The Fiber Knob Protein of Human Adenovirus Type 49 Mediates Highly Efficient and Promiscuous Infection of Cancer Cell Lines Using a Novel Cell Entry Mechanism

Adenoviruses are powerful tools experimentally and clinically. To maximize efficacy, the development of serotypes with low preexisting levels of immunity in the population is desirable.

IMPORTANCE Adenoviruses are powerful tools experimentally and clinically. To maximize efficacy, the development of serotypes with low preexisting levels of immunity in the population is desirable. Consequently, attention has focused on those derived from species D, which have proven robust vaccine platforms. This widespread usage is despite limited knowledge in their basic biology and cellular tropism. We investigated the tropism of HAdV-D49, demonstrating that it uses a novel cell entry mechanism that bypasses all known HAdV receptors. We demonstrate, biologically, that a pseudotyped HAdV-C5/D49K vector efficiently transduces a wide range of cell lines, including those presenting no known adenovirus receptor. Structural investigation suggests that this broad tropism is the result of a highly basic electrostatic surface potential, since a homologous pseudotyped vector with a more acidic surface potential, HAdV-C5/D30K, does not display a similar pantropism. Therefore, HAdV-C5/D49K may form a powerful vector for therapeutic applications capable of infecting difficult to transduce cells. KEYWORDS adenoviruses, surface receptor, oncolytic viruses, anticancer therapy cells have previously suggested that HAdV-D49 may engage CD46 as a cellular receptor, although the effects observed were small (23).
Despite these studies and the development of HAdV-D49 as a therapeutic agent, there remains little information surrounding the basic biology of HAdV-D49 and its means of cellular engagement. Here, we investigate the tropism of HAdV-D49, focusing on the fiber knob protein as the major mediator of cellular attachment and evaluate the potential utility of a pseudotyped HAdV-C5/D49K vector to infect a range of cancer cell lines.

RESULTS AND DISCUSSION
HAdV-C5/D49K is not dependent on any known adenovirus receptor for cell entry. To investigate the receptor usage of human adenovirus type 49 fiber knob protein, we generated a replication-incompetent HAdV-C5 vector pseudotyped with the fiber knob protein of HAdV-D49 (HAdV-C5/D49K), expressing either green fluorescent protein (GFP) or luciferase as transgenes. We also produced a replication-deficient HAdV-C5-based pseudotyped vector with the whole fiber protein, including both the fiber shaft and the fiber knob of HAdV-D49, expressing luciferase (HAdV-C5/D49F). This pseudotyping approach is a well-established means to investigate the fiber knob in the context of a well-understood, replication-incompetent virus (1,24). Using these pseudotyped vectors, we performed transduction assays in CHO cells expressing common adenovirus receptors (Fig. 1). CHO-K1 cells do not express any known adenovirus receptor, while CHO-CAR cells express the HAdV-C5 receptor, coxsackievirus and adenovirus receptor (CAR), and CHO-BC1 cells express the BC1 isoform of CD46, the major receptor for species BI adenovirus, which includes HAdV-B35.
D49K engages an alternative cellular receptor. Interestingly, we also observed, in this and later experiments, that HAdV-C5/D49K was less efficient at infecting CHO-CAR cells (Fig. 1A) compared to non-CAR-expressing CHO cell types ( Fig. 1B and C). These data indicate that the presence of CAR may actively reduce the efficiency of transduction of HAdVC5/D49K compared to levels of transduction in the absence of CAR in the same cell line background.
At 385 amino acids in length, the native HAdV-D49 protein is significantly shorter than the equivalent HAdV-C5 fiber protein, which is 581 amino acids in length. This manifests as a naturally shorter and less flexible fiber shaft in HAdV-D49 compared to that of HAdV-C5. This shortened fiber shaft length may impact upon viral infectivity, resulting in trapping of adenoviral particles within late endosomes due to the decreased endosomolytic activity of shorter shafted adenoviral particles (27), reviewed elsewhere (28). To assess the impact of pseudotyping the entire short fiber protein from HAdV-D49 on viral infectivity, we performed similar transduction assays using CHO-K1 cells. Consistent with engaging an alternative receptor on CHO-K1 cells, the HAdV-C5/D49F whole-fiber pseudotyped vector efficiently transduced CHO-K1 cells, where HAdV-C5 was unable. Also consistent with previous observations of shortershafted HAdVs potentially displaying reduced infectivity due to altered or less efficient intracellular trafficking postentry, the HAdV-C5/D49F was less efficient than the "knobonly" pseudotype HAdV-C5/D49K at infecting CHO-K1 cells (Fig. 1D).
