Main

Metabolic coupling and paracrine communication between cancer cells and CAFs can determine the response of tumors to a variety of cancer therapeutics2,3,4,5,6,7,8. A number of virus based anticancer therapeutics9,10 are in clinical development, and we sought to determine whether CAFs impact the ability of OVs to infect and destroy cancer cells within the tumor bed.

The transition from a normal fibroblast (NF) to a CAF phenotype is an epigenetically regulated phenomenon that occurs following NF exposure in vitro to tumor-secreted factors such as TGF-β (refs. 11, 12, 13). When grown as monocultures, normal human GM38 foreskin fibroblasts were refractory to infection by OVs; however, they became sensitive to virus infection (Supplementary Fig. 1a–g) after culture in the presence of either cancer cell–conditioned medium or TGF-β1. This suggested that CAFs may have increased sensitivity to OV infection when compared to their normal counterparts. We treated CAFs and NFs (Supplementary Fig. 2a) isolated from two individuals with lung cancer with OVs. CAFs were significantly (P < 0.0005) more sensitive than NFs to infection with several OVs, including vaccinia virus (VV)14, vesicular stomatitis virus (VSVΔ51)15 and Maraba MG1 virus16 (Fig. 1a,b). This led us to hypothesize that normal stromal cells growing in the presence of tumor cells have an altered transcriptional response to virus infection. Gene expression profiling of NFs and CAFs revealed that in the absence of infection, CAFs exhibited reduced expression of genes involved in innate antiviral and type I interferon (IFN-I) signaling relative to NFs (Fig. 1c). After VSVΔ51 infection, we observed 63% induction of these repressed genes in CAFs to levels equivalent to, or greater than, those observed in NFs at baseline (Fig. 1d,e). We validated the expression of several antiviral response genes (DDX58, OAS1, RSAD2, BST2 and IFI44L) in two pairs of donor-matched NFs and CAFs (Fig. 1f,g). We also observed downregulation of RIG-I protein levels in CAFs and activated fibroblasts in the absence of OV infection (Supplementary Fig. 2a,b). Thus, we attribute the increased virus sensitivity of CAFs relative to NFs to baseline differences in the transcription of key antiviral response genes. Although following infection we observed induction of these genes in CAFs, rapidly replicating viruses were able to outpace the antiviral response. In contrast, we speculate that elevated baseline transcription of the same antiviral genes protected NFs from infection.

Figure 1: CAFs, but not NFs, are sensitive to virus-based therapy due to rewired antiviral networks.
figure 1

(a) Microscopy images from autologous NFs and CAFs infected with VSVΔ51-GFP, MG1-GFP or VV-GFP. Nuclei were stained with DAPI. Mean ± s.e.m. of GFP-positive cells for three independent fields are indicated. Scale bar, 100 μm. (b) VSVΔ51, MG1 and VV titers resulting from infection of two autologous NFs and CAFs (PFU/ml). (c) Gene ontology (GO) analysis of biological processes associated with genes repressed in CAFs versus NFs. Illustrated GO terms represent all those biological processes significantly different (hypergeometric test; P < 0.001) after correction for multiple hypothesis testing (Benjamini–Hochberg). (d) Fold change in gene expression as a function of previously documented induction by type I IFNs for genes repressed in CAFs but induced over threefold by VSVΔ51. (e) Fold change in gene expression as a function of previously documented induction by type I IFNs for genes repressed in CAFs but not induced by VSVΔ51. (f,g) Differential expression (P < 0.05 or 0.005 depending on the gene) of select antiviral genes before or after VSVΔ51 infection of CAFs and NFs derived from two individuals with lung cancer: individual 1 (f) and individual 2 (g). Error bars indicate s.d. among replicates (n = 3). All results shown are representative of three independent experiments (t-test, *P < 0.05, **P < 0.005, ***P < 0.0005).

Source data

We compared the infection by OVs of CAFs and cancer cells grown in co-culture systems or in isolation. We cultured human renal carcinoma cells expressing a red fluorescent protein (dsRED-786-0) alone or in the presence of CAFs and then infected the cultures with VSVΔ51-GFP. We observed both increased GFP expression (Fig. 2a) and augmented virus titers (Fig. 2b) from CAF–786-0 co-cultures compared to 786-0 cell or CAF monocultures. In co-cultures, we observed in a 15- and a 11-fold increase in VSV-infected 786-0 cells and CAFs, respectively, compared to infection in cells grown independently (Fig. 2c–f). We found similar results in 786-0 cells grown with CAF-like fetal fibroblasts such as WI-38 cells and human fetal pancreatic fibroblasts (PaCFs)17,18,19 (Supplementary Fig. 3a–e). Virus infection was also promoted by CAFs in human pancreatic (MiaPaca-2) and ovarian (OVCAR8) cancer cell lines (Supplementary Fig. 3f,g). The effect was not restricted to VSV, as VV replication in 786-0 cells was enhanced more than tenfold in the presence of CAFs (Supplementary Fig. 4a–c). We also found that MiaPaca-2 cells became more sensitive to OV infection when exposed to CAF-conditioned medium (Fig. 2g,h and Supplementary Fig. 5), indicating that a soluble factor specifically produced by CAFs mediates this effect. Moreover, whereas NF-conditioned medium was unable to enhance VSV growth in 786-0 cells, conditioned medium from TGF-β–activated NFs or CAF-like fetal fibroblasts also increased virus production in 786-0 cells (Fig. 2i and Supplementary Fig. 4d,e).

