Analysis of the interaction between RGD-expressing adenovirus type 5 fiber knob domains and {alpha}vbeta3 integrin reveals distinct binding profiles and intracellular trafficking

Adenovirus (Ad) infection and Ad-mediated gene transfer depend upon efficient cell attachment, internalization and the subsequent orchestrated disassembly of the Ad capsid that allows the release of the packaged Ad genome and its transportation through the nuclear-pore complex. For most Ad serotypes, infection is initiated after binding to their attachment receptor, the coxsackie and adenovirus receptor or CAR (Bergelson et al., 1997). Receptor binding occurs on the lateral surface of the Ad5 fiber knob domain (Bewley et al., 1999; Kirby et al., 2000; Roelvink et al., 1999) and is also influenced by the length of the fiber shaft domain that determines the overall flexibility of the fiber protein (Shayakhmetov & Lieber, 2000; Wu et al., 2003). Following cell attachment of subgroup C Ad serotypes 2 (Ad2) and 5 (Ad5), a series of events is initiated that includes fiber detachment, cell signalling and Ad cell entry (Greber, 2002; Greber et al., 1993, 1994), mediated by the interaction between the arginineglycineaspartic acid (RGD) motif of the penton base in the Ad capsid and its integrin cell-surface receptors (Davison et al., 1997, 2001; Li et al., 2001; Wickham et al., 1993). Subsequent steps in Ad2 and Ad5 trafficking and disassembly lead to the shedding of the penton base protein and escape of the partially disassembled virion from endosomes to the cytoplasm (Greber et al., 1993; Martin-Fernandez et al., 2004). In the cytosol, the Ad particle utilizes microtubules and microtubule-dependent motors to be transported towards the cell nucleus, where it docks at nuclear-pore complexes (Leopold et al., 2000). The Ad2 capsid is then disassembled and the viral genome is imported into the nucleus (Greber et al., 1993). Intracellular trafficking of the CD46-binding subgroup B Ad serotypes 3 (Ad3), 7 (Ad7) and 35 (Ad35) appears to differ from that of Ad2 in some important respects (Miyazawa et al., 1999, 2001; Shayakhmetov et al., 2003). Ad3 and Ad7 do not appear to lose their fibers at the cell surface and are retained in late endosomal and lysosomal compartments, with consequentially lower infection rates (Miyazawa et al., 1999, 2001).

Ad2, Ad5 and, more recently, the CD46-binding Ad35 have been investigated as vectors for gene therapy. However, the efficiency of Ad-mediated gene transfer is often found to be poor and there is consequently great interest in overcoming biological barriers to efficient virus uptake by the development of Ad vectors that bind to alternative cell-surface receptors. These targeted Ad vectors also hold the promise of localizing gene transfer to specific cell types. The successful development of such Ad vectors may result in reduced vector-associated toxicity, as lower viral doses may be required to achieve therapeutic gene transfer. Genetic targeting strategies have focused predominantly on incorporation of heterologous ligands into the Ad5 fiber knob domain, the crystal structure of which revealed an eight-stranded antiparallel sandwich composed of two β-sheets, with loops and turns connecting the strands (Xia et al., 1994). The HI loop, a surface-located, flexible loop that is not involved with molecular interactions that may affect the stability or trimerization of the fiber, has been utilized extensively as a site for the insertion of heterologous peptide ligands of varying sizes. For example, the incorporation of the cyclic, cysteine-constrained, integrin-binding peptide RGD-4C (CDCRGDCFC) into the HI loop of an Ad5 vector (Dmitriev et al., 1998; Reynolds et al., 2000) resulted in a recombinant Ad vector capable of increased gene delivery to cells that would otherwise be refractory or poorly permissive to Ad infection. Whether this is the optimal site for ligand insertion in terms of structural stability, biophysical interactions with cellular receptors and subsequent internalization is currently unknown. For the purpose of this study, we selected three surface-exposed loops (CD, HI and IJ loops) on the Ad5 fiber knob domain as sites for insertion of the RGD-4C peptide. Three RGD-containing Ad5 fiber knob-domain mutants were produced as recombinant soluble proteins and all were shown to interact with soluble αvβ3 integrin by using biomolecular cell-free assays. Cell adsorption and subsequent internalization and intracellular trafficking of each of these proteins were assessed by confocal microscopy.

