Coagulation Factor Binding Orientation and Dimerization May Influence Infectivity of Adenovirus-Coagulation Factor Complexes

Abstract

Adenoviruses (Ads) are promising vectors for therapeutic interventions in humans. When injected into the bloodstream, Ad vectors can bind several vitamin K-dependent blood coagulation factors, which contributes to virus sequestration in the liver by facilitating transduction of hepatocytes. Although both coagulation factors FVII and FX bind the hexon protein of human Ad serotype 5 (HAdv5) with a very high affinity, only FX appears to play a role in mediating Ad-hepatocyte transduction in vivo. To understand the discrepancy between efficacy of FVII binding to hexon and its apparently poor capacity for supporting virus cell entry, we analyzed the HAdv5-FVII complex by using high-resolution cryo-electron microscopy (cryo-EM) followed by molecular dynamic flexible fitting (MDFF) simulations. The results indicate that although hexon amino acids T423, E424, and T425, identified earlier as critical for FX binding, are also involved in mediating binding of FVII, the FVII GLA domain sits within the surface-exposed hexon trimer depression in a different orientation from that found for FX. Furthermore, we found that when bound to hexon, two proximal FVII molecules interact via their serine protease (SP) domains and bury potential heparan sulfate proteoglycan (HSPG) receptor binding residues within the dimer interface. In contrast, earlier cryo-EM studies of the Ad-FX interaction showed no evidence of dimer formation. Dimerization of FVII bound to Ad may be a contributing mechanistic factor for the differential infectivity of Ad-FX and Ad-FVII complexes, despite high-affinity binding of both these coagulation factors to the virus.

INTRODUCTION

Viral vectors based on adenoviruses (Ads) specific to human and animal species have been adapted widely for gene transfer applications both in vitro and in vivo. Extensive in vitro analyses have revealed a model of Ad cell infection whereby the initial virus attachment to a plasma membrane-localized receptor via the fiber protein is followed by virus penton interaction with integrins that mediate virion internalization into the cell (1, 2). To date, a number of cellular proteins that serve as functional high-affinity virus attachment receptors have been identified. For the most common vectors, based on species C human adenovirus serotype 5 (HAdv5), as well as for HAdv of species A, D, E, and F, it was found that a tight junction protein designated coxsackie and adenovirus receptor (CAR) can serve as a high-affinity attachment receptor (3–5). The species B Ads may utilize CD46 and/or DSG2 proteins to gain entry into host cells (6–8), while several species D Ads may utilize CD46, sialic acid, or GD1a glycan to enter the cell (9, 10). It was also shown that HAdv can bind a variety of integrin classes that interact with a penton RGD amino acid motif to trigger internalization of cell-bound virus particles into the cell (2).

Although this canonical pathway of HAdv cell entry operates efficiently in vitro and explains the topology and functional interdependence between viral capsid proteins (11, 12), the biodistribution of HAd5-based vectors in mice after intravascular administration revealed no correlation with tissue levels of CAR expression (13, 14). Instead, it was found that Ad particles accumulate in the liver and that virus entry into hepatocytes is mediated by hexon interaction with vitamin K-dependent blood coagulation factors (15–18). While earlier studies showed that several homologous blood coagulation factors, including FVII, FIX, and FX, can support virus infection of susceptible cells in vitro via this mechanism (16), specific inactivation of FX alone was sufficient to completely abrogate hepatocyte transduction with Ad in mice after intravenous virus administration (18). Furthermore, although analyses of binding affinities of different blood coagulation factors for hexon showed that FIX binds poorly to Ad, both FVII and FX bound the virus hexon protein with very high affinities, in the low-nanomolar range (15). Using moderate-resolution cryo-electron microscopy (cryo-EM), we and others previously showed that FX interacts with solvent-exposed hypervariable (HVR) loops of HAdv5 hexon via the GLA domain (15, 18). Substitution of HVR loops in HAdv5 hexon for HVR loops from HAdv26 hexon, which does not bind FX, or introduction of mutations into solvent-exposed regions of Ad hexon resulted in a loss or significant reduction of the FX binding affinity for hexon (19). However, using this extensive mutagenesis approach, the specific amino acids that form the FX-Ad hexon binding interface were not defined.

Using high-resolution cryo-EM followed by molecular dynamic flexible fitting (MDFF) simulations, we recently modeled the FX interaction with HAdv5 hexon and identified the T423-E424-T425 amino acid motif in HVR7 as critical for high-affinity FX binding to adenovirus. We further demonstrated that a single amino acid substitution, T425A, completely abrogated FX binding to Ad (20). To reconcile the discrepancy between high-affinity binding of FVII to the virus and its poor capacity to support virus entry into the cell, in this study we analyzed the HAdv5-FVII interaction interface by using high-resolution cryo-EM followed by MDFF simulations. Our analyses revealed that although hexon amino acids T423, E424, and T425 are also involved in mediating binding of FVII, the FVII GLA domain sits within the surface-exposed hexon trimer depression, in a different orientation from that found for FX. The MDFF simulations indicated that when two proximal FVII molecules are bound to hexon, they interact via their serine protease (SP) domains and bury potential heparan sulfate proteoglycan (HSPG) receptor binding residues within the dimer interface. In contrast, earlier cryo-EM studies showed no indication that FX interaction with hexon leads to the formation of SP domain dimers and indicated that virus attachment to cells is efficient in the presence of FX. Dimerization of FVII bound to Ad may be a contributing mechanistic factor for the differential infectivity of Ad-FX and Ad-FVII complexes despite high-affinity binding of both these coagulation factors to the virus.

