Secretome Screening Reveals Fibroblast Growth Factors as Novel Inhibitors of Viral Replication

Viruses infect human cells in order to replicate, while human cells aim to resist infection. Several cellular antiviral programs have therefore evolved to resist infection. Knowledge of these programs is essential for the design of antiviral therapeutics in the future. The induction of antiviral programs is often initiated by secreted proteins, such as interferons. We hypothesized that other secreted proteins may also promote resistance to viral infection. Thus, we tested 756 human secreted proteins for the capacity to inhibit two pseudotypes of vesicular stomatitis virus (VSV). In this secretome screen on viral infection, we identified fibroblast growth factor 16 (FGF16) as a novel antiviral against multiple VSV pseudotypes as well as coxsackievirus. Subsequent testing of other FGF family members revealed that FGF signaling generally inhibits viral infection. This finding may lead to the development of new antivirals and may also be applicable for enhancing oncolytic virus therapy.

ABSTRACT

Cellular antiviral programs can efficiently inhibit viral infection. These programs are often initiated through signaling cascades induced by secreted proteins, such as type I interferons, interleukin-6 (IL-6), or tumor necrosis factor alpha (TNF-α). In the present study, we generated an arrayed library of 756 human secreted proteins to perform a secretome screen focused on the discovery of novel modulators of viral entry and/or replication. The individual secreted proteins were tested for the capacity to inhibit infection by two replication-competent recombinant vesicular stomatitis viruses (VSVs) with distinct glycoproteins utilizing different entry pathways. Fibroblast growth factor 16 (FGF16) was identified and confirmed as the most prominent novel inhibitor of both VSVs and therefore of viral replication, not entry. Importantly, an antiviral interferon signature was completely absent in FGF16-treated cells. Nevertheless, the antiviral effect of FGF16 is broad, as it was evident on multiple cell types and also on infection by coxsackievirus. In addition, other members of the FGF family also inhibited viral infection. Thus, our unbiased secretome screen revealed a novel protein family capable of inducing a cellular antiviral state. This previously unappreciated role of the FGF family may have implications for the development of new antivirals and the efficacy of oncolytic virus therapy.

IMPORTANCE Viruses infect human cells in order to replicate, while human cells aim to resist infection. Several cellular antiviral programs have therefore evolved to resist infection. Knowledge of these programs is essential for the design of antiviral therapeutics in the future. The induction of antiviral programs is often initiated by secreted proteins, such as interferons. We hypothesized that other secreted proteins may also promote resistance to viral infection. Thus, we tested 756 human secreted proteins for the capacity to inhibit two pseudotypes of vesicular stomatitis virus (VSV). In this secretome screen on viral infection, we identified fibroblast growth factor 16 (FGF16) as a novel antiviral against multiple VSV pseudotypes as well as coxsackievirus. Subsequent testing of other FGF family members revealed that FGF signaling generally inhibits viral infection. This finding may lead to the development of new antivirals and may also be applicable for enhancing oncolytic virus therapy.

INTRODUCTION

In order to infect their host, enveloped viruses bind to cell surface receptors and enter target cells by using virus-encoded glycoproteins that drive membrane fusion. After fusion of the virus envelope with the cell's plasma membrane or endosomal membrane, the viral genome is delivered into the cytoplasm, where it can be transcribed and replicated. Replication of some viruses, such as lentiviruses and influenza viruses, requires translocation of the viral genome to the nucleus.

Cells are equipped with sophisticated programs to combat virus infections. These programs include antiviral signaling cascades downstream of ligand-activated receptors. Several ligands secreted locally or systemically convey antiviral activity. For example, tumor necrosis factor alpha (TNF-α), interleukin-6 (IL-6), IL-34, and IL-1β inhibit hepatitis B virus infection, IL-6^−/− macrophages have been shown to be more vulnerable to herpes simplex virus 1 (HSV-1) infection, and IL-27 decreases shedding of HSV-1 in vitro (1–4). Yet the best-studied and most potent secreted antiviral proteins are alpha interferon (IFN-α) and IFN-β, also known as type I interferons (IFNs). Type I IFNs activate STAT signaling through the IFN-α/β receptor (IFNAR), leading to transcription of hundreds of interferon-stimulated genes (ISGs) (5). Only a small subset of the ISGs (e.g., myxovirus resistance protein 1 [MxA; also called MX1] and RNase L [RNASEL]) are direct antiviral effectors. MxA is located in the endoplasmic reticulum, where it can bind viral components, which leads to their inactivation (6–9). RNASEL is activated by 2′,5′-oligoadenylate synthetase 1 (OAS1), which recognizes double-stranded RNA. After its activation, RNASEL degrades the viral RNA (10, 11). Furthermore, the physiological importance of type I IFNs is demonstrated by the fact that most viruses are capable of counteracting the IFN pathway. Viruses have evolved to inhibit the production of IFNs, block the signaling downstream of IFNAR, and/or affect the synthesis of IFN effector molecules (12). Although several cytokines have been described to induce antiviral signaling programs, potential antiviral properties of most secreted proteins have not yet been explored.