Construction of Ad5 fiber knob RGD mutants (Ad5 FK RGD).
Ad5 fiber knob-domain mutants bearing RGD peptide insertions were created by insertion of a cysteine-constrained 9 aa sequence (CDCRGDCFC) (Dmitriev et al., 1998) into the Ad5 fiber knob wild-type (WT) prokaryote expression plasmid created as described previously (Kirby et al., 1999, 2000, 2001). The sequence was inserted immediately after amino acids at positions Gly450 in the CD loop (Ad5 FK^CD-RGD), Thr546 in the HI loop (Ad5 FK^HI-RGD) and Gly560 in the IJ loop (Ad5 FK^IJ-RGD) in the Ad5 fiber knob WT sequence by using a QuikChange site-directed mutagenesis kit (Stratagene), which utilizes complementary pairs of DNA oligonucleotides encoding the mutant sequence. All mutations and the integrity of the remaining sequence were confirmed by automated DNA sequencing. The following pairs of complementary oligonucleotide primers were used to create each mutation in Ad5 FK: Ad5 FK^HI-RGD, 5'-gtcccatgaaaatgacatagagtatgcacttgggcagaaacagtctccgcggcagtcacaagttgtgtctcctgtttcctgtgtaccgtttag and 5'-ctaaacggtacacaggaaacaggagacacaacttgtgactgccgcggagactgtttctgcccaagtgcatactctatgtcattttcatgggac; Ad5 FK^CD-RGD, 5'-gctgttaaaggcagtttggctccaatatctggatgtgactgccgcggagactgtttctgcacagttcaaagtgctcatcttattataagattt and 5'-aatcttataatttgatgagcactttgaactgtgcagaaacagtctccgcggcagtcacatccagatattggagccaaactgcctttaacagc; Ad5 FK^IJ-RGD, 5'-actctatgtcattttcatgggactggtctggctgtgactgccgcggagactgtttctgccacaactacattaatgaaatatttgccacatcc and 5'-gatgtggcaaatatttcattaatgtagttgtggcagaaacagtctccgcggcagtcacagccagaccagtcccatgaaaatgacatagagta.

Expression and purification of histidine (His)-tagged recombinant RGD-containing Ad5 fiber knob-domain proteins.
His-tagged Ad5 FK^CD-RGD, Ad5 FK^HI-RGD and Ad5 FK^IJ-RGD proteins were expressed in bacteria and each purified by nickel affinity chromatography and size exclusion as described previously (Kirby et al., 1999, 2000, 2001). For the final step in the purification protocol, traces were obtained from a Superdex-75 (S-75) (Pharmacia) size-exclusion column. The elution times representing the maximal peak height of the eluted proteins as measured by A₂₈₀ were similar for the recombinant Ad5FK His-tagged WT and RGD mutant proteins and compatible with trimeric knob domains. Subsequent proteins were labelled for confocal microscopy analysis where appropriate by using the Cy3 dye following the manufacturer's instructions (Amersham Biosciences).

Surface plasmon resonance (SPR) analysis of proteinprotein interactions
Human vitronectin and soluble αvβ3 integrin were obtained from commercial sources (Chemicon).

Preparation of SPR sensor surface.
Purified soluble recombinant CAR (sCAR) and soluble recombinant αvβ3 integrin were coupled to the CM5 sensor chip by using the amine coupling reaction as described previously (Kirby et al., 1999, 2000, 2001). Immobilization densities of 6001000 resonance units (RU) were used for sCAR and 20003000 RU for soluble integrins.

SPR analysis of Ad5 FK RGD proteins and sCAR.
All interactions were carried out as described previously for WT Ad5 fiber knob domain.

SPR analysis of vitronectin with soluble αvβ3 integrin.
All interactions were carried out at 25 °C using HBS-P [10 mM HEPES (pH 7.4), 150 mM NaCl, 0.005 % P-20 surfactant] with 1 mM MgCl₂ as the continuous flow buffer. Soluble purified human vitronectin (Chemicon) was diluted in HBS-P with 1 mM MgCl₂ to a concentration of 100 µg ml¹ and injected for 120 s at 30 µl min¹ over soluble αvβ3 integrin immobilized to the CM5 sensor chips. This was followed by injection of the HBS-P with 1 mM MgCl₂ buffer to observe dissociation of the analyte.