MATERIALS AND METHODS

Cells and viruses.

293 cells were obtained from Microbix (Toronto, Canada). CHO-K1 cells (expressing HSPG; ATCC CCL-61) were obtained from the American Type Culture Collection. CHO-CAR cells were kindly provided by Jeffery Bergelson and were described earlier (3). All cell lines were grown on Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum. The E1- and E3-region-deleted replication-defective HAdv5-based vectors Ad-WT, Ad-GAT, Ad-TAT, Ad-TEA, and Ad5S were previously constructed and described in detail elsewhere (20–22). The Ad5S vector possesses a short Ad9-derived fiber shaft domain but the fiber tail and knob domains of HAdv5. This vector was used in lieu of Ad-WT virus to obtain high-contrast cryo-EM images of virus-FVII complexes. All these vectors express the green fluorescent protein (GFP) gene under the control of the human cytomegalovirus (CMV) early promoter. The GFP gene expression cassette is located in the E3 region of the Ad genome. Per previously published designations, Ad-WT possesses the wild-type HAdv5 capsid without any modifications, Ad5-GAT possesses T423G and E424A mutations in the hexon HVR7 loop, Ad5-TAT possesses the E424A mutation, and Ad-TEA possesses the T425A mutation in the hexon HVR7 loop (20). For Ad amplification, 293 cells were infected under conditions that prevented cross-contamination. Viruses were banded in CsCl gradients, dialyzed, and stored in aliquots as described earlier (22). Ad genome titers were determined by measuring the optical density at 260 nm (OD₂₆₀). For each Ad used in this study, at least two independently prepared virus stocks were obtained. Each produced virus stock was tested for endotoxin contamination by using Pyrotell Limulus amebocyte lysate (Cape Cod Inc., Falmouth, MA). For in vivo experiments, only virus preparations confirmed to be free of endotoxin contamination were used.

Cryo-electron microscopy, image processing, and modeling.

For cryo-electron microscopy analyses, we utilized a modified HAdv5-based vector, referred to earlier as Ad5S (21), which contains HAdv5 hexon, the HAdv5 penton base, and a modified fiber protein. In this fiber protein, the knob and tail domains are from HAdv5, but the fiber shaft domain is derived from HAdv9, rendering the fiber about 3-fold shorter than that of unmodified HAdv5. This form of HAdv5 was chosen to produce thinner ice on cryo-EM grids and to improve image contrast in cryo-electron micrographs. Samples were prepared by mixing 50 fmol of modified HAdv5 with a short-shafted fiber with 200 pmol of FVII in 50 mM Tris, pH 7.6, 150 mM NaCl, 2 mM CaCl₂, and 2 mM MgCl₂ for 15 min at room temperature. After incubation, 3.5-μl aliquots of virus particles with FVII bound were applied to Quantifoil grids (Quantifoil Micro Tools GmbH) and suspended in thin vitrified ice layers by use of a homebuilt vitrification device. Digital micrographs of the HAdv5-FVII complex were acquired under low-dose conditions at liquid nitrogen temperature on an FEI Polara 300-kV FEG transmission cryo-electron microscope with a Gatan UltraScan 4000 charge-coupled device (CCD) camera. A total of 1,503 particle images were collected at an absolute magnification of ×397,878, corresponding to a pixel size of 0.4 Å on the molecular scale. The defocus values of the micrographs ranged from −1 to −4 μm. Individual particles were selected from micrographs by use of in-house scripts calling IMAGIC subroutines (23) and were binned to generate particle image stacks with pixel sizes of 4.5, 2.2, and 1.5 Å that were used for initial, intermediate, and final refinement rounds, respectively. The program CTFFIND3 (24) was used to determine estimates for the microscope defocus and astigmatism parameters. A cryo-EM structure of Ad5.F35 (25) was used as the starting model for FREALIGN refinement (26). After the final round of refinement, the resolution of the icosahedral capsid (radii of 325 to 460 Å) at the Fourier shell correlation (FSC) 0.5 threshold was calculated to be 8.7 Å. The final map included 837 particles selected from the total data set.