Secretome libraries have successfully been employed to establish roles of secreted proteins in various biological systems (13, 14). For instance, secretome screens led to the identification of IL-34 as a promoter of monocyte viability and of pigment epithelium-derived factor as an inducer of human embryonic stem cell proliferation (15, 16). In the present study, we generated and employed a library containing 756 secreted proteins to perform a secretome screen for modulators of viral entry and/or replication. We specifically used recombinant vesicular stomatitis viruses expressing the glycoprotein from Lassa virus (VSV-LASV) or Ebola virus (VSV-EBOV), since the glycoproteins of these viruses initiate distinct entry pathways (17–19). This allowed for straightforward discrimination between modulators at the level of viral entry versus viral replication. Using this approach, we identified fibroblast growth factor 16 (FGF16) as a novel inhibitor of viral replication. Its antiviral activity was recapitulated in multiple cell lines and was not limited to VSV pseudotypes, as it also decreased infection by coxsackievirus. Treatment with FGF16 induced an FGF-like transcriptional response, and several other members of the FGF family inhibited viral replication, indicating that the induction of FGF signaling represents a novel way to combat certain virus infections.

RESULTS

Generation of a secreted protein library.

To identify new secreted proteins that affect viral infection, we generated a large secretome library. To this end, we acquired a cDNA collection representing 756 different transcripts derived from 679 genes. Ninety-six percent of the transcripts in this collection encode established secreted proteins, while the rest of the encoded proteins are predicted to be secreted. Our secretome library covers a broad spectrum of physiological functions (Fig. 1A; see Table S1 in the supplemental material) and includes cytokines and chemokines but also peptide hormones, extracellular matrix proteins, neuropeptides, and enzymes. Each of these categories consists of a mix of broadly studied secreted proteins and proteins with hitherto undiscovered functions. Our library enables the discovery of new functions for well-described proteins as well as identification of the functions of less-well-explored proteins. To produce protein from each transcript within this secretome library, each individual cDNA was transfected into HEK293T cells, and cell-free conditioned medium was collected and stored for use in a separate viral infection assay (Fig. 1B). Quantification of secreted proteins in the conditioned medium showed average levels of IFN-γ above 30 ng/ml, while TNF-α levels reached up to 1.4 μg/ml (Fig. 1C). The production was performed in seven rounds, for which the efficiency was monitored by separate transfections of IFN-γ and TNF-α cDNAs. The amounts of released cytokines in the medium were consistently within the same range, indicating that the transfection performance was robust over time (Fig. 1C). Next to the secretion of IFN-γ and TNF-α, we observed secretion of biologically active concentrations of nine other proteins (our unpublished observations), indicating that we generated a functional secretome library that can be used for high-throughput screening.

FIG 1.

Generation of a secreted protein library covering broadly diverse physiological functions. (A) Our cDNA library consists of 679 genes encoding 756 transcripts. The 679 genes were categorized according to the Panther and Gene Ontology classification systems and additionally colored from top to bottom to belong to the supercategories immunology, development, signaling, extracellular matrix, transport, enzymes, and other/undefined (57–59). (B) Schematic overview of conditioned medium generation. HEK293T cells were plated, incubated overnight, and then transfected with single plasmid DNAs. The cell-free secreted protein-conditioned medium was collected at day 3. Aliquots were diluted 4× in medium and stored in ready-to-screen plates at −80°C. (C) Transfection consistency across different transfection rounds was determined by ELISA with control IFN-γ- and TNF-α-conditioned media. IFN-γ and TNF-α levels in empty vector-conditioned medium were below the detection limit of 20 pg/ml.

Secretome screening reveals FGF16 as a novel modulator of viral replication.

