An Oncolytic Virus Expressing a T-cell Engager Simultaneously Targets Cancer and Immunosuppressive Stromal Cells

Cancer-associated fibroblasts (CAF) facilitate invasion (1), coordinate angiogenesis (2), and maintain an immunosuppressive microenvironment in solid carcinomas (3). Their immunomodulatory functions include production of indoleamine 2,3-dioxygenase (IDO) and regulatory cytokines such as VEGF, FGF, IL10, and TGFβ (4–8). Notably secreted TGFβ can accumulate in the stromal matrix, exerting a powerful immunosuppressive effect on newly infiltrating naïve immune cells (9, 10), while CAF-produced CXCL12 can block entry of CD8⁺ cells into the tumor and attract regulatory T cells, inhibiting effector T-cell proliferation (11, 12).

CAFs are pivotal to tumor immunology, making it difficult to envisage cancer immunotherapy achieving its full potential without addressing their deleterious effects. CAF depletion can reverse local immune suppression and improve tumor immunotherapy. Genetic-based CAF depletion in an autochthonous pancreatic cancer model uncovered the ability of anti-PD-L1 to inhibit tumor growth and improve survival (13). While such an approach has a strong therapeutic rationale, implementation can be difficult due to the lack of unique target antigens on the CAF surface, with most of their known surface markers also present on normal fibroblasts.

One promising target antigen is fibroblast activation protein (FAP), which is upregulated on CAFs across a broad range of solid malignancies (14) but also found on normal fibroblasts in connective tissue in the muscle, gall bladder, bladder, and bone marrow stromal cells (BMSC; ref. 15). Elimination of FAP-positive cells with mAbs or FAP-targeted CAR-T cells demonstrated the potential to reverse tumor-associated immune suppression, particularly when combined with immunotherapeutic strategies such as cancer vaccines (16, 17). However, FAP expression on extratumoral cells is concerning, with previous FAP-targeting preclinical studies showing extensive bone marrow toxicity and cachexia that would caution against clinical development of systemic FAP-targeted treatments (15, 18).

Bispecific T-cell engagers (BiTE) show powerful targeted killing of cancer cells, but can also be deployed against stromal targets such as CAFs. BiTEs crosslink T cells (via CD3ϵ) to antigen-positive target cells, independent of HLA presentation, and can activate any T-cell to engage with and destroy adjacent target cells (19). Moreover, BiTE-mediated T-cell activation can overcome elements of tumor-associated immunosuppression that limit physiologic immune responses, leading to reactivation and proliferation of exhausted tumor-specific T cells (20–22). BiTEs targeted to a CAF marker such as FAP could be a potent strategy to activate intratumoral T cells to attack and deplete CAFs. However, systemic delivery of the BiTE would likely mediate significant toxicity by activating circulating T cells to attack normal fibroblasts and BMSCs. Accordingly, this potentially powerful approach is frustrated by challenges of site-specific delivery.

With their ability to encode and specifically express biologics in disseminated tumors, oncolytic viruses (OV) are an ideal solution. One promising candidate is enadenotucirev (EnAd), which has demonstrated good blood stability and systemic bioavailability in several early-phase clinical trials (23–25). An encoded FAP-specific BiTE would be produced and secreted only upon virus infection of tumor cells, allowing it to access tumor-infiltrating lymphocytes (TIL). This approach has been validated using BiTEs to target T-cell cytotoxicity to tumor cell antigens (21, 26–28). However, a virus-encoded BiTE that activates T cells to kill tumor stromal fibroblasts would provide a “multimodal” therapeutic agent that simultaneously targets two distinct cell types within the tumor. Alongside direct OV-mediated cytolysis of tumor cells, which is often proinflammatory (29), secretion of FAP-specific BiTEs should activate TILs to attack and deplete CAFs, acting to reverse CAF-induced immune suppression. This approach combines direct cytotoxicity, immune stimulation, and reversal of local immunosuppression, thereby transforming an immunologically inactive “cold” tumor into one that is “hot,” that is, with greater immune infiltration, yielding an integrated and more effective immunotherapeutic response.

DLD, SKOV3, A549, HEK293A (ATCC), and normal human dermal fibroblast (NHDF; Lonza) cells were cultured in DMEM (Sigma-Aldrich). Chinese Hamster Ovary (CHO, ATCC), normal human bronchial epithelial (NHBE) cells (Lonza) were cultured in RPMI1640 (Sigma-Aldrich). All cells were authenticated by short tandem repeat profiling (CRUK Cambridge Institute, United Kingdom) and routinely tested each month for Mycoplasma (MycoAlert Mycoplasma Detection Kit, Lonza). Cell lines were passaged no more than ten passages after thawing before use in experiments. Growth medium was supplemented with 10% (v/v) FBS (Thermo Fisher Scientific). Cells were incubated at 37°C and 5% CO₂. A FAP-expressing stable CHO cell line was generated using the FAP gene sequence (ID: 1149, NCBI) as described previously (21).

A FAP-targeted BiTE was produced by joining the DNA encoding two single-chain antibody fragments (scFvs) recognizing human FAP and CD3ϵ with a sequence encoding a flexible glycine-serine (GS) linker. An N-terminal immunoglobulin signal sequence for mammalian secretion and C-terminal decahistidine tag for detection were added. DNA sequences were synthesized and inserted into a CMV promoter–driven expression vector (pSF-CMV-Amp; Oxford Genetics) by standard cloning techniques. Recombinant BiTE protein was produced by transfecting HEK293A cells with polyethylenimine [PEI, linear, MW 25000, Polysciences; DNA:PEI ratio of 1:2 (w/w)]. Cells maintained in serum-free DMEM. Supernatants were harvested, concentrated 50-fold using 10,000 MWCO Amicon Ultra-15 Filter Units (Millipore), and stored at −80°C. BiTE protein concentration was determined by dot blot using decahistidine-tagged cathepsin D (BioLegend) as a standard. Specific binding of the FAP BiTE to recombinant FAP protein was confirmed by ELISA (data not shown).

