Reducing off target viral delivery in ovarian cancer gene therapy using a protease-activated AAV2 vector platform

Abstract

Gene therapy is a promising strategy for treating metastatic epithelial ovarian cancer (EOC). However, efficient vector targeting to tumors is difficult and off-target effects can be severely detrimental. Most vector targeting approaches rely on surface receptors overexpressed on some subpopulation of cancer cells. Unfortunately, there is no universally expressed cell surface biomarker for tumor cells. As an alternative, we developed an adeno-associated virus (AAV) based “Provector” whose cellular transduction can be activated by extracellular proteases, such as matrix metalloproteinases (MMP) that are overexpressed in the tumor microenvironments of the most aggressive forms of EOC. In a non-tumor bearing mouse model, the Provector demonstrates efficient de-targeting of healthy tissues, especially the liver, where viral delivery is less than 1% of AAV2. In an orthotopic HeyA8 tumor model of EOC, the Provector maintains decreased off-target delivery in the liver and other tissues but with no loss in tumor delivery. Notably, approximately 10% of the injected Provector is still detected in the blood at 24h while more than 99% of injected AAV2 has been cleared from the blood by 1h. Furthermore, mouse serum raised against the Provector is 16-fold less able to neutralize Provector transduction compared to AAV2 serum neutralizing AAV2 transduction (1:200 vs 1:3200 serum dilution, respectively). Thus, the Provector appears to generate less neutralizing antibodies than AAV2. Importantly, serum against AAV2 does not neutralize the Provector as well as AAV2, suggesting that pre-existing antibodies against AAV2 would not negate the clinical application of Provectors. Taken together, we present an EOC gene delivery vector platform based on AAV with decreased off-target delivery without loss of on-target specificity, and greater immunological stealth over the traditional AAV2 gene delivery vector.

Graphical Abstract

Introduction

Epithelial ovarian cancer (EOC) is the most lethal gynecologic malignancy in industrialized countries.[1] Most women are not diagnosed until widespread intraperitoneal metastasis has occurred, complicating surgical approaches to treatment that result in poor survival rates of <27% after 5 years.[2,3] Despite decades of extensive effort, little advancement has been made in developing therapeutic alternatives to conventional chemotherapy and surgery.

Gene therapy is a promising therapeutic strategy with demonstrated ability to correct genetic defects,[4] kill tumor cells,[5] increase immunogenicity,[6] and enable molecular imaging.[7] Unfortunately, the lack of tight on-target specificities and high risk of off-target toxicity has limited the translation of cytotoxic gene therapy-based treatments. Some approaches have involved the regulation of gene expression specifically in tumor cells with tumor-specific promoters.[8,9] While safer, the delivery and sequestration of vectors in non-target tissues makes it difficult to achieve the effective therapeutic dose in tumors. Higher vector dosages are required, which could increase other problems, such as dose-dependent vector immunogenicity/cytotoxicity and more expensive treatments due to higher vector manufacturing costs. Therefore, programming the delivery vector itself to better distinguish between healthy and diseased cells would be highly beneficial.

Recent precision therapy approaches have attempted tumor delivery by targeting overexpressed receptors and signaling pathways in cancer cells.[10] However, due to a high degree of interpatient heterogeneity, and both inter- and intratumoral heterogeneity, no ubiquitous receptor or pathway exists in all tumor cells.[11,12] Furthermore, increasing evidence demonstrates non-malignant cell types, such as cancer associated fibroblasts (CAF), help EOC thrive and spread.[13] This means that multiple cell types may need to be ‘hit’ to achieve effective tumor clearance. Thus, using precision therapies that go after overexpressed tumor receptors and pathways will likely only target a subset of cells and have limited therapeutic efficacy.

An alternative approach may be to create therapeutics that use a tumor microenvironment-specific feature as the target biomarker.[12,14] Specifically, matrix metalloproteinases (MMPs) are overexpressed in metastatic EOC to help facilitate the degradation of mesothelium during tumor cell invasion.[15] As such, extracellular secretion of MMPs into the tumor microenvironment and stroma is a strong prognosticator of poorer overall survival.[16–19] Thus, exploiting MMP overexpression may be a reliable biomarker for EOC tumor targeting.

We previously developed an AAV-based gene delivery vector, called the Provector, which requires activation by extracellular MMPs in order to deliver transgenes into cells in vitro. [20,21] The AAV2-based Provector design involves a tetra-aspartic acid motif inserted after G586 in the capsid, adjacent to the receptor binding domain, and acts as a charge- and steric-based impedance to AAV2’s interaction with its cellular receptor,[22] heparan-sulfate proteoglycan (HSPG). Flanking this motif on both sides is an MMP-cleavable sequence, PLGLAR, which when recognized by MMPs, results in the peptide lock being removed from the capsid surface and allowing for virus transduction.[20–23] Here we demonstrate our Provector platform has improved EOC tumor specificity in an orthotopic ovarian cancer mouse model.

