Cell motility is fundamental for the development, maintenance and immune integrity of multicellular organisms. Its dysregulation is linked to defects in embryonic development, immune cell trafficking and wound healing, as well as the induction of metastasis and tumour cell invasion3,4. Cell motility is a repetitive multistep process under the control of G protein-coupled or growth factor receptors5. Receptor activation initiates leading edge extension, followed by adhesion to the cell substrate, cell body translocation and finally rear edge retraction. Underlying this complex set of movements are cell-adhesion molecules and the Rho GTPases, Rac1/Cdc42 and RhoA, which regulate actin dynamics for leading edge extension and rear edge retraction, respectively3,5. Viruses, including Kaposi sarcoma herpesvirus, herpes simplex virus, Epstein–Barr virus, human papillomavirus and the poxvirus vaccinia virus (VACV) are known to activate this essential cell process1,2,6,7,8,9. For poxviruses, a single viral protein, F11, is known to be needed for virus-induced cell motility2,10,11. F11 has been shown to inhibit RhoA signalling by bringing together RhoA and the Rho GTPase-activating protein, myosin-9a, using its functional PDZ-like domain12. The subsequent loss of RhoA signalling to Rho-associated protein kinase and mDia facilitates infected cell detachment and tail retraction2. Viruses lacking F11 display reduced infected cell motility, a reduction in virus spread and are attenuated in murine infection models2,10,11. Surprisingly, given the critical importance of VACV-induced cell motility for efficient virus spread, to date, the viral factor(s), cell receptor(s) and downstream effectors required for activating virus-mediated cell motility have not been identified. In addition, the dynamics of infected cell migration, how this is influenced by the loss of viral and cellular factors that control motility and how this affects virus spread have not been determined.

Many poxviruses encode homologues of epidermal growth factor (EGF)13 (Supplementary Fig. 1a). VACV encodes two copies of vaccinia growth factor (VGF), which, when cleaved from infected cells by an unknown protease, can activate EGF receptor (EGFR)14,15,16,17. VACV deleted for both copies of VGF can no longer induce cell proliferative responses in chicken eggs and is attenuated in mice18,19,20. To determine whether VGF is needed for VACV-induced cell motility and virus spread, we generated viruses deleted for VGF (ΔVGF), F11 (ΔF11) or VGF and F11 (ΔVGF/ΔF11) in a parental western reserve (WR) strain expressing a nuclear-targeting NP-eGFP21. With its role in virus-induced cell motility well established2,10,11, ΔF11 virus was used as the control. Plaque assays, which measure the efficiency of cell-to-cell spread, indicated that ΔVGF, ΔF11 and ΔVGF/ΔF11 viruses were attenuated compared to WR (Fig. 1a). Measurements of plaque diameters indicated that ΔVGF, ΔF11 and ΔVGF/ΔF11 plaques are 38.6%, 53.1% and 62.6%, respectively, smaller than those of WR (Supplementary Fig. 1b,c). VACV produces two types of infectious particles: single membrane mature virions and double membrane enveloped virions. Mature virions are released from host cells by lysis and enveloped virions by membrane wrapping and exocytosis of mature virions. Enveloped virions are released from the cell as extracellular enveloped virions (EEVs) or remain associated with the plasma membrane as cell-associated enveloped virions (CEVs), a subset of which can direct the formation of actin tails22. Mature virions are thought to mediate host-to-host transmission and EEVs facilitate long-range spread in an organism, whereas CEVs and actin tails contribute to cell-to-cell spread. Analysis of virus production over a 24-h period showed a 3.3-fold and 4.4-fold defect in mature virion yield with ΔVGF and ΔVGF/ΔF11 viruses, respectively (Fig. 1b). More relevant to virus spread, no defect in the formation of EEVs, CEVs or actin tails was seen with ΔVGF (Fig. 1c–f). Minor reductions in EEV (1.6-fold) and CEV (1.2-fold) formation with ΔF11 and actin tails (1.4-fold) with ΔVGF/ΔF11 were observed (Fig. 1c–f). These results indicate that the dramatic attenuation of virus plaque formation and spread of infection could not be attributed to a major defect in virus production. To assess virus-induced cell motility, live-cell imaging and single-cell tracking of WR, ΔVGF, ΔF11 and ΔVGF/ΔF11 plaque formation were performed (Supplementary Video 1). Tracks of WR plaques showed kinetically synchronized waves of cell migration out from the plaque centre, whereas ΔVGF, ΔF11 and ΔVGF/ΔF11 cells appeared to cover less distance and to move more sporadically (Fig. 1g). To quantify this, we measured the distance cells moved from the plaque origin over time (radial velocity) and the degree to which cells migrate in the straightest possible manner between their origin and end point (directional migration efficiency). When assessed relative to WR, both the radial velocity and the directional migration efficiency of cells infected with mutant viruses were impaired, with ΔVGF/ΔF11 being the most severe (Fig. 1h,i). These results identify VGF as a new viral factor that facilitates VACV-induced cell motility to enhance cell-to-cell virus spread.