A composite structural model of the zymogenic form of FVII (residues 1 to 406) was built using existing X-ray crystallographic data (27, 28) as well as a computational model of the full-length protein (29). Atomic-resolution information is available for HAdv5 hexon (11, 12). In order to model the interface between HAdv5 and FVII, atomic coordinates for one or two hexon trimers and one or two copies of the FVII model were docked into the HAdv5-FVII cryo-EM density at the 2-fold axis of the icosahedral capsid. Weak density for the EGF1 and EGF2 domains between the GLA and SP domains helped to guide the overall position. Attempts to use the GLA domain position from the previous HAdv5-FX model (20) to guide positioning of the FVII GLA domain resulted in models that did not match the experimental HAdv5-FVII cryo-EM density. The docked FVII coordinates served as input for multiple 100-ps MDFF simulations (30), with implicit solvent and a g-scale parameter of 0.3. The robustness of the MDFF-refined FVII orientation was tested by rotating the FVII starting orientation by ±10° or ±20° about the 3-fold molecular axis of hexon. Many of the same protein-protein interactions were recovered compared to the 0° starting orientation, lending support to the MDFF-derived fit of FVII within the cryo-EM density. Nonbonded interaction energies were evaluated with the NAMD Energy plug-in in VMD (31). The MDFF simulations were performed on the Case Western Reserve University High-Performance Computing Cluster.

Ad infection in vitro.

Unless noted otherwise, 2.5 × 10⁵ CHO-K1 or CHO-CAR cells were infected at a multiplicity of infection (MOI) of 2,000 virus particles (v.p.)/cell in 400 μl saline, with or without FVII or FX. Coagulation factors were added at concentrations ranging from 0.5 μg/ml to 25 μg/ml (each). The 12-μg/ml concentration of FX corresponds to its physiological concentration in the blood (1 U/ml). The physiological concentration of FVII in the blood is 0.5 to 0.6 μg/ml, which is about 17-fold lower on a molar basis than that of FX. We used the same amounts of coagulation factors in all comparative analyses. Two hours after the addition of viruses to cells, the virus-containing saline was replaced by fresh growth medium. Reporter gene activity was analyzed 48 h later by flow cytometry. Affinity-purified human blood coagulation factors FX and FVII were purchased from Hematologic Technologies, Inc.

Ad attachment assay.

Ad attachment studies were performed based on a protocol published elsewhere (22). Briefly, 5 × 10⁵ cells were incubated for 1 h on ice with equal amounts of [³H]thymidine-labeled HAdv5 or Ad5S at an MOI of 2,000 v.p./cell, either alone or in the presence of human FVII or FX (at 12 μg/ml) in 150 μl of ice-cold adhesion buffer (DMEM supplemented with 2 mM MgCl₂, 1% bovine serum albumin [BSA], and 20 mM HEPES). Next, the cells were pelleted by centrifugation for 4 min at 1,000 × g and washed two times with 0.5 ml of ice-cold phosphate-buffered saline (PBS). After the last wash, the cells were pelleted at 1,500 × g, the supernatant was removed, and the cell-associated radioactivity was determined by use of a scintillation counter. The number of viral particles bound per cell was calculated using the virion-specific radioactivity and the number of cells as described earlier.

SPR analyses.

All surface plasmon resonance (SPR) analyses were carried out on a Biacore T100 machine (15). HAdv5 and FVII samples were dialyzed against HBS-N buffer (10 mM HEPES, 150 mM NaCl, 1 mM CaCl₂, 0.5 mM MgCl₂, pH 7.4) twice, using 1 liter buffer and 0.1- to 0.5-ml dialysis cassettes (molecular weight cutoff [MWCO], 10,000) (Pierce Co.). Samples were diluted 1:1 with dialysis buffer to ensure that the volume was greater than 100 μl prior to loading into the cassette, and samples were dialyzed for at least 2 h at 4°C prior to changing the buffer once. Dialyzed samples were clarified in 1.5-ml centrifuge tubes at 14,000 × g prior to SPR analysis.

Dialyzed virus samples were diluted 1:9 in 10 mM sodium acetate, pH 4.0, and immobilized onto research-grade CM5 Biacore chips (Biacore Inc., Piscataway, NJ). Once immobilized, the chips were allowed to equilibrate in running buffer for 10 min prior to testing binding of FVII and regeneration. The running buffer was 10 mM HEPES, 150 mM NaCl, 1 mM CaCl₂, 0.5 mM MgCl₂, 0.1% (wt/vol) BSA, and 0.005% (vol/vol) polysorbate 20, pH 7.4. All buffers were filtered through a 0.2-μm membrane prior to use. The regeneration buffer was the same as the running buffer, without calcium or magnesium and including 3 mM EDTA. After immobilization, the chips were tested with a manual run to confirm function and to establish the best protocols for kinetic analysis. From this analysis, the operating conditions chosen were a flow rate of 50 μl/min, a 1-min association time, a 5-min dissociation time, and two 2-min regeneration cycles, with 1 min of stabilization. Samples were run in triplicate with buffer injections for double referencing. All data were collected at 1 Hz, using two or three replicate injections for each concentration of analyte. The dissociation time of 5 min was chosen to aid in regeneration rather than being necessary for the kinetic fits, so for data analysis the dissociation curve was truncated at 90 s. Fits to a 1:1 kinetic model were good for the wild-type Ad-WT virus and adequate for two of the mutants, Ad-GAT and Ad-TAT. The Ad-TEA mutant showed no significant binding to FVII within the range of tested concentrations of FVII (0 to 500 nM). Data processing and kinetic analysis were performed using Origin 8.5 software.

Statistical analysis.

Unless otherwise indicated, statistical analysis in each independent experiment was performed with an unpaired, two-tailed Student's t test. Data are reported as means ± standard deviations. P values of <0.05 were considered statistically significant.