Methods

Plasmid Cloning

The scAAV2-CMV-GFP-(miR-1BS)₃ transgene plasmid was generated using the scAAV2-CMV-GFP plasmid and the rAAV9CBnLacZ-(miR-1BS)₃ plasmid that has three miR-1 binding sites (miR-1BS)₃.[24] Site-directed mutagenesis was used to insert a SpeI restriction site between the GFP stop codon and the SV40 polyadenylation signal of the scAAV2-CMV-GFP plasmid. PCR was used to amplify (miR-1BS)₃ from the rAAV9CBnLacZ-(miR-1BS)₃ plasmid while adding a NotI restriction site to the 5’ end of the amplicon and a SpeI restriction site to the 3’ end. The amplicon was ligated into the backbone followed by bacterial transformation into NEB10 competent cells. Plasmids were extracted and purified using the Zyppy Miniprep Kit (Zymo Research) then sequence verified. All Provector plasmids were generated as described previously.[20] Briefly, ‘peptide lock’ inserts were prepared for ligation insertion through the annealing and phosphorylation of synthesized oligos (see Table S1 for sequences, Integrated DNA Technologies). The Scrambled peptide sequence was generated by randomly scrambling the PLGLAR peptide cleavage sequence.

Cell Culture

Human Embryonic Kidney 293T (HEK293T) cells were used for AAV and lentivirus vector production, and HeyA8, HeyA8luc, HeyA8luc/RFP/MMP9, and OVCAR-8 were used to characterize in vitro transduction behavior. HEK293T and all HeyA8 variant cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Lonza) supplemented with 10% fetal bovine serum (FBS, Gemini Bio) and 1% penicillin and streptomycin (Life Technologies). OVCAR-8 cells were maintained in Roswell Park Memorial Institute Medium (RPMI, Lonza) supplemented with 15% FBS and 0.1% gentamicin sulfate (Gemini Bioproducts). Cells were grown as adherent cultures in 5% CO₂ at 37°C, subcultured after treatment with 0.25% trypsin−EDTA (Life Technologies) for 2−5 min at 37°C, and resuspended in complete medium.

HeyA8luc and HeyA8luc/RFP/MMP9 Cell Line Generation

Both human MMP9 and firefly luciferase lentiviral vectors were generated using a 3^rd generation lentiviral packaging system. MMP9/RFP lentiviral vector plasmid (pLenti-GIII-CMV-RFP-MMP9–2A-Puro) or luciferase lentiviral vector plasmid (pLenti-GIII-Luc) (10μg) (ABM Inc) and the packaging plasmids pMDLg/pRRE (5μg), pRSV/REV (2.5μg), and pMD2.G (3μg) were transfected into HEK293T cells using polyethylenimine (PEI). After 48 h, supernatants were collected, filtered, and stored at −80°C. Vectors were concentrated using a 50% polyethylene glycol (PEG) solution and a 30 min spin @ 3000rpm. The supernatant was aspirated and the pellet carefully resuspended in serum-free cell media. Lentivirus was added at increasing concentration to cells plated in 24-well plates (Greiner). For the HeyA8luc cell selection, puromycin was used to select for positively transduced cells. For HeyA8luc/RFP/MMP9 cell selection, lentivirus encoding MMP9-RFP was added to HeyA8luc cells and RFP+ cells were selected using cloning rings and expanded to enrich their populations. Several rounds of enrichment were performed followed by flow sorting based on RFP+ populations (BD FACSAria™ II).

Virus Production and Purification

To produce purified AAV vectors as previously described[25] the Provector rep/cap plasmids were used as packaging plasmids, along with a self-complementary GFP transgene (scAAV-GFP), with or without (miR-1BS)₃, and a helper plasmid (XX6–80). Briefly, linear PEI was used to triple transfect the Provector rep/cap plasmid (10μg), scAAV-GFP (10μg), and XX6–80 (20μg) into 90% confluent HEK293T cells on poly-L-lysine-coated plates. Media was replaced at 24 h post-transfection and at 48 h post-transfection cells were harvested and separated through an iodixanol gradient. Genomic titers were determined by qPCR using SYBR Green (Life Technologies) using primers against the CMV promoter (see Table S2 for primer sequences). Virus samples were concentrated and buffer exchanged through Amicon Ultra 4 (Millipore) columns into gradient buffer (GB, 50mM Tris, pH 7.6, 150mM NaCl, 10mM MgCl₂) with 0.001% Pluronic-F68 (PF68).

Proteolysis of Provectors

MMP-2, MMP-7, and MMP-9 proteases were purchased from Enzo Life Sciences. Protease activity was calibrated prior to each experiment to reduce variability due to enzyme storage, as previously described.[20] Briefly, the activity of 5nM MMP on 5mM of the fluorogenic substrate Mca-PLGL-Dpa-AR (Calbiochem) was measured using a Tecan M1000 plate reader in a buffer containing 50mM Tris, pH 7.4, 150mM NaCl, 5mM CaCl₂, and 0.005% Brij-35. The amount of enzyme added to each proteolysis was standardized to the initial reaction velocity as previously described.[20] The MMPs were diluted to 9x in MMP storage buffer (50mM Tris, pH 7.5, 1mM CaCl₂, 300mM NaCl, 5μM ZnCl₂, 0.1% Brij-35 and 15% glycerol). MMP was added (final concentration 317 nM) to purified virus samples [1.5×10¹² viral genomes VG/ml] diluted in GB-PF68. The reactions were incubated at 37°C for 20 h and stopped with EDTA (25mM final concentration).