We next studied the effects of secreted proteins on viral infection by using enhanced green fluorescent protein (EGFP)-expressing VSV-LASV and VSV-EBOV. These two viruses have different modes of viral entry due to their respective glycoproteins, while they share the same cytoplasmic replication mechanism. The level of viral infection on HAP1 cells was determined by the total EGFP fluorescence intensity as quantified using a sensitive microplate reader. Using infections with and without preincubation with IFNB1-conditioned medium, a known inhibitor of VSV replication, we set up a reproducible assay which allowed us to detect inhibition and potentially also enhancement of viral infection (Fig. 2A and B). Notably, the Z-factor of this assay was higher than 0.5 for both VSV-LASV and VSV-EBOV, indicating that the dynamic range was large enough for effective high-throughput arrayed screening (Fig. 2C) (20).

FIG 2.

Secreted protein screening reveals FGF16 as an inhibitor of VSV-EBOV and VSV-LASV infection. (A) Viral infection assay set-up. One day after HAP1 target cells were plated, they were exposed to different secreted proteins for another 24 h. Finally, cells were infected with VSV-EBOV or VSV-LASV for 5 or 6 h, respectively. GFP levels were detected using a fluorescence microplate reader. (B) HAP1 cells were pretreated with empty vector- or IFNB1-conditioned medium for 24 h. These cells were then infected with serial dilutions of VSV-LASV supernatant. IFNB1-treated cells nearly completely resisted infection. For the sake of simplicity, we therefore used the residual GFP signal in IFNB1-pretreated wells as the background of the assay. Medium treatment was plotted as a white line. We chose to use an 8× dilution of VSV-LASV in the screen because, at this dilution, the dynamic range to find potential inhibitors (blue arrow) or augmenters (red arrow) of viral infection seemed optimal. (C) HAP1 cells were preincubated as described for panel A, followed by infection with either VSV-LASV or VSV-EBOV. Forty-eight wells were analyzed for each condition. The Z-factors for both assays were subsequently calculated. (D) The secretome screens were performed in this screen-compatible assay, including additional control empty vector- and IFNB1-conditioned media on each plate. Calculation of B-scores for the GFP levels is described in Materials and Methods. The averages for two replicates for each screen are depicted in readout order. Hits beyond three times the SD (dashed line) are shown in blue. FGF16 is colored red in the VSV-LASV screen because its value is below the threshold for hit selection here. (E) All hits were tested again, using new conditioned media. The statistical significance of the difference between each secreted protein and its corresponding empty vector (EV) was determined by analysis of variance (ANOVA) followed by Tukey's post hoc test. (F) Specified images of VSV-EBOV infections. Nuclei were stained with DAPI. (G) Target cells were treated with medium, 1 μg/ml IFN-α, or a serial dilution of commercially available purified recombinant FGF16, followed by infection with VSV-LASV. (H) HAP1, U2OS, 2A14, and HepG2 cells were pretreated with control medium, 1 μg/ml IFN-α, or 5 μg/ml FGF16 and infected with VSV-LASV. Statistically significant differences were identified by ANOVA followed by Tukey's post hoc test. For panels E, G, and H, GFP fluorescence intensities were normalized to the mean for all empty vector or medium controls. Data from a representative experiment of 2 (G) or 3 (H) experiments are shown. For panel H, background fluorescence intensities of noninfected cells were subtracted before normalization to improve the comparability between cell lines.

