Ad5_NULL-A20: A Tropism-Modified, αvβ6 Integrin-Selective Oncolytic Adenovirus for Epithelial Ovarian Cancer Therapies

This article is featured in Highlights of This Issue, p. 4057

Ovarian cancer remains the deadliest gynecologic cancer with global 5-year survival rates below 50% (1). The early stages of the disease are commonly asymptomatic, with the result that most patients have advanced, incurable disease, at presentation. Ovarian cancer metastasizes with large volumes of malignant, intraperitoneal ovarian ascites (OAS) providing a protumorigenic microenvironment (2). Chemoresistance rapidly develops during treatment, requiring alternative regimens. Epithelial ovarian cancer (EOC) is the most common (90%) ovarian cancer type (3). One-third of patients with EOC have cells expressing an epithelial cancer-specific marker, αvβ6 integrin (4). Upregulation of αvβ6 expression in cancer has been linked to aggressive transformation, metastasis, and poor prognosis (5–8). αvβ6 is absent in healthy epithelium (5, 9) but widely overexpressed in plethora of cancers, including ovarian, lung, skin, esophageal, cervical, and head and neck cancers (4), thus making it a promising target for therapeutic vectors. αvβ6 is an activator of TGFβ1 signaling that promotes metastasis by enhancing angiogenesis, immune cell suppression, and epithelial-to-mesenchymal transition (reviewed in ref. 10).

Cancer virotherapy is undergoing renewed interest, including recent regulatory approval for clinical use of herpes simplex type 1–based talimogene laherparepvec (T-VEC), the first oncolytic immunotherapy approved for advanced melanoma (11). Very recently, oncolytic viruses were shown to sensitize difficult-to-treat tumors, including triple-negative breast cancer (TNBC; ref. 12) and glioblastoma (13) to subsequent immunotherapies with immune checkpoint inhibitors. This highlights the potential of virotherapies for combination studies in the clinical setting, and the scope for generating a vector capable of systemically targeting tumors following intravenous introduction. Adenovirus serotype 5 (Ad5) has been commonly deployed in clinical trials of cancer and gene therapies (14), due to ease of genetic manipulation and capacity for large transgenes (15). However, this serotype has suboptimal features that hamper its wider clinical use. As a common respiratory virus with high seroprevalence rates (16), efficient neutralization of vector by neutralizing antibodies (nAb) limits efficacy. Other limitations include significant and rapid off-target sequestration to spleen and liver via complexing of the virion with human coagulation factor 10 (FX; ref. 17) and potentially other coagulation factors (reviewed extensively in ref. 18), “bridging” the complex to heparan sulphate proteoglycans (HSPG), abundant on hepatocytes (19). In vitro, Ad5 enters host cells via coxsackie and adenovirus receptor (CAR; ref. 20) that is ubiquitous within tight junctions on polarized epithelial cells (reviewed in ref. 21) but commonly downregulated in progressive cancers (22–26), limiting use of wild-type Ad5 for tumor therapy.

We have generated a novel virotherapy vector, Ad5_NULL-A20, with altered, tumor-selective tropism. We ablated all native tropisms of Ad5 by mutating key residues in the three main capsid proteins (hexon, fiber, and penton) and retargeted the resulting vector, Ad5_NULL, to the tumor-selective integrin αvβ6 through incorporation of an αvβ6-binding peptide (A20, NAVPNLRGDLQVLAQKVART) within the fiber knob domain HI loop, generating the novel vector Ad5_NULL-A20. A20 peptide was originally derived from foot-and-mouth disease virus (FMDV) capsid protein VP1, and has high affinity for its native receptor, αvβ6 integrin (27, 28). We have investigated potential clinical utility of an oncolytic variant of Ad5_NULL-A20 (Δ24/T1) for intraperitoneal treatment of ovarian cancer by investigating its biodistribution, tumor-selective oncolytic capabilities and avoidance of immune neutralization using in vitro and in vivo models of human EOC.