Heparin Binding Assay

Viruses were tested for their ability to bind to heparin. AAV2, Provector, and Scrambled were treated with either the storage buffer (sham) or MMP9 as detailed above. 1×10⁸ VG of virus was then added to 100μL of heparin-agarose beads (Sigma Aldrich) for each condition. The viruses were incubated with the beads for 15 min at room temperature while rotating. After incubation, the beads were centrifuged for 5 min at 6,000 x g and any unbound virus was collected by removing the supernatant. The beads were then exposed to buffer solutions of increasing concentrations of NaCl (150mM, 500mM, 700mM and 1M) to elute bound virus. Eluted virus was collected from each salt wash and titered by qPCR.

Silver Staining

Virus samples (1.5 × 10⁹ VG) were denatured in NuPAGE LDS Sample Buffer, run on a 4–12% Bis-Tris NuPAGE gel for 2 h at 100V, and stained using the Silverquest staining kit (Life Technologies) according to the manufacturer’s instructions. Stained gels were imaged with an Epson scanner.

Transduction and Transfection Assays

For miRNA silencing assays, HEK293T cells at 70% confluency (approximately 3×10⁵ cells/well) were double-transfected with varying ratios (1:0, 1:3, 1:5, and 1:10) of scAAV2-CMV-GFP-(miR-1BS)₃ plasmid and the plasmid encoding miR-1 using linear PEI with nitrogen:phosphate ratio of 14:1. At 24 h post-transfection, GFP fluorescence was quantified via flow cytometry (FACSCanto II flow cytometer, BD Biosciences). For in vitro transduction assays characterizing Provector function, cells at 95% confluency were transduced at varying MOI in serum-free media. Media was replaced with complete serum-containing media 8–12 h post-transduction. At 48 h post-transduction, cells were harvested and GFP fluorescence was quantified via flow cytometry. Transduction index (TI) was calculated as %GFP-positive cells X geometric mean fluorescence intensity (gMFI).

BALB/c Biodistribution

Ten-week-old female BALB/c mice were injected intravenously through the tail vein with 1×10¹¹VG of AAV2, Provector, or Scrambled vectors. Three weeks following injection, mice were euthanized. Brain, heart, lung, liver, kidney, spleen, and muscle (obtained from quadricep and hamstring) were acquired and immediately flash frozen in liquid nitrogen. Tissues were kept frozen at −80°C until further downstream processing. All animal studies were reviewed, approved, and conducted in accordance with Rice University Institutional Animal Care and Use Committee whose animal research facilities uphold the terms of the Animal Welfare Act (AWA) and PHS Policy on Humane Care and Use of Laboratory Animals. IACUC approval number: 987597–9.

Tumor-Bearing Mouse Biodistribution

Ten-week-old female BALB/c nude mice were injected intraperitoneally with 5×10⁶ HeyA8luc or HeyA8luc/RFP/MMP9 cells. Tumor take was determined by imaging of luciferase activity on IVIS Kinetic III by intraperitoneal injection of D-luciferin potassium salt at 150mg/kg (Perkins Elmer). Approximately 1–1.5 weeks following IP tumor cell injection, tumor-bearing mice were injected with 1×10¹¹VG of AAV2, Provector, or Scrambled vectors. One or two weeks following virus injection, mice were euthanized and tissues harvested and stored as described above. For biodistribution analysis, one Scrambled data point was removed as an outlier using 3x SEM criteria.

RNA/DNA Extraction and Quantification

Mouse tissues were homogenized using the BeadBug™ microtube homogenizer (Benchmark Scientific). 10–25 mg of tissue was used for RNA and DNA extraction using RNeasy or DNeasy kits (Qiagen), respectively, according to the manufacturer’s instructions. For MMP9 mRNA detection, HeyA8 or HeyA8luc/RFP/MMP9 cells were seeded into PLL-coated 6-well plates (Greiner). At confluency, the cells were harvested and RNA was extracted using the RNeasy kit. To quantify mRNA expression, 1μg RNA per sample was reverse-transcribed to cDNA using the Verso cDNA synthesis kit (Thermo Fisher) according to manufacturer’s instructions using a 3:1 blend of random hexamers and anchored oligo-dT. Gene expression (GFP or MMP9) was then quantified by qPCR using SYBR Green and primers against GFP or MMP9 and the housekeeping gene 18s (see Table S2 for primer sequences) on a BIO-RAD CFX96. 250ng-1μg cDNA was added per well, and each sample run in triplicate. Transgene mRNA expression was normalized to 18s expression. To quantify viral genomes, 100ng isolated DNA was added per reaction and qPCR was performed using SYBR Green and primers against the CMV promoter. Each organ sample was run in triplicate.

Neutralizing Antibody Assay

Ten-week-old female BALB/c mice were injected intravenously through the tail vein with 1×10¹¹VG of AAV2, Provector, or Scrambled vectors. Three weeks post-injection, mice were euthanized and cardiac puncture was immediately performed to collect blood. Blood was incubated at room temperature for 20 min to allow for coagulation then centrifuged at 1000g for 20 min at 4°C to separate serum. Sera were stored at −20°C until further processing. HEK293T cells were seeded at 75,000 cell/well of a 96 well plate. The following day, serial dilutions of serum in serum-free media were prepared and vectors at MOI 1000 were added to a total volume of 50μl. Virus and serum were incubated on ice for 2 h and then added to cells. At 4 h post-transduction, virus and serum were removed and replaced with complete media. At 24 h post- transduction, cells were harvested for flow cytometry.