Figure 2: Paracrine factors secreted by cancer cells and CAFs enhance OV replication.
figure 2

(a) Representative microscopy images of monotypic cultures or co-cultures of dsRED-786-0 cells and CAFs derived from two individuals with lung cancer (P1 and P2) infected with VSVΔ51-GFP. Nuclei were stained with DAPI. Scale bar, 100 μm. (b) Virus titer in cell-associated supernatants from a. (cf) Cells were harvested and percentage of GFP-positive cells determined by flow cytometry. (c,d) Representative plots (c) and quantification (d) gated on dsRED-negative CAFs. (e,f) Representative plots (e) and quantification (f) gated on dsRED-786-0 cells. Histograms show mean ± s.e.m. (n = 3). (g) MiaPaca-2 cells were cultured alone or in combination with activated fibroblasts grown on transwell inserts. After 72 h, inserts were removed and MiaPaca-2 cells were infected with VSVΔ51-GFP. Representative GFP and bright-field (BF) images are shown. Fold increase in mean GFP fluorescence ± s.e.m. is indicated. (h) Quantification of infectious virus particles from g. (i) VSVΔ51 titers from 786-0 cells cultured with regular growth medium (GM) or in conditioned medium from normal GM38 fibroblasts (NFCM) or activated fibroblasts (NFACM). Error bars show ± s.e.m. of triplicate infections (*P < 0.05, **P < 0.005, ***P < 0.0005 by one-way analysis of variance). NS, not significant. All results shown are representative of three independent experiments.

Source data

To identify the secreted factors that mediate the virus sensitization of cancer cells, we tested several cytokines and growth factors (epidermal growth factor (EGF), fibroblast growth factor 1 (FGF1), FGF2, TGF-β1, TGF-β2, platelet-derived growth factor (PDGF), hepatocyte growth factor (HGF), stromal cell–derived growth factor 1 (SDF1), interleukin-6 (IL-6) and IL-8) known to be commonly overexpressed within the tumor milieu11,20,21,22,23,24,25. Although all the candidate cytokines tested were able to activate their corresponding signaling pathways in 786-0 cells (Supplementary Fig. 6a,d), only FGF2 was able to significantly (P < 0.05) enhance VSV replication (Supplementary Fig. 6b). Although TGF-β1 treatment of NFs could sensitize them to OV infection (Supplementary Fig. 1), it had no impact on the ability of 786-0 cells to support virus growth (Supplementary Fig. 6b,c). VSVΔ51 multistep growth curves in 768-0 cells showed that FGF2-dependent virus sensitization increased the rate of virus replication and the maximum virus yield per cell (Supplementary Fig. 7a–c). This effect was inhibited by FGF2-specific neutralizing antibodies in a dose-dependent manner (Supplementary Fig. 7d). The activity of FGF2 was independent of the OV tested (Fig. 3a,b), occurred in several different cancer cell lines (Fig. 3c) and was not a consequence of increased cell proliferation following growth factor stimulation (Supplementary Fig. 6e). FGF2 was unable to enhance VSVΔ51 replication in NFs but did enhance virus replication in activated fibroblasts (Fig. 3c). In both human-derived CAFs and those activated by TGF-β-mediated differentiation, FGF2 expression was higher than in NFs (Fig. 3d,e). siRNA-mediated FGF2 silencing in human activated fibroblasts (Supplementary Fig. 8a,b) markedly reduced VSVΔ51 infectivity and replication in CAF-cancer cell co-cultures (Fig. 3f,g and Supplementary Fig. 8c,d).

Figure 3: FGF2 induces replication of various clinically relevant OVs in a panel of cancer cells and activated fibroblasts.
figure 3

(a) Representative fluorescent images of 786-0 cells pretreated with FGF1 or FGF2 and subsequently infected with various OVs are shown. The cell nuclei were stained with DAPI. Mean percentage ± s.e.m. (n = 3) of infected cells is indicated. Scale bar, 100 μm. (b) Virus yields from 786-0 cells treated and infected as indicated in a. (c) The indicated human cancer cell lines and normal fibroblasts were treated with FGF1 or FGF2 prior to VSVΔ51 infection (multiplicity of infection (MOI) = 0.01). Means of virus yields from three independent experiments were calculated and plotted. Error bars represent ± s.e.m. (d,e) Immunoblot analyses showing expression of α-SMA, FGF2, and tubulin in human-derived NFs and CAFs (d) or in GM38 cells treated with TGF-β1 or vehicle control (e). Fold increase in total FGF2 expression; results are representative of three independent experiments. (f) Representative microscopy images from at least three independent experiments of mono- or co-cultures of siRNA scramble control (siSC)- or siRNA for FGF2 (siFGF2)-treated human pancreatic fibroblasts and dsRED-786-0 cells infected with VSVΔ51. Nuclei were stained with DAPI. Scale bar, 100 μm. (g) Mean ± s.e.m. (n = 3) of virus titers from mono- and co-cultures treated and infected as described in f. ((*P < 0.05, **P < 0.005, ***P < 0.0005, by t-test (g) or one-way analysis of variance (a and c)). NS, not significant.

Source data

Given that FGF2 enhanced the replication of a variety of OVs in cancer cells, we investigated its effect on antiviral programs. FGF2 specifically delayed the secretion of IFN-β by 786-0 cells (Supplementary Fig. 9a) in response to virus. In addition, we found that when pretreated with IFN-α, FGF2-treated cells were still substantially more sensitive to OV replication than PBS- or FGF1-treated cells (Supplementary Fig. 9b–g). Given the key role that the RIG-I double-stranded RNA helicase26,27,28 plays in the recognition of cytosolic viruses and the primary activation of IFN response pathways, we focused our attention on the impact of FGF2 treatment on the expression of this critical pattern recognition receptor. Cancer cells treated with FGF-2 or co-cultured with CAFs had reduced RIG-I expression and decreased induction of downstream antiviral proteins compared to untreated cells (Supplementary Fig. 10a–g).