All vectors generated in this study included a luciferase (Luc) reporter gene. Genetic modifications were carried out by AdZ homologous recombineering methods (29) as described previously (30). Viruses were produced in T-REx-293 or HEK293-β6 cells (for A20-modified viruses) and purified as described previously (30, 31). A triply detargeted vector genome, Ad5_NULL, was generated by introducing mutations in key genes encoding of each of the major capsid proteins to preclude cellular uptake by all known native Ad5 pathways. Ablation of binding to CAR was achieved via the KO1 mutation in the AB loop of the L5 fiber knob gene; ablation of binding to coagulation factor 10 (FX) via a mutation in hypervariable region 7 of the L3 hexon gene; and ablation of αvβ3/5 integrin binding via RGD-to-RGE mutation in the L2 penton base gene. αvβ6 retargeting was achieved by insertion of sequences encoding peptide A20 (NAVPNLRGDLQVLAQKVART) into the fiber knob HI loop (between residues G546 and D547) of Ad5_NULL, generating Ad5_NULL-A20. Replication-deficient variants of Ad5_NULL-A20 carry a complete E1/E3 deletion. Oncolytic variants have a 24-base pair deletion (dl922–947) in the retinoblastoma protein (pRB) binding domain of E1A (Δ24; ref. 32) and a single adenine insertion at position 445 within the endoplasmic reticulum (ER) retention domain of E3/19K (T1 mutation; ref. 33). Additional details of genetic modifications are provided in Supplementary Methods.

Homology modeling was performed using the previously published Ad5 fiber knob structure (PDB ID: 1KNB; ref. 34) and FMDV O PanAsia VP1 protein in complex with ανβ6 (PDB ID: 5NEM; ref. 35). The peptide sequence forming the interaction with ανβ6 (NVRGDLQVLAQKVART) was edited to conform with the A20 peptide sequence used in this study (NAVPNLRGDLQVLAQKVART), docked to the Ad5 fiber knob structure in the HI loop and the KO1 mutation added using WinCoot (36) and PyMol 2.0 (37). The crude Ad5_NULL-A20 structure was aligned with the existing 5NEM structure and the complex energy minimized using the YASARA algorithm (38). Binding energy calculations were performed using PISA (39), and surface charge was calculated using APBS tool in PyMol 2.0 (37).

αvβ6-high/CAR⁺ SKOV3-β6 cell line was generated in-house by retroviral transfection of SKOV3 cells (that natively express the αv subunit; ref. 40) with integrin beta6 pBABE puro plasmid to express the β6 subunit. Primary EOC cells from ascites were obtained through Wales Cancer Bank under existing ethical permissions (WCB 14/004). Cells were processed and subcultured as described previously (30, 31), and tested regularly for Mycoplasma infection by commercially available PCR-based methods.

Cell surface receptor expression was assessed by flow cytometry (30). The presence of anti Ad5 antibodies in OAS and serum was determined by ELISA as previously reported (41). Antigen specificity of the antibodies was assessed by Western blot. Transduction efficiency was assessed by standard luciferase assays, described previously (30, 31). Animal experiments were approved by Institutional Care and Use Committee (IACUC) and performed at Mayo Clinic, Rochester, MN. Animals were age and sex matched. Animal handling and injections were performed by a veterinary technologist. In vivo experiments are further described in detail in Supplementary Methods.

Figures and statistical analyses were generated using GraphPad Prism v6.03. In vitro and ex vivo assays were analyzed by two-tailed unpaired t tests or one-way ANOVA with the Dunnett multiple-comparisons post hoc test. In vivo data were normalized and analyzed by one-way ANOVA or Kruskal–Wallis test with Sidak or Dunn multiple-comparisons post hoc test, respectively. Overall survival (%) following oncolytic treatment is shown as a Kaplan–Meier survival curve; survival proportions were analyzed by the Gehan–Breslow–Wilcoxon test.

We generated and produced to very high viral titers replication-defective and oncolytic variants of a novel Ad5_NULL-A20 vector (Fig. 1A) with three detargeting mutations and an A20 peptide insertion that retargets the vector to αvβ6 integrin-expressing cells (Fig. 1B). Additionally, we generated replication deficient and oncolytic versions of Ad5.A20, which harbors the αvβ6 targeting-peptide A20 insertion, in the absence of any detargeting modification. The multiple genetic manipulations did not have a significant impact on viral titer (Fig. 1A).