Blood-Half Life Assay

Ten-week-old female BALB/c mice were injected intravenously through the tail vein with 1×10¹¹VG of AAV2, Provector, or Scrambled vectors. Ten microliters of blood were collected at 5 min, 10 min, 20 min, 40 min, 1 h, 4 h, 24 h, 48 h, and 72 h post-injection through saphenous vein prick. Blood was combined with 10μl of 3.2% sodium citrate anticoagulant (Greiner Bio-One) and stored at −20°C until further processing. DNA isolation was performed followed by qPCR against the CMV promoter to quantify viral genome copy number.

Results

Provector Exhibits Protease-Activatable Behavior in Response to MMPs

The first design criterion for the protease-activatable AAV is that the inserted peptide lock is accessible to and cleavable by MMPs. We previously characterized genome packaging of Provectors by transmission electron microscopy and silver stain densitometry and did not observe any significant structural, packaging, or titer differences between Provector and AAV2 (Table S3).[20–22] Therefore, in order to evaluate viral protein (VP) cleavage by proteases, the viruses were exposed to MMP-2, −7, or −9 and were loaded based on qPCR titer to visualize cleavage fragments on a gel via silver stain. Similar to our previous reports, proteolysis of AAV2 demonstrates no cleavage, while exposure of Provector to MMPs results in a 17kDa c-terminal protein fragment - consistent with cleavage of the lock located at position G586 in all VPs (Figure 1A). [20–22] The Scrambled vector shows no VP cleavage as desired.

Figure 1. Provector is cleavable by MMPs and can transduce ovarian cancer cells in vitro.

(A) Silver stain of AAV2, Provector, and Scrambled vectors proteolyzed with MMP −2, −7, −9, or sham. AAV2 VP1, 2, 3 subunits are visible at expected sizes (87kDa, 73kDa, and 62kDa, respectively). Slight size shifts can be observed in sham-treated Provector and Scrambled vectors corresponding to peptide lock insertions in each VP subunit. Cleaved VP fragments are observable at approximately 66 kDa (VP1 N-terminus, VP1’), 50 kDa (VP2 N-terminus, VP2’), 44 kDA (VP3 N-terminus, VP3’), and 17kDa (shared C-terminal fragment, c-term). (B) Heparin binding affinity of vectors in response to MMP9 proteolysis. Vectors treated with MMP9 or sham were bound to heparin-coated beads and eluted with increasing concentrations of NaCl. qPCR was used to quantify the number of viral genomes in the various fractions and data was normalized to total virus eluted from the beads. Data reported are an average of three independent experiments performed in triplicate. Error bars are SEM. (C) Transduction efficiencies of viral vectors from flow cytometry measurements 48 h post-transduction following proteolysis with MMP-2, −9, or sham. HEK293T (MOI 500, N=4), HeyA8 (MOI 3000, N=3), and OvCar8 (MOI 3000, N=3) cell lines were tested. Transduction index (TI) is calculated by multiplying geometric mean fluorescence intensity (gMFI) by percent GFP-positive cells (%GFP+). Error bars are SEM, and multiple t-tests with a Holm-Sidak correction method determined significance. *p<0.05.

Ablation of receptor-binding is the second design criterion for creating a Provector specific to tissue with upregulated MMPs. AAV2, when exposed to MMP-9 or sham buffer, demonstrates tight binding to the heparin column, with less than 10% of virus coming out in the elution fraction and the majority of virus found in the 500nM NaCl wash fraction (Figure 1B). For AAV4, which does not bind to heparin, over 80% of virus is in the unbound fraction. The Provector shows weaker binding to heparin in the sham condition (highest elution in 300nM NaCl fraction), and tighter binding in the MMP-9 condition (highest elution in 500nM NaCl fraction). As expected, the Scrambled vector shows weak binding regardless of MMP9 exposure.

Finally, MMP-activatable transduction ability of the Provector was tested in cells. HEK293T, HeyA8, and OvCar8 cells were transduced with vectors incubated with either MMP-2, −9, or sham buffer. In all of the cell lines the Provector shows a marked reduction in transgene delivery in the sham condition (less than 1% of AAV2) (Figure 1C). Notably, in the MMP-2 and MMP-9 conditions, the transduction ability is rescued to about 10% of AAV2. Compared to the sham condition, protease-activation results in increased transduction by 10.4-, 12.7-, and 77-fold for HEK293T, HeyA8, and Ovcar8 cells, respectively. The Scrambled vector also shows reduction in transduction, which is not rescued when proteolyzed with MMPs. The in vitro data demonstrates protease-activatable receptor binding and cell transduction behaviors by the engineered Provector.