SPR analysis of Ad5 FK RGD proteins with soluble αvβ3.
All interactions were carried out as for human vitronectin by using RGD-containing Ad5 fiber knob proteins diluted in HBS-P and 1 mM MgCl₂ to concentrations of 12 µM at flow rates of 3040 µl min¹. The binding was also assessed in the presence of 1 mM MnCl₂ in the continuous flow buffer and 10 mM EDTA.

Kinetic evaluation of SPR interactions.
The association (K_a) and dissociation (K_d) rate constants for a monophasic model of binding were obtained by using the BIAevaluation analysis package (version 2.1) as described previously (Kirby et al., 2000, 2001). Biacore Antibody supplied the BIAcore system, CM5 sensor chip and amine coupling kit.

Immunocytochemistry.
NIH 3T3 and MCF7 cells grown on glass coverslips were incubated with specified Ad5 FK RGD mutant proteins (50 µg ml¹) labelled with Cy3 dye in PBS at 4 °C for 1 h. Cells were then placed at 37 °C for up to 60 min, after which they were washed in cold PBS and fixed by using 4 % paraformaldehyde. Cells were then permeabilized and stained by using the specified antibodies in 3 % bovine serum albumin. Post-staining, samples were mounted by using Immunofluor (ICN) containing DACO (Sigma) onto glass microscope slides.

Confocal microscopy.
Confocal images were acquired on a confocal laser-scanning microscope (model LSM 510; Carl Zeiss Inc.) equipped with both x40/1.3Plan-Neofluar and x63/1.4Plan-Apochromat oil immersion objectives. Each image represents a two-dimensional projection of two to three slices in the z series, taken across the mid-depth of the cell at 0.2 µm intervals.

Molecular modelling.
In order to explore the range of conformations available to the inserted loop peptides, 50 homology models were generated for each of the loops based on the Ad5 fiber knob domain as a template (PDB code 1KNB) by using MODELLER 6.2 (Martí-Renom et al., 2000). The structure of αvβ3 complexed with an RGD peptide (PDB code 1L5G[PDB] ) was superimposed on the RGD-containing loop of each of the homology models and viewed by using INSIGHT II (Accelrys).

Antibodies.
Anti-His antibodies were obtained from Invitrogen and were detected by using Alexa Fluor 555-conjugated secondary antibody (Molecular Probes). Anti-transferrin (Tf) receptor antibodies (Transduction Laboratories) and anti-lysosome-associated membrane protein 1 (LAMP1) antibody (Santa Cruz Biotechnology) were used at the recommended concentration and detected by using Alexa Fluor 488-conjugated secondary antibodies (Molecular Probes).

Cells.
NIH 3T3 fibroblasts and MCF7 human breast carcinoma cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10 % fetal calf serum. MCF7 human breast carcinoma cells were specifically selected for these studies because they do not express CAR, as assessed by fluorescence-associated cell sorting using the RMCB mAb, and also because they overexpress αvβ3 on their cell surface. MCF7 cells also express, at low levels, the αvβ5 and α5β1 cell-surface integrins (J. Marshall, personal communication).

Insertion of heterologous RGD peptides in the Ad5 fiber knob domain
Three regions on the Ad5 fiber knob were identified as suitable for insertion of the cyclic, cysteine-constrained RGD peptide (Fig. 1). These positions were located after Gly450 in the CD loop, Thr546 in the HI loop and Gly560 in the IJ loop and were all considered suitable for peptide insertion on the basis of their accessibility for interaction with soluble integrins. The nucleotide sequence of all mutant fiber knob domains was determined; each contained the appropriate insert without any additional sequence alterations. All mutant fiber knob domains were expressed as soluble proteins in bacteria and were demonstrated to be stable trimers by size-exclusion chromatography.