This assay was then used to identify novel secreted proteins with the capacity to modulate viral infection. The total arrayed library of 756 cell-free conditioned media was tested and was controlled by 56 empty vector-conditioned media and 28 IFNB1-containing media equally distributed over the library plates. The screens were performed independently with VSV-LASV and VSV-EBOV, both as biological duplicates. Data were normalized using the B-score, a method that is well suited for analysis of multiplate high-throughput screens (21). The majority of the conditioned medium-treated wells had GFP levels similar to those of the empty vector controls for both viruses, whereas IFNB1-containing media showed clear inhibition (Fig. 2D). Importantly, this unbiased analysis highlighted known inhibitors of viral replication within our library (IFNA2, IFNB1, and IFNG) in both virus screens (Fig. 2D). The most significant novel infection inhibitors in either of the screens were ADCK1, FGF16, LAS2, and TTR, while SCUBE3 and ENHO2 seemed to enhance virus infectivity. Subsequently, we set out to validate these hits by using both VSV recombinants and found that only FGF16 reproducibly inhibited infection (Fig. 2E and F). Interestingly, FGF16 significantly inhibited both viruses, with the strongest inhibition against VSV-EBOV. Further analysis of the data from the VSV-LASV screen revealed that FGF16 was just below the stringent cutoff for hit selection (Fig. 2D), while nonvalidated hits were all within the single standard deviation (SD) range in the screen for the other VSV pseudotype. All these experiments were performed using FGF16-conditioned medium, in which the exact concentration of FGF16 was unknown. The FGF16-conditioned medium used for the screens may have contained a suboptimal concentration of FGF16. Consequently, we investigated the effect of commercially available purified recombinant FGF16 to determine its active concentration. More than 40% reductions in viral infection were detected at FGF16 concentrations of 35 ng/ml and higher, confirming the initial findings using conditioned medium (Fig. 2G). Moreover, at higher concentrations, FGF16 almost completely abolished infection, similar to recombinant IFN-α. To determine whether FGF16's inhibitory effect would apply broadly to cell types other than our model HAP1 cell line, we additionally pretreated the osteosarcoma cell line U2OS, the uveal melanoma cell line 2A14, and the liver cancer cell line HepG2. FGF16 exhibited significant antiviral effects on U2OS and 2A14 cells but not on HepG2 cells, indicating that multiple—but not all—cell types can be protected against viral infection by FGF16 (Fig. 2H).

FGF16 inhibits replication of multiple cytosolic viruses.

FGF16 apparently affected general VSV replication, since infection by both VSV-LASV and VSV-EBOV was inhibited. While the cell entry mechanisms of these viruses differ, their replication machineries are identical. To further confirm this notion, we analyzed the infectivity of VSV expressing its native glycoprotein or one of the glycoproteins of the New World arenaviruses Guanarito virus (VSV-GTOV), Machupo virus (VSV-MACV), and Junin virus (VSV-JUNV), which utilize yet another entry pathway. Indeed, next to inhibiting VSV-LASV and VSV-EBOV infection, FGF16 robustly inhibited infection by wild-type VSV and three other VSV pseudotypes (Fig. 3A and B). Furthermore, FGF16 did not seem to block a step in viral entry, as the antiviral activity of FGF16 was pronounced only after prolonged pretreatment (Fig. 3C). These data indicate that a downstream replication process, not viral entry, was inhibited by FGF16.

FIG 3.

FGF16 inhibits replication of cytoplasmic viruses. (A) HAP1 cells were pretreated with medium, 1 μg/ml IFN-α, or 5 μg/ml FGF16, followed by infection with the indicated VSV pseudotypes. (B) Representative images of DAPI-stained cells after VSV infection. (C) HAP1 cells, pretreated for 24 h, 5 h, or 0.5 h, as indicated, were infected with VSV-LASV. (D) Infection of HAP1 cells with GFP-expressing coxsackie B3 virus for 5 h. (E) Representative images of coxsackie B3 virus-infected cells. (F) HAP1 cells were pretreated as described above, followed by infection with GFP-expressing Lenti-VSV for 24 h. Percentages of GFP⁺ cells were determined by flow cytometry. (G) Quantification of GFP⁺ cells. For panels A, C, and D, GFP intensities were normalized to the medium control. Statistical significance was determined by ANOVA followed by Tukey's post hoc test.

To gain more insight into the mechanism of the antiviral activity of FGF16, we tested whether the antiviral effect of FGF16 extended beyond the VSV replication machinery. Strikingly, coxsackievirus, a nonenveloped cytoplasmic RNA virus, was also inhibited by FGF16 treatment (Fig. 3D and E). Since both VSV and coxsackievirus replicate in the cytosol, we next wondered whether FGF16 targets only cytosolic viral replication. As a model virus for nuclear replication, we utilized a VSV-pseudotyped lentivirus (Lenti-VSV) which consequently has the same viral entry mechanism as wild-type VSV. Interestingly, Lenti-VSV infection was not inhibited by FGF16 (Fig. 3F and G), which further substantiated the notion that FGF16 does not inhibit viral entry. Taken together, these data suggest that FGF16 mainly affects cytosolic replication of multiple RNA viruses.

FGF signaling inhibits viral replication.

