Identification of a Conserved Receptor-Binding Site on the Fiber Proteins of CAR-Recognizing Adenoviridae

See allHide authors and affiliations

Science  19 Nov 1999:
Vol. 286, Issue 5444, pp. 1568-1571
DOI: 10.1126/science.286.5444.1568


The human adenovirus serotype 5 (Ad5) is used widely for applications in human gene therapy. Cellular attachment of Ad5 is mediated by binding of the carboxyl-terminal knob of its fiber coat protein to the Coxsackie adenovirus receptor (CAR) protein. However, Ad5 binding to CAR hampers the development of adenovirus vectors capable of specifically targeting (diseased) tissues or organs. Through sequence analysis and mutagenesis, a conserved receptor-binding region was identified on the side of three divergent CAR-binding knobs. The feasibility of simultaneous CAR ablation and redirection of an adenovirus to a new receptor is demonstrated.

The human adenovirus family has 49 viral serotypes that are associated with a wide range of pathologies and tissue tropisms (1, 2). The serotype Ad5 has been studied intensively and is currently used as a vector in human gene therapy (3). Cellular attachment is mediated by binding of the COOH-terminal knob of the adenovirus fiber coat protein to the CAR protein by serotypes from five of the six adenovirus subgroups (1, 4–11). The fiber knob amino acid sequences among these serotypes vary widely. These serotypes, despite the divergence in the knob sequence of 29 to 66% relative to the fiber 5 knob (F5K), all recognize CAR (10,12), which implies a conservation of the amino acids or motifs involved in receptor binding. We aligned the knob amino acid sequences of 14 adenoviruses from the subgroups that have CAR-binding serotypes, and scored the identity and similarity (13). As expected, only very few amino acids, in similar positions, were conserved in the structure (Fig. 1, A and B). Unexpectedly, this analysis identified a conserved cluster of amino acids on the side of the fiber trimer. The crystallographic model for the Ad5 knob predicts that this cluster is mostly surface-exposed (14, 15). Very little amino acid conservation was observed for the exposed receptor face (R sheet) on top of the protein, which has been hypothesized to interface with CAR (14, 15).

Figure 1

(A and B) Variability and conservation analysis of CAR-binding fiber protein sequences. The amino acid sequences of 14 fiber knobs were aligned by the Megalign program, which uses the Jotun-Hein algorithm with a PAM (point accepted mutations per 200 aligned positions) 250-residue weight table. Colors were assigned on the basis of sequence identity score in each of the positions of the alignment (red, 93 to 100%; orange, 71 to 86%; yellow, 57 to 64%; green for any lower score). (Cand D) Mutants in loops and β sheets. Point mutations that reduce the fiber-CAR interaction by two to more than three orders of magnitude, as measured by IC50 in a competition assay, are red. The mutation r474-476, which resulted in an IC50 value five times that of wild-type F5K, is shown in yellow. Mutations that did not affect CAR binding are green. Residues that were not mutated are gray.

We have mutagenized the majority of the amino acids present in the exposed loops and β sheets on the top and the side of the F5K protein, including the conserved cluster, and expressed all mutants in insect cells (16). The majority of mutant proteins were expressed as soluble proteins, as evaluated by protein immunoblot and detection by an affinity-purified polyclonal antibody to the knob (4). Trimerization was evaluated by binding with mouse monoclonal antibodies (mAbs) that only recognize the native knob trimer (17). All mutants that were found to be soluble and trimeric are shown on the F5K map (Fig. 1, C and D).