To determine whether FGF2 has a role in sensitizing tumors to virus-based therapy in vivo, we monitored VSVΔ51-luciferase infection in mice bearing bilateral MiaPaca-2 tumors pretreated or not with intratumoral injections of recombinant FGF2 (Supplementary Fig. 11a). FGF2 pretreatment increased viral titers and transgene expression in MiaPaca-2 tumors (Fig. 4a,b). We observed similar results in OVCAR8 and 786-0 xenograft models (Supplementary Fig. 11b,c). Subcutaneous co-injection of MiaPaca-2 cells with pancreatic fibroblasts (PaCFs) into the flank of severe combined immunodeficiency (SCID) mice produced tumors with a high density of α-smooth muscle actin (α-SMA)- and FGF2-positive cells (Supplementary Fig. 12a). Co-injected tumors became highly sensitized to VSVΔ51 infection relative to control MiPaca-2 tumors (Supplementary Fig. 12c,d). This was not due to differential tumor growth, as at the time of virus treatment there was no significant (P = 0.72) size difference between MiaPaca-2 tumors enriched or not with PaCFs (Supplementary Fig. 12b). Immunohistochemical analysis using antibodies to visualize viral proteins, RIG-I, FGF2, and CAF markers demonstrated that VSVΔ51 infection occurred primarily in areas of the tumor that were enriched in fibroblasts (Supplementary Fig. 12e) and showed FGF2 overexpression and decreasedRIG-I protein expression (Fig. 4c). Infection of tumor explants ex vivo revealed that FGF2-mediated sensitization is not the result of improved virus delivery, nor of immune recruitment, but rather of impaired innate immune responses associated with RIG-I circuit inhibition (Supplementary Fig. 13).

Figure 4: FGF2 induces sensitivity to OV therapy in vivo.
figure 4

(a) VSVΔ51-luciferase was delivered intravenously into mice bearing MiaPaca-2 tumors previously supplemented with FGF2 or vehicle control. After 72 h, tumors were homogenized and titered (n = 8 per group). (b) Representative bioluminescence images of mice in a. (c) Immunostaining showing expression of FGF2, VSV and RIG-I proteins in tumor tissues from the mice in a. Scale bar, 250.4 μm. Arrows show co-distribution of VSVΔ51 with FGF2-rich areas. (d) Immunohistochemical assessment of endogenous FGF2 expression in tumor grafts derived from subjects with pancreatic cancer. Tumors were arbitrarily classified as low or high FGF2 protein expressors. Percentage of FGF2-positive cells is indicated. (e) Quantification of infectious particles from FGF2 low or high expression tumor cores (shown in d) infected with VSVΔ51 ex vivo. (f) VSVΔ51-luciferase was delivered intravenously into mice, bearing bilateral tumors derived from individuals with pancreatic cancer, that were previously supplemented with FGF2 or vehicle control. After 72 h, virus titers were assessed (n = 6 per group). (g) GM38, 786-0, and MiaPaca-2 cells were infected with MG1 or MG1-FGF2. At indicated time points, secreted virions were quantified (n = 3). (h) Quantification of MG1 or MG1-FGF2 infectious particles from four primary cultures (AF-P1 to AF-P4) derived from ascitic fluid from individuals with ovarian cancer. (i) A single dose of PBS, MG1 or MG1-FGF2 was delivered intravenously into mice bearing subcutaneous MiaPaca-2 tumors. Tumor size was regularly measured (n = 4 or 5 per group). All graphs show mean values ± s.e.m. of at least two independent samples (*P < 0.05, **P < 0.005, ***P < 0.0005, by t-test). NS, not significant.

Source data

To explore the relevance of these findings to the treatment of human tumors with OVs, we obtained tissue samples from individuals with pancreatic cancer and determined the correlation of VSVΔ51 replication following ex vivo infection with FGF2 expression within the tumor (Fig. 4d). Human samples with high tumor FGF2 expression were much more susceptible to VSVΔ51 infection than those with low expression, which suggests that FGF2 is a potential biomarker for OV infectivity (Fig. 4e). Consistent with this idea, two samples derived from subjects with pancreatic cancer expressing low or undetectable FGF2 protein levels (Supplementary Fig. 14) could be rendered sensitive to VSVΔ51 infection in vivo by pretreating the tumors with recombinant FGF2 (Fig. 4f).

Because locally administered FGF2 enhances OV tumor-specific replication, we hypothesized that inclusion of an FGF2 transgene in a viral genome would improve the replication and tumor-specific spread of the OV product. We engineered a Maraba MG1 virus encoding human FGF2 (MG1-FGF2, Supplementary Fig. 15a,b). We found that the MG1-FGF2 virus was a more potent killer of cancer cells than the parental MG1 virus, yet it did not replicate in or kill normal cells (Fig. 4g–i). As predicted, FGF2 expressed from MG1 dampened antiviral responses in infected cells (Supplementary Fig. 15c,d). We tested the therapeutic effect of intravenous MG1-FGF2 treatment of subcutaneous MiaPaca-2 tumors and found that although MG1 virus partially controlled tumor growth, the MG1-FGF2 virus was more effective at reducing tumor burden and caused complete tumor regression in some animals (Fig. 4i). We observed a similar therapeutic effect of MG1-FGF2 in an immunocompetent mouse renal cell carcinoma model (Supplementary Fig. 15e). We did not detect infectious particles in normal tissues of immunocompetent mice treated with MG1 or MG1-FGF2 viruses, demonstrating that the transgene-expressing virus maintains its tumor selectivity. We monitored animal weight following intravenous virus treatment and found that transient weight loss and recovery was indistinguishable between animals treated with the parental or FGF2-expressing virus (Supplementary Fig. 15f). FGF2 is a known proangiogenic cytokine, and thus it was possible that an FGF2-expressing OV could promote tumor metastasis. This was not the case, however, as neither intravenous administration of MG1-FGF2 nor intratumoral injection of recombinant FGF2 induced tumor metastasis to distal organs (Supplementary Fig. 15g).