Modified EnAd were produced by direct insertion of the BiTE cassette into the parental EnAd cloning plasmid pEnAd2.4 using Gibson assembly (30, 31). Additional viruses with FAP BiTE expression linked to red fluorescent protein (RFP) via a P2A site were also generated. Plasmid DNA was linearized by restriction digest with AscI (New England Biolabs) and transfected into HEK293A cells for virus production in DMEM (2% FBS). Upon extensive plaque formation, cells were harvested, and virus released by three freeze–thaw cycles. Single clones were selected by serial dilution and amplified by serial infection, followed by double CsCl banding to produce concentrated virus stocks. Stocks were titered by the Quant-iT Picogreen dsDNA assay (Thermo Fisher Scientific) and infectious dose determined by serial titration on A549 cells.

PBMCs were isolated from leukocyte cones (NHS Blood and Transplant, UK) by density-gradient centrifugation. CD3⁺ cells were extracted by depleting non-CD3 cells using the Pan T-cell Isolation Kit (Miltenyi Biotec). For CD4⁺ and CD8⁺ cells, CD4⁺ Microbeads were used (Miltenyi Biotec). Primary human malignant ascites samples and human prostate tissue samples were obtained from the Churchill Hospital (Oxford University Hospitals NHS Foundation Trust) following written informed patient consent and approval by the institutional review board and research ethics committee of the Oxford Centre for Histopathology Research (Reference 09/H0606/5+5) in accordance to the UK Human Tissue Act 2004 and the Declaration of Helsinki. For ascites, samples were immediately processed with cells and fluid separated by centrifugation (300 × g), with the cellular fractions treated with red blood cell lysis buffer (Qiagen). For ex vivo T-cell activation and cytotoxicity, cells were used immediately, or adherent cells were expanded by serial passage. For human prostate tissue specimens, tissue was transported in RPMI and stored on ice until slicing within two hours of surgery. Tissue cores were embedded in UltraPure low melting-point agarose (4% w/v, Thermo Fisher Scientific), and 300-μm tissue slices were prepared using a vibratome (Leica VT 1200S, Leica Microsystems). Each ex vivo tissue slice was transferred to a 0.6 cm² PTFE insert (Millipore) in 24-well plates containing 1 mL of cultivation media for prostate tissue (Supplementary Material). After overnight culture, the media were replaced, and tissue slices were treated with BiTE or recombinant virus. On day 0, 4, and 7 postinfection, 30% of the supernatant was collected, frozen, and replaced. On day 7, slices were fixed in paraformaldehyde (4%) and embedded in paraffin for IHC.

For in vitro coculture studies, PBMCs were seeded with the appropriate target cells (E:T ratio, 5:1) in flat-bottom 96-well plates in 100-μL medium. Target cell lines were prepared with cell dissociation buffer to preserve cell surface antigens. For ex vivo experiments, unpurified total cells from bone marrow or ascites were seeded in culture medium or fluid from the same exudate sample, respectively. To assess T-cell activation by virus-infected cells or BiTE-containing supernatants (300 ng/mL), cocultures were treated with 100-μL supernatant or infected with 100 vp/cell in 100-μL medium. Where appropriate, CD3/CD28 Dynabeads (Thermo Fisher Scientific) were included as positive controls for T-cell activation. T cells were harvested by pooling the culture media and a subsequent PBS wash. If adherent cells are also required, cell dissociation buffer was used to detach from plate surface and cells were pooled with nonadherent cells.

T-cell activation was measured by staining for surface expression of activation markers (CD69, CD25) and analyzed by flow cytometry. To study T-cell proliferation, T cells were labeled with 5 μmol/L carboxyfluorescein succinimidyl ester (CFSE) dye (Thermo Fisher Scientific) prior to culturing with target cells. After five days, T cells were harvested and analyzed by flow cytometry. As a surrogate for proliferation in mixed cell populations (e.g., whole ascites samples), total T-cell number per well was determined using precision counting beads (BioLegend). To measure T-cell degranulation, the externalization of CD107a was assessed by adding a CD107a antibody directly to the well at the start of the experiment. After 1-hour incubation, GolgiStop (6 μg/mL, BD Biosciences) was added, followed by flow cytometry analysis after an additional five hours. IL2 and IFNγ quantities were measured using the Human IL-2 ELISA MAX kit (BioLegend) or Human IFNγ ELISA MAX Kit (BioLegend). A flow cytometric multiplex bead immunoassay was performed using LEGENDplex Th Kit (BioLegend).

To assess target cell cytotoxicity by free BiTE or virus, release of LDH into the supernatant (CytoTox 96 Non-Radioactive Cytotoxicity Assay; Promega) or MTS viability assay (CellTiter 96 Cell Proliferation Assay, Promega) were used. To determine viability of specific cell types, total cells were harvested by cell dissociation buffer, and residual number of viable target cells measured by flow cytometry using an amine-reactive fluorescence live–dead stain. For observation of cell viability in real-time, xCELLigence technology (Acea Biosciences) was used. TGFβ and VEGF quantities were measured using TGF beta-1 Human/Mouse ELISA Kit (Thermo Fisher Scientific) and LEGENDplex Growth Factor Kit (BioLegend), respectively.