We next hypothesized that the mechanism by which FGF16 inhibits viral replication may have similarities to that of type I IFN-mediated antiviral activity. To determine the expression signature of cells, we performed transcriptome sequencing (RNA-Seq) of FGF16-treated cells. Importantly, genes which have been shown to play a role in viral infection after type I IFN stimulation were not significantly upregulated after FGF16 stimulation, arguing for a distinct mechanism of action compared to that of type I IFN-mediated viral inhibition (Fig. 4A) (22). To confirm that FGF16's mechanism of action is indeed independent of the type I IFN pathway, we either neutralized type I IFNs by use of antibodies or inhibited IFNAR signaling by use of JAK inhibitor I (23, 24). As expected based on the transcriptome analysis, FGF16 inhibited viral infection regardless of type I IFN pathway interference, while the antiviral activity of IFN-α was significantly reduced (Fig. 4B and C). Thus, the mechanism by which FGF16 inhibits viral infection is independent of type I IFN production or (IFN) signaling via JAK/STAT.

FIG 4.

FGF signaling induces resistance to viral infection. (A) RNA-Seq was performed on three replicates of HAP1 cells that were treated with either medium or 6 μg/ml FGF16 for 24 h. Box plots show the log₂-transformed ratios of gene expression of FGF16 over the medium control [log₂(fold change)]. Data for genes that are induced by type I IFN or FGF signaling, according to the literature (22, 25, 26, 28–30, 60), were plotted in separate box plots. The data were analyzed by ANOVA followed by Tukey's post hoc test. (B and C) HAP1 cells were pretreated with 50 μg/ml anti-IFN-α and 50 μg/ml anti-IFN-β (B), 1 μM JAK inhibitor I (C), or medium (control) for 1 h before treatment with either medium, 5 μg/ml FGF16, or 1 μg/ml IFN-α for 24 h, followed by infection with VSV-LASV. The data were analyzed by two-way ANOVA followed by Tukey's post hoc test. (D) HAP1 cells were pretreated with the indicated concentrations of IFN-α together with FGF16 for 24 h before infection with VSV-LASV. (E) Volcano plot showing the expression change versus adjusted P value for each FGF signaling-induced gene (gray dots) as determined by the RNA-Seq analysis described for panel A. (F) The promoter region (−400 to +100) of the 949 significantly upregulated genes (P < 0.05) was further analyzed for enrichment of known transcription factor binding motifs by use of Homer (56). The 38 significant motifs (P < 0.05) are shown in Table S2 in the supplemental material. The pie charts depict, for each indicated family (top three families), the proportion of transcription factor motifs within the Homer database that were significantly enriched (colored) in the upregulated promoters. (G) HAP1 cells were pretreated with either medium, 1 μg/ml IFN-α, or 5 μg/ml of at least one FGF of each canonical and endocrine subfamily of FGFs. The corresponding subfamilies are depicted below each FGF (27). Data are shown for a representative experiment of two experiments. For panels B, C, D, and G, data were normalized to the medium control after subtraction of background fluorescence.

Since FGF16 and type I IFNs thus probably activate distinct antiviral pathways, their effects may be additive. To investigate this, we performed serial dilution of IFN-α in the presence of different concentrations of FGF16. FGF16 increased the inhibition of viral infection at suboptimal levels of IFN-α, arguing that FGF16 may act in conjunction with type I IFNs to inhibit viral replication (Fig. 4D). Note that the 50% inhibitory concentration (IC₅₀) for inhibition of VSV-LASV infection was 76.4 ng/ml (3.2 nM) for FGF16 and 0.15 ng/ml (6.7 pM) for IFN-α (calculated from the data in Fig. 2G and 4D, respectively).

We next analyzed the RNA-Seq data set for other obvious transcriptional signatures and found that—as could be expected after FGF16 stimulation—FGF receptor (FGFR)-regulated genes were significantly upregulated (Fig. 4A) (25, 26). These genes included those for the transcription factors ETV4 and ETV5, DUSP6, and SPRY family members, all of which are known targets of E26 transformation-specific (ETS) transcriptional activity downstream of FGFR signaling (Fig. 4E) (27). Moreover, in-depth analyses of all upregulated genes after FGF16 treatment showed that their promoter regions were significantly enriched for the ETS binding site motif (Fig. 4F; see Table S2 in the supplemental material) (28–31). Thus, the transcriptional profile of FGF16-treated cells suggests that inhibition of viral replication may be directed via signaling through FGFR. If general FGF signaling induces resistance to viral infection, then other FGFs may also cause this effect. To test this hypothesis, we pretreated cells with members of each of the five subfamilies of secreted FGFs, including FGF9 and FGF20 of the FGF16-containing subfamily. Subsequently, we inoculated these cells with virus and quantified the levels of infection. Similar to FGF16, both FGF9 and FGF20 strongly inhibited viral replication (Fig. 4G). Several other FGF subfamily members were also capable of inducing protection against viral infection, as illustrated by marked reductions of infection after preincubation with FGF1 and FGF8B (Fig. 4G). These results reinforce the notion that FGF signaling triggers an antiviral program.