Although previous studies have demonstrated that specific cell types in the tumor microenvironment, including immune and endothelial cells, are important in determining the success of OV therapies29,30,31, we show here, for what is to our knowledge the first time, that reciprocal communication between cancer cells and CAFs promotes OV growth and killing in both cell types. Although it is well known that cancer cells condition their local microenvironment to promote tumor survival, chemoresistance and escape from immune surveillance32, we discovered that interactions between cancer cells and CAFs promote OV infection and killing of 'normal' stromal cells. Contributions from other soluble factors within the tumor microenvironment cannot be excluded, but our findings suggest that TGF-β1 is a primary factor involved in promoting OV infection of CAFs. In turn, we observed that FGF2 secreted by CAFs is responsible for deregulating antiviral circuits within cancer cells, overall creating a niche composed of OV-sensitive tumor cells and CAFs.

Oncolytic viruses alone and in combination with immune checkpoint inhibitors are advancing through clinical trials33 and are poised to become an important component of the oncologist's armament. Our observation that overexpression of FGF2 within the tumor milieu may predict sensitivity to OV infection provides a candidate biomarker for the stratification of patients who are most likely to benefit from virus therapy. In addition, our demonstration that FGF2 enhances the growth of multiple OV types argues that clinical development of OV therapeutics expressing FGF2 merits serious consideration. Further studies are necessary to determine the safety of this approach and to elucidate the specific molecular mechanism by which FGF2 suppresses RIG-I and increases susceptibility of tumor cells to OV. These findings may be particularly relevant for pancreatic cancers, which have a preponderance of CAFs3,11,34,35 and which we have shown are readily sensitized by FGF2 stimulation.

Methods

Reagents.

General laboratory chemicals were purchased from Sigma Chemical Co. (St. Louis, MO). MiaPaca-2, OVCAR8, 786-0, Vero, U2OS, L929, RENCA, and WI38 cells were from the American Type Culture Collection (Manassas, VA). Human normal GM00038 foreskin fibroblasts (abbreviated as GM38) and human fetal pancreatic fibroblasts (PaCFs) were obtained from the Coriell Cell Repositories (Camden, NJ) and Vitro Biopharma (Golden, CO), respectively.

Cell culture and viruses.

786-0, OVCAR8, U2OS, Vero, PaCF, WI38 and CAFs were cultured in Dulbecco's minimal essential medium (DMEM) containing 10% FBS (FBS) and 1 mM HEPES. MiaPaca-2 cells were cultured in DMEM supplemented with 10% FBS and 2% horse serum. GM38 fibroblasts and subject-derived normal fibroblasts (NFs) were cultured in DMEM containing 2% FBS and 1 mM HEPES. Cells were incubated at 37 °C in a humidified atmosphere with 5% CO2 and 80% humidity. All cell lines were routinely tested for mycoplasma contamination. Engineering and rescuing of the interferon-sensitive version of VSV, termed VSVΔ51, has been described before15. The oncolytic version of Maraba virus, named MG1, was previously described16. To construct and rescue the MG1-FGF2 virus, the human FGF2 cDNA was PCR-amplified from the pWPI-spbFGF plasmid (a gift from J. Kiss and P. Salmon, Addgene plasmid #2962-76)37 to add MluI sites at both ends using forward (5′-TGCAACGCGTTGCATGAAAAAAAACTCATCAGCCATCATGCC-3′) and reverse (5′-TGCAACGCGTTCAGCTCTTAGCAGATTGGAAG-3′) primers and cloned at the gene junction between the G and the L protein into a modified pSEAP2 shuttle vector plasmid containing part of the MG1 genome. Subsequently, the shuttle vector was digested with KpnI and NheI, and the FGF2-containing fragment was subcloned into LC-Kan vector encoding the viral complementary DNA sequence. All constructs were verified by sequencing (StemCore laboratory, Ottawa Hospital Research Institute). To rescue MG1-FGF2 virus, A549 lung carcinoma cells were plated at 3.0 × 105 cells/well in 6-well plates and infected 24 h later with vaccinia virus (MOI = 10) expressing the T7 RNA polymerase. After 1.5 h, vaccinia virus was removed and cells were co-transfected with LC-KAN Maraba encoding FGF2 (2 μg) and pCI-Neo constructs encoding Maraba N (1 μg), P (1.25 μg), and L (0.25 μg) genes using Lipofectamine 2000 (5 μl per well) as per manufacturer's instructions. After 24 h, medium was replaced with DMEM containing 10% FBS. Forty-eight hours post-transfection, medium was collected, and 1 ml was used to infect Vero cells. Recombinant virus was filtered (0.2 μm) and subjected to three rounds of plaque purification on Vero cells, after which MG1-FGF2 production was scaled up for further testing and validation. HSV-1 N212 expressing GFP14,36 was a gift from Dr. Karen Mossman (McMaster University, Canada). The oncolytic version of vaccinia virus encoding GFP was previously described38. Reovirus was a gift from Dr. Patrick Lee (Dalhousie University, Halifax, Canada). All virus stocks were diluted with cell culture media and then added to cells that had been washed with phosphate-buffered saline (PBS). Cells were incubated with the virus inoculum for 1–3 h at 37 °C, after which time the inoculum was replaced with normal growth medium. Infected cultures were kept at 37 °C until experimental analysis.

Isolation of normal and cancer-associated fibroblasts from individuals with lung cancer.