(45K): Fig. 1. Stereotactic images showing the course of the polypeptide chain (Cα atoms only) within the Ad5 fiber knob domain. Top: the trimer is viewed along the trimer axis looking from the outside the virus. One hundred and fifty models are shown superimposed upon a single Ad5 fiber knob monomer. Fifty models were generated per loop (HI, CD and IJ loops) in order to explore the range of conformations available at each position. Side: the Ad5 fiber knob-domain trimer is viewed at 90° to the trimer axis with the shaft domain at the bottom.

Analysis of proteinprotein interactions by SPR
Analysis of the interaction between soluble CAR and RGD-bearing Ad5 fiber knob-domain proteins.
Our previous kinetic analysis by SPR of the wild-type Ad5 fiber knob interaction with sCAR revealed a high-affinity interaction with a K_d value of 14 nM (Kirby et al., 2000, 2001). sCAR was immobilized to a sensor chip and Ad5 FK RGD mutants were used as the analyte at concentrations of 400 nM. The three RGD-containing Ad5 fiber knob-domain mutants were all found to interact specifically with sCAR; the K_d values for Ad5 FK^HI-RGD, Ad5 FK^CD-RGD and Ad5 FK^IJ-RGD mutants were 17 nM (K_a, 3.5x10⁴; K_d, 1.4x10⁴), 3 nM (K_a, 4.3x10⁴; K_d, 5.8x10⁴) and 10 nM (K_a, 5.5x10⁴; K_d, 5.7x10⁴), respectively (Fig. 2a), demonstrating that each of the three RGD-containing Ad5 fiber knob domains bound to sCAR with similar association and dissociation kinetics. Based on these findings, it can be concluded that the insertion of the RGD peptide motif in the HI, CD and IJ loops of the Ad5 fiber knob domain did not alter the quaternary structure of the knob domain, nor did it disrupt the kinetics of binding to sCAR.

(14K): Fig. 2. RGD-containing Ad5 fiber knob domains bind to soluble CAR and to soluble αvβ3 as determined by SPR. (a) SPR sensograms for RGD-containing Ad5 fiber knob domains binding to immobilized sCAR. All three RGD-expressing Ad5 fiber knob domains (Ad5 FKHI-RGD, black line; Ad5 FKCD-RGD, blue line; Ad5 FKIJ-RGD, green line) and WT Ad5 fiber knob domain (red line) were injected over the sensor surface at the concentrations stated. A 120 s association phase was followed by a similar dissociation phase with HBS-P buffer and 1 mM MgCl2 flowing over the sensor surface at 30 µl min1. Similar results were obtained in at least three independent experiments for each RGD-containing fiber knob. (b) SPR sensograms for RGD-containing Ad5 fiber knob domains binding to immobilized integrin αvβ3. All three RGD-expressing Ad5 fiber knob domains (Ad5 FKHI-RGD, solid line; Ad5 FKCD-RGD, dotted line; Ad5 FKIJ-RGD, dashed line) were injected over the sensor surface at the concentrations stated. A 120 s association phase was followed by a similar dissociation phase with HBS-P buffer and 1 mM MgCl2 flowing over the sensor surface at 30 µl min1. Similar results were obtained in at least three independent experiments for each RGD-containing fiber knob.

SPR analysis of vitronectin binding to soluble αvβ3 integrin.
Soluble αvβ3 was immobilized onto a sensor chip and human vitronectin was used as the analyte at a concentration of 100 µg ml¹, using HBS-P and 1 mM MgCl₂ as the continuous flow buffer. Specific binding with αvβ3 was demonstrated and the binding profile was similar to that reported by Takagi et al. (2002). The binding was not altered by the use of MnCl₂ in the continuous flow buffer and binding was not observed in the presence of 10 mM EDTA.