Intravenous Injection of Provector Leads to Lower Delivery in Off-Target Organs

We next investigated the biodistribution of the Provector as compared to AAV2 and Scrambled in both a healthy mouse model and in an orthotopic ovarian cancer mouse model created by IP injection of HeyA8luc cells. We incorporated miRNA-1 (miR-1) binding sites in the vector transgene cassette as done previously[24] in order to reduce the possibility of off-target heart expression (Supplemental Figure S1). AAV2 has the highest viral genome delivery in the liver of healthy mice (Figure 2A), consistent with previous literature.[26] Notably, there are fewer Provector viral genomes in all of the organs compared to AAV2, with statistically significant reductions in the liver, heart, and brain. The Scrambled vector also shows reduction of viral genomes in all organs except the lung and the muscle, but to a lesser extent than the Provector.

Figure 2. Provector demonstrates successful de-targeting of several tissues in both healthy non-tumor as well as tumor-bearing mice.

(A) Biodistribution of viral genomes from healthy female BALB/c mouse tissues. AAV vectors (1×10¹¹VG/mouse) were injected intravenously into the tail vein, and 3 weeks later the animals were sacrificed and organs harvested to quantify the number of viral genomes per nanogram of total DNA extracted. Several tissues demonstrate significantly reduced delivery by Provectors over AAV2, most markedly in the liver. (B) Biodistribution of viral genomes in orthotopic HeyA8luc tumor-bearing female BALB/c nude mice. IVIS imaging of luciferase activity was performed to confirm tumor take in the animals and virus injection immediately followed (approximately 1 week after cancer cell injection). AAV vectors (1×10¹¹VG/mouse) were injected intravenously into the tail vein, and 1 week later the animals were sacrificed and organs harvested. Greatest de-targeting of tissues by the Provector was observed in the liver. (A) and (B) Data shown are the means with error bars representing SEM of 5 mice per treatment group and multiple t-tests with a Sidak-Bonferroni correction method were used to determine significance. *p<0.05, ***p<0.001, ****p<0.0001 (C) Ratio of viral delivery in the tumor to delivery in the liver in the HeyA8luc tumor-bearing mouse model. Horizontal lines represent the experimental group average (AAV2=3.5, Provector=41.5, Scrambled=19.0).

In the orthotopic ovarian cancer model (Figure 2B), significantly fewer viral genomes of Provector and Scrambled vectors are delivered to the liver and brain. Interestingly, all vectors yield a similarly high level of viral genomes in the tumor tissue. We observed overall increases in total viral genome delivery in the tumor-bearing model compared with non-tumor bearing mice. This difference in viral genome delivery may be due in part to difference in mouse strains used (BALB/c for non-tumor bearing versus BALB/c nude for tumor bearing). Specifically, the antiviral immune responses in the BALB/c model may impact delivery levels. To further quantify the ability of the vectors to arrive at the on-target tissue, the tumor-to-liver ratio was calculated (Figure 2C). AAV2 achieves ~4-fold more VG/ng DNA in the tumor compared to the liver. Scrambled delivers 20-fold more to the tumor than the liver. Finally, the Provector demonstrates a 40-fold increase in VG/ng DNA in the tumor than the liver, with 2 of the 5 mice showing an increase of over 75-fold in the tumor-to-liver ratio.

Provector Shows Consistent Liver De-Targeting in Tumor Model with Even Higher MMP9 Expression

We next sought to evaluate any differences in delivery when a tumor cell line with higher MMP expression is used. Toward this goal, we generated a stable MMP-9 overexpressing cell line from the parent line, HeyA8luc. The stable HeyA8luc/RFP/MMP9 cells display a 1.7-fold increase in MMP-9 mRNA expression as quantified by qPCR (Figure 3A). The lentiviral vector used to make the stable cell line encoded the MMP-9 gene in tandem with a reporter RFP gene driven by the same promoter; therefore, RFP expression serves as an easy surrogate for MMP-9 expression. Flow cytometry measuring RFP expression in the stable cell line population demonstrates a distinct shift, suggesting increased MMP-9 expression (Figure 3B).

Figure 3. Provector demonstrates de-targeting of the liver in ovarian cancer model made to artificially overexpress MMP9.

(A) qPCR of MMP9 cDNA normalized to 18S cDNA generated from HeyA8luc parental cell line and HeyA8luc/RFP/MMP9 cell line. Statistical significance determined by an unpaired t-test with *p<0.05. (B) Flow cytometry data quantifying RFP expression (driven by the same promoter as MMP9) of both parental HeyA8luc and HeyA8luc/RFP/MMP9 cell lines. (C) Biodistribution of viral genomes in HeyA8luc/RFP/MMP9 tumor-bearing BALB/c nude female mice tissues (N=4 and 5, AAV2 and Provector, respectively). Approximately 1 week after cancer cell injection into the mice, 2×10¹¹VG/mouse was injected intravenously via the tail vein. Tissues were collected 2 weeks post virus injection. Compared to AAV2, decreased delivery of Provector viral genomes to the off-target liver is observed. Multiple t-tests with a Sidak-Bonferroni correction method were used to determine significance. *p<0.05. (D) Ratio of viral delivery in the tumor to delivery in the liver. Horizontal lines represent the experimental group average (AAV2=0.9, Provector=45.7).