Tumor and surrounding normal tissue specimens from two individuals with lung carcinoma (77 and 51 years old; both male) were surgically resected and placed in sterile DMEM medium supplemented with 10% FBS and 100 ng/ml of placenta growth hormone (complete media). Samples were then diced using sterile scalpel blades and transferred to a T25 flask with complete media. All tissues were obtained from individuals with non-small-cell lung carcinoma undergoing thoracotomy with approval of the University of California, Los Angeles Institutional Review Board and following written informed consent from the individuals. The cultures were incubated for two weeks at 37 °C with 5% CO2. Attached cells were serially passaged four additional times to obtain pure cultures of fibroblasts. Expression profiles for vimentin, laminin, fibronectin, esterase, factor 8, cytokeratin, and αSMA were determined to confirm purity of cultures.

Establishment of primary cultures from ascites fluids from women with ovarian cancer.

Ascites fluids were obtained from four women with ovarian cancer (adenocarcinomas, stage III–IV, 51 years or older) undergoing routine paracentesis with approval of the Ottawa Hospital Research Ethical Committee and with written consent from the individuals (Ottawa General Hospital, Canada). Ascitic fluid was centrifuged at 450 × g for 10 min. Primary ascites-derived cancer cells were cultured in RPMI-1640 containing 10% FBS, 7.5% autologous ascitic fluid and 2 mmol/L glutamine. Tumor cells were identified by positive staining of CA125 using flow cytometry and used at low passage numbers. The corresponding primary cultures were named AF-P1, AF-P2, AF-P3, and AF-P4, respectively.

Cancer cell and fibroblast co-cultures.

For direct in vitro co-culture, cancer cells were plated with fibroblasts at a 1:3 ratio in reduced-serum medium (2% FBS). The cultures were incubated for 3 days, after which cells were infected with the indicated OV for 24 h (rhabdoviruses) or 48 h (VV, HSV-1, and reovirus). For indirect co-culture, cancer cells were seeded in 6-well dishes and activated fibroblasts were grown in cell culture transwell inserts (0.4-μm pore, BD Falcon, San Jose, CA). Similarly, for conditioned media transfer experiments, cancer cells or fibroblasts were grown in reduced-serum medium (2% FBS) for 3 days. Importantly, all cell lines used in co-culture experiments were grown in minimal standard medium as indicated above to minimize confounding effects of supplements. The supernatants were then collected, filtered and applied immediately to the indicated cancer cells or fibroblasts and grown for 2 days before OV infection or viable cell counting by Trypan blue staining.

Cell treatment with various growth factors and cytokines.

786-0, OVCAR8, MiaPaca-2 or WI38 cells were grown until 80% confluent in DMEM containing 2% FBS and treated with either FGF1 (20 ng/ml), FGF2 (20 ng/ml), EGF (50 ng/ml), TGF-β1 (10 ng/ml), TGF-β2 (10 ng/ml), SDF1 (50 ng/ml), IL-6 (50 ng/ml), or IL8 (50 ng/ml). All growth factors and cytokines were obtained from R&D Systems (Minneapolis, MN). Twenty hours post-treatment, cells were infected with the indicated OV for 24 h (rhabdoviruses) or 48 h (VV, HSV-1 and reovirus).

FGF1 and FGF2 signaling was abrogated by pretreating cells with a FGF receptor 1 inhibitor, PD173074 (25 μM, Sigma-Aldrich). Similarly, FGF2 activity was diminished by adding a specific anti-FGF2 neutralizing antibody at the final concentration of 2 μg/ml to 50 μg/ml (AB-233-NA, R&D Systems, Minneapolis, MN).

Analysis of cell proliferation and cell death.

Cell viability was calculated by trypan blue exclusion assay using a ViCell automated viable cell analyzer (Beckman Coulter, Fullerton, CA). Cell death was measured with a cytotoxicity assay measuring lactate dehydrogenase (LDH) released as per the manufacturer's protocol (Promega, Madison, WI). All experimental conditions included three biological replicates, and all experiments were independently repeated at least three times and reported by mean values.

Enzyme-linked immunosorbent assay.

786-0 cells were pretreated with FGF2 (20 ng/ml) or vehicle control (PBS) 24 h before infection with VSVΔ51 (MOI = 0.01). Supernatants were collected at the indicated time points and secreted IFN-β quantified using a Verikine human IFN-β ELISA kit as per manufacturer's instructions (PBL Interferon Source, Piscataway, NJ).

siRNA transfections.

For FGF2-specific silencing, human WI38 or pancreatic fibroblasts were transfected with small interfering RNAs (10 μM) against FGF2 (siGENOME Smart pool for human FGF2 siRNA, Thermo Scientific) or with a non-targeting siRNA (siGENOME non-targeting control, Thermo Scientific). Transfections were carried out according to manufacturer's protocol (Lipofectamine RNAiMAX, Life Technologies). Interference of target gene expression was confirmed by immunoblot analysis using FGF2-specific antibodies (1:1,000 mouse anti-FGF2, Millipore, clone bFM2, Temecula, CA).

Plaque assays.

For rhabdoviruses. Vero cells (5 × 105 cells) were infected with serial dilutions of virus containing samples in 6-well dishes for 1 h. Cells were then washed and overlaid with warm 0.5% (w/v) agarose in culture medium and incubated for 1 d. Viral plaques were visualized by staining with 0.05% (w/v) crystal violet in 17% (v/v) methanol for 2 h at room temperature.

For vaccinia virus. U20S cells (7.5 × 105) were infected with serial dilutions of virus containing samples in 35-mm dishes for 2 h. Cells were then washed and overlaid with warm 1.2% (w/v) methylcellulose in culture medium and incubated for 3 d. Viral plaques were visualized by staining with 0.05% (w/v) crystal violet in 17% (v/v) methanol for 2 h at room temperature.