Protein structure accession number.

The cryo-EM structure from this study has been deposited in the EM Data Bank under accession number EMD-5594.

RESULTS

Cryo-EM structural analysis of adenovirus-FVII interaction.

Previous cryo-EM structural studies indicated that FX binds in the central depression of the 240 hexons within the adenoviral capsid (15, 18, 20) and led to identification of key interaction residues on hexon (20). Here we applied the same approach to investigate the interaction between FVII and HAdv5. A subnanometer (8.7-Å)-resolution cryo-EM structure was determined for the HAdv5-FVII complex (Fig. 1A and B). In the structure, the FVII density is observed to extend outward from each hexon, creating a network of density surrounding the HAdv5 capsid. It is clear from the structure that FVII does not interact with either the penton base or fiber capsid protein. Atomic models of FVII and hexon were docked into a region of cryo-EM density taken from near the icosahedral 2-fold axis of the HAdv5 capsid. A good fit was obtained for the GLA domain, in the density within the hexon central depression, and for the distal SP domain, within the density above the capsid (Fig. 1C). Presumably the full-length FVII molecule is somewhat flexible at the domain linker regions. This could explain why not much density is observed for the intervening EGF1 and EGF2 domains and why the SP domain density is weaker than that of the GLA domain. An MDFF analysis revealed a stable interaction between the FVII GLA domain and hexon. In particular, an Arg residue of FVII (R28) formed interactions with a patch of polar and negatively charged residues on hexon, i.e., T423-E424-T425 (Fig. 1D to F). While the MDFF simulation indicated additional interactions between the GLA domain and hexon, we noted these three residues in particular because they were also predicted to be involved in the interaction with FX (15, 20).

Fig 1.

Cryo-EM structure of the HAdv5-FVII complex and simulation of the FVII-hexon interface by molecular dynamic flexible fitting. (A) Cryo-EM structure of HAdv5 in complex with FVII. The density is filtered to 16 Å and is shown with the hexon capsid in blue, the penton base in gold, the fiber in green, and FVII in purple. (B) Enlarged view of the HAdv5-FVII complex showing the network of FVII density above the hexons of the viral capsid. The inset shows the density near an icosahedral 2-fold axis of the capsid, which is indicated by a white oval. (C) The best MDFF-refined model of zymogenic FVII (purple ribbon) within the FVII cryo-EM density (transparent light purple). This FVII density was selected from above a hexon near an icosahedral 2-fold axis of the capsid. (D) Coordinates from the final frame of the best MDFF simulation that show hexon residue E424 forming a salt bridge with residue R28 of the FVII GLA domain. The side chains of these two residues are shown in space-filling representation and colored by element. (E) FVII GLA domain and associated Ca²⁺ ions (green) in the central depression of the hexon trimer. Residue R28 of the FVII GLA domain is shown, along with the nearby hexon residues T423, E424, and T425. (F) Enlarged view of residue R28 near T423, E424, and T425 as shown in panel E. Bars, 100 Å.

Mutation of the HVR7 TET amino acid motif reduces the affinity of FVII binding to the virus.

Prior to the analysis of FVII binding to a set of HVR7-mutated Ad vectors, we evaluated the integrity and infectivity of a virus preparation of each individual hexon-mutated vector. Polyacrylamide gel electrophoresis showed that all major proteins for all vectors were present in similar amounts (Fig. 2A). The analysis of vector infectivities on susceptible CHO-CAR cells further confirmed that all the hexon-mutated vectors exhibited similar transduction properties (Fig. 2B). Because MDFF simulations of the Ad-FVII hexon interaction interface showed that the T423-E424-T425 amino acids are likely to contribute to the formation and/or stabilization of the complex, we next analyzed the FVII binding affinities of Ad-WT and Ad mutants possessing amino acid substitutions in this region of hexon by using surface plasmon resonance (Biacore, Piscataway, NJ). To avoid complications with interpretation of the results caused by the multivalent nature of the virus particle, Ad-WT and the previously described Ad mutants Ad-GAT, Ad-TAT, and Ad-TEA (20) were immobilized on a CM5 sensor chip by cross-linking. Next, various concentrations of FVII, ranging from 0 nM to 500 nM, were injected over sensor surfaces with immobilized virus, and binding responses were recorded and analyzed. Figure 2 shows a representative data set obtained from the analysis of the Ad-FVII interaction together with global fits to a 1:1 interaction model. The kinetic dissociation constant (K_D) for the Ad-WT–FVII complex was determined to be 2.99 nM, in agreement with our previously published measurements (15). Using the same experimental approach, we found that mutations of the TET amino acids within the hexon HVR7 region drastically reduced the K_D of FVII-Ad complexes (Table 1). Specifically, the kinetic K_D determined for the FVII–Ad-GAT complex was 32.7 nM, and that for the FVII–Ad-TAT complex was 28.5 nM, while FVII failed to bind to the Ad-TEA mutant immobilized on the sensor chip. Collectively, these data confirm that the TET amino acid motif in HVR7 predicted by MDFF to interact with the FVII GLA domain is indeed critical for mediating high-affinity binding between FVII and Ad hexon.