SPR analysis of the binding interaction between RGD-containing Ad5 fiber knob domains and soluble αvβ3 integrin.
Soluble αvβ3 was immobilized onto a sensor chip and Ad5 FK RGD mutants were used as the analyte at concentrations of 1 and 2 µM, using HBS-P and 1 mM MgCl₂ as the continuous flow buffer. Specific binding was demonstrated for all three mutants, with different association and dissociation profiles (Fig. 2b). Ad5 FK^CD-RGD and Ad5 FK^HI-RGD mutants behaved similarly in relation to association and, in particular, the very slow dissociation rates. In contrast, Ad5 FK^IJ-RGD bound with slower association and faster dissociation rates (Fig. 2b). All binding was abolished in the presence of EDTA in the continuous flow buffer. No binding was demonstrated by using Ad5 FK WT as the analyte at concentrations of up to 200 µM. These results demonstrate specific binding for all three Ad5 fiber knob-domain mutants to αvβ3, although with different binding profiles. It was not possible to obtain reliable kinetic measurements, because the binding surface proved to be highly sensitive to the regeneration buffer. Consequently, a new binding surface had to be prepared for each run. Nevertheless, the experiment was repeated at least three times for each RGD-containing fiber knob domain and the same pattern as seen in Fig. 2(b) was obtained each time and at each concentration.

Molecular modelling
Inspection of the structure of αvβ3 complexed with an RGD-containing Ad5 fiber knob loop revealed that only a small proportion of the loop conformations were capable of allowing an integrin molecule to bind to the RGD-containing Ad5 fiber knob domain without clashes with the rest of the structure. Of the 50 loop conformations generated for each loop insertion, the RGD sequence in the HI loop could support integrins bound to 19 of the possible conformations, but only eight for the CD loop and only one for the IJ loop.

In addition, whilst the RGD insertion in the HI and CD loops could support integrins bound to each of the subunits of the Ad5 fiber knob trimer, the RGD in the IJ loop could support only one integrin per trimer (Fig. 3). An integrin bound to the IJ loop overlaps the other two equivalent loops within the trimer, occluding them sterically. This is simply a consequence of the spacing between loops in adjacent subunits. Within the native fiber knob trimer, the interloop distances are 6.8 nm for the HI loop, 4.5 nm for the CD loop and 3.4 nm for the IJ loop.

(37K): Fig. 3. Stereotactic images showing the course of the polypeptide chain (Cα atoms only) of αvβ3 complexed with an RGD-containing Ad5 fiber knob loop. CD: three αvβ3 integrin molecules bound to an Ad5 FKCD-RGD trimer (shown in bold) containing the CD loop RGD peptide. HI: three αvβ3 integrin molecules bound to an Ad5 FKHI-RGD trimer (shown in bold) containing the HI loop RGD peptide. IJ: only one αvβ3 integrin molecule can bind to an Ad5 FKIJ-RGD trimer (shown in bold) containing the IJ loop RGD peptide.

Ad5 fiber knob domains presenting the RGD sequence in different loops are internalized into MCF7 and NIH 3T3 cells
Membrane-binding capability of all three mutant proteins labelled with Cy3 fluorescent dye was demonstrated with Ad5 FK^CD-RGD, Ad5 FK^HI-RGD and Ad5 FK^IJ-RGD on fixed MCF7 human breast carcinoma cells expressing high levels of αvβ3 after incubation for 60 min at 4 °C in PBS (Fig. 4). This was also demonstrated on NIH 3T3 fibroblasts (data not shown). In contrast, the WT Ad5 fiber knob domain was not detected at the cell membrane. In order to visualize the localization of the integrin and RGD-expressing fiber knob domains at the membrane, MCF7 cells were fixed and co-stained with an antibody against αvβ3 conjugated directly to Oregon green. Fig. 4 shows confocal images acquired from cells incubated with all three proteins at 4 °C for 60 min. The data indicate a high degree of co-localization between αvβ3 integrin and all Ad5 fiber knob-domain mutant proteins. The specificity of binding was tested by using blocking antibodies specific for αvβ3 preceding incubation with the Ad5 fiber knob-mutant proteins. No binding of the Ad5 mutant proteins occurred in the presence of the blocking antibody, or indeed when pre-incubated with a molar excess of synthetic, cysteine-constrained RGD-4C peptide (Fig. 4).