We generated a homology model of the Ad5_NULL-A20 fiber knob protein in complex with the ανβ6 dimer (Fig. 1C). The A20 peptide (dark blue) occupies space spanning both αν (green) and β6 (purple) subunits. The predicted Ad5_NULL-A20 interacting residues of the A20 peptide (dark blue) and the native knob structure (cyan) against the approximated charge surface of the ανβ6 (red is negative, blue is positive; Fig. 1D). The ανβ6 has mostly negative surface potential in this region (1D), complementary to the predominantly positive charge of the Ad5_NULL-A20 interface (Supplementary Fig. SI A). The adjacent CD loop of the native Ad5 fiber knob contributes two polar residue interactions from Lys-442 and Gly-443 (1D), binding to an additional three αν residues (Supplementary Fig. SI B). The binding energy of the ανβ6-Ad5_NULL-A20 fiber knob complex is calculated to be –24.3 Kcal/mol, suggesting an exceptionally stable interface (Supplementary Fig. SI C), providing confidence that our αvβ6 targeting strategy was feasible.

The transduction efficiency of replication-deficient vectors was assessed in cell lines expressing variable levels of CAR and αvβ6 integrin. The detargeting mutation triplet of Ad5_NULL-A20 completely abolished entry via CAR in CHO-CAR cells (CAR⁺), while Ad5 transduced these cells at expectedly high efficiency (Fig. 2A). The HVR7 mutation abolished Ad5 vector transduction via FX pathway (Fig. 2B; ref. 42). As expected, FX significantly increased transduction of Ad5 into CHO-K1 cells as compared with FX-free culture conditions (Fig. 2B, left). Conversely, addition of human FX in culture medium had no effect on the transduction efficiency of the FX binding-ablated Ad5.HVR7 control vector in these cells (Fig. 2B; right). Furthermore, the enhanced transduction seen for Ad5 was reversed by the addition of a 3:1 molar excess of Gla-domain interacting protein, anticoagulant X-bp, which binds and inactivates FX in the medium (ref. 19; Fig. 2B, left). On the contrary, FX depletion did not affect the transduction of Ad5.HVR7 vector (Fig. 2B, right).

We confirmed αvβ6 integrin as the primary entry receptor for the triply detargeted, integrin retargeted Ad5_NULL-A20 vector (Fig. 3). Ad5_NULL-A20 transduced αvβ6^+/CAR⁻ BT-20 breast cancer cells with 305-fold higher efficiency (Fig. 3A; P = 0.0270) and primary, patient-derived EOC004 cells (αvβ6⁺/CAR⁻) at 69-fold increased efficiency (Fig. 3B; P = 0.0090) relative to Ad5. Competition assays using a function-blocking anti-αvβ6 antibody (10D5) significantly inhibited cell transduction by Ad5_NULL-A20 vector in SKOV3-β6 cells (αvβ6⁺/CAR⁺; Fig. 3C; P = 0.0010), confirming the vector's selectivity for αvβ6 integrin.

We next evaluated the ability of the Ad5_NULL-A20 vector to retain its infectivity in the highly neutralizing environment presented by OAS. To this end, freshly isolated clinical OAS samples from 20 patients with ovarian cancer were screened for the presence of anti-Ad5 antibodies by direct ELISA. The titers of anti-Ad5 abs in malignant OAS were scrutinized against the serum anti-Ad5 antibody titer of a healthy adult male volunteer (Fig. 4A). Equal proportions of patients were found to have lower and higher antibody titers than the control serum (Fig. 4A, black dashed line). Ascites from patient 001 (OAS001) were chosen for subsequent neutralization assays due to its similar antibody titer with the control serum. Antibodies in OAS001 and control serum appeared specific for the viral fiber protein, whilst the most abundant capsid protein—hexon—was recognized only at very low levels in Western blot using denatured whole viral particles (Fig. 4B). The neutralizing effect of OAS001 on transduction efficiency of Ad5_NULL-A20 was assessed in αvβ6⁺/CAR⁻ EOC004 primary cells. Ad5_NULL-A20 showed up to 902-fold higher transduction efficiency in primary human EOC cultures relative to Ad5 at OAS concentrations of 2.5%, 5%, and 10%, whilst Ad5 was not capable of transducing these cells at detectable levels (Fig. 4C).