Fig. 1: VGF is required for VACV-induced cell motility and virus spread.
figure 1

a, Plaque formation by VACV WR and mutants (ΔVGF, ΔF11 and ΔVGF/ΔF11). Nuclei (red) and infected cells (green) are shown. b,c, 24-h mature virion (MV; b) and EEV (c) yields from WR and mutant VACVs. d,e, CEVs (d) and actin tails (e) per cell during WR and mutant VACV infections. f, Representative images of cells infected with WR or VACV mutants for 10 h and stained with antibody directed against the VACV EV protein B5 and actin. g, Single-cell tracking of WR and mutant VACV plaque formation (20–48 hpi). Tracks are colour coded by time (hpi). h,i, The radial velocity (h) and directional migration efficiency (i) of cells migrating from the centre of plaques in g. Data represent three or more biological replicates (ai). Images are representative of three biological replicates (a,f,g). Scale bars, 500 μm (a,g) or 20 μm (f). Bars represent means ± s.d. (b,c) or means ± s.e.m. (h,i) of n = 5 plaques per condition and replicate. Lines represent the means of 15–20 cells per condition and replicate (d,e). A paired (b,c) or unpaired (d,e,h,i) t-test was applied (****P < 0.0001, **P < 0.01, *P < 0.05, NS, not significant). See Supplementary Table 1 for exact statistics.

To simultaneously stimulate cell growth and prevent cell death, VGF activates EGFR and the downstream signalling molecules phospholipase C-γ1 (PLC-γ1) and extracellular signal-regulated kinase 1/2 (ERK1/2) (refs 23,24,25,26,27). Given the diversity of EGFR signalling, activation arrays were employed to analyse the full compendium of EGFR signalling in WR-infected or ΔVGF-infected cells, as well as cells treated with purified VGF or EGF (Supplementary Fig. 2a). Significant increase in VGF-dependent steady-state phosphorylation of EGFR, MEK1/2, Akt and ERK1/2 was seen (Supplementary Fig. 2b–d). This suggested that VGF-mediated activation of EGFR may be triggering ERK–mitogen-activated protein kinase (MAPK)-mediated cell migration28. To extend these findings, the phosphorylation of EGFR, Raf, MEK1/2, p44/42 MAPK (ERK1/2), p90RSK and focal adhesion kinase (FAK) were compared during WR and ΔVGF infections. As opposed to WR, which showed robust activation of this pathway between 2 and 8 h post-infection (hpi), phosphorylation of each of these proteins was markedly lower in ΔVGF-infected cells (Fig. 2a–c). The addition of exogenous VGF confirmed that it could effectively mimic EGF by activating EGFR–MEK–FAK signalling (Fig. 2a–c).

Fig. 2: VGF activates cell motility through EGFR–MAPK–FAK signalling.
figure 2

ac, Immunoblot analysis of EGFR (a), MAPK (b) and FAK (c) phosphorylation during WR and ΔVGF infections. d, Plaque formation in the presence of VGF signalling inhibitors. DMSO, dimethylsulfoxide. e, Diameter of plaques from d. f, Single-cell tracking of VACV plaque formation in the presence of EGFR, MEK or FAK inhibitors (24–48 hpi). Tracks are colour coded by time (hpi). g,h, The radial velocity (g) and directional migration efficiency (h) of cells migrating from the centre of plaques in f. Data represent three or more biological replicates (ah). Images are representative of three biological replicates (ad,f). Lines represent the means of 100 plaques per condition and replicate (e). Scale bar, 500 μm (f). Bars represent means ± s.e.m. of n = 5 plaques per condition and replicate (g,h). An unpaired t-test was applied (****P < 0.0001). See Supplementary Table 1 for exact statistics.