Fig 2.

Integrity, infectivity, and kinetic response data and dissociation constants for FVII binding to wild-type Ad and Ad vectors with mutated hexons. (A) Twofold dilutions of purified virus preparations for the indicated vectors were loaded onto a polyacrylamide gel, resolved, and stained with Coomassie blue to visualize major virus capsid proteins (indicated on the right). M, molecular size ladder. (B) Infectivity of hexon-mutated vectors on virus-susceptible CHO-CAR cells, analyzed by flow cytometry at 24 h postinfection. Error bars indicate standard deviations of the means (n = 6). (C) Experimentally obtained data are shown by multicolored lines, with different colors representing different concentrations of FVII used for the analyses. Global fits of these data to a 1:1 single-site interaction model are shown in gray. The responses are shown in instrument response units (RU) versus time in seconds (s). Representative data obtained from three independent experiments are shown.

Table 1.

Binding of FVII to adenovirus vectors and summary of SPR fitting parameters for 1:1 kinetic model^a

Sample	ka	kd	KD
Ad-WT	5.37E+05	1.60E−03	2.99E−09
Ad-GAT	5.13E+05	1.68E−02	3.27E−08
Ad-TAT	6.76E+05	1.93E−02	2.85E−08
Ad-TEA	NA	NA	NA

^ak_a, association rate constant; k_d, dissociation rate constant; K_D, kinetic dissociation constant; NA, not available.

FVII is inefficient at supporting virus attachment and cell transduction, despite efficient binding to hexon.

The observation of high-affinity FVII binding to Ad hexon (in the low-nanomolar range) and the lack of an FVII contribution to transduction of hepatocytes in vivo after intravenous Ad administration (16, 18) are puzzling. These findings suggest that if the Ad-FVII complex is formed in vivo, the cell transduction properties of such a complex are likely to be inferior to those of Ad complexed with FX. To directly compare the infectivities of the Ad-FVII and Ad-FX complexes, we infected CHO-K1 cells with Ad-WT in the presence of human FVII or FX at concentrations ranging from 0.5 μg/ml to 25 μg/ml. This analysis demonstrated that the addition of FX to Ad-WT prior to infection of this cell line with the virus significantly increased the percentage of virus-transduced cells and the GFP mean fluorescence intensity (Fig. 3A and B). In contrast, the addition of FVII to Ad-WT prior to cell infection did not increase the percentage of virus-transduced cells or the GFP mean fluorescence intensity compared to that in experimental settings where cells were infected with the virus alone. We hypothesized that differential virus attachment to cells in the presence of FVII and FX could explain the reduced infectivity of the Ad-WT–FVII complex compared to the Ad-WT–FX complex. To directly assess this possibility, we incubated [³H]thymidine-labeled Ad-WT vector with CHO-K1 cells in the presence of either FVII or FX and analyzed the efficacy of virus-cell attachment as described earlier (22). This analysis confirmed that the Ad-FX complex bound to cells significantly more efficiently than the Ad-FVII complex did (Fig. 3C). Because we utilized the Ad5S variant for collection of high-contrast cryo-EM images of virus-FVII complexes, we also analyzed whether Ad5S attachment to cells was differentially affected by FX and FVII, using a [³H]thymidine-labeled Ad5S vector. This analysis demonstrated, similar to the case with Ad-WT, that only the addition of FX could increase the number of attached Ad5S particles to cells, while the addition of FVII failed to increase virus attachment (Fig. 3D).

Fig 3.

Transduction of CHO-K1 cells with Ad-WT vector and attachment of Ad-WT and Ad5S vectors to CHO-K1 cells in the presence of FVII and FX. (A and B) CHO-K1 cells were infected with an Ad-WT vector that expresses the GFP reporter gene under the control of the CMV early promoter, with or without the addition of FX or FVII to the infection medium, and were analyzed by flow cytometry 48 h after infection. In control settings, cells were treated with saline instead of the virus (mock). (C and D) Attachment of [³H]thymidine-labeled Ad-WT (C) and Ad5S (D) to CHO-K1 cells in the presence of FVII or FX. Representative data from two independent experiments done in triplicate are shown. The attachment of virus particles to CHO-K1 cells with or without the addition of coagulation factors was analyzed as described in Materials and Methods and is expressed as the number of virus particles per cell (vp/cell). *, P < 0.05; n.s., not statistically significant.

FVII binds HAdv5 hexon in an altered orientation compared to that of FX.

Comparison of the network of factor density in the HAdv5-FVII complex (Fig. 1A and B) with the cryo-EM density observed for the HAdv5-FX complex (20) shows a different distribution for each coagulation factor despite interactions with the HAdv5 capsid via homologous GLA domains. The FVII and FX GLA domains have 60% identity. Our MDFF analyses of the two HAdv5 factor complexes indicated that several of the same hexon residues play a role in mediating binding (Fig. 1) (20). In addition, our Biacore data for the two complexes showed that mutation of hexon residue T425 to alanine ablates binding of both factors (Fig. 2) (20). Interestingly, despite these similarities, the cryo-EM and MDFF analyses indicated that the FVII GLA domain sits within the hexon trimer, in a different orientation from that found for FX (Fig. 4A and B). This leads to a different angle for the rest of the factor molecule with respect to the viral capsid and to a different presentation of the SP domain (Fig. 4C and D). While an Arg residue (R28) in FVII appears to form critical interactions with the hexon TET sequence (Fig. 1D to F), the key factor residue in the modeled FX interaction appears to be a Lys residue (K10) (20). These different interaction residues result in the FVII GLA domain sitting deeper within the central hexon depression and in a different orientation, resulting in an altered positioning of FVII compared to FX (Fig. 4C and D).