We next evaluated biodistribution of virus infection in immunocompetent, non–tumor-bearing mice. Mice were injected intravenously with replication-defective vectors to assess in vivo tropism (Fig. 5A), in particular the effect of the three detargeting mutations on biodistribution of virus infection. As expected and as previously documented, the Ad5 vector showed intense localization in the area of liver and spleen, while luminescence by the Ad5_NULL-A20 vector was completely undetectable at the 72-hour time point (Fig. 5B). Animals inoculated with Ad5 vector had significantly higher whole-body luminescence than the control animals (P < 0.0001) or the Ad5_NULL-A20 vector (P < 0.0001; Fig. 5C). The liver, spleen, lungs, ovaries, and heart were resected postmortem and quantified for ex vivo luminescence (for luminescence heatmaps, see Supplementary Fig. SIIA–C). The livers of Ad5-challenged animals emitted significantly more luminescence than the PBS control or Ad5_NULL-A20 groups (both P < 0.0001; Fig. 5D). Similarly, Ad5_NULL-A20 had significantly decreased transgene expression in the spleen, lungs, ovaries and heart, relative to Ad5 (Fig. 5E–H; P < 0.0001 for all). For fold changes in luminescence intensity in each off-target organ, see Supplementary Fig. SIID.

Confirmation that the modifications in Ad5_NULL-A20 resulted in reduced sequestration of virus in multiple normal tissues was performed via quantitation of viral load by qPCR. Genome copy-number of the Ad5_NULL-A20 vector was 10 million times lower in the liver relative to the Ad5 (Fig. 6A; P < 0.0001). Similarly, Ad5_NULL-A20 genome copy-number was over 700-fold lower in the spleen compared with Ad5 (Fig. 6B; P < 0.0001). In addition, the Ad5_NULL-A20 vector showed improved off-target profiles in all organs relative to Ad5, with viral load 10⁵, 10⁴, and 10³ lower in the lungs, heart, and ovaries, respectively (Fig. 6C–E). Successful detargeting of the liver being due to our genetic modifications of Ad5 is supported by immunohistochemical staining of liver sections, which showed high expression levels of CAR, whilst αvβ6 was undetectable (Supplementary Fig. SIIIA). Confirmation of the detargeting effects of genetic modifications in Ad5_NULL-A20 is provided by the observation that liver sections from mice showed positive staining for Ad capsid proteins in the Ad5 group, but not in livers of mice that had been challenged with the Ad5_NULL-A20 vector (Supplementary Fig. SIIIB).

To evaluate efficacy in a human EOC model in vivo, SKOV3 human ovarian cancer xenografts were established in immunocompromised NOD/SCID mice. Animals developed large solid tumors at the cell injection site and at various sites within the peritoneal cavity within 14 days after intraperitoneal implantation of SKOV3 cells (for tumor localization and take rate; see Supplementary Fig. SIV) and by day 49, tumors were spread throughout the peritoneal cavity with accumulation of large volumes of ascites. Based on these observations, we performed virotherapy efficacy studies by delivering three intraperitoneal doses of oncolytic variants of Ad5, Ad5.A20, and Ad5_NULL-A20 vectors on days 14, 16, and 18 after implantation of SKOV3 cells.

IVIS imaging at 48 hours after first virotherapy treatment dose (day 16) showed widespread luminescence throughout the abdominal region in animals with SKOV3 xenografts and treated with the oncolytic Ad5 vector, with highest intensity in the liver/spleen region (Fig. 7B). This distribution was maintained, but at lower intensity, until 5 days later, day 21 (Fig. 7B). In contrast, the oncolytic Ad5_NULL-A20 vector, however, showed selective tumor localization, with significantly reduced overall luminescence relative to Ad5, consistent with successful detargeting of non-tumor tissues. The distribution of infection mediated by the oncolytic Ad5.A20 vector was intermediate between the Ad5 and Ad5_NULL-A20. Quantitation of total body luminescence showed uptake of the Ad5_NULL-A20 vector to be significantly lower than Ad5 both on day 16 (Fig. 7C; P < 0.05 and <0.01, respectively) and on day 21 (Fig. 7D; P < 0.0001), while there was no statistically significant difference in the uptake of Ad5.A20 as compared with Ad5.