The mutant proteins were tested in competition assays for their ability to inhibit the transduction of CAR-expressing A549 cells by an Ad.CMV-βgal vector (9, 18). Because transduction first involves binding to CAR, the ability of mutant knob proteins to compete for this receptor, and thus block transduction, can be assessed indirectly. By determining the amount of protein needed to inhibit 50% of transduction for the wild-type F5K protein (IC50) and comparing that to the IC50 values obtained for the mutant proteins, the reduction in affinity for CAR can be estimated. Mutation of six critical residues of F5K, all located on the side of the fiber knob protein structure (Fig. 1, B and D), either completely or partially failed to block transduction of A549 cells by Ad.CMV-βgal (Fig. 2, A and B): S408E or G, P409A, K417G or L (AB loop), K420A (B β sheet), Y477A or T (DE loop), and Y491A (FG loop) (19). In addition, a four–amino acid deletion of the sequence TAYT (d489-492) or replacement of the sequence EGTAY in the FG loop with glycine residues (r487-491) also completely failed to inhibit Ad.CMV-βgal transduction of the A549 cells. A replacement mutant where the amino acids DPE that precede the critical residue Y477 in the DE loop are replaced with glycines (r474-476) had a partially inhibitory effect. Three point mutations in the AB loop, and one in the proximal B β sheet, overlap with the conserved cluster on the side of the fiber knob (Fig. 1, B and D) and yield mutant proteins with IC50 values two to more than three orders of magnitude greater than that for wild-type F5K protein (Fig. 2A). This suggests conservation of critical residues among CAR-binding adenoviruses and demonstrates that the AB loop is a crucial element of the CAR-binding site. Mutational analysis of the FG loop sequence EGTAY showed that only Y491A resulted in an IC50 that was two orders of magnitude greater than that of wild-type F5K protein (Fig. 2B). As the Y491 residue is partially buried (Fig. 1B), this mutation possibly alters structural conformation of the CAR-binding site. This implies that the FG loop may play an indirect role in binding the CAR protein. All point mutants having an IC50 that was more than two orders of magnitude greater than that of F5K were generally conserved between different serotypes (Fig. 1, B and D). All mutations to the R sheet blocked CAR-mediated transduction as efficiently as wild-type F5K protein, indicating that this structure is not critical to receptor binding (14, 15, 20).

Figure 2

Competition-expression assay analysis of mutant fiber knob proteins. (A) F5K AB loop and B β sheet mutants. (B) F5K DE and FG loop mutants. (C) F9K mutants. (D) F41LK mutants.

Analysis of amino acid conservation suggested that the CAR-binding site would likely be localized in the same region of other CAR-binding fiber knobs (Fig. 1B). We generated point mutants in the Ad9 fiber knob (F9K) and the Ad41L fiber knob (F41LK), which also bind to CAR (10). All residues in F9K corresponding to the critical residues identified in F5K (that is, F9K residues S189, P190, K198, K201, and Y262) are conserved. In contrast, F41LK has only three of five critical residues conserved (S395, P396, and K407); instead of F5K residues K417 and Y477, F41LK has, respectively, L404 and T470.

Five residues in F9K and four in F41LK were mutagenized. From our analyses, it followed that all the residues corresponding to those that were found to be critical to CAR binding in F5K also were critical to CAR binding in F9K and F41LK. Thus, the F9K mutations S189E, K198G, K201A, and Y262A (Fig. 2C)—corresponding to F5K residues S408, K417, K420, and Y477—all have IC50values that are at least three orders of magnitude greater than that of their native wild-type protein. The same is true for the F41LK mutations S395E and L407G (Fig. 2D), corresponding to F5K residues S408 and K417, respectively. The F9K mutation P190A and the F41LK mutations P396A and T470A (corresponding to F5K residues P409, K420, and Y477, respectively) resulted in mutants with a 25- to 100-fold increase in IC50 relative to that of wild-type F9K or F41LK (Fig. 2, C and D). Thus, all the residues corresponding to those that were found to be critical to CAR binding in the F5 knob also were critical to CAR binding in the F9 and F41L knobs.

To test the hypothesis that the amino acids identified as critical are truly part of a putative CAR-binding site, we analyzed the binding of F5K mutant proteins to soluble CAR protein [sCAR (10)] and to the non-neutralizing mAb 3D9, which recognizes only the F5K trimer (17). The mutations that knocked out sCAR binding, as well as the mutations that defined the epitopes of mAb 3D9, were mapped on the F5K protein (Fig. 3). The sCAR protein failed to bind to, or had reduced affinity for, the trimeric F5K mutants that failed to compete in competition assays (lane 1). Three mutations—K506R, the partial FG loop replacement rSHGKTA-GSGSGS 507-512 (r507-512), and the HI loop deletion/replacement rGTQETGDTTPSA-GSGG 538-549 (dr538-549)—did not abrogate sCAR binding. These mutations did abrogate mAb 3D9 binding (Fig. 3, lane 2), indicating that the FG and HI loops are part of its epitope. These mutant proteins, however, were recognized by mAbs 2C9 and 3F10, which also only recognize the native F5K trimer (17).