Fig 4.

Cryo-EM and MDFF simulations indicate that FVII and FX adopt distinct binding orientations relative to hexon. (A) FVII GLA domain (purple) docked into the central depression of hexon, shown with (upper) and without (lower) hexon. Residue R28 of the GLA domain is shown in space-filling representation and colored by element. (B) The FX GLA domain (red) is shown similarly, with residue K10 of the GLA domain (20). (C) Overlay showing the distinct orientations of the zymogenic forms of FVII (purple) and FX (red) relative to hexon. Basic residues, including those that constitute heparin-binding exosites and are known to contribute to heparan sulfate proteoglycan binding in the FX serine protease domain (R273, K276, R306, R347, K351, K420, and R424) and the FVII serine protease domain (H249, R271, R277, K389, R392, R396, and R402), are shown. (D) Perpendicular view of panel C.

FVII dimerizes via SP domain interactions when bound to HAdv5 hexon.

Visual inspection of the cryo-EM density assigned to FVII near the icosahedral 2-fold axis suggested that two SP domains might interact (Fig. 5). The volume of the SP density at this site is larger than expected for one copy of SP and nearly large enough to encompass two copies of the SP domain (Fig. 5B). We suspect that flexibility between FVII domains leads to weaker and more diffuse SP domain density than that of the GLA domains or hexon. In general, we found a much weaker SP domain density in the HAdv5-FVII cryo-EM structure than in the HAdv5-FX cryo-EM structure (20). This is likely due to flexibility within the FVII molecule combined with formation of different SP domain interactions over the capsid. We assume that each hexon trimer can bind one FVII molecule in three different orientations related by 120° rotations. Therefore, an FVII molecule bound to a specific hexon may have a variety of different neighboring FVII molecules with which it could dimerize, leading to poorly defined SP domain density in the cryo-EM structure.

Fig 5.

Modeling of two molecules of FVII at the icosahedral 2-fold axis. (A) MDFF-refined model of neighboring capsid-bound FVII molecules. These FVII proteins are shown as light and dark purple ribbons, and two hexon trimers are shown as blue ribbons docked within the cryo-EM density. The density is contoured to reveal a small region of density for the FVII GLA domain within the central depression of hexon (purple). (B) Similar to panel A, but with the density contoured to include the distal FVII serine protease domains. (C and D) Side views perpendicular to panels A and B.

To simulate the interaction between neighboring FVII molecules, we docked two copies of FVII and two hexon trimers into the cryo-EM density extracted from a region near the 2-fold axis. This region displays the strongest SP domain density in the HAdv5-FVII cryo-EM structure. The hexon trimers could be docked with a high level of precision because strong density rods were observed for α-helices within hexon. Both FVII molecules were initially docked using the MDFF-refined orientation found for a single FVII molecule and then subjected to further MDFF simulations (Table 2). Although the two FVII models were positioned in the same way with respect to hexon at the beginning of simulation 1, at the end of this simulation the orientations of the two FVII molecules diverged. In addition, the two FVII molecules showed a significant interaction via their SP domains, as indicated by a relatively large and favorable calculated nonbonded interaction energy (−367 kcal/mol). Also, at the end of simulation 1, we found that one FVII-hexon pair had a less favorable interaction than the other (−638 versus −1,027 kcal/mol). After an additional simulation (simulation 2), the two FVII molecules showed a stronger favorable interaction between them, and the nonbonded interaction energies for the two FVII-hexon pairs diverged even farther. These results indicate that neighboring FVII molecules interact with and influence each other. In order to remove the influence of one FVII molecule on the other, a third simulation (simulation 3) was performed with only one FVII molecule and one hexon trimer present. This resulted in a strengthening of the calculated nonbonded interaction energy between FVII and hexon. These results for the HAdv5-FVII complex are clearly different from the previous cryo-EM and MDFF findings for the HAdv5-FX complex, in which no evidence for FX dimerization was observed (20).

Table 2.

Intermolecular nonbonded energies between FVII molecules and hexons at the icosahedral 2-fold axis of HAdv5 at the end of three 100-ps MDFF simulations

Interaction	Intermolecular nonbonded energy (kcal/mol)a
	Simulation 1	Simulation 2	Simulation 3
FVII 1-FVII 2	−367	−413	NA
FVII 1-hexon 1	−638	−281	NA
FVII 2-hexon 2	−1,027	−804	−1,065
Avg for FVII 1-hexon 1 and FVII 2-hexon 2	−842	−563	NA

^aNote that negative values for nonbonded interaction energies are favorable.