We performed similar transduction assays using CHO-K1 and SKOV-3 ovarian cancer cells with or without pretreatment with either heparinase or neuraminidase to determine the ability of HAdVC5/D49K to bind heparan sulfate proteoglycans (HSPGs) or sialic acid, respectively, to mediate cellular infection (Fig. 2). As a positive control for heparinase activity we compared HAdV-C5/D49K infectivity to that of HAdV-C5 in the presence or absence of coagulation factor X (FX), a blood coagulation factor which can facilitate infection of some adenovirus by binding to the viral hexon and cellular HSPGs (29). We observed in CHO-K1 and SKOV-3 cells that transduction levels of HAdV-C5 alone were poor ( Fig. 2A and B) but were significantly enhanced by the presence of FX, enabling cell entry through cellular HSPGs (30)(31)(32). Treatment with heparinase to cleave HSPGs reduced transduction efficiency to that of HAdV-C5 alone. HAdV-C5/ D49K transduction efficiency was unaffected by treatment with heparinase ( Fig. 2A and B), indicating that HAdV-D49 is unlikely to utilize HSPGs for cell entry.
Treatment with neuraminidase to remove cellular sialic acid did not alter the ability of any of the viruses to transduce CHO-K1 cells (Fig. 2C), as we previously demonstrated to show the involvement of sialic acid in HAdV-C5/D26K infection (24). In SKOV-3 cells, removal of sialic acid actually enhanced the transduction mediated by HAdV-C5/D49K and HAdV-C5/B35K, an effect which we have previously observed by neuraminidase treatment in SKOV-3 cells (Fig. 2D) (24,33,34). This effect could be a result of the removal of sialic acid enhancing nonspecific charge-based interactions between the cell surface and viral capsid. Regardless, these data do not support a role for sialic acid in HAdV-C5/D49K cell infection.
The transduction affinity of HAdV-C5/D49K in the experiments in Fig. 2 was noticeably weaker than in the CHO cell experiments (Fig. 1). This is due to the methodology used in each experiment. In the transduction experiments ( Fig. 1), cells were incubated with virus at 37°C for 3 h. For studies evaluating the role of sialic acid and HSPGs, cells were pretreated with enzyme for 1 h at 37°C. The virus was then incubated with cells on ice for 1 h after enzymatic digestion to prevent repair and reconstitution of the cleaved heparin/sialic acid. This incubation on ice (and for a shorter period of time) likely decreases viral internalization during the absorption step, seemingly more profoundly for HAdV-C5/D49K than for the HAdV-C5, suggesting weaker binding at the cell surface or a comparatively low frequency of cell surface receptor.
Desmoglein 2 (DSG2) is the other remaining well-established adenovirus receptor. DSG2 is described to interact with species BII adenovirus, including HAdV-B3K, via a low-affinity, avidity-dependent mechanism (35). We investigated whether HAdV-D49K might also interact with DSG2 by utilizing surface plasmon resonance (SPR), which we have previously used to establish a 66.9 mM affinity between DSG2 and HAdV-D3K (36). HAdV-D49K had no detectable affinity for DSG2 (Fig. 3A), an unsurprising finding since DSG2 has never been observed as a receptor for any adenovirus outside species B.
HAdV-D49 fiber knob can interact with CAR but does not require it for cell entry. We also used SPR to further probe the binding affinity of HAdV-D49K, and a mutant version, HAdV-D49.KO1.K, for CD46 and CAR. This HAdV-D49.KO1.K mutant, harbors the KO1 mutations S408E and P409A in the fiber knob AB loop, previously shown to ablate CAR binding in HAdV-C5K (37). The structure of the HAdV-D49.KO1.K fiber knob is also presented ( Table 1, PDB 6QPO) (38). As predicted, we did not observe binding between either fiber knob protein and CD46. However, we did observe HAdV-D49K binding to CAR with a detectable 0.19 mM affinity which was ablated by the KO1 mutation (Fig. 3A).