To classify different cellular populations, antibodies specific for CD11b (ICRF44), EpCAM (9C4), FAP (427819, R&D Systems, USA), CD3 (HIT3a), CD4 (OKT4), CD8a (HIT8) were used. To analyze T-cell populations, the following antigens were used: CD69 (FN50), CD25 (BC96), IFNγ (4S.B3), CD107a (H4A3), PD1 (H4A3). To analyze macrophage populations, cells were treated with Fc receptor block (Miltenyi Biotec) and stained with CD163 (GHI/61), CD206 (15-2), CD64 (10.1), and CD86 (IT2.2). The appropriate isotype control antibody was used in each case. All antibodies were acquired from BioLegend unless stated otherwise. Analysis was performed on a FACSCalibur flow cytometer (BD Biosciences) and data processed with FlowJo v10.0.7r2 software (TreeStar Inc.).

Automated staining was carried out with the Leica BOND-MAX autostainer (Leica Microsystems). Antigen retrieval was performed at 100°C using Epitope Retrieval Solution 2 (Leica Biosystems), followed by incubation with antibodies for CD8 (Agilent Technologies), CD25 (Atlas Antibodies), EpCAM (BioLegend), FAP (R&D Systems) or adenoviral hexon (Millipore). Detection was performed using the BOND Polymer Refine Detection System (Leica Biosystems). Alternatively, for the FAP primary antibody only, anti-sheep HRP-DAB Staining Kit (R&D Systems) was used. Sections were incubated with hematoxylin and imaged (Aperio CS2 slice scanner, Leica Microsystems).

Adenovirus genomes were measured by qPCR using primers and probe against hexon (primers: 5′-TACATGCACATCGCCGGA-3′/5′-C GGGCGAACTGCACCA-3′, probe: 5′-FAM-CCGGACTCAGGTACTCCGAAGCATCCT-TAMRA-3′). At the specific timepoint, total cell and supernatants were harvested, and DNA extracted (PureLink Genomic DNA Mini Kit; Thermo Fisher Scientific). In brief, primers and probe were mixed with DNA samples and added to QPCRBIO Probe Mix Hi-Rox (PCR Biosystems) Master Mix. To measure levels of FAP mRNA in ascites, reverse transcription qPCR was performed. The total cell fraction was harvested after 72 hours of treatment, RNA was extracted (RNAqueous-Micro Total RNA Isolation Kit; Thermo Fisher Scientific), and cDNA prepared (Superscript III First-Strand Synthesis SuperMix; Thermo Fisher Scientific). FAP expression was quantified using FAP-specific primers (5′-TCAGTGTGAGTG CTCTCATTGTAT-3′/5′-GCTGTGCTTGCCTTATTGGT-3′) and 2xqPCRBIO SyGreen Blue Mix Hi-ROX Master Mix (PCR Biosystems). Expression of the 18S gene was also measured as a normalization control (5′-GCCCGAAGCGTTTACTTTGA-3′/5′-TCCATTAT TCCTAGCTGCGGTATC-3′). All qPCR was run on ABI PRISM 7000 (Applied Biosystems).

Gene expression analysis was performed using the nCounter PanCancer Immune Profiling Panel (NanoString Technologies). The nSolver Advanced Analysis module was used for data analysis, in accordance with NanoString guidelines. Background thresholding was performed, followed by normalization of the data via the mean of the internal NanoString positive controls, and differential expression determined, with reference to uninfected cells. A gene set's directed global significance score for a covariate measures the cumulative evidence for the up- or downregulation of genes in a pathway and is calculated as the square root of the mean squared t-statistic of genes, with t-statistics generated from the linear regression algorithm within the nSolver Advanced Analysis module.

Brightfield and fluorescence images were obtained on a Zeiss Axiovert 25 microscope and captured with a Nikon DS5M camera. For time-lapse sequences, images were obtained on a Nikon TE 2000-E Eclipse inverted microscope and captured with a Hamamatsu Orca-ER C4742-95 using MetaMorph imaging software. Images were collected at 15-minute intervals with videos (12 frames/second) generated using ImageJ software (NIH, Bethesda, MD). Where appropriate, cells were stained with CellTracker Orange CMTMR Dye (Thermo Fisher Scientific) and CellTrace Violet Cell Proliferation Kit (Thermo Fisher Scientific). Apoptosis was visualized using CellEvent Caspase 3/7 Detection Reagent (Thermo Fisher Scientific).

Where experiments produced two datasets, significance was evaluated using a Student two-tailed t test. In all cases of more than two experimental conditions, statistical analysis was performed using a one-way ANOVA test with Tukey post hoc analysis or two-way ANOVA test using Bonferroni post hoc analysis. All data are presented as mean ± SD. Significance levels used were P = 0.01–0.05 (*), 0.001–0.01 (**), 0.0001–0.001 (***). Experiments were performed in biological triplicate, unless stated otherwise.

A FAP-targeted BiTE was engineered recognizing human FAP and CD3ϵ. A BiTE specific for CD3ϵ and an irrelevant antigen, filamentous hemagglutinin adhesin (FHA) of Bordetella pertussis, was used to control for unspecific binding. An N-terminal immunoglobulin signal sequence and C-terminal decahistidine tag were added for mammalian secretion and detection. BiTE protein production was assessed by transfection of HEK293A cells.