FVII SP domain dimerization obscures putative receptor-interacting residues within the dimer interface.

In the FX SP domain, seven basic amino acids (R273, K276, R306, R347, K351, K420, and R424) have been shown to mediate surface attachment of HAdv-FX complexes to HSPGs on hepatocytes (32). Previous analyses identified that K420 and R424 are the most critical residues for heparin binding to FXa, and together with the other charged amino acids, they form the so-called heparin-binding exosite on the surface of the FX SP domain (33). Efficient heparin binding was also confirmed for coagulation factors FVII (34) and FIX (35), and the locations of heparin-binding exosites were proposed to be similar on these factors, based on structural homology analyses (33, 34). Highlighting positively charged amino acids, including those constituting the experimentally confirmed heparin-binding exosite on the available crystal structures for FXa (R306, K420, R424, K427, R429, K433, and K435) and the suggested heparin-binding exosite on FVII (H249, R271, R277, K389, R392, R396, and R402), showed that these amino acids cluster on one and the same side of the SP domain for both FX and FVII (Fig. 6). However, highlighting these amino acids on the three-dimensional surface of the FVII SP domain in complex with HAdv5 revealed that these potential HSPG-binding amino acids are mostly buried in the middle of the modeled SP domain dimerization interface (Fig. 7A). This finding suggests that dimerization between FVII molecules bound to HAdv5 may shield putative receptor-interacting amino acids from HSPGs on the cell surface and contribute to the inability of FVII to mediate virus attachment via HSPGs. This is consistent with our data demonstrating that the addition of FVII to the virus does not increase its attachment to CHO-K1 cells (Fig. 3C and D). Furthermore, our earlier cryo-EM analysis of Ad-FX revealed that the known HSPG-binding residues of the FX SP domain are solvent accessible on the surface of the HAdv5-FX complex (Fig. 7B). The structural analysis indicates that FX bound to HAdv5 should be capable of mediating virus attachment, consistent with our experimental data on HSPG-expressing cells (Fig. 3). At the icosahedral 2-fold axis, we measured the distance of closest approach between FX SP domains to be ∼40 Å. The absence of FX dimerization when FX is bound to HAdv5, as well as a favorable binding orientation for FX, may therefore explain the efficient cell transduction by Ad in the presence of FX. Our data also indicate that in contrast to the case for FX, the poor efficiency of FVII in supporting Ad cell transduction in vitro can be explained, at least in part, by an unfavorable binding orientation leading to dimerization of SP domains and inaccessibility of HSPG-binding residues at the SP domain dimer interface.

Fig 6.

Localization of positively charged amino acids on the surfaces of serine protease domains of coagulation factors FX (A) and FVII (B). Positively charged amino acids are highlighted in yellow within the crystal structures of FXa (structure 1HCG in the structure database [SDB]) and FVIIa (structure 1KLJ in SDB). Side views of the SP domains of FX and FVII are shown in the left panels. The right panels show positively charged amino acids, highlighted in yellow, in front views of the SP domains of FX and FVII. For FX, amino acids K420 and R424 contribute to the heparin-binding exosite and were experimentally confirmed as the most critical residues for heparin binding (33). For FVII, amino acids R392, R396, and R402 were suggested to be part of the heparin-binding exosite, based on biochemical analyses of heparin binding to FVIIa and on structural homology analyses of the exosite location in thrombin and FIXa (34).

Fig 7.

Proximal FVII molecules interact and bury potential heparan sulfate proteoglycan binding residues that are exposed on FX. (A) Top and side views of the MDFF-modeled FVII interactions with two hexon trimers at the icosahedral 2-fold axis. HAdv5 hexons are shown in blue, and FVII molecules are shown in dark and light purple space-filling representations. Seven basic residues (H249, R271, R277, K389, R392, R396, and R402) in the FVII serine protease domain, localized on the side that was proposed to bind heparin and heparan sulfate proteoglycans (24), are shown in yellow. A side view with one transparent hexon and FVII molecule reveals that these basic residues are mostly buried within the modeled serine protease dimer interface. (B) Similar views for the MDFF-modeled FX interactions (20), with two hexon trimers at the icosahedral 2-fold axis. FX molecules are shown in red and do not interact. Seven basic residues in the serine protease domain (R273, K276, R306, R347, K351, K420, and R424) that bind heparan sulfate proteoglycans (4) are shown in green. The residues of both FVII and FX are numbered according to the zymogen models. Bar, 100 Å.