(53K): Fig. 4. RGD-expressing Ad5 fiber knob proteins co-localize at the cell surface with β3 integrin. MCF7 human breast carcinoma cells were incubated with Ad5 FKHI-RGD, Ad5 FKCD-RGD and Ad5 FKIJ-RGD proteins coupled directly to Cy3 (red fluorescence) at 4 °C to allow cell-surface binding to occur. Cells were then fixed and stained with antibodies to β3 integrin coupled directly toOregon green 488 (green fluorescence). Yellow areas indicateregions where the two proteins co-localize. Binding of Ad5 FKHI-RGD in the presence of αvβ3-blocking antibody or with molar excess of synthetic, cysteine-constrained RGD-4C peptide (+GRGDS peptide) to MCF-7 cells was abolished, as shown.

After fixing and permeabilizing cells following 3060 min incubation at 37 °C, increasing intensity of staining was demonstrated intracellularly, with internalization having occurred by 30 min with all three proteins (Fig. 5). There were, however, differences in localization at 60 min. Whereas Ad5 FK^CD-RGD and Ad5 FK^HI-RGD mutants localized to perinuclear structures by 60 min, the Ad5 FK^IJ-RGD mutant remained in a submembrane position. (Fig. 5). Co-staining experiments using mutant proteins labelled directly with Cy3 and the recycling early endosome marker anti-Tf receptor antibody, detected with Cy5-conjugated secondary antibody showed co-localization with all three Ad5 fiber knob domain mutants at 30 min. By 60 min, however, the Ad5 FK^CD-RGD and Ad5 FK^HI-RGD mutants were localized to the perinuclear region, whereas the Ad5 FK^IJ-RGD mutant still remained co-localized with the endosomal marker (Fig. 5). These results suggest either a delay in the rate of Ad5 FK^IJ-RGD escape from the endosome or that entrapment and recycling of this protein within the endosome are occurring. Co-staining experiments using anti-His antibody and the lysosomal marker LAMP1, detected with Alexa Fluor 488-conjugated secondary antibody, showed no co-localization with Ad5 FK^CD-RGD and Ad5 FK^HI-RGD mutants at 60 min (data not shown), suggesting that neither of these two mutant knob domains was being processed through the lysosomal pathway. Upon endosome escape, they were released into the cytoplasm and trafficked to a perinuclear position. In contrast, the Ad5 FK^IJ-RGD mutant is not degraded, but it is uncertain whether it is being recycled or merely retarded in the endosome.

(91K): Fig. 5. RGD-expressing Ad5 fiber knob domains are internalized via the endocytic pathway. MCF7 human breast carcinoma cells incubated with Ad5 FKHI-RGD, Ad5 FKCD-RGD and Ad5 FKIJ-RGD proteins coupled directly to Cy3 were co-stained for markers of intracellular organelles. Confocal images demonstrate a time course of Ad5 FKHI-RGD, Ad5 FKCD-RGD and Ad5 FKIJ-RGD protein (red fluorescence) binding and internalization (at37 °C) in conjunction with Tf receptor expression stained with a secondary antibody coupled directly to Cy5 (shown as green fluorescence). Areas of yellow fluorescence in the merged images represent regions where the two fluorescent dyes coincide. Parallel samples were also co-stained with antibodies to LAMP1 and demonstrated no co-localization of any proteins with the lysosome (data not shown). HI, CD and IJ correspond to Ad5 FKHI-RGD, Ad5 FKCD-RGD and Ad5 FKIJ-RGD, respectively.

The efficiency of Ad-mediated gene transfer may be improved by overcoming biological barriers to efficient virus uptake, such as deficiency of attachment and internalization receptors, by incorporating novel receptor ligands to target the desired tissue into the virus. Understanding the kinetics and affinity of such novel interactions and their effect on subsequent intracellular events, such as transport to the nucleus, is an essential first step in designing appropriate targeted Ad vectors. In this study, we focused specifically on the fiber knob domainreceptor interaction to address these questions.