We performed 50% inhibitory concentration (IC 50 ) binding studies using recombinant HAdV-D49K protein on CHO-CAR and CHO-BC1 cells and assessed the ability of the recombinant fiber knob protein to inhibit the binding of anti-CAR or CD46 antibodies, respectively (Fig. 3B). HAdV-D49K was able to block anti-CAR antibody binding to CHO-CAR cells in a dose-dependent manner (IC 50 = 0.16 mg/10 5 cells, Fig. 3B). However, no IC 50 could be derived by using HAdV-D49K to block CD46 on CHO-BC1 cells, where HAdV-D49K was unable to achieve .20% inhibition of antibody binding, suggesting weak or incidental CD46 interactions (Fig. 3B). Therefore, these data support the findings from SPR and transduction experiments that HAdV-D49K may bind CAR with low affinity but does not bind CD46.
One potential explanation for this activity is that the unknown alternative receptor to CAR has a lower affinity for HAdV-D49K than CAR. Therefore, in the presence of CAR the recombinant fiber knob would be sequestered on the higher-affinity CAR receptor, leaving the alternative receptor free to interact with the virus. A low-affinity receptor would also likely depend upon avidity and so might not be observed with single trimers of HAdV-D49K; a similar effect has previously been observed with HAdV-B3K and DSG2 (39) and CD46 (40). This is supported by the observation that HAdV-D49K cannot transduce cells as efficiently when incubated on ice in the absence of CAR, whereas HAdV-C5 and HAdV-C5/B35K, which form high-affinity receptor interactions, are unencumbered (Fig. 2).
HAdV-D49K may bind cells through a charge-dependent mechanism. To investigate other closely related HAdV with homologous fiber knob proteins, we performed a BLASTp search using the HAdV-D49K amino acid sequence. This search revealed that the HAdV-D30K protein is highly homologous to HAdV-D49K, differing in just 4 amino acid residues (Fig. 4A).
This profound difference in transduction efficiency between HAdV-C5/D30K and HAdV-C5/D49K must be dependent upon the three surface-exposed amino acid differences. We investigated the effect of the opposing charges at residue substitutions 238 and 331 (Fig. 4A to E) by modeling the electrostatic surface potential of the two fiber knob proteins, based on our crystal structures (Fig. 5). The surface potential maps reveal that while structural homology was high, they present radically different electrostatic surface potential distributions. HAdV-D30K is significantly more acidic (pI = 5.57) than HAdV-D49K (pI = 8.26) (Fig. 5).
Thus, it seems probable that the interaction with the unknown cell surface receptor requires basic electrostatic potential. This is commensurate with the previous inference that its receptor is likely to be low affinity, since electrostatic interfaces are often FIG 4 HAdV-D49K differs from HAdV-D30K in only three surface-exposed amino acids but demonstrates radically altered cellular tropism. (A) Clustal X sequence alignment (numbering is based on the whole fiber sequence) of HAdV-D30K and HAdV-D49K. (B and C) Viewed from the apex down the 3-fold axis, as if toward the viral capsid, the crystal structures of HAdV-D30K (B) and HAdV-D49K (C) reveal that three of these residues are surface exposed. These residues can be seen projecting into the solvent from loops on the apex of HAdV-D30K (D) and HAdV-D49K (E). Residue numbers and names correspond to the fiber knob protein depicted in that frame. Sticks representing residues belonging to HAdV-D30K and HAdV-D49K are seen in pink and green, respectively. (F and G) Transduction assays were performed to assess tropism of HAdV-C5/D30K, HAdV-C5/D49K, and HAdV-C5/D49.KO1.K in CHO-CAR cells (F) and CHO-K1 cells (G). Cells were infected with 5,000 viral particles per cell of replication-deficient HAdV-C5 or HAdV-C5/D49K expressing a luciferase transgene, with or without blockade by 20 mg of recombinant HAdV-C5 or HAdV-D49 fiber knob protein.