To confirm specificity of the FAP-BiTE for surface FAP, we established a human FAP-positive stable cell line using FAP-negative CHO cells. Peripheral blood mononuclear cell (PBMC)-derived T cells were activated 24 hours after coculturing with CHO-FAP cells and FAP BiTE–containing supernatants from transfected HEK293A (Fig. 1A) and mediated CHO-FAP cell lysis (Fig. 1B). Neither T-cell activation nor lysis were observed in cultures with parental CHO cells or control-BiTE, indicating that surface FAP expression is required for T-cell activation, presumably via surface CD3 clustering and pseudoimmunologic synapse formation.

FAP-BiTE–induced T-cell activation was also evaluated in coculture with NHDFs, which express surface FAP when cultured in high serum (10% FBS, Supplementary Fig. S1A). Incubation of NHDF- and PBMC-derived T cells from six donors with FAP-BiTE-containing supernatants for 24 hours induced significant T-cell activation and NHDF lysis, with the control-BiTE having no effect (Fig. 1C; EC₅₀, 2.5 ng/mL). CFSE-labeled PBMC T cells cocultured with NHDF and FAP-BiTE underwent multiple rounds of T-cell proliferation (Fig. 1D) and showed at least 10-fold increase in IFNγ, IL2, TNFα, IL17F, IL22, and IL10 (Fig. 1E), with IFNγ production 10-fold higher than that induced by physiologic anti-CD3/CD28 activation beads (Fig. 1F). Interestingly, FAP-BiTE induced activation and degranulation of CD4 and CD8 T cells, directing both subsets to kill NHDF cells with similar potency (Fig. 1G–I). Importantly, no induction of activation markers, proliferation and cytokines was observed with control-BiTE or in the absence of NHDF target cells, confirming that CD3 clustering is essential for T-cell activation.

EnAd is a conditionally replicating chimeric group B adenovirus generated by bioselection (Fig. 2A; ref. 32). The FAP-BiTE and control-BiTE sequences were inserted downstream of the fiber gene under transcriptional control of either an exogenous CMV promoter or a splice acceptor (SA) site for the adenoviral major late promoter (MLP). The former drives immediate transgene expression upon successful cell infection, whereas MLP-driven expression occurs only in cells permissive to virus replication, such as human tumor cells. Viruses were rescued and purified from HEK293A cells (Supplementary Table S1).

Colorectal adenocarcinoma (DLD) cells were infected at 100 vp/cell of the parental or recombinant viruses to assess replication kinetics. Viral genome copies reached between 3–6 × 10¹² genomes/mL (Fig. 2B), indicating that BiTE expression did not impair replication relative to the parental virus. Cytotoxicity of all recombinant viruses was also comparable with parental EnAd (Fig. 2C). Therefore, modification of the viral genome to incorporate the BiTE transgene had little effect on viral replication or oncolytic activity.

FAP-BiTE secretion by virus-infected HEK293A cells was demonstrated by immunoblotting of the supernatant (Supplementary Fig. S1B). Functionality of these secreted virus-encoded BiTEs was assessed by adding supernatants to cocultures of PBMC-derived CD3⁺ T cells and either CHO or CHO-FAP cells. T cells cocultured with CHO-FAP cells showed strong CD25 induction and target cell lysis when incubated with supernatants from EnAd-CMV-FAP-BiTE– or EnAd-SA-FAP-BiTE–infected cells (Fig. 2D and E). No activation or cytotoxicity was observed with supernatants from cells infected with unmodified or control-BiTE–expressing EnAd, in the presence of parental CHO cells, or in the absence of T-cells. FAP-BiTE yield from DLD cells infected with modified viruses was measured by comparing T-cell–mediated NHDF cytotoxicity induced by 72-hour–infected DLD supernatants to a standard curve. Following a 24-hour cytotoxicity assay, we measured FAP-BiTE at 9.8 and 49.2 μg/10⁶ cells after 72 hours for EnAd‐CMV-FAP-BiTE and EnAd‐SA-FAP-BiTE, respectively. This was consistent with previous reports, suggesting that while transcriptional initiation is delayed, there is superior total transgene expression when driven by the endogenous MLP compared with the CMV promoter (30). The FAP-BiTE showed impressive potency, with cytotoxicity detectable in supernatants diluted 10,000-fold (Supplementary Fig. S1C).

EnAd kills carcinoma cell lines by direct oncolysis (33), but does not effectively replicate in, or directly kill, fibroblasts or other nonepithelial stromal cells (23). However, FAP-targeted BiTE from infected tumor cells should allow T-cell activation and mediate targeted killing of FAP-expressing stromal fibroblasts. Cocultures of fibroblasts, moderately permissive SKOV3 ovarian carcinoma cells killed by EnAd 5–7 days postinfection (acting as BiTE producers), and PBMC-derived T cells were measured in real-time by cell index, a unitless measure of cell viability (Fig. 2F). In the absence of T cells, tumor cells and fibroblasts cells persisted for 100–120 hours, independent of virus infection. In the presence of T cells, FAP-BiTE expression from infected SKOV3 cells led to complete NHDF cytotoxicity, with lysis observed within 22 hours postinfection (hpi) by EnAd-CMV-FAP-BiTE and 42 hpi for EnAd-SA-FAP-BiTE. Crucially, no cytotoxicity was observed in cultures infected with EnAd or EnAd expressing the control-BiTE.

NHDF lysis was confirmed by lactate dehydrogenase (LDH) release in similar coculture experiments (Supplementary Fig. S1D). The kinetics of T-cell activation paralleled that of NHDF cytotoxicity (Fig. 2G; Supplementary Fig. S1E). Importantly, FAP-BiTE–encoded viruses failed to induce CD25 expression in the absence of NHDF, further demonstrating the requirement of FAP⁺ cells for BiTE-mediated T-cell activation (Supplementary Fig. S1F).