DISCUSSION

Viral vectors based on HAdv5 are highly efficient at infecting both dividing and nondividing cells and are now being used in clinical trials to deliver therapeutic genes to various diseased cell and tissue types in vivo. However, upon intravascular administration, Ads are rapidly trapped in the liver and transduce hepatocytes via a mechanism that relies on binding of the virus hexon protein to blood coagulation factors (15, 18, 36). Although several related and structurally homologous vitamin K-dependent blood coagulation factors, including FVII, FIX, and FX, were shown to be able to mediate virus entry into cells of hepatic origin in vitro, with various efficacies (16), the selective inactivation of FX alone was fully sufficient to ablate virus-mediated hepatocyte transduction in vivo (18). Using surface plasmon resonance, we demonstrated that FIX binds poorly to HAdv5 particles immobilized on the surface of a sensor chip. However, both FVII and FX bind to immobilized HAdv particles or purified hexon trimers with a very high affinity, in the low-nanomolar range (15). Although the physiological molar concentration of FVII in the blood is lower than that of FX, the lack of any contribution of FVII to hepatocyte transduction in vivo is puzzling. This is true even under conditions where FX has selectively been inactivated in vivo (18), indicating that FVII lacks transduction-supporting capability, at least in mouse models of intravascular HAdv5 delivery. We hypothesized that the lack of contribution of FVII to mediating hepatocyte transduction by Ad after intravascular virus administration may be due in part to differences in presentation of HSPG receptor-interacting amino acids on SP domains of FX and FVII upon their binding to the virus.

To address this hypothesis experimentally, we utilized high-resolution cryo-EM followed by MDFF to model the HAdv5-FVII interaction. We recently obtained a subnanometer-resolution cryo-EM structure of the HAdv5-FX complex by using the same approach (20). Subsequent MDFF simulations revealed a potential FX-hexon interaction interface with a single dominant orientation of the FX GLA domain within the central depression of the hexon trimer, involving K10 in the FX GLA domain and HVR7 amino acids E424 and T425 in the HAdv5 hexon protein. Indeed, a single amino acid substitution, T425A, completely abrogated FX binding to hexon and resulted in a loss of hepatocyte transduction by a mutated vector possessing this T425A substitution in hexon. The cryo-EM visualization of the HAdv5-FVII complex presented here confirms that FVII interacts with virus hexons via its GLA domain (Fig. 1). Furthermore, our MDFF analysis indicates that HVR7 amino acids T423, E424, and T425 appear to be critical for the formation of the complex between hexon and the FVII GLA domain, consistent with experimental data showing that mutation of these individual amino acids significantly reduces FVII's binding affinity for the virus. Surprisingly, attempts to use the GLA domain position from the previous HAdv5-FX model (20) to guide positioning of the FVII GLA domain resulted in models that did not match the experimental HAdv5-FVII cryo-EM density. The de novo cryo-EM density-guided MDFF simulations revealed that the FVII GLA domain is positioned deeper within the hexon trimer central depression and in a different orientation from that of FX (Fig. 4). The most surprising finding, however, was that upon binding to the virus via their GLA domains, two adjacent FVII molecules appear to form stable dimers via their SP domains, a feature that was not observed in the HAdv5-FX complex.

Although MDFF simulations were performed only for FVII molecules bound to the hexons at the icosahedral 2-fold axis, manual docking of FVII molecules at other hexons in the capsid was performed. This indicated that a relatively small amount of flexibility between FVII domains, combined with three possible binding orientations of FVII at each trimeric hexon, could lead to dimerization of most, if not all, of the bound FVII molecules on Ad. Averaging of multiple possible dimerization patterns for FVII bound at the hexons other than those at the 2-fold axis would lead to weak SP domain density except at the 2-fold axis. Indeed, a poorly defined SP domain density was observed over most of the Ad capsid in the cryo-EM structure. The hexons at the icosahedral 2-fold axis provide a unique environment with one favored FVII dimerization pair, which leads to stronger SP domain density at this particular site. The structural and functional data on localization of HSPG-interacting residues within FVII (34), combined with our cryo-EM-based model of HAdv5-FVII, indicate that most of the HSPG residues are hidden within the interacting interface of modeled SP domain dimers formed by adjacent FVII molecules bound to the virus (Fig. 7). The suboptimal presentation of these basic residues on the surfaces of SP domains of virus-bound FVII may contribute to the poor capacity of FVII in supporting Ad cell infection. Indeed, the direct assessment of attachment of Ad to the cell surface in the presence of FX and FVII showed that FVII failed to increase the number of attached virus particles to the cell.

Collectively, our structural and functional analyses revealed an unexpected divergence of properties for Ad complexes with FX and FVII. Although several highly homologous vitamin K-dependent coagulation factors and protein C were shown to bind to the virus and support cell infection in isolated systems (15, 16), our study reveals unanticipated and potentially mechanistic insight into the highly efficient and selective role of FX in mediating efficient virus entry into HSPG-expressing cells. Although coadministration of HAdv5 and mouse coagulation factor FVII (at concentrations similar to those of FX) into warfarin-treated mice could reveal the potential role of FVII in supporting virus entry into hepatocytes in a mouse model in vivo, mouse FVII is not commercially available. In addition, administration of human FVII in this setting would not provide definitive information on the role of mouse FVII in virus entry into hepatocytes. Our structural and computational analyses of the HAdv5-FX and HAdv5-FVII complexes indicate a favorable SP domain orientation only for FX, which leads to presentation of heparan sulfate proteoglycan receptor-interacting residues on the virus-bound coagulation factor. Our findings may have important implications for strategies of cell type-specific Ad targeting via FX GLA domain-containing heterologous bifunctional ligands (37). When bound to virus, these novel targeting ligands can interact with each other in unexpected ways, reducing the efficacy of virus cell infection, as we demonstrated here for Ad-FVII complexes.