DISCUSSION

In the present study, we generated and employed a secreted protein library to discover novel cell-intrinsic pathways that can modulate viral infection. Using this unbiased and unique forward screening approach, we identified FGFs as a novel family of secreted proteins with antiviral properties. The major advantage of forward screening approaches is that hits have a causal relationship with the observed phenotype. In our screen, perturbations—consisting of hundreds of secreted proteins—were tested for their effect on the observed outcome—virus infection—leading to the discovery that FGFs can cause a cellular antiviral state. In the past, secretome screens have led to other fascinating discoveries, but to our knowledge, they have not been performed before to study viral infection (13–16).

FGFs are required for organ development and regeneration but are also involved in metabolism (27). However, little is known about their role during viral infection. The human FGF family consists of 18 secreted and four intracellular FGFs, categorized into seven subfamilies (27). For each of the six secreted FGF subfamilies, we tested the antiviral capacity of a representative member and found that at least three subfamilies can induce an antiviral state in cells. The secreted subfamilies are known to signal via one or more of the four FGFRs, each also having several splice variants (32). The expression of FGFRs and downstream molecules differs between cell lines, which may have accounted for the difference in antiviral efficacy of FGF16 in the cell lines examined here. In addition, FGFR binding and downstream signaling may be enhanced by cofactors, such as heparin/heparan sulfate proteoglycans or the Klotho protein family, but these did not act synergistically with FGF16 (data not shown). We next analyzed the relationship between the antiviral activities of the FGF and type I IFN families. The transcriptional profile of FGF16-treated cells lacks upregulation of IFN antiviral effector molecules. Furthermore, neither neutralization of potentially produced type I IFNs nor inhibition of JAK/STAT signaling inhibited FGF16's antiviral capacity. These data indicate that the antiviral mechanism induced by FGF16 is distinct from that of IFNs. FGF signaling may therefore represent an innate strategy by which host cells can combat IFN-resistant viruses (12). It is still unclear which intracellular effector molecules exert the antiviral effect of FGF signaling, although the requirement for more than 5 h of preexposure to FGFs suggests that the antiviral effect induced by FGF16 covers a transcription activation program. A more detailed molecular investigation of the effectors targeting the viral replication cycle is required to further understand the FGF-induced antiviral pathway and to elucidate why infection by cytoplasmic but not nuclear RNA viruses is inhibited by FGF signaling.

Interestingly, HSV-1 specifically uses FGFRs for docking and entry (33), and blockade of FGFRs with the high-affinity ligand FGF2 can inhibit HSV-1 infection in vitro and in vivo (33, 34). Importantly, the antiviral effect of FGF16 is not through blockade of FGFRs, since FGFRs are not the entry receptors for the viruses that we tested (18, 35–37). Furthermore, a short preincubation with FGF16 before infection still allowed efficient viral infection, arguing for a mechanism that entails receptor-mediated signaling. Since the antiviral effect was established for several different viruses with diverse cell entry strategies, the mechanism is distinct from FGF-mediated inhibition of HSV-1 infection. In this study, the cytoplasmic viruses VSV and coxsackievirus, but not the nuclear virus Lenti-VSV, were inhibited by FGF16. Others have found that nuclear adenovirus type 2 is not inhibited by FGF2, supporting the notion that the antiviral activity of FGFs may be exerted in the cytoplasm (33). However, adenovirus type 2 is a DNA virus, as opposed to the RNA viruses studied here. It would be of considerable interest to investigate whether FGFs can also potentiate antiviral responses against DNA viruses, especially those with cytoplasmic replication, such as vaccinia virus (38).