BiTE-induced cytotoxicity of stromal fibroblasts by T cells was observed by time-lapse microscopy using cocultures of T cells, fibroblasts, and DLD cells, which are more susceptible than SKOV3 cells to EnAd-mediated lysis (Fig. 2H; Supplementary Movie S1–S3). While infection with EnAd induced dramatic DLD killing within 48 hours, NHDFs remained viable throughout. In contrast, EnAd-CMV-FAP-BiTE infection induced both direct DLD killing and T-cell–mediated fibroblast cytotoxicity. Quantification of DLD and NHDF cells in parallel cultures showed complete elimination of both cell types upon treatment with EnAd-CMV-FAP-BiTE or EnAd-SA-FAP-BiTE 72 hpi (Supplementary Fig. S1G).

Conventional FAP-targeted therapeutics given intravenously are reported to induce FAP⁺ cell toxicity within the bone marrow compartment (18). Coupling BiTE expression to virus replication via the viral MLP restricts expression to the tumor compartment, minimizing unwanted toxicity to FAP⁺ fibroblasts in normal physiologic sites. To compare selectivity of virally encoded CMV- and MLP-driven BiTE expression, NHDFs were incubated with EnAd, EnAd-CMV-FAP-BiTE or EnAd-SA-FAP-BiTE in the presence of primary T cells only. At 72 hpi with EnAd-CMV-FAP-BiTE, we observed cytotoxicity in 80% of NHDF cells (Fig. 3A). No lysis was observed in EnAd-SA-FAP-BiTE–infected cells, consistent with the inability of EnAd to complete its life cycle in nonepithelial tumor cells (23).

To better simulate the multiple cell types present in a healthy tissue, NHDFs were cultured with exogenous PBMC-derived T cells and either NHBE cells or SKOV3 and subsequently infected with EnAd-SA-FAP-BiTE or EnAd-SA-control-BiTE. While EnAd-SA-FAP-BiTE infection of SKOV3 allowed T-cell activation and target cell lysis (Fig. 3B–D), NHBE cells did not.

Finally, for maximum clinical relevance, EnAd-SA-FAP-BiTE activity was evaluated in three fresh human whole bone marrow samples from healthy donors. Despite literature reports of bone marrow toxicity of FAP-targeted antibodies, no FAP⁺ cells were detected in any samples. Accordingly, FAP⁺ NHDF ‘target’ cells were added prior to infection to determine whether our armed viruses triggered uncontrolled toxicity against FAP⁺ cells. Neither EnAd-SA-FAP-BiTE nor EnAd-SA-control-BiTE induced endogenous T-cell activation (Fig. 3E) or targeted cytotoxicity (Fig. 3F) in the absence of tumor cells. The addition of SKOV3 cells led to T-cell activation and cytotoxicity following EnAd-SA-FAP-BiTE infection, with the latter thought to reflect predominantly BiTE-mediated T-cell lysis of FAP⁺ NHDFs. These data confirm that MLP-driven FAP-BiTE production is restricted to tumor cells and suggest there should be no systemic toxicity against FAP⁺ cells within normal bone marrow.

Malignant peritoneal ascites are frequent in several advanced carcinoma types, including ovarian, pancreatic, breast, and lung cancers (34), and often associated with a poor prognosis (34, 35). The fluid is routinely drained from some patients as a palliative treatment, providing a convenient and informative liquid biopsy. There is mounting evidence that malignant ascites are sites of substantial immunosuppression (36). We therefore assessed the effects of cell-free ascites fluid on PBMC-derived T-cell activation using anti-CD3/CD28 beads and FAP-BiTE. Bead-mediated T-cell activation was significantly inhibited by 3 of 5 ascites samples (Fig. 4A). In contrast, FAP-BiTE-mediated T-cell activation was not suppressed by any ascites fluids compared with levels observed in normal serum (Fig. 4B).

Human ascites biopsy samples typically contain tumor cells, fibroblasts, lymphocytes, and macrophages, representing a unique tumor-like model system to assess endogenous tumor-associated T-cell activation. Figure 4C shows the cellular composition of a representative sample containing CD3⁺, EpCAM⁺, CD11b⁺, and FAP⁺ cells (see also Supplementary Fig. S2A and S2B; Supplementary Table S2). Ascites-associated CD3⁺ T cells were, on average, 63% (up to 92.5%) positive for the exhaustion marker PD1 compared with only 10%–20% of PBMC-derived T cells (Fig. 4D). We assessed the ability of BiTE-encoding EnAd to infect ascites cancer cells and secrete sufficient amounts of BiTE, leading to endogenous T-cell activation and killing of autologous cancer-associated fibroblasts within an ascites sample. Total ascites cells from four patient biopsies were incubated in 50% ascites fluid with free FAP-BiTE or EnAd encoding FAP-BiTE. After five days, endogenous T cells were strongly activated in all ascites biopsy samples (30%–80% of total CD3⁺ cells; Fig. 4E), combined with CD3⁺ T-cell proliferation (Fig. 4F). Parental and control-BiTE viruses did not induce T-cell activation or proliferation.