Figure 3

Binding of soluble CAR and 3D9 mAb to mutant knob proteins. Proteins (5 μg) were blotted on nitrocellulose and incubated for 2 hours in phosphate-buffered saline, pH 7.4, with 5% milk powder. The blots were incubated overnight with sCAR protein (2 μg/ml, lane 1), mAb 3D9 (lane 2), or a 1:5000 dilution of a rabbit polyclonal antibody to the F5K protein, as a control (lane 3) (4). Further procedures were as described (10).

We next sought to demonstrate that the mutations we have identified can be used to construct an adenovirus vector that lacks native CAR binding and is also redirected to a novel receptor. We introduced the CAR ablating mutation d489-492 TAYT into the vector Adf, which expresses green fluorescent protein (GFP) as a marker. In addition, the tag YPYDVPDYA, which corresponds to the epitope recognized by a mAb raised against the hemagglutination (HA) protein of influenza virus (21), was inserted into the HI loop of the fiber. The resultant vector Adf.F(TAYT-HA) was generated using standard techniques (22) and has a particle/plaque-forming unit ratio comparable to that of Ad5 vectors (23). We evaluated the transduction by this new vector of the cell lines 293 (which expresses CAR) and 293-HA (which expresses both CAR and a novel receptor, a membrane-tethered single-chain antibody to the HA epitope) (23). When we added equal numbers of particles of Adf.F(TAYT-HA) or Adf to 293 cells, we observed that transduction of the cells by Adf.F(TAYT-HA) was reduced by two orders of magnitude relative to Adf (Fig. 4, A and B). In contrast, transduction of 293-HA cells by Adf.F(TAYT-HA) was equal to transduction by Adf (Fig. 4, C and D). Transduction was quantified by counting the number of transduced cells in a microscopic field (∼1000 cells per field, eight fields counted). The vector Adf transduced 293 cells at 117 ± 12 cells per field, whereas Adf.F(TAYT-HA) transduced only 0.7 ± 0.3 cells. For 293-HA cells, transduction by Adf and by Adf.F(TAYT-HA) was comparable (105 ± 26 and 114 ± 22 cells per field, respectively). This low-level residual transduction can be blocked by penton base protein, but not by recombinant fiber protein (20). This finding suggests that the penton base mediates the majority of this residual transduction.

Figure 4

Transduction of 293 and 293-HA cell lines by Adf (A and C) and the CAR ablated vector Adf.F(TAYT-HA) (B and D). Cells at a density of 106 cells per 35-mm well were infected with the adenovirus vector at 10 physical particles per cell and incubated for 1 hour at 37°C. Cells were briefly washed with DMEM + 5% FCS and incubated overnight in the same medium. GFP was visualized on a Nikon fluorescent microscope equipped with the appropriate filters. Scale bar, 50 μm.

We conclude that the CAR-binding site consists of residues from the AB loop, the B β sheet, and the DE loop—that is, F5K residues S408, P409, K417, K420, and Y477. This site is located on the side of the knob, maps to a single monomer, and does not overlap two adjacent monomers. This finding implies that the trimeric knob binds to three CAR molecules independently, and it agrees with a determination of the molecular mass of the Ad12 fiber knob complexed with the IgV domain of the CAR protein (24).

The fiber knobs of the CAR-binding serotypes display profound dissimilarity of the loops (distal DE, distal FG and HI) that surround the critical residues, which indicates a potential role for these structures in avoiding neutralizing immunity. An analogous situation may exist for the human immunodeficiency virus gp120 envelope glycoprotein (25–27) and the sialic acid binding site of the hemagglutinin protein of influenza viruses (28,29).

Our experiments with the vector Adf.F(TAYT-HA) demonstrate the feasibility of combined CAR-binding ablation and simultaneous retargeting to a new receptor by introduction of new targeting sequences to an adenovirus vector fiber protein. Incorporation of this concept into adenoviral gene therapy vectors will lead to the development of safer, targeted vectors capable of delivering therapeutic genes to tissues with the highest possible specificity.

Note added in proof: Bewley et al. (30) have analyzed the crystal structure of the CAR-recognizing Ad12 fiber knob complexed with the D1 domain of the CAR protein. Their study confirms that the fiber AB loop plays an essential role in the interaction with the cellular receptor protein, CAR.

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


View Abstract

Navigate This Article