Different members of the Herpesviridae as well as measles virus have been reported to induce expression of FGF2 (39). Transfection with a plasmid encoding the Epstein-Barr virus latent membrane protein 1 leads to increased FGF2 expression (40). Higher levels of FGF2 have also been reported following herpes simplex virus 1 (HSV-1) infection in both mice and humans (41, 42). These data together suggest that FGFs may participate in the innate antiviral response. Although for many viral infections the induction of FGF expression has not been determined, it is clear that FGFs are expressed upon tissue damage to promote tissue regeneration. Our data suggest inhibition of viral replication as an added benefit of FGF expression. FGF signaling may therefore be harnessed to strengthen this effect. However, this requires further investigation of the pathways involved in this process in order to determine whether molecular determinants of FGF signaling may be a suitable therapeutic target.

Since the secretome library contains seven additional classical FGFs besides FGF16, we wondered why they were not identified as hits in the screens. Interestingly, these included five FGFs from the subfamilies that did not affect viral infection (FGF4, FGF7, FGF10, FGF21, and FGF23) (Fig. 4D), possibly explaining a lack of antiviral efficacy in the screen. Furthermore, subfamily members FGF1 and FGF2 in the library may not have been potent enough, since purified recombinant FGF1 only partially inhibited viral infection. Alternatively, it is possible that the FGFs in the library were not expressed at sufficient (bioactive) levels to inhibit viral replication, as the recombinant protein levels in the library conditioned media varied considerably (from nanograms per milliliter to micrograms per milliliter) (Fig. 1C and data not shown).

Recently, the excellent capacity of oncolytic viruses—such as VSV and coxsackievirus—to lyse cells and elicit a systemic immune response has been harnessed clinically for immunotherapy of cancer (43–45). However, intracellular antiviral signaling may inhibit successful infection of tumor cells by these viruses. Indeed, mice treated with IFN-α showed a weaker response to oncolytic alphavirus M1 treatment (46). These results may be extrapolated to FGFs, as the expression of various FGFs is upregulated in ovarian cancer, breast cancer, prostate cancer, and colon cancer (47–51). In tumor microenvironments with high FGF levels, replication of oncolytic VSV or coxsackievirus is possibly naturally inhibited. Such oncolytic virus therapy may benefit from additional inhibition of FGF signaling. Since in our experiments not all viruses are affected by FGFs, the use of FGF-resistant oncolytic viruses may be considered for cancers which highly express FGFs.

In conclusion, we identified FGFs as novel inhibitors of virus infection. This finding may have several implications: promoting FGF signaling may have potential for antiviral therapy, while inhibition of FGF signaling provides opportunities for the improvement of oncolytic viral therapy.

MATERIALS AND METHODS

Cells and viruses.

The HEK293T (provided by J. Neefjes, NKI, Amsterdam, The Netherlands; HLA-A and -B typed as a control for authenticity), HAP1 (Horizon Genomics), HepG2 (ATCC), U2OS (ATCC), and 2A14 (provided by M. Griffioen, LUMC, Leiden, The Netherlands) (52) cell lines were cultured in IMDM (Lonza) supplemented with 10% fetal calf serum (FCS) and 1% penicillin-streptomycin at 37°C and 5% CO₂. VSV-LASV, VSV-EBOV, VSV-GTOV, VSV-MACV, and VSV-JUNV expressing EGFP as well as GFP-expressing coxsackievirus B3 (CV-B3-GFP; strain Nancy) were propagated as described previously (17, 18, 35, 36). A replication-incompetent lentivirus pseudotyped with the VSV glycoprotein (VSV-G) and expressing GFP was produced using the puc2CL6IPwo plasmid (kindly provided by H. Hanenberg) in combination with packaging constructs (53).

Recombinant secreted proteins.

Plasmids for the secreted protein library were obtained from Origene and GE Healthcare. HEK293T cells (40,000) were plated in wells containing 0.5 ml medium in 24-well plates and incubated overnight. Transfection mixes were prepared by mixing 50 ng plasmid DNA with 1.5 μg polyethylenimine (Polysciences) in 50 μl serum-free medium per transfection, followed by 30 min of incubation at room temperature. Fifty microliters of transfection mix was added per well. After 6 h of incubation, the medium was replaced by 1 ml fresh medium. After three more days, conditioned medium for each individual transfection was collected and cleared of cells by multiple rounds of centrifugation. Levels of IFN-γ and TNF-α were determined by enzyme-linked immunosorbent assay (ELISA) according to the manufacturer's protocol (Sanquin). Purified recombinant human proteins IFN-α, FGF1, FGF5, FGF8b, FGF9, FGF10, FGF16, FGF20, and FGF21 were all from PeproTech. Anti-IFN-α clone M710 was obtained from Thermo Scientific, and anti-IFN-β was produced in-house. JAK inhibitor I was obtained from Calbiochem.