Ascites T-cell activation and cytotoxicity toward endogenous FAP⁺ fibroblasts was assessed by measuring the change in FAP⁺ cell number during treatment (Fig. 4G). Free FAP-BiTE and EnAd-CMV-FAP-BiTE induced significant depletion of FAP⁺ fibroblasts in all samples, typically to levels below 1% of those in untreated or control samples, consistent with marked falls in FAP mRNA, VEGF secretion, and elimination of cells with a fibroblast-like morphology (Supplementary Fig. S3A–S3C). A similar trend was observed upon infection with EnAd-SA-FAP-BiTE, although one patient (patient biopsy 1) demonstrated neither T-cell activation nor fibroblast depletion (Fig. 4E and G). Infection of this sample with EnAd-SA-GFP also showed no GFP-positive cells (Supplementary Fig. S3D; Supplementary Table S2) and no EpCAM⁺ tumor cells at the outset (Supplementary Fig. S3E). The sample likely had insufficient tumor cells to support virus replication, demonstrating the strict necessity of tumor cells for virus replication and MLP-drive BiTE expression, suggesting one predictor for potency. A cytokine array of patient biopsy 1 demonstrated that EnAd-CMV-FAP-BiTE infection induced at least 10-fold increases in IL17A, IL17F, IL22, IFNγ, and IL10 expression (Fig. 4H). Parallel experiments using expanded mixed cultures of ascites-derived fibroblasts and tumor cells showed that fibroblast depletion led to 50%–70% lower TGFβ levels in supernatants (Fig. 4I), suggesting FAP⁺ cells to be a major source of immunosuppressive TGFβ within tumor ascites. Altogether, these data show that treatment of malignant ascites with free or virally encoded FAP-BiTE is able to polyclonally activate anergic T-cells, leading to targeted depletion of autologous tumor-associated fibroblasts.

To assess the impact of CAF depletion, T-cell activation and virolysis of cancer cells on the tumor microenvironment, we used NanoString to determine changes in the expression of 730 cancer and immune pathway genes in three primary ascites samples, selected to represent the spectrum of clinical possibilities. Biopsy 4 had a high ratio of FAP⁺ cells relative to EpCAM⁺ cells (likely epithelial cancer cells; Supplementary Table S3), while biopsy 5 had a similar proportion of EpCAM⁺ and FAP⁺ cells, and biopsy 6, a relatively low ratio of FAP⁺ to EpCAM⁺ cells. All samples had similar levels of CD3⁺ T cells and CD11b⁺ myeloid cells.

Significant T-cell activation was observed in all samples three days postinfection with EnAd-SA-FAP-BiTE, but not EnAd-SA-control-BiTE (Supplementary Fig. S4A). Approximately 40% of all genes showing significant changes in mRNA levels of at least two-fold (Supplementary Fig. S4B); only mRNA basally expressed above a minimum threshold level were included in the analysis. Considerably more genes showed changes following exposure to EnAd-SA-FAP-BiTE than EnAd-SA-control-BiTE, except biopsy 6, where the high number of EpCAM⁺ cancer cells may have resulted in extensive EnAd-specific gene changes due to BiTE-independent direct virolysis. Changes were grouped by immune response category as an average of three samples (Fig. 5A) or individual samples (Supplementary Fig. S4C). Although individual biopsies showed some variation, all EnAd-SA-FAP-BiTE–infected samples demonstrated increased gene expression in numerous immune groupings including cytotoxicity, pathogen defense, and T-cell, B-cell, and NK-cell function. While T-cell- and NK-cell–attractant chemokines (CXCL9, CXCL10, CXCL11) were also upregulated in all biopsies, strong decreases in fibroblast-associated CXCL6, CXCL12, and CXCL14 induced a downregulation in overall chemokine expression for biopsy 4 (Supplementary Fig. S4D).

Figure 5B shows the dataset for biopsy 4, which had the highest number of FAP⁺ cells. The most highly upregulated genes (up to 100-fold) following treatment with EnAd-SA-FAP-BiTE included T-cell markers (GZMB, IFNG, IL2RA, PRF1, TNF). The greatest decreases (up to 1,000-fold) were in fibroblast-associated genes, such as COL3A1, FN1, THY1, CXCL12, and IL13RA2. Similar trends were seen across all three samples, with the most modest changes in fibroblast markers observed in biopsy 6, which had the lowest levels of FAP⁺ cells. Changes in expression of key genes compared with untreated samples are shown by individual biopsy in Fig. 5C–F. For example, expression of the fibroblast marker collagen type III (COL3A1) is dramatically reduced upon infection with EnAd-FAP-BiTE compared with EnAd-control-BiTE in all three samples, while THY1 and IL13RA2 (also used as fibroblast markers; Fig. 5C) showed FAP-BiTE–dependent decreases in two biopsies. Basal expression of these genes in biopsy 6 did not pass the minimum threshold for analysis. T-cell activation markers (GZMB, PRF1, and IL2RA; Fig. 5D), checkpoint markers (PDCD1, CTLA4, and LAG3; Supplementary Fig. S4E), T-cell–recruiting chemokine CXCL9, and DC maturation/antigen presentation markers LAMP3 and TAP1 (Fig. 5E) all increased in a FAP-BiTE–dependent manner. These latter findings are particularly encouraging, raising the possibility of immunosuppression reversal in the tumor microenvironment following EnAd-SA-FAP-BiTE infection.