For reovirus. L929 cells (1 × 106) were infected with serial dilutions of virus-containing samples in 35-mm dishes for 3 h. Cells were then washed and overlaid with warm 1% (w/v) agar (Sigma) in culture medium and incubated for 3 d. Viral plaques were visualized by adding neutral red to 0.01% (w/v) final concentration.

For oncolytic HSV1. Samples containing HSV1 were serially diluted and titered on Vero cells as previously described14.

For virus titering, all experiments were performed blinded, and number of infectious virus particles was expressed as plaque-forming unit (PFU) per milliliter (ml) or per milligram of tissue (mg).

Measurement of virus infection by flow cytometry.

Monotypic cultures or co-cultures of dsRED 786-0 and fibroblasts were incubated for 3 days and then infected with VSVΔ51-GFP (MOI = 0.01, 24 h) or VV-GFP (MOI = 0.01, 48 h) before analysis by flow cytometry (Cyan ADP 9, Beckman Coulter, Fullerton, CA). For each sample, at least 10,000 events were acquired. dsRED fluorescence was detected through the FL-2 channel through a 585-nm filter, and eGFP fluorescence was measured using the FL-1 channel equipped with a 489-nm filter. Data were analyzed using Kaluza Flow Cytometry Analysis software (Beckman Coulter). The percentage of infected cancer cells was equated to the percentage of dsRED-positive cells that were also GFP positive, whereas the number of VSVΔ51-GFP– or VV-GFP–positive fibroblasts was equated to the percentage of dsRED-negative cells that were GFP positive. Results are representative of three independent experiments.

Indirect immunofluorescence microscopy.

Cells cultured on glass coverslips were infected with either VSVΔ51 (MOI = 0.01), MG1 (MOI = 0.01), oncolytic VV (MOI = 0.01), oncolytic HSV1 (MOI = 0.01) or Reovirus (MOI = 1). 24 or 48 h post-infection, cells were fixed in 4% paraformaldehyde for 20 min, followed by quenching with 50 mM ammonium chloride. Cell membranes were permeabilized with 0.2% Triton-X-100 for 5 min before incubation with primary and secondary antibodies. All the washes were done in PBS containing 0.1 mM CaCl2 and 1 mM MgCl2. VSV and MG1 proteins were detected with 1:1,000 rabbit anti-VSV39. Reovirus was detected using a rabbit anti-Reovirus T3 antibody (1:300, gift from Dr. Earl Brown, University of Ottawa, Canada). RIG-I and IRF1were detected using a mouse anti-RIG-I (1:250; ENZO Life Sciences, Alme-1) and a specific mouse monoclonal antibody for IRF1 (1:300; Abcam, ab26109), respectively. Primary antibodies were detected with Alexa Fluor 594 chicken anti-mouse (A-21201), Alexa Fluor 488 donkey anti-rabbit (A-21206), and/or Alexa Fluor 488 donkey anti-mouse (A-21202), or Alexa 594 goat anti-rabbit (A-11012) (Molecular Probes, Invitrogen, Carlsbad, CA). All secondary antibodies were diluted 1:500 with 1% (w/v) BSA/PBS. Coverslips were mounted onto microscope slides using ProLong Gold antifade reagent with 4′-6-Diamidino-2-phenylindole (Molecular probes, Invitrogen). Samples were then examined using a Zeiss Imager.M1 microscope or a Zeiss Axioskop2 microscope equipped with an AxioCam HRm digital camera (Zeiss, Germany). Image quantification was performed using ImageJ software (http://rsb.info.nih.gov/ij/index.html). For quantifications, data were averaged from a pool of three independent fields counting at least 100 nuclei per pool.

Immunohistochemistry.

Formalin-fixed, paraffin-embedded tumors were sectioned (8 μm) for immunohistochemistry. After rehydration, sections were treated with sodium citrate (pH: 6.0) for 17.5 min in a pressure cooker and cooled for antigen retrieval. Sections were then blocked with 10% goat serum for 1 h at room temperature and then incubated with rabbit antibodies to either human FGF2 (1:300, Santa Cruz Biotechnology, Dallas, TX), VSV39, alpha smooth muscle actin (1:500, Abcam, ab5694, Cambridge, MA) or RIG-I (1:200, ENZO Life Sciences, Alme-1, Farmingdale, NY) overnight at 4 °C. Dilution and specificity of each antibody were confirmed by omission of the primary antibody. The secondary antibodies used were biotinylated goat antibody to rabbit or mouse IgG (1:300; Vector Laboratories, Burlingame, CA). Sections were briefly counter stained with hematoxylin. Quantitative analysis of immunostaining was done with the Aperio ImageScope software (Leica Biosystems).

Immunoblot analysis.