Antitumor activity was observed for oncolytic Ad5, oncolytic Ad5.A20, and oncolytic Ad5_NULL-A20 in the SKOV3 xenograft model (Fig. 7E). Consistent with an enhanced tumor-selective effect of Ad5_NULL-A20, all mice treated with Ad5_NULL-A20 were still alive and tumor-free at the final time point of 101 days, while animals treated with either Ad5 or Ad5.A20 almost identical (and statistically not significantly different) survival curves with median survival of around 60 days.

We describe here an exquisitely refined and tumor-selective oncolytic adenoviral vector, Ad5_NULL-A20, which is ablated for all known native tropisms and retargeted to an overexpressed, prognostic cancer marker—αvβ6 integrin (43). Integrin αvβ6 is a promising target for therapeutic cancer applications due to its overexpression in aggressively transformed cancers (4). A20 peptide is a feasible tool for a variety of clinical applications and has been used for imaging diagnostics in an αvβ6⁺ pancreatic tumor model (44) and in a humanized single-chain Fv antibody B6-2 (45). αvβ6 is emerging as a promising target for a range of advances therapies, including those based on chimeric antigen receptor (CAR) T-cell immunotherapies (reviewed in ref. 46), where efficacy in the αvβ6 expressing SKOV3 cell lines has been demonstrated. Furthermore, the αvβ6-blocking antibody 264RAD showed promising in vivo efficacy in HER2⁺/αvβ6⁺ breast cancers in combination with the monoclonal antibody trastuzumab (47) and is being developed for phase I clinical trials. αvβ6 therefore represents a highly appealing target for cancer treatment across a range of technologies and therapeutic applications.

In silico evaluation of the Ad5_NULL-A20 interface with ανβ6 by homology modeling (Fig. 1; Supplementary Fig. SI) predicts the Ad5_NULL-A20 fiber knob domain to form a low entropy interface with ανβ6. A20 possesses the putative RGD integrin interacting motif (48), but specificity to the β6 subunit is derived from the helical motif C-terminal of RGD. It is further stabilized by electrostatic interactions across the interface and polar bonds between αν and the Ad5 CD loop. Each fiber trimer possesses three copies of the A20 peptide, with 12 trimeric fibers per adenovirus capsid, thus Ad5_NULL-A20 possesses 36 potential ανβ6 interaction sites per viral particle. While not all these sites will be utilized in a single cellular interaction, it is extremely likely that the virus benefits from a potent avidity effect when interacting with a cell possessing multiple ανβ6 copies.

In the present study, we presented the Ad5_NULL-A20 as a highly selective vector platform. A replication-defective form of Ad5_NULL-A20 vector successfully detargeted viral uptake by cells via native viral uptake pathways (Fig. 2), instead selectively retargeting αvβ6⁺ cells, in vitro and ex vivo (Fig. 3). Although the efficacy-limiting interactions that occur in the systemic delivery of adenoviral vectors can, theoretically, be bypassed by intracavity administration of the vector via the i.p. route, in practice this approach presents challenges because wild-type Ad5 is sequestered by preexisting anti-Ad5 immunity in the form of neutralizing antibodies (nAb) in ascitic fluid (41, 49, 50). We therefore assessed the transduction efficiency of Ad5_NULL-A20 in the presence of freshly isolated clinical OAS from patients with ovarian cancer with confirmed high levels of anti-Ad5 nAbs (Fig. 4A). Unlike the Ad5 vector, Ad5_NULL-A20 retained its ability to transduce αvβ6⁺ cells, even at relatively high OAS concentrations (Fig. 4C).

Clinical efficacy of therapeutic Ad5 vectors with unmodified capsids is also significantly limited by off-target tissue sequestration, particularly in the liver. We demonstrate that Ad5_NULL-A20 significantly altered the biodistribution of the Ad5 vector in vivo by reducing the sequestration in remarkable magnitudes. In tumor-free mice, replication-deficient Ad5_NULL-A20 demonstrated significantly reduced viral transgene expression the liver, spleen, and lungs compared with the parental Ad5 (Fig. 5), and lower viral genome copy-number in all off-target organs relative to the Ad5 vector (Fig. 6).