Macrophages are known to show plasticity between proinflammatory “M1” and wound-healing “M2” phenotypes, with tumor-associated macrophages (TAM) usually skewed toward “M2.” To investigate the influence of FAP-BiTE on macrophage polarization, patient-derived malignant ascites samples were treated with free or EnAd-encoded FAP-BiTE to determine activation of endogenous T cells and CD11b⁺CD64⁺ cells. Treatment with free FAP-BiTE or EnAd-SA-FAP-BiTE induced strong T-cell activation and IFNγ secretion (Supplementary Fig. S4F). We observed simultaneous induction of an activated M1-like macrophage phenotype, manifested by strong decreases in CD206 and CD163 (Fig. 5F) and increased CD64 expression (Fig. 5G). Infection with EnAd-SA-FAP-BiTE induced a large increase in CD86 expression, while free FAP-BiTE, EnAd-SA-control-BiTE, or IFNγ alone had no effect (Fig. 5G).

We obtained seven fresh paired punch biopsies of malignant prostate tissue from patients undergoing radical prostatectomy, cut into thin sections for ex vivo cultures. Prostate tissue slices showed a characteristically complex architecture, with glandular structures of (malignant) EpCAM⁺ epithelial cells interspersed with large regions of intervening stroma containing scattered CD8 T cells (Supplementary Fig. S5A). FAP expression was generally weak in benign prostate tissue and high in malignant prostate tissue (Fig. 6A). Fibroblasts showing the strongest FAP expression were often adjacent to malignant epithelial cells (Supplementary Fig. S5A).

To facilitate assessment of virus activity, we developed reporter viruses linking FAP-BiTE and RFP expression (Supplementary Fig. S5B). Following infection with EnAd-SA-FAP-BiTE-RFP, malignant tissue slices showed RFP expression, demonstrating successful viral infection, replication and BiTE expression (Supplementary Fig. S5C). Positive staining for viral hexon confirmed EnAd replication, apparently limited to malignant epithelial cells (Fig. 6B). Malignant prostate tissue infected with EnAd-SA-FAP-BiTE showed an increase in activated endogenous TILs seven days postinfection (Fig. 6C). Slices from all patients showed significant induction of IFNγ production, with IL2 levels also increasing in four samples (Fig. 6D and E); both cytokines are associated with activated CD4 Th1 and CD8 cytotoxic T cells (37). Neither untreated slices nor those infected with EnAd-SA-control-BiTE-RFP showed detectable T-cell activation, although some samples demonstrated modest increases in IFNγ and IL2 following EnAd-SA-control-BiTE-RFP infection, likely a direct result of virolysis. In benign prostate tissue, there was little increase in IFNγ or IL2 with any treatment (Supplementary Fig. S5D and S5E).

BiTE-mediated activation of T cells is expected to lead to fibroblast killing. Cells were visualized undergoing apoptosis in real-time within ex vivo prostate tissues (Fig. 6F). EnAd-SA-FAP-BiTE-RFP–infected cells strongly associated with apoptotic nuclei, suggesting BiTE-mediated induction of proximal cytotoxicity of surrounding cells. FAP-BiTE–mediated cytotoxicity was observed in all patient biopsy samples, with intrinsic EnAd activity also inducing small increases, potentially due to a greater number of cancer cells in some samples (Fig. 6G). Indeed, regions of high T-cell activation showed an absence or degradation of surrounding tissue or stroma, with tissue architecture remaining intact in uninfected samples (Fig. 6H). Crucially, FAP^low benign prostate tissue showed negligible increases in cytotoxicity within the duration of the study (Supplementary Fig. S5F).

Here, we have developed an armed oncolytic adenovirus combining three distinct therapeutic strategies: direct virus-mediated cytotoxicity toward cancer cells, creation of a proinflammatory immune environment, and removal of a key stromal cell mediator of tumor immunosuppression. Encoding BiTEs within OVs exploits the strengths of both virotherapy and immunotherapy while overcoming limitations of each agent alone. When expressed locally, the short plasma half-lives of BiTEs will become advantageous, minimizing systemic exposure and avoiding “on-target, off-tumor” toxicities (38). Conversely, arming an OV with a BiTE provides an additional mechanism of cell killing and broadens the range of target cells to include OV-resistant stromal cells.

Combining OVs and BiTEs also has a potentially synergistic effect on infiltrating T-cells. Increased CD8⁺ T-cell infiltration into the tumor bed has been observed in several OV clinical trials, including studies using EnAd and FDA-approved Imlygic (24, 39), likely providing more effector cells for BiTE-mediated cytotoxicity. Simultaneously, BiTE-mediated redirection of TILs (potentially virus-specific) toward chosen targets may delay viral clearance and increase intratumoral spread.

For maximal translational relevance, our current and future studies will focus on primary human tumor biopsies maintained ex vivo, rather than compromise with imperfect murine models that may not provide the desired tumor heterogeneity or realistic levels of immune suppression (23, 40). Clinical samples retain the heterogeneous and multifaceted cellular interactions of advanced human cancer, and, in the case of organotypic prostate tumor slices, the stromal architecture and extracellular matrix of a solid tumor. Treatment of both solid and liquid tumor biopsies with EnAd-FAP-BiTE led to tumor-associated T-cell activation and destruction of endogenous FAP⁺ fibroblasts, alongside secretion of large quantities of proinflammatory cytokines and chemokines, including IFNγ, IL2, TNFα, IL17, and CXCL9. Crucially, this demonstrated that the patient's own tumor-associated T cells can be used for therapeutic purposes in the realistic environment of an advanced human tumor. It was particularly encouraging that T cells within all tested patient biopsies, shown to be PD1⁺ and likely anergic, were readily activated and rendered functional by the BiTE to mediate cytotoxicity. This may reflect the high level of activating stimuli each T-cell can receive using a BiTE, where in principle every CD3 can be crosslinked to the target antigen (>100,000 copies per cell) without being limited by the relatively small number of HLA-presented TCR-cognate peptides available, likely to be less than 100 per cell. Indeed, the efficacy of BiTE-mediated T-cell stimulation is augmented when targeting a high-density receptor like FAP, a result also seen with other antigens (21, 22).