observed to be less stable than their ionic counterparts. It is possible that the electrostatic potential differences explain the reduced transduction affinity observed in HAdV-C5/D30K compared to HAdV-C5/D49K in CHO-CAR cells. Should the strong charge on HAdV-D49K be opposed to that on the surface of CAR, this could enhance the interaction stability and therefore the overall virus affinity. It seems unlikely that the residue substitutions themselves would strongly influence CAR affinity since they occur at the apex of the fiber knob, an area that is not critical for the CAR interface (1).
HAdV-C5/D49K is able to efficiently infect a large range of cancer cell lines. Given that HAdV-C5/D49K infects cells independently of known adenovirus receptors, we hypothesized that it may form the basis of an efficient vector for cancer virotherapy applications. We therefore compared its transduction efficiency to that of HAdV-C5 in panels of pancreatic, breast, esophageal, colorectal, ovarian, and lung cancer cell lines ( Table 2).
In pancreatic cancer cell lines, HAdV-C5/D49K was consistently more efficient at cellular transduction than HAdV-C5. This improved activity ranged between 4.2Â more efficient in MiaPaCa2 cells to 210.9Â more efficient in BxPc3 cells. The most effectively transduced cell line was Panc10 cells, producing 7.2 Â 10 6 relative light units (RLU)/mg of fluorescence, compared to the least efficient at just 4.0 Â 10 5 RLU/mg in Panc0403 cells. This suggests that the large differences between HAdV-C5/D49K and HAdV-C5 transduction levels are likely due to the variability in the expression of CAR. A similarly broad range of different relative infection efficiencies was observed in the breast cancer cell lines studied. In MCF7 cells and BT20 cells, HAdV-C5/D49K was nearly 500-fold more efficient at transducing the cells due to these cells expressing low levels of CAR, with consequent poor levels of HAdV-C5-mediated transduction ( Table 2).
Conclusions. Previous experiments using whole HAdV-D49 virus concluded that it utilizes CD46 as its primary cellular receptor (23). The data presented, generated using either purified HAdV-D49 fiber knob protein or a pseudotyped HAdV-C5/D49K vector, clearly demonstrate CD46 to be implausible as a receptor for HAdV-D49.
We demonstrate that the HAdV-D49 fiber knob has a weak affinity for CAR, although it is not dependent upon this interaction to mediate efficient cell entry, and in fact the presence of CAR may inhibit cellular transduction. In support of this, we show that a mutant vector, HAdV-C5/D49.KO1.K, containing mutations within the fiber knob domain which ablate CAR affinity, retained the ability to efficiently transduce cells in the absence of any detectable binding to CAR. Based on the low efficiency by which HAdV-C5/D49K transduced cells when absorbed on ice and the observation that HAdV-D49K is only capable of inhibiting HAdV-C5/D49K transduction in the absence of CAR, we tentatively suggest that the unknown receptor is likely to be bound with weak affinity and virus attachment may be avidity dependent.
Regardless of the mechanism of interaction, this study strongly suggests there is an as-yet-unknown adenovirus receptor or mechanism of cell entry which mediates efficient transduction of a broad range of cell lines. This is demonstrated by its ability to efficiently infect every cell line tested throughout this study. The weakest observed transduction was in Panc0403 cells, where it achieved 4.0 Â 10 5 RLU/mg of luminescence. Although this is not a particularly strong transduction efficiency, it is still significantly higher (33.0Â, P , 0.05) than that of HAdV-C5. It is likely, therefore, that the HAdV-C5/D49K vector described here may be useful in biotechnology applications to efficiently express proteins in difficult-to-transduce cell lines.