Viral infection assay.

Target cells were plated in black flat clear-bottomed 96-well tissue culture plates (Greiner). HAP1 and U2OS cells were plated at 15,000 cells/well, and other cell lines were plated at 20,000 cells/well. The next day, conditioned medium or purified recombinant proteins (at the indicated concentrations) were added. The final dilution of conditioned medium in the screen was 11×, as this dilution resulted in the maximal difference between empty vector and control IFNB1 samples in optimization experiments. Individual screening plates contained two control wells with IFNB1-conditioned medium and four control wells with empty vector-conditioned media covering the different backbones of the secreted protein-encoding plasmids. After overnight incubation, cells were infected using a concentration of virus that enabled significant detection of inhibition and induction of viral infection, using a microplate reader for readout (Fig. 2B). After 5 to 7 h, cells were fixed using 4% formaldehyde for 30 min. Plates were washed twice with phosphate-buffered saline (PBS). Total GFP fluorescence per well was measured with a Clariostar microplate reader (BMG Labtech).

Screen analysis.

The quality of an assay to be used for screening can be assessed by the Z-factor (20). The Z-factor for the viral infection assay was calculated from the GFP fluorescence intensities of the negative (empty vector) and positive (IFNB1) controls by using the following formula: Z′ = 1 − [3(σ_{empty vector} + σ_IFNB1)/(μ_{empty vector} − μ_IFNB1)], in which σ equals the standard deviation and μ equals the mean.

GFP fluorescence intensities in the secretome screens were normalized per plate by B-score normalization in Excel (Microsoft), as described by Malo et al. (21). This method normalizes for plate and row confounding effects by iterative subtraction of median row and column values (excluding those for IFNB1 controls) from each individual well value. After this median polishing step, the B-score for each well was determined by a plate normalization using the median absolute deviation, as follows: MAD = median[∣polished_well_va − median(polished_wells_plate)∣]. Each virus screen was performed in duplicate. The final B-score for a condition was the mean of the two values. The cutoff for hit selection was set at 3 times the standard deviation for all conditions, excluding IFNB1 controls.

Fluorescence microscopy and flow cytometry.

Imaging of infected cells (i.e., GFP-positive cells) was performed after visualizing cellular nuclei by use of 4′,6-diamidino-2-phenylindole (DAPI; Invitrogen) and a fluorescence microscope (Zeiss, Oberkochen, Germany). For lentivirus infections, HAP1 cells were inoculated with virus for 24 h before the GFP-positive cells could be determined by flow cytometry using an LSR flow cytometer (BD Biosciences). Data were analyzed using FlowJo VX (Treestar).

RNA-Seq.

HAP1 cells were plated at a density of 0.5 million cells/well of a 6-well plate. The following day, cells were treated with either medium or 6 μg/ml purified recombinant FGF16 (n = 3; total of six samples). After 24 h of incubation, cells were harvested and lysed in TRIzol reagent (Invitrogen). RNA libraries were prepared for sequencing using the standard manufacturer's protocols and sequenced on a HiSeq 2000 platform (Illumina). The single-read sequences were mapped against the human genome (hg38) by use of Tophat software. Read counts were determined using HTseq-count (22). In edgeR, counts per million were calculated from the read counts, and lowly expressed genes, with 1 cpm or less in three or more samples, were discarded. A multidimensional scaling (MDS) plot was made to confirm whether a paired analysis was appropriate. In limma, VOOM normalization was applied, and a linear model was fitted using a paired design (54, 55). The false-discovery rate (FDR) was used to correct for multiple testing, and adjusted P values of <0.05 were considered significant. For the significantly upregulated genes, enrichment of known transcription factor binding motifs in the transcription start site region of −400 to +100 was identified using the findMotifs.pl script of Homer (v4.9.1) (56; http://homer.ucsd.edu/homer/).

Statistics.

P values were determined by the indicated statistical tests, using R. The following symbols are used to indicate statistical significance in the figures: *, P < 0.5; **, P < 0.1; ***, P < 0.01; ****, P < 0.001; and n.s., not significant. IC₅₀ values were calculated using the online IC₅₀ calculator tool of AAT Bioquest (https://www.aatbio.com/tools/ic50-calculator).

Accession number(s).

The RNA-Seq data have been deposited in the Sequence Read Archive under accession no. SRP151416.