By modifying the HI loop of the Ad5 fiber knob domain genetically, it has proved possible to redirect Ad5 to novel receptors and to achieve Ad5-mediated gene transfer to cells and tissues that are otherwise refractory or relatively resistant to Ad5 infection (Curiel, 1999a, b, 2000). Despite these successes, it remains unclear whether the HI loop represents the optimal or only site for heterologous peptide-ligand insertion. Based on the structure of the Ad5 fiber knob domain (Xia et al., 1994), we identified, in addition to the HI loop, two other structurally distinct, surface-exposed loops, the CD and IJ loops, as candidate sites for the insertion of heterologous ligands. The CD and IJ loops are surface-accessible regions at the apex of the Ad5 fiber knob-domain structure that are not flexible. Both loops are located away from the CAR-binding site (Bewley et al., 1999; Kirby et al., 2000; Roelvink et al., 1999). The CD loop includes a turn of α-helix, whilst the IJ loop is random coil; these conformations would best accommodate peptide ligands that are likely to be functional when extended. In contrast, the HI loop is highly mobile, flexible and located on the side of the protein (Xia et al., 1994) away from the CAR-binding site (Bewley et al., 1999; Kirby et al., 2000; Roelvink et al., 1999). Its β-turn structure is well suited to accommodate loop or linear peptides and its flexibility is such that long peptides may be inserted without affecting the secondary structure of the knob domain adversely. In view of these differences in CD, IJ and HI loop structure, it is likely that a heterologous ligand inserted in each of the three loops will acquire distinct conformations and any variation in structure may alter the binding interaction between the heterologous ligand and its receptor.

To address some of these issues, we inserted a cyclic, cysteine-constrained RGD peptide in the CD, HI and IJ loops and assessed the CAR- and αvβ3 integrin-binding activity of each of these recombinant Ad5 fiber knob domains by SPR. SPR was chosen to evaluate binding of each of the three mutant proteins to soluble CAR and αvβ3 integrin as this allows detailed evaluation of the binding interaction in a cell-free system. We found that recombinant fiber knob domains expressing the cyclic RGD peptide in each of the three loops bound to soluble CAR with WT affinities; this indicates that the insertion of the RGD peptide in the three loops did not alter the quaternary structure of the fiber knob domain, nor did it disrupt the CAR-binding interaction, as we had predicted. Each recombinant knob domain also bound specifically to αvβ3 with different association and dissociation profiles. For example, the mutant knob domain expressing the RGD sequence in the HI loop demonstrated the slowest dissociation rate and the fastest association rate. In contrast, the mutant knob domain expressing the RGD peptide in the IJ loop demonstrated the slowest association and fastest dissociation rate. The RGD peptide, therefore, when presented in the HI loop will remain bound to αvβ3 more tightly than when expressed in the IJ loop. It is well known that expression of an RGD peptide alone is not sufficient to confer biological activity (Torshin, 2002) and that the precise spatial arrangement and flexibility of the region surrounding the RGD peptide may enhance interaction with cell-surface integrins (Chiu et al., 1999). It has also been suggested that positioning the RGD peptide on a loop or β-turn with high surface accessibility will enhance the likelihood of creating an active RGD conformation (Torshin, 2002). It is therefore highly likely that the observed variation in binding profiles between the RGD-expressing fiber knob-domain mutants and αvβ3 is due to the cyclic RGD peptide acquiring distinct conformations in each of the three loops.

By using molecular modelling, we predicted that the mutant Ad5 fiber knob domain in which the RGD peptide was expressed in the IJ loop was only capable of binding to one αvβ3 integrin molecule per trimer. In contrast, fiber knob domains in which the RGD peptide was expressed in the HI and CD loops were capable of binding to one integrin molecule per monomer. The short intersubunit distance between adjacent IJ loops explains this difference; with αvβ3 bound to the RGD peptide in one IJ loop, the RGD peptides in the other two IJ loops are inaccessible, due to steric interference. In contrast, the larger intersubunit distance between adjacent CD and HI loops is sufficient to allow association between the RGD peptide and αvβ3 in each monomer without steric interference. These observations are in keeping with our predictions based on the structure of the Ad5 fiber knob domain and suggest that the HI loop represents a suitable site for targeting receptorligand interactions that require receptor clustering to generate the necessary signals for attachment and internalization. In contrast, the IJ loop may be the preferred site if receptor clustering is best avoided.

