S421, which lengthens the AB loop, and the deletion of E425 and L426, which shortens the AB loop, all abrogate CAR D1 binding (13). In addition, we made single amino acid substitutions outside this loop, the majority of which behaved in a manner that is consistent with the structure we constructed (14). The results are consistent with the hypothesis that the AB loop is a primary determinant for the specificity of most Ad serotypes for CAR. Unexpectedly, the deletions of G550 and I551 in the HI loop significantly reduced the affinity of CAR D1 binding even though the residues do not interact directly with the receptor or model. In the native structure, there are water-mediated hydrogen bonds between G550N and G550O in the HI loop and R518O and A521N in the FG loop that stabilize the FG loop in a conformation through which it can make direct contacts with CAR D1. Thus, a likely explanation for the decreased affinity of Ad12 for CAR D1 when G550 and I551 are deleted is that the loss of these hydrogen bonds, because of the removal of G550, allows the FG loop more flexibility in conformation. Although this result highlights the limitations of using a purely mutational strategy for identifying knob's receptor-binding sites, it shows that altering the amino acid sequence in the AB loop may be one way of changing the tropism of Ad-based gene therapy vectors.

In protein-protein complexes, the interacting surfaces are typically well conserved and usually display a high degree of surface complementarity. The CAR D1 binding site in Ad12 knob lies along a ridge of residues primarily from the AB loop that are generally highly conserved, although some sequence variability is seen (14). It is located on the side of the molecule at the interface between adjacent monomers (Fig. 2A). In total, the complex formation buries ∼1880 Å2 of mixed hydrophobic and hydrophilic surfaces at each knob-CAR interface: 960 Å2 of surface are contributed by knob (820 Å2 from one monomer and 140 Å2 from the second monomer) and 920 Å2 by CAR D1 (Fig. 2, B and C) (15). These values are within the range typically observed for protein-protein interactions. However, the interface is atypical because the Ad12 knob and CAR D1 residues each produce a surface that, when united with its partner, creates a prominent channel and two adjacent cavities. A consequence of this discontinuous interface is that the actual footprints of the interacting surfaces in the complex are ∼15% larger than would be predicted from the values of buried surface area.

Figure 2

A molecular surface representation of the interface in the Ad12 knob-CAR D1 complex. (A) Sequence conservation surface diagram of two knob monomers viewed at the CAR interface. The molecules are colored on a sliding scale from white (conserved) to red (nonconserved). Conservation analysis was based on an alignment of all human Ad knob sequences available in GenBank. A white strip of conservation transects the surface of the molecule. Upon binding, the CAR D1 molecule occludes the conserved strip on Ad12 knob. (B) Surface diagram of two adjacent Ad12 knob monomers shown in the same view as (A). The molecules are colored on a sliding scale from yellow (contact) to red (no contact). Atoms in contact with CAR D1 are shared between monomers. (C) Surface diagram of CAR D1. The molecules are colored on a sliding scale from magenta (contact) to cyan (no contact). This figure was generated with GRASP (30).

The high affinity of the knob-CAR D1 complex directly conflicts the general observation that shape mismatches in the interface of protein complexes tend to correlate with low-affinity binding (16). In the Ad12 knob–CAR D1 complex, the lack of van der Waals interactions between the protein surfaces may be partially compensated for by solvent molecules within the cavities and interfacial channel (Fig. 3A). The two equally sized (total volume ∼120 Å3) cavities are separated by P418 in the AB loop of Ad12 knob interacting with residues in CAR D1, and the cavities are lined with a mixture of hydrophobic and polar groups, >60% of which either are backbone atoms or are conserved in sequence. From a thermodynamic standpoint, empty cavities of this size are energetically unfavorable and are thus likely to be filled with solvent. The cavities could accommodate ∼four water molecules, although in the current structural model only one well-ordered water molecule was observed that formed a bridging interaction with the conserved or backbone atoms of E37 Oɛ1 and K102 Nζ in CAR and D415 O, P416 O, and K429 Nζ in Ad12 knob. In addition, there is a solvent-lined interfacial channel, created by the packing of residues N415, K429, V450, L455, N520, S522, and E523 in Ad12 knob (Fig. 2C) against residues V48, N49, K102, V51, Y64, and P66 in CAR D1 (Fig. 2D). In the current model, four ordered water molecules markedly increase the interfacial contact surface and may thus act as molecular glue. Weaker electron density, also observed throughout the cavities and channel, may be indicative of additional mobile water molecules. The identification of the key residues involved in the Ad12 knob–CAR D1 interface may aid in the design of Ad-based gene targeting vectors that no longer bind CAR but specifically target other cell surface receptors at a second site.

Figure 3

CPK model of the region around the cavity. The three consecutive proline residues in Ad12 knob partially shape the cavity, which is colored magenta. The AB loop, whose carbon atoms are colored yellow, lines one side of the cavity. The carbon atoms from the remainder of the monomer are colored red, those of the second knob monomer, green, and those of CAR D1, cyan. All oxygen and nitrogen atoms are colored light red and blue, respectively. The cavity is lined with atoms from residues D415, P416 (backbone),K429 (side), V448 (side), G449 (backbone), V450, L455 (side), Q535 (side), P573 (side), and S575 (side) from one Ad12 knob; S514 (backbone), A515 (backbone), P517 (side), N520 (side), A524 (main), E523, K525, and S526 (side) from the other Ad12 knob; and L39 (side), K47 (backbone), V48 (backbone), D49, Q50, V51, and K102 (side) from CAR. The underlined residues are conserved or similar in all CAR-binding Ad serotypes.

The results presented here, and those of others, suggest that viruses have developed at least two structural means of successfully binding their receptors. The receptor-binding sites in picornoviruses, such as poliovirus, and, by extension, the CAR-binding site on coxsackievirus B, are located in deep crevices or canyons on the capsid surface (17, 18). Burying the receptor-binding site in this way may act as an antigenic shield for the conserved residues that define receptor-binding specificity (19). By contrast, the structures and biochemical studies of the structurally unrelated Ad12 knob and the HIV gp120 in complex with their receptors (20,21) show that their receptor-binding faces are surface loops and thus are exposed to immunoselective pressure. Both viruses bury a similar amount of surface area to create cavities and channels that may serve a dual purpose as both a water buffer and as molecular glue, because they mediate hydrogen bonds between backbone and conserved atoms. By using a noncomplementary interface that traps water molecules, the virus can maintain its receptor specificity while altering its sequence. Because this type of interface has been observed in two very different viruses, the water-buffer hypothesis may represent a second general mechanism by which viruses can successfully achieve the first stage of infection.

  • * To whom correspondence should be addressed. E-mail: flanagan{at} and Freimuth{at}


View Abstract

Navigate This Article