Cells (2 × 105) in 35-mm culture dishes were infected with the indicated OV and/or treated with indicated cytokine and then incubated for 24 h at 37 °C before lysis. Cells were lysed in 1% NP-40, 150 mM NaCl, 2 mM EDTA, 50 mM Tris, pH 7.4 containing Complete EDTA-free protease inhibitors (Roche). Cell lysates were clarified by centrifugation at 10,000 × g for 10 min at 4 °C. Secreted FGF2 was recovered from conditioned media using StrataClean Resin (Agilent Technologies, Santa Clara, CA). Briefly, conditioned media was incubated for 2 h with beads (10% v/v) and then pelleted by centrifugation at 500g for 2 min at 4 °C. Bound proteins were eluted by heating at 95 °C for 5 min in sample buffer. Proteins were then separated by SDS-PAGE and transferred to polyvinylidene fluoride (PVDF) membranes (Immobilon-P Millipore, Bedford, MA). Membranes were incubated for 1 h at room temperature with the following antibodies and dilutions: 1:3,000 rabbit anti-VSV39, 1:1,000 goat anti-tubulin (Abcam, ab21057), 1:1,000 rabbit anti-GAPDH (Abcam, ab37168), 1:1,000 rabbit anti-ERK (Santa Cruz, sc7383), 1:1,000 mouse anti-phospho ERK (Santa Cruz, sc154), 1:1,000 mouse anti-FGF2 (Santa Cruz, sc79) or 1:1,000 mouse anti-FGF2 (Millipore, clone bFM2, Temecula, CA). After three washes with Tris-Buffered-Saline-Tween (TBS-T), the membranes were incubated with goat anti-rabbit (111-005-003), goat anti-mouse (115-005-146) or bovine anti-goat (805-005-180) horseradish peroxidase-conjugated IgG (Jackson ImmunoResearch Laboratories, Inc.) for 1 h. All secondary antibodies were diluted 1:3,000 in 5% (w/v) skim milk/TBS-T. Membranes were washed four times with TBS-T, and immunoreactive proteins were detected using Supersignal West Pico chemiluminescent substrate (Pierce Biotechnology, Rockford, IL) followed by exposure to X-ray film (Fuji Photo Film Co, LTD, Tokyo, Japan). Densitometry of western blots was carried out with ImageJ software. For all quantifications, the level of protein of interest was normalized to the expression level of respective loading control.

Mitochondria isolation and crosslinking.

786-0 cells (1 × 106 cells) treated with growth factor control (FGF1, 20 ng/ml) or with FGF2 (20 ng/ml) were cultured for 12 h, followed by VSVΔ51 infection (MOI = 1) for 12 h. Cells were then homogenized in ice-cold mitochondria isolation buffer containing 200 mM mannitol, 70 mM sucrose, 10 mM Hepes, and 1 mM EGTA (pH: 7.5) using a Dounce homogenizer with a loose-fitting pestle. Unbroken cells and nuclei were pelleted by centrifugation at 500 × g for 10 min at 4 °C. The supernatants were then centrifuged at 10,000g for 20 min at 4 °C to obtain crude mitochondrial pellets that were cross-linked with 10 mM bis(maleimido)hexane (BMH; Thermo Fisher Scientific, Wilmington, DE) for 30 min at room temperature. Samples then were separated on 4–12% polyacrylamide gels (BioRad, Hercules, CA) and processed for immunoblotting with rabbit polyclonal antibodies to MAVS (1:1,000; Abcam, ab31334, Cambridge, MA) and Complex II Subunit 30 KDa (1:1,000; MitoSciences, MS203, Eugene, OR).

Affymetrix microarrays analysis and quantitative real time PCR.

NFs or CAFs (1 × 106 cells) were seeded in a 6-well dish. The following day, cells were treated with VSVΔ51 at MOI 0.01 or left untreated. Eighteen hours following treatment, total RNA was extracted using an RNAeasy kit (Qiagen Inc., Valencia, CA) according to the manufacturer's protocol, and sample quality was verified on a bioanalyzer (Agilent Technologies). Triplicate samples were pooled and hybridized to a GeneChip Human Gene 1.0 ST Array according to manufacturer's instructions. Data was processed using AltAnalyze40 under default parameters. Probeset filtering implemented a DABG threshold of 70 with P < 0.05. Only constitutively expressed exons for a given locus were used to quantify gene expression. Gene ontology (GO) enrichments were performed using Gorilla41, and identification of genes induced by Type-I interferons was performed using data obtained from the interferome database42. Genes selected for GO analysis are the subset of genes repressed over threefold in CAFs (n = 87/25465). qRT -PCR was performed on the non-pooled triplicate samples. Following conversion to cDNA by Superscript RT II (Invitrogen, Carlsbad, CA), qRT-PCR was carried out using SYBR Green (Invitrogen) according to the manufacturer's instructions. Analysis was performed on a Rotor-Gene RG-3000A machine (Corbett Research) according to the manufacturer's recommended protocols. The primer pairs specific for various gene products used in our experiments are listed in Supplementary Table 1. qRT-PCR measurements were normalized to the GAPDH gene using the Pfaffl method43.

Animal and xenograft studies.

All animal studies were approved by the Institutional Animal Care Committee of the University of Ottawa and carried out in accordance with guidelines of the National Institutes of Health and the Canadian Council on Animal Care. BALB/c, SCID and NOD-SCID mice were purchased from Charles River Laboratories (Wilmington, MA). Female mice, 6–8 weeks of age, were randomized to the different treatment groups according to tumor size in all experiments. For assessment of the acute effect of activated fibroblasts in virus-based therapy, human pancreatic cancer xenograft models were established in SCID mice by subcutaneously injecting MiaPaca-2 cells alone (1 × 107 cells) or MiaPaca-2 cells (5 × 106 cells) in combination with human activated pancreatic fibroblasts (5 × 106 cells). It has been previously shown that CAFs engage in “metabolic coupling” with cancer cells and thereby promote tumor growth44. We initially optimized the MiaPaca + fibroblast co-engraftment model by testing different numbers of cancer cells and PaCFs to be injected to minimize the impact that tumor size could have on virus replication. Tumor growth was assessed on a weekly basis using calipers. When tumor sizes reached 1.2–1.4 cm in diameter, mice received VSVΔ51 encoding Luciferase at a dose of 1 × 106 PFU by injection in the tail vein (IV). Luminescence imaging of mice was performed 48 and 72 h post-virus delivery using the IVIS imaging System Series 200 (Xenogen Corporation). Data acquisition and analysis was performed using Living Image v2.5 software. Mice were killed 72 h after virus treatment, and tumors were removed for histological analysis and virus quantification by plaque assay. Tumor measurements, imaging, histological analysis, and virus titers were performed blinded.