In our HeyA8 MMP9-overexpressing tumor model, we increased viral doses to 2×10¹¹ (vg/mouse) in an attempt to see if the tumor:liver ratios could be improved. Furthermore, we extended the biodistribution experiment by 1 week (from 1 week in Figure 2B to 2 weeks in Figure 3C). With these experimental variables in mind, Figure 3 biodistribution data (overall genome delivery) is almost 1-log higher compared to data in Figure 2B but similarly efficient at tumor-targeting and liver de-targeting. Similar to the biodistribution in the previous HeyA8luc model (Figure 2B), in the HeyA8luc/RFP/MMP9 tumor model there is a significant reduction of Provector genomes in the liver, while both vectors show a comparable level of delivery in the tumor tissue (Figure 3C). AAV2 displays a tumor-to-liver ratio equal to 1 (Figure 3D). The Provector demonstrates an ~45-fold increase in delivery to the tumor than the liver, with 1 of the 5 mice showing an increase of 180-fold. Taken together, the increase in MMP-9 expression observed in the tumor cells in vitro did not yield a drastic shift in tumor targeting with our Provector in vivo. This result may suggest that maximal MMP-dependent activation of Provector has already been reached in the regular HeyA8 tumor bearing model. Furthermore, the result demonstrates that liver de-targeting efficiency in tumor bearing animals is consistent at two different Provector doses and is maintained for at least 2 weeks post vector injection.

Provector Demonstrates Increased Blood Circulation and Decreased Neutralizing Antibody Response

We next sought to characterize virus properties in vivo in relation to circulation time and immunogenicity compared to AAV2 vectors. Blood was collected at several time points following vector injection and viral genomes quantified. The Provector and Scrambled vectors demonstrate longer circulation times in mice compared to AAV2 (Figure 4A). At 4h post injection, we observe fewer than 1% of AAV2 genomes remaining in circulation, while greater than 20% of Provector and 10% of Scrambled remain (Figure 4B). These observations of prolonged circulation time are consistent with the results showing reduced Provector and Scrambled vector delivery into tissues as compared to AAV2.

Figure 4. Provector displays longer circulation time than AAV2.

(A) Blood was collected at 5 min, 10 min, 20 min, 40 min, 1 h, 4 h, 24 h, 48 h, and 72 h post-virus injection (1×10¹¹VG/mouse). The number of vector genomes in the serum was quantified via qPCR on isolated DNA. The data is represented as a fraction of the injection dose. (B) Percent of injected viral genomes remaining in blood circulation at 4h post-injection. Error bars for (A) and (B) are SEM, and one-way ANOVA determined significance. ***p<0.001, ****p<0.0001.

Immunogenicity of vectors was evaluated by in vitro neutralizing antibody (NAb) assay. Serum raised in vivo from intravenous injection of AAV2, Provector, and Scrambled vectors were serially diluted, incubated with AAV2, and AAV2 transduction efficiency was measured in vitro (Figure 5A). As expected, NAb generation against AAV2 is robust, yielding a neutralizing antibody titer of 1:3200. Interestingly, serum raised against Provector and Scrambled vectors are much less able to neutralize AAV2 in vitro compared to serum from AAV2 infected mice, with NAb titers of approximately 1:64. Similar NAb studies were conducted testing serum neutralization of Provector transduction in vitro (Figure 5B). Notably, antibodies raised against Provectors are much less efficient at neutralizing Provector transduction in vitro with titers of 1:200 compared to anti-AAV2 serum neutralizing AAV2 transduction (1:3200). Moreover, NAb responses from anti-AAV2 serum on Provector transduction (1:200) are similar to that of anti-Provector serum against Provector virus (1:200). Perhaps more importantly, anti-AAV2 serum is substantially less able to neutralize Provector transduction in vitro compared to AA2 transduction (Figure 5C), suggesting individuals with pre-existing antibodies against AAV2 may still benefit from Provector-based therapy.

Figure 5. In vitro neutralizing antibody assay demonstrates Provector is immunologically distinct from AAV2.

(A) AAV2 was incubated with serial dilutions of serum from AAV2, Provector, or Scrambled injected mice (BALB/c, N=5, 1×10¹¹VG/mouse) and used for transduction of HEK293T cells at MOI 1000 in vitro. Shown are the TI values (gMFI x %GFP+ cells) of AAV2 at various serum dilutions normalized to the TI of the no-serum positive control (N=3). Sera generated against Provector and Scrambled vectors weakly neutralize AAV2. (B) Provector was incubated with serum from AAV2 or Provector injected mice and used for transduction of HEK293T cells at MOI 1000 in vitro. Shown are the TI values of the Provector at various serum dilutions normalized to the TI of the no-serum positive control (N=3). The Provector is equally neutralized by both anti-AAV2 and anti-Provector serum. (C) Either AAV2 or Provector was incubated with anti-AAV2 serum and used to transduce cells at MOI 1000. Shown are the TI values at each serum dilution normalized to the TI of each virus’ respective no-serum positive control (N=3). Anti-AAV2 serum weakly neutralizes the Provector.