HAdV-C5/D49K represents a highly efficient gene transfer vehicle that is not restricted by any known adenovirus tropism. It possesses a broad range of infectivity and has potential as both a laboratory reagent for the transient expression of transgenes and as a therapeutic vaccine or oncolytic virus. For oncolytic applications, it is likely The fold change was determined as the RLU/mg of HAdV-C5/D49K divided by the RLU/mg of HAdV-C5. b Statistical significance (P), determined using a Student t test, is indicated as follows: *, P , 0.05; **, P , 0.01; ***, P , 0.001; ****, P , 0.0001; or ns, not significant.
that a further refinement, such as the introduction of mutations known to confer tumor selective replication, such as a dl24 mutation (41)(42)(43), or the use of tumor-specific promoters, such as hTERT (44) or Survivin (45), to drive transgene therapeutic expression selectively within tumor cells will be necessary to ensure tight tumor selectivity.

MATERIALS AND METHODS
GFP transduction assay. Adherent cells were seeded into a Nunc delta surface 96-well cell culture plate (Thermo Fisher) at a density of 5 Â 10 4 cells/well in 200 ml of cell culture media and left to adhere overnight at 37°C in a 5% CO 2 humidified atmosphere. The medium was removed, and cells were washed twice with 200 ml of phosphate-buffered saline (PBS). Virus was added at the desired concentration in 200 ml of serum-free RMPI 1640, followed by incubation for 3 h. The virus-containing medium was then removed and replaced with complete cell culture medium, and the cells were incubated for a further 45 h. The cell culture medium was then removed, and the cells were washed twice with 200 ml of PBS, trypsinized in 50ml of 0. Luciferase transduction assay. Luciferase infectivity assays were performed using the luciferase assay system kit (Promega). Cells were seeded into a Nunc delta surface 96-well cell culture plate (Thermo Fisher) at a density of 2 Â 10 4 cells/well in 200 ml of cell culture media and left to adhere overnight at 37°C in a 5% CO 2 humidified atmosphere. The medium was removed, and the cells were washed once with 200 ml of PBS. Luciferase transgene encoding replication-incompetent viruses were added to the wells at the required titer in 200 ml of serum-free RMPI 1640, followed by incubation for 3 h. The virus-containing medium was then removed and replaced with complete cell culture medium, and the cells were incubated for a further 45 h. The cell culture medium was then removed, and the cells were washed twice with 200 ml of PBS and then lysed in 100 ml of cell culture lysis buffer (part of the Promega kit) diluted to 1Â in ddH 2 O. The plate was then frozen at 280³C.
After thaw, 10 ml of lysate from the cell culture plate mixed was then transferred to a white Nunc 96microwell plate (Thermo Fisher), and 100 ml of luciferase assay reagent (Promega kit) was added. The luciferase activity was then measured as RLU by a plate reader (Clariostar; BMG Labtech). The total protein concentration was determined in the lysate by using a Pierce BCA protein assay kit (Thermo Fisher) according to the manufacturer's protocol, and the absorbance was measured on an iMark microplate absorbance reader (Bio-Rad).
The relative virus infection was determined by normalizing the measured luciferase intensity to the total protein concentration (the RLU was divided by protein concentration). This gave a final infectivity readout in RLU/mg of protein.
Blocking of virus infection with recombinant fiber knob protein. This assay was also performed using the luciferase assay system kit (Promega). Cells were seeded into a Nunc delta surface 96-well cell culture plate (Thermo Fisher) at a density of 2 Â 10 4 cells/well in 200 ml of cell culture media and left to adhere overnight at 37°C in a 5% CO 2 humidified atmosphere. The medium was removed, the cells were washed twice with 200 ml of cold PBS, and the plate was cooled on ice. Then, 20 pg/cell of recombinant adenovirus fiber knob was added to each well in 200 ml of cold PBS, followed by incubation on ice in a 4°C cold room for 1 h. The medium was then removed, and luciferase transgene encoding replicationincompetent viruses were added to the necessary wells at the required titer in 200 ml of cold serum-free RPMI 1640, followed by incubation on ice in a 4°C cold room for 1 h. The virus-containing medium was then removed and replaced with complete cell culture medium, and the cells were incubated for a further 45 h under normal cell culture conditions. From this point forward, the assay is identical to the GFP and luciferase transduction assays.