To determine the effect of recombinant FGF2 on virus replication within the tumor, MiaPaca-2 (1 × 107 cells), Ovcar8 (7.5 × 106 cells) or 786-0 (1 × 107 cells) bilateral tumors were generated in SCID mice. Tumor growth was assessed weekly using calipers. When tumor sizes reached 1.2–1.4 cm in diameter, mice received a single intratumoral (IT) injection of recombinant FGF2 (3 μg) or vehicle control (PBS). Twenty-four hours later, a dose of 1 × 106 PFU VSVΔ51-Luciferace was delivered by IV injection. At 48 and 72 h post-virus delivery, the mice were injected intraperitoneally with D-luciferin (Molecular Imaging Products Company) and imaged using the IVIS imaging System Series 200 (Xenogen Corporation). The mice were killed after the last IVIS imaging and tumors removed for histological analysis and virus quantification by plaque assay.

For efficacy studies using MG1 or MG1-FGF2 viruses, MiaPaca-2 or RENCA (mouse renal cell carcinoma cell line) tumors were established in SCID or BALB/c mice, respectively (at least 5 animals per group) as described above. On day 14 (tumor size: 0.5–0.7 cm in diameter) MG1 or MG1-FGF2 (1 × 106 PFU per mouse) was delivered by IV injection while PBS was administered to the control group. Animals were monitored daily for weight loss and morbidity. Tumor size was measured three times weekly during the course of the experiment using calipers. Tumor volume was calculated by the modified ellipsoidal formula [Tumor volume = 1/2(length × width2)] (ref. 45).

Measurement of tumor spread in vivo.

To determine whether MG1-FGF2 virus or recombinant FGF2 potentiate tumor spread, we established a MiaPaca luciferase-expressing stable cell line using the Lenti-X-tet-On advanced inducible expression system (Clontech Laboratories, La Jolla, CA). Lentivirus production in HEK 293T and transduction of MiaPaca-2 cells were performed as per the manufacturer's instructions. At 48 h post-transduction, cells were split 1:2 into medium containing G418 (500 μg/ml) and puromycin (0.5 μg/ml). Surviving cells were tested for inducible expression of Luciferase. MiaPaca-Luciferase cells were used to establish subcutaneous tumors as described above. When tumor sizes reached 1.0–1.2 cm in diameter, mice received 3 doses of PBS, recombinant FGF2 (3 μg/mouse) or MG1-FGF2 (1E6 pfu/mouse) by IT or IV injection, respectively. After 10 days, tumors and normal tissues were recovered. After genomic DNA extraction using a QIAamp DNA kit (Qiagen), samples were evaluated by qRT-PCR for the presence of cells containing the Luciferase gene (See Supplementary Table 1 for primer sequences).

Establishment of tumor grafts derived from individuals with pancreatic cancer.

All tissue samples were collected with informed consent from individuals being treated/diagnosed at the Ottawa Hospital General Campus under a protocol approved by the Institutional Ethics Board. Fresh or frozen tumor samples were subcutaneously implanted in NOD-SCID mice with 100 μl of Cultrex high protein concentration basement membrane extract (Trevigen, Gaithersburg, MD). Tumor growth was measured weekly. When tumors reached 1–1.5 cm3, animals were killed and tumors were excised for ex vivo infection or for histological analysis. For ex vivo infection, tumors were cut into 2 × 2 mm cores and infected with VSVΔ51 (1 × 104 PFU) for 48 h. Subsequently, the secreted infectious particles were titered by plaque assay. For in vivo experiments, when tumors reached 1–1.2 cm3, VSVΔ51 (1 × 107 PFU/mouse) was injected I.V. Forty-eight hours later, the animals were euthanized and tumor-associated infectious virus particles were quantified by plaque assay.

Statistical analyses.

Microarray study analyses are described above. In vitro studies of cell viability and cell proliferation were performed at least twice, and data are reported as mean values. In vitro, ex vivo and in vivo infections were carried out a minimum of three times, and data are reported as ± s.e.m. Statistical analysis was performed on raw data by one-way analysis of variance or Student's t-test (two-tailed). P values < 0.05 were considered significant.

Strength of correlation (r) between αSMA and FGF2 immunohistochemical staining in Supplementary Figure 12 was calculated among five independent quantifications of FGF2 (x) and αSMA (y) staining using a coefficient of correlation calculator (http://www.alcula.com/calculators/statistics/correlation-coefficient/).

To compare the expression levels of FGF2 among different Pancreatic cancer human samples (Fig. 4d), the relative FGF2 expression was measured using the Aperio Imagescope software. Human samples were arbitrarily dichotomized as FGF2-low (bottom 25%) and FGF2-high (top 75%) based on the relative FGF2 protein expression level. A hypergeometric test (P value = 0.01428) was used to assess the significance of FGF2 protein expression level and virus titer in samples derived from individuals with pancreatic cancer. The differences in virus titers between the high and the low FGF2 protein expression groups (Fig. 4e) were compared using Student's t-test. The investigators were blinded during experiments. No samples were excluded from the analysis.

In all animal experiments, n ≥ 5 was used following the calculation of sample size requirements under the equation n =1 + 2C (S/D)2 (ref. 46). S, which is defined as the s.d. of viral titers in the population, was set to 1/3 log. D, which is the magnitude of difference in viral titers between the two populations, was set to 1 log. The constant C was set to 17.8 for a significance level of α = 0.01 and a power of β = 0.95.

Accession codes.

All microarray data are deposited in ArrayExpress under accession code E-MTAB-2359.