Discussion

Effective cancer gene therapy strategies depend upon both enhanced specificities of vector toward tumor cells and limited off-target delivery to minimize toxicity. One way this goal can be pursued is by restricting transgene expression with tissue- or tumor-specific promoters.[9,27] However, this method suffers from low delivery efficiency overall due to vector sequestration in off-target tissues. Therefore, we approached tumor-targeting from a capsid engineering perspective and programmed oncotropism into an otherwise broadly tropic gene therapy vector, AAV, in order to minimized virus sequestration in off-target tissues. In this study, we demonstrate effective oncotargeting of EOC tumors in vivo with an AAV-based Provector platform that becomes activated by MMPs found in the tumor microenvironment. Importantly, this approach is unlike other rationally-designed capsids where tumor-associated ligands are incorporated to re-direct tropism.[28,29] Such monotargeted strategies have demonstrated limited success due to high levels of tumor heterogeneity.[30] By designing our vector to engage extracellularly secreted biomarkers that are associated with the most aggressive forms of disease, this strategy may circumvent current therapeutic targeting challenges, such as tumor cell heterogeneity. While a handful of other groups have engineered murine leukemia virus, measles viruses, and retroviruses to be activated by proteases, our approach based on AAV is a potentially safer oncotargeting platform as it is non-replicative, has reduced potential for genome integration following transduction, and does not require cell division for long-term gene expression, a feature that may be desirable, for example, if long-term tumor growth monitoring was intended.[31–33] We generated our Provector platform by incorporating peptide locks in the capsid such that virus interaction with its native cellular receptor is restricted and transduction is ‘switched off’. In the presence of several MMPs, receptor binding is rescued, ‘switching on’ transduction of ovarian cancer cells in vitro.

Using a disease-relevant orthotopic model of EOC, we show the Provector mediates delivery to tumors with reduced off-target delivery in other tissues (Figure 2b, 3c). Of note, the in vitro transduction experiments show MMP-proteolyzed Provectors transduce cells at 10% efficiency of AAV2 (Figure 2). We have yet to understand the effects of the amino acids that remain on the virus capsid surface, or scars, after MMP cleavage on virus transduction. It is likely these amino acid scars can impact virus engagement of cellular receptors. Future studies will alter the amino acid sequence of the lock such that the impact of the residual scar on transduction efficiency can be further characterized. Therefore, Provector delivery to tumors in vivo at levels slightly higher than AAV2 could suggest increased tumor uptake (due to reduced virus sequestration in other tissues) by over 10-fold. Future experiments that correct for differences in transduction efficiency observed between AAV2 and Provectors and deliver equivalent infectious titers (i.e., inject 10-fold more Provector than AAV2) could potentially demonstrate even greater tumor delivery of the Provector over AAV2. However, it remains unclear whether 10% in vitro transduction efficiency is truly the maximum level of Provector activation attainable. It is possible greater levels of activation may be reached in vivo where Provectors are simultaneously exposed to a variety of MMPs that are each independently able to recognize the cleavage sites in the peptide locks. While our previous kinetic studies of MMP9 proteolysis of Provectors in vitro suggest that we reach near complete cleavage,[20] our threshold for detection of uncleaved subunits remains high.

Importantly, the Provector biodistribution data show vector delivery is decreased across several tissues in non-tumor bearing mice accompanied by prolonged blood-circulation over AAV2, further supporting the premise of decreased virus sequestration in off-target tissues in congruence with decreased receptor binding behavior observed in vitro. Provector de-targeting of off-target tissues was also observed in tumor-bearing animals - most distinctly in the liver where delivered viral genomes are roughly 1% of AAV2. Taken together, the data suggests Provector tumor targeting is the same or better than AAV2 and could likely be administered at similar doses to achieve comparable therapeutic outcomes but with decreased off-target delivery, particularly in the liver. Furthermore, administration of Provector at similar infectious titers to AAV2 might be used to achieve greater tumor delivery.

As a platform technology, Provectors can be easily made to package a diversity of transgenes for therapeutic applications in EOC ranging from long-term tumor imaging to immunotherapy. In particular, Provectors may be used to deliver cytotoxic transgenes, a strategy otherwise largely avoided due to the high likelihood of off-target toxicity. Several AAVs packaging suicide genes such as TRAIL or BikDD, driven by tumor-associated promoters, like hTERT, have shown effective tumor reduction in several cancer models including liver, lung, and colorectal cancers.[8,34,35] Such apoptosis inducing genes could be packaged into Provectors and evaluated for reduction in tumor burden either alone or in combination with tumor-associated promoters. Alternatively, herpes simplex virus-thymidine kinase (HSV-tk) could be packaged in Provectors and used in combination with ganciclovir to achieve restricted cytotoxic prodrug activation in tumor tissues, a system that has previously been demonstrated to work in EOC models but with high toxicity.[36],37] Therefore, Provector-based treatments may offer new avenues for cancer gene therapy.