To test the efficacy of an oncolytic form of our detargeted/retargeted Ad5_NULL-A20 vector, we established an orthotopic i.p. xenograft model of human EOC SKOV3 in immunocompromised mice. The more localized biodistribution of virally encoded transgene expression of oncolytic Ad5_NULL-A20 following intraperitoneal administration was consistent with reduced off-target sequestration and/or tumor-selective virus uptake (Fig. 7B–E). This was supported by the superior survival of animals treated with Ad5_NULL-A20 relative to Ad5 in a SKOV3 xenograft model (Fig. 7E), although extended survival (compared with unmodified Ad5) was not observed in mice treated with the oncolytic Ad5.A20 variant. This observation highlights that efficacy in vivo depends upon both the combination of complete ablation of all native means of cellular uptake via hCAR, αvβ3/5 integrins and FX, coupled with an efficient and selective retargeting mechanism to tumor-associated ligands, such as the αvβ6: A20 receptor:ligand interaction. This observation likely explains previous studies (51, 52), which described no improved efficacy (compared with oncolytic Ad5) of virotherapies targeted to αvβ6 integrin, because the vectors used in those studies lacked modifications in at least two of the three native infectious pathways (the hexon: FX and penton base: αvβ3/5 interactions). Additional studies will be needed to fully evaluate αvβ6⁺ cancer retargeting in vivo, as well as to dissect the fate in tissues and immunologic responses to the Ad5_NULL-A20 vector.

Local, i.p. Ad5_NULL-A20 administration presents a promising treatment option for advanced, chemotherapy-resistant, αvβ6⁺ ovarian cancer. Here, we describe a novel vector that can be further manipulated for various clinical applications, with the scope of selective targeting to αvβ6 integrin-expressing cells and minimal off-target effects that limit current Ad5-based therapies. Ad5_NULL-A20 vector provides an agile and versatile platform that could ultimately be modified for precision virotherapy applications by various innovative approaches, potentially providing a platform for the local, tumor-selective overexpression of additional, virally encoding therapeutic modalities, such as immunotherapies.

No potential conflicts of interest were disclosed.

Conception and design: R. Jones, R.G. Vile, A.L. Parker

Development of methodology: H. Uusi-Kerttula, J.A. Davies, P. Wongthida, K.G. Shim, A.T. Baker, A.L. Parker

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Uusi-Kerttula, J.A. Davies, J.M. Thompson, P. Wongthida, L. Evgin, K.G. Shim, A. Bradshaw, A.T. Baker, R. Jones, L. Hanna, E. Hudson, R.G. Vile

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Uusi-Kerttula, J.A. Davies, A. Bradshaw, A.T. Baker, P.J. Rizkallah, R.G. Vile, J.D. Chester, A.L. Parker

Writing, review, and/or revision of the manuscript: H. Uusi-Kerttula, J.A. Davies, P. Wongthida, A.T. Baker, R. Jones, L. Hanna, E. Hudson, J.D. Chester, A.L. Parker

Study supervision: A.L. Parker

We are most grateful to patients at Velindre Cancer Centre, Cardiff, UK, who donated ascites samples. We would like to thank Mrs. Dawn Roberts for technical assistance, Dr. Richard Stanton for his expertise in AdZ recombineering, Dr. Alexander Greenshields-Watson for his assistance with homology modeling and structural biology, the team who maintain the YASARA energy minimization server, Dr. Edward Wang for his guidance in immunohistochemistry, Dr. Lisa Spary and clinicians at Velindre Cancer Centre for access to clinical ascites samples, Mr. William Matchett for his help with the IVIS 200 software, and Prof. Gavin Wilkinson, Dr. Michael Barry, and Dr. John Marshall for insightful discussions.

H. Uusi-Kerttula was supported by a Cancer Research Wales PhD studentship to A.L. Parker, J.A. Davies is supported by a Cancer Research UK Biotherapeutics Drug Discovery Project Award to A.L. Parker (project reference C52915/A23946). A.T. Baker is supported by a Tenovus Cancer Care PhD studentship to A.L. Parker (project reference PhD2015/L13). A.L. Parker is funded by Higher Education Funding Council for Wales.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.