To determine whether activation of this pathway is critical for cell-to-cell virus spread and cell motility, inhibitors of EGFR (Iressa (also known as gefitinib)), Ras (salirasib), Raf (sorafenib), MEK1/2 (U0126) and FAK (PF573228) were assessed for their ability to retard plaque formation (Fig. 2d). Inhibitors of EGFR, Ras and Raf reduced plaque size by ≥50.0% and inhibitors of MEK and FAK by 33.3% (Fig. 2e). Reduced plaque size with EGFR and MEK1/2 inhibitors is in accordance with previous findings29,30. As these inhibitors may disrupt various stages of the virus life cycle, resulting in attenuated virus spread, their effect on WR mature virion, EEV, CEV and actin tail formation was assessed. No major defect in mature virion production was seen in the presence of any inhibitor tested, whereas minor defects in EEV formation were seen upon Raf (1.6-fold) and MEK1/2 (1.4-fold) inhibition (Supplementary Fig. 2e,f). Most pronounced, inhibition of EGFR resulted in a 1.6-fold decrease in CEVs and 1.6-fold decrease in actin tail formation, and inhibition of FAK reduced actin tail formation by 2.8-fold (Supplementary Fig. 2g,h). Owing to their effect on virus production and/or actin tail formation, inhibitors of EGFR, MEK and FAK were directly assessed for their effect on VACV-induced cell motility by live-cell imaging and single-cell tracking (Fig. 2f and Supplementary Video 2). Despite their respective defects in virus formation, both radial velocity and directional migration efficiency of infected cells were impaired by more than fourfold upon inhibition of EGFR and nearly twofold upon inhibition of MEK and FAK (Fig. 2g,h). Collectively, these results indicate that activation of the EGFR–Ras–Raf–MEK–FAK signalling pathway by VGF is critical for VACV-induced cell motility and virus spread.

VACV-mediated cell motility was first described using a classical scratch assay, in which infection of an entire monolayer facilitated cell crawling into the wound1. To investigate this process during the formation of a plaque, cell monolayers were scratched and single-cell tracking was performed on plaques adjacent to wounds (Fig. 3a and Supplementary Video 3). Vector field analysis showed that motile cells within a plaque that approach a wound had a high propensity to halt or turn and move parallel to the scratch, whereas cells that did not encounter a wound tended to maintain their kinetically synchronized outward movement (Fig. 3a). This phenotype led us to hypothesize that VGF expressed in early infected cells at the leading edge of a plaque might establish a chemotactic gradient towards which virus-producing cells crawl. If this were the case, with no cells in the wound and therefore a lack of VGF signal to follow, motile virus-producing cells would either halt or follow their closest early infected neighbour. To test the chemoattractant activity of VGF, transwell migration assays were employed. Neither soluble VGF nor VGF-secreting infected cells were capable of directing chemotaxis of ΔVGF-infected cells (Supplementary Fig. 3a,b).

Fig. 3: ADAM10-mediated VGF release triggers cell motility in a paracrine manner.
figure 3

a, Analysis of directional motility by live-cell imaging, single-cell tracking and vector field analysis during plaque formation adjacent to wounds. b, Spatial analysis of VGF expression within a plaque using immunofluorescence. c, Supernatant and cell transfer assays to investigate paracrine-mediated and juxtacrine-mediated activation of EGFR signalling by VGF. d, Metalloprotease inhibitors of ADAM10 (GI254023X (GI)) and ADAM10/17 (GW280264X (GW)) prevent VGF shedding and VGF paracrine signalling activity. e, RNA interference-mediated silencing of ADAM10, but not ADAM17, prevents VGF shedding and VGF paracrine signalling activity. KD, knockdown. N.C., AllStars negative control siRNA. f, Single-cell tracking of VACV plaque formation in the presence of the ADAM10 inhibitor (GI254023X) (24–48 hpi). Tracks are colour coded by time (hpi). g,h, The radial velocity (g) and directional migration efficiency (h) of cells migrating from the centre of plaques in f. i,j, 24-h mature virion (MV; i) and EEV (j) yields from dimethylsulfoxide (DMSO)-treated or GI254023X-treated WR-infected cells. k,l, CEVs (k) and actin tails (l) per cell during WR infections in the presence of GI254023X for 10 h. Data represent three or more biological replicates (al). Images are representative of three biological replicates (af). Scale bars, 500 μm (a,b,f). Bars represent means ± s.d. (i,j) or means ± s.e.m. (g,h) of n = 5 plaques per condition and replicate. Lines represent the means of 15–20 cells per condition and replicate (k,l). A paired (i,j) or unpaired (g,h,k,l) t-test was applied (****P < 0.0001, NS, not significant). See Supplementary Table 1 for exact statistics.