Unexpectedly, we found the Scrambled vector is near equivalent at delivering to tissues as AAV2 in vivo despite the in vitro data showing locked behavior. Given that we successfully demonstrated that insertion of the peptide locks in the Scrambled vector blocks transduction in vitro, we postulate the Scrambled vector may be becoming unlocked and activated by other proteases in vivo. Several MMP family members exist in both secreted and membrane bound forms. However, our in vitro activation studies were limited to the use of MMPs 2, 7, or 9 and thus activation by other MMP family members cannot be excluded. Interestingly, we observed two distinct phenotypes with the Scrambled vector in the livers of tumor and non-tumor bearing mice, locked and unlocked respectively. This phenotypic difference in Scrambled behavior in the two models may provide insights into what is activating the Scrambled vector. Future work in proteomic profiling of secreted proteases from the livers of tumor-bearing and non-tumor bearing mice may help to elucidate potential proteolytic enzymes. While the Scrambled results differ from its desired phenotype (i.e. low delivery in all tissues), it demonstrates that the MMP-recognizable cleavage sequence is critical for the locking and de-targeting mechanism and the observed biodistribution profile of the Provector. However, given that we observed MMP-independent activation of the Scrambled in vivo, we cannot rule out the possibility of the Provector being activated by both MMP-dependent and MMP-independent processes as well. In future work, Scrambled vectors with different peptide lock sequences need to be designed and validated in vivo.

Since Provectors appeared to be unlocked in vivo we performed tumor biodistribution studies with HeyA8 cells overexpressing MMP9 (HeyA8luc-MMP9) to see if tumor targeting could be enhanced. We observed AAV2-like levels of viral genome delivery to the tumor and significantly reduced viral delivery compared to AAV2 in the liver. Of note, the reduction in the liver is less drastic than that of the parental cell line biodistribution data. This result may suggest vectors are re-entering circulation post-cleavage following the initial saturation of virus uptake in tumors. However, the lack of significant improvements in viral delivery to tumors may reflect a biologically inconsequential increase in MMP9 levels over the HeyA8luc parental line. QPCR of MMP9 mRNA transcript levels shows a modest 1.6-fold increase over the parental line. RFP quantification of HeyA8 MMP9 cells indirectly confirms increased protease expression as both RFP and MMP9 are driven by the same promoter. However, baseline levels of MMP9 protein expression in the HeyA8luc cells cannot be verified by RFP quantification. Furthermore, distinguishing between pro- and active MMPs would need to be quantified to measure effective differences in MMPs between HeyA8luc and HeyA8luc/MMP9 cell lines. Quantification of active MMP is complicated by extracellular release of the protein from cells, particularly in an in vivo context. Future studies could quantify active MMPs in vivo using fluorometric FRET-based MMP substrate cleavage assays in mice bearing HeyA8luc or HeyA8luc/MMP9 tumors.[38] However, these assays are subject to their own limitations in signal resolution since active MMPs may be secreted by other tissues.

Finally, the humoral immune responses against AAV vectors pose a significant barrier to widespread clinical application for gene therapy. NAbs against AAVs are highly prevalent across the human population, with anti-AAV2 serum NAbs being most common, at an estimated 72%.[39] Anti-AAV NAbs can reduce gene delivery efficiencies by orders of magnitude leading to poor biodistribution and limited efficacy.[40] We therefore investigated the NAb production profile of Provectors compared with AAV2. The results demonstrate serum from mice injected with either Provector or Scrambled vector are less able to neutralize AAV2 transduction than serum from AAV2 injected mice. This may suggest that NAbs from AAV2 and Provector injected mouse serum recognize different AAV2 capsid epitopes. More specifically, it may suggest serum NAbs are raised against the surface-exposed peptide locks of the Provector and thus unable to neutralize AAV2 virus. Alternatively, the low Provector NAb titer for AAV2 may also imply reduced immunogenicity against Provectors in vivo. Previous studies of NAb responses against AAV2 mapped several NAb epitopes near and at the 3-fold spikes and HSPG binding domain which are proximal to the peptide lock insertion site. [41,42]. Thus, it is plausible that insertion of peptide locks at this site interferes with antibody generation. Decreased immunogenicity of Provectors is further supported by the data showing that Provector serum neutralizes Provector transduction at titers of 1:200, a titer 16-fold lower than the AAV2 NAb titer against AAV2. These same Provector NAb assays revealed that AAV2 and Provector serum have similar recognition of the Provector. Given the stark differences in anti-AAV2 NAbs against AAV2 and Provector (1:3200 vs 1:200), these results imply that Provector lock insertion adjacent to the receptor binding domain interferes with anti-AAV2 NAb recognition of the 3-fold spike epitope. In fact, Huttner et al. identified residue 587, adjacent to our peptide lock insertion site at 586, as a potentially important site for human serum neutralization of wtAAV2.[43] Importantly, these findings suggest pre-existing NAbs against AAV2 may not negate the clinical application of Provectors. Specifically, the results from Figure 5B would imply that the efficacy of a single administration of Provector in patients with pre-existing NAbs against AAV2 would have similar gene delivery efficiency as a second administration of Provector in patients with no prior exposure to AAV2 virus. Future studies with Provectors in vivo could examine vector biodistribution in mice pre-immunized with AAV2.

Conclusions

We present a gene delivery vector platform designed to recognize MMPs for delivery to ovarian tumors. The Provector delivers to tumors at equivalent levels to AAV2 but with significantly decreased delivery to the liver. Promisingly, we demonstrate decreased neutralizing antibody responses against the Provector compared to AAV2 and reduced recognition of Provectors by anti-AAV2 serum. To evaluate the potential application of Provectors clinically, future investigation of Provectors will test repeated administration strategies, impact of anti-AAV2 pre-existing immunity against Provector performance, and therapeutic outcomes of delivering cytotoxic transgenes with the engineered platform.