Integrins, including αvβ3, anchor cells to the extracellular matrix, promote cellular functions and also act as surface receptors for many viruses, including Ad. Binding of host as well as viral ligands can induce integrin-mediated signalling and, in the case of viruses, this may initiate or modulate virus entry and trafficking (Greber, 2002). Clustering of integrins upon ligand binding is an important mechanism for generation of intracellular signals. Our molecular-modelling studies suggest that αvβ3 clusters are more likely to form when the RGD peptide is expressed in the HI or CD loops than in the IJ loops. This, as well as the very different association and dissociation rates of each knob domain for αvβ3, may influence how each mutant fiber knob is internalized upon binding to its receptor and subsequently trafficked through the cell.

We found that, for each of the three RGD-containing Ad5 fiber knob domains, fiber was internalized into MCF7 human breast carcinoma cells and NIH 3T3 fibroblasts, both naturally expressing αvβ3. As expected, unmodified WT Ad5 fiber knob domain did not demonstrate any binding to the cell membrane. As internalization of these recombinant fiber knob domains was inhibited totally in the presence of excess RGD peptide as well as αvβ3 blocking antibody, we conclude that attachment and internalization were mediated predominantly by an interaction between the RGD domain and cell-surface αvβ3. Imaging data also demonstrate a high degree of co-localization between the Ad5 fiber knob-domain mutants and αvβ3, suggesting strongly that the proteins are using this integrin as a binding and internalization receptor. We found that the site of insertion of the RGD peptide altered the intracellular transport of individual Ad5 fiber knob mutants. Although all three mutant knob domains were localized within endosomal structures 30 min following binding to the cell, by 60 min, the knob domains bearing the RGD peptide sequence in the CD and HI loops were seen to be residing in the perinuclear region. In contrast, the Ad5 mutant bearing the RGD sequence in the IJ loop was still co-localized within the endosome at 60 min. This suggests that Ad5 FK^IJ-RGD mutant had not escaped the endosome, whereas the other two mutants had undergone endosomal escape. LAMP1 staining demonstrated no co-localization of any of the mutant knob domains, indicating that they are not handled via the lysosomal degradation pathway. These findings demonstrate similar internalization rates for each RGD-containing knob domain, but a distinct pattern of endosomal transport and escape for Ad5 FK^IJ-RGD. This would suggest that the mechanisms responsible for αvβ3-mediated internalization of RGD-containing fiber knob domains are not the same as those responsible for endosome escape. Differences in the interactions between each mutant and αvβ3 may explain this observation. The inability of the fiber knob domain bearing the RGD insert in the IJ loop to escape from the endosome at 60 min may, for example, be a consequence of its much faster dissociation rate, in contrast to the Ad5 FK^CD-RGD and Ad5 FK^HI-RGD mutants that remain bound more tightly to αvβ3. The fact that the Ad5 FK^IJ-RGD fiber knob domain appears capable of binding to only one αvβ3 molecule per trimer may also contribute to the observed variation in intracellular transport, due to reduced αvβ3 clustering.

Multiple Ad serotypes utilize αv integrins during viral entry into cells, indicating that this may be a common pathway in the Ad life cycle (Mathias et al., 1994). The RGD sequence in different Ad penton base proteins is located in the middle of a stretch of polypeptide chain of variable length and sequence, implying that the length and conformation of the exposed RGD loop varies among serotypes. Flexibility and accessibility of these loops may represent structural determinants of integrin-binding specificity and affinity (Chiu et al., 1999). Recent studies have directly implicated the RGD sequence not only in Ad5 internalization, but also in endosomal escape (Shayakhmetov et al., 2005). In view of our findings that variation in RGDintegrin binding affinities can influence endosomal escape, we speculate that Ad serotypes that share the RGD motif, but bind to soluble integrin with different efficiencies, may enter the cell and escape the endosome at different rates, with consequently different infection rates.

Integrins, particularly αvβ3, are upregulated in tumour vasculature and metastatic tumour cell lines; they can therefore be exploited as target receptors for tumour-specific gene transfer. The development of such vectors is an important goal for gene therapy and our results contribute to this process by providing a biophysical basis for their rational design.

R. L. was the recipient of a research training fellowship for clinical graduates by the Guy's and St Thomas' Hospitals Charitable Foundation. The work was funded in part by BBSRC project grant to G. S.