NanoString analysis confirmed the extensive depletion of fibroblast-associated RNA in human malignant ascites samples treated with EnAd-FAP-BiTE, together with strong induction of genes involved in T-cell function. Despite their varying cellular compositions, similar immune-activating trends were seen in all samples following EnAd-FAP-BiTE infection, with stimulation of RNAs involved in leukocyte trafficking, dendritic cell maturation, and antigen presentation. Surface markers on ascites TAMs revealed a clear shift from an M2-like phenotype to one that is more proinflammatory. We expect that newly infiltrating monocytes, recruited by OV-mediated induction of monocyte-attractant chemokines, such as CCL2, CCL7, and RANTES, will acquire an M1 “activated” phenotype. Significantly, surface expression of costimulatory ligand CD86 on TAMs was only induced by the combination of virus and FAP-BiTE (EnAd-FAP-BiTE). We hypothesize that virus stimulation of pathogen-associated molecular patterns, IFN signaling, or STING and removal of CAF-mediated suppression are required for CD86 activation. These findings indicate that coupling CAF depletion with potent activatory stimuli (T-cell activation and viral-mediated immunogenic cell death) synergistically repolarize the tumor microenvironment toward promotion of an effective anticancer immune response (41). Similarly, although we expect that suppressive markers will increase in tandem with activation markers, this potential barrier to continued virus activity could be counteracted by combining OVs with checkpoint inhibitors.

Using OVs for cancer-targeted transgene expression has now been validated both preclinically and clinically (24, 42). Here, we regulated BiTE expression using the adenoviral MLP, limiting BiTE production to cells permissive to the virus life cycle. In the absence of cancer cells, we observed no BiTE production or cytotoxicity (Fig. 3). This is particularly important in light of the “on-target off-tumor” toxicities observed with FAP-targeted antibodies or CAR-T cells toward FAP⁺ bone marrow cells. Infection of primary cultures of freshly isolated human bone marrow by EnAd-SA-FAP-BiTE showed no T-cell activation or bone marrow cell toxicity in the absence of tumor cells. Endogenous FAP⁺ cells also likely occur at frequencies too low (∼0.01%) to identify in the mononuclear cell fraction. Hence, our elegant targeted expression strategy is expected to avoid such toxicities while exploiting the potent effects of the FAP-BiTE within the microenvironment of each tumor deposit.

We therefore believe that arming OVs to express BiTEs targeting stromal elements, such as CAFs, can provide a powerful new multimodal approach to cancer therapy. In this way, a single-agent actively kills two different cell types using two distinct, yet targeted, cytotoxic mechanisms. EnAd provides a particularly promising virus platform to achieve targeted BiTE expression in disseminated tumors, exploiting the blood stability and systemic bioavailability of the virus, which has been studied in several early phase clinical trials. This strategy to induce proinflammatory cell death while reversing TME-mediated immunosuppression may be what is ultimately required to turn intransigent, stromal-rich carcinomas into targets for a complete and durable immunotherapeutic response.

B.R. Champion is a chief scientific officer and has ownership interest (including stock, patents, etc.) in PsiOxus Therapeutics. L.W. Seymour reports receiving a commercial research grant, has ownership interest (including stock, patents, etc.), and is a consultant/advisory board member for PsiOxus Therapeutics. K.D. Fisher is a scientific advisor at PsiOxus Therapeutics and has ownership interest (including stock, patents, etc.) in PsiOxus Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Freedman, M.R. Duffy, J. Hagel, B.R. Champion, L.W. Seymour, K.D. Fisher

Development of methodology: J. Freedman, J. Hagel, R.J. Bryant

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Freedman, A. Muntzer, E.M. Scott, J. Hagel, L. Campo, R.J. Bryant, P. Miller

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Freedman, J. Lei-Rossmann, J. Hagel, L.W. Seymour

Writing, review, and/or revision of the manuscript: J. Freedman, J. Lei-Rossmann, A. Muntzer, R.J. Bryant, C. Verrill, B.R. Champion, L.W. Seymour, K.D. Fisher

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Lambert

Study supervision: J. Freedman, L.W. Seymour, K.D. Fisher

Other (sample processing): A. Muntzer

Other (provided primary tumor material): C. Verrill

The authors gratefully acknowledge support from the Medical Research Council (MRC-Oxford Doctoral Training Partnership, MR/K501256/1 to J.D. Freedman) and Cancer Research UK (grant #C552/A17720 to J. Lei-Rossmann, K. Fisher, L. Seymour; studentship C5255/A20936 to E.M. Scott). M.R. Duffy is funded by the Kay Kendall Leukaemia Fund (grant KKL1050). J. Lei-Rossmann is supported by Linacre College, Oxford. pEnAd2.4 was kindly provided by PsiOxus Therapeutics. We are grateful to Egon Jacobus (University of Oxford, Oxford, United Kingdom) for the use of his primers. Special thanks to Alison Carr and her team for their helpful collection of ascites. C. Verrill's research time is part-funded by the Oxford NIHR Biomedical Research Centre (Molecular Diagnostics Theme/Multimodal Pathology Subtheme). We acknowledge the contribution to this study made by the Oxford Centre for Histopathology Research and the Oxford Radcliffe Biobank, which are supported by the NIHR Oxford Biomedical Research Centre.

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.