Heparinase and neuraminidase transduction assays. Cells were seeded at a density of 5 Â 10 4 cells/well in a flat-bottom 96-well cell culture plate and incubated overnight at 37°C to adhere. Cells were washed twice with 200 ml of PBS. Then, 50 ml of neuraminidase (from Vibrio Cholera, Merk) at a concentration of 50 mU/ml or 50 ml of heparinase III (from Flavobacterium heparinum [Merck]) at a concentration of 1 U/ml was diluted in serum-free media, added to the appropriate wells, and incubated for 1 h at 37°C. Cells were cooled on ice and washed twice with 200 ml of PBS. GFP-expressing, replicationincompetent viruses were added to the appropriate wells at a concentration of 5,000 viral particles per cell in 100 ml of serum-free media at 4°C, followed by incubation on ice for 1 h. Serum-free medium alone was added to uninfected control wells. Cells were washed twice with 200 ml of cold PBS, and complete medium was added (Dulbecco modified Eagle medium, 10% fetal calf serum), followed by incubation for a further 48 h at 37°C. Cells were then trypsinized and transferred to a 96-well V-bottom plate, washed twice in 200 ml of PBS, and fixed in 2% paraformaldehyde containing PBS for 20 min before washing and resuspension in 200 ml of PBS.
Samples were analyzed by flow cytometry on an Attune NxT (Thermo Fisher), and voltages were set prior to each experiment, for each cell type, using an uninfected cell population treated identically. Data were analyzed using FlowJo, gating sequentially on singlets, cell population, and GFP-positive cells. The levels of transduction were defined as the percentage of GFP-positive cells (% 1ve), and/or total fluorescence (TF), defined as the percentage of GFP-positive cells multiplied by the MFI of the GFP-positive population. These measures are distinct in that % 1ve describes the total proportion of cells infected, and TF describes the total efficiency of transgene delivery.
Surface plasmon resonance. Surface plasmon resonance was performed, in triplicate, as previously described, using recombinant HAdV-D49K protein (36). Approximately 5,000 RU of recombinant human desmoglein-2 Fc chimera protein (R&D Systems, catalogue no. 947-DM-100) was amine coupled to a CM5 sensor chip at a slow flow rate of 10 ml/min to ensure uniform distribution on the chip surface.
Competition inhibition assay. Competition inhibition assays of antibody binding to cell surface receptors were performed as previously described (36).
Generation of recombinant fiber knob proteins. Recombinant fiber knob proteins used in transduction inhibition, antibody blocking, and crystallization experiments were produced as previous described (24,36). Briefly, pQE-30 vectors containing the sequence of the relevant fiber knob protein, spanning from 13 amino acids preceding the TLW motif to the stop codon, were transformed into SG13009 Escherichia coli harboring the pREP-4 plasmid. Portions (1 liter) of these E. coli were grown to and optical density of 0.6, and protein expression was induced with a final concentration of 0.5 mM IPTG (isopropyl-b-D-thiogalactopyranoside). E. coli was harvested by centrifugation and resuspended in 50 ml of lysis buffer (50 mM Tris [pH 8.0], 300 mM NaCl, 1% [vol/vol] NP-40, 1 mg/ml lysozyme, 1 mM b-mercaptoethanol). Sample was then loaded onto a HisTrap FF Crude column and eluted by using imidazole. Fractions determined to contain protein of interest were then concentrated to ,1 ml total volume and purified by size exclusion chromatography using a Superdex 200 10/300 GL Increase column.
Calculation of electrostatic surface potentials and pIs. Electrostatic surface potential and isoelectric points were calculated at pH 7.2 using the PDB2PQR Server (V 2.1.1) (46) as previous described (24).
RMSD calculation, sequence alignment, and imaging of crystal structures. Alignments were performed using the Clustal Omega multiple sequence alignment algorithm and visualized with BioEdit (47,48). RMSD calculations were performed using the "align" command in PyMOL 2.0, which was also used to visualize protein structures (49).
Data availability. The following proteins have been deposited in the Protein Data Bank (PDB) under the indicated accession numbers: adenovirus 30 fiber knob protein (6STU), adenovirus species D serotype 49 fiber knob (6QPN), and adenovirus species D serotype 49 fiber knob KO1 mutant (6QPO).