As VGF seemed to harbour no chemoattractant activity, we asked which cells, in the context of a VACV plaque, were expressing VGF. Immunofluorescence staining of WR or ΔVGF plaques showed that VGF was expressed only in infected cells at the leading edge where cell motility is ongoing and not in cells at the plaque centre where motility has ceased (Fig. 3b). We reasoned that the expression of VGF in early infected cells at the leading edge of plaques serves to spatially and temporally regulate infected cell motility. Consistent with this notion, the activation pattern of EGFR within VACV plaques, a hallmark of VGF-mediated cell motility, mirrors VGF expression29.

Cellular growth factors exert their effects by activating their cognate receptors on cells from which they are produced (autocrine), on direct neighbouring cells (juxtacrine) or on cells in the vicinity (paracrine)31. VGF is an early gene product of VACV infection produced between 2 and 8 hpi as a 25-kDa precursor and is released from cells as a 22-kDa soluble ligand14 (Supplementary Fig. 3c–e). To test whether VGF could induce cell motility in an autocrine manner, individual uninfected, WR-infected or ΔVGF-infected cells were sorted into single wells, monitored by live-cell imaging and subjected to single-cell tracking (Supplementary Fig. 3f). Infection with WR significantly increased the radial velocity and directional migration efficiency of single cells, whereas cells infected with ΔVGF showed no radial velocity increase and only a minor increase in directional migration efficiency, over uninfected cells (Supplementary Fig. 3g). This indicates that VGF secreted from an infected cell can trigger its motility. As both autocrine signalling and paracrine signalling are directed by soluble ligands, this result is consistent with supernatant transfer experiments showing that secreted VGF can activate EGFR in a paracrine manner15.

However, as juxtacrine signalling is mediated by membrane-bound ligand, and infected cells within a viral plaque are in close proximity, we sought to determine the relative contribution of juxtacrine and paracrine signalling in VGF-mediated EGFR activation. For this, WR-infected or ΔVGF-infected supernatants and cells were collected and separately transferred to naive cells, which were then probed for activation of EGFR and MEK1/2. Although control ΔVGF supernatants and cells showed neither paracrine nor juxtacrine activity, WR-infected supernatants showed potent paracrine activity and WR-infected cells showed no juxtacrine activity (Fig. 3c). These results indicate that VGF can only promote motility of infected cells in an autocrine and paracrine manner once it is shed from infected cells.

With one exception, shedding of EGFR ligands from the cell surface is mediated by a disintegrin and metalloprotease 10 or 17 (ADAM10/ADAM17)32. To assess whether VGF is released from cells by one of these, cells were infected in the presence of ADAM inhibitors: GW280264X or GI254023X. GW280264X is an inhibitor of ADAM10 and ADAM17, whereas GI254023X inhibits ADAM10 with 100-fold more potency33. Both GI254023X and GW280264X effectively blocked the shedding of VGF into the supernatant, resulting in its accumulation on infected cells (Fig. 3d, left). Transfer experiments showed that supernatants from GI254023X-treated or GW280264X-treated infected cells could not trigger EGFR and MEK1/2 signalling (Fig. 3d, right). RNA interference-mediated silencing of ADAM10 and ADAM17, followed by supernatant transfer experiments, confirmed that ADAM10, and not ADAM17, was largely responsible for VGF shedding and subsequent activation of EGFR and MEK1/2 signalling (Fig. 3e). These results identify ADAM10 as the metalloprotease responsible for shedding of VGF. In line with this, the addition of GI254023X resulted in reduced plaque size (Supplementary Fig. 4a,b), which correlated with shortened tracks, a 1.3-fold reduction in cell velocity and a 3.9-fold reduction in directional migration efficiency (Fig. 3f–h and Supplementary Video 4). No defects in WR mature virion, EEV, CEV or actin tail formation were seen in the presence of GI254023X (Fig. 3i–l and Supplementary Fig. 4c). Thus, ADAM10-mediated shedding of VGF is a pre-requisite for its autocrine or paracrine activation of EGFR signalling, virus-induced cell motility and virus spread.

An in vivo hallmark of poxvirus infection is the formation of cutaneous lesions. As plaque formation may serve as a two-dimensional (2D) in vitro surrogate for this, the role of VGF in VACV lesion formation was addressed. Mice ear pinnae were epicutaneously infected with WR or ΔVGF viruses and lesions were visualized using multiphoton microscopy. By 6 days post-infection, WR had formed large multi-foci lesions, whereas ΔVGF lesions were less numerous and 3.8-fold smaller (Fig. 4a,c). Analysis of lesion cross-sections revealed that the depth of ΔVGF lesions was also reduced by 3.7-fold (Fig. 4b,d). That ΔVGF displays no major defects in virus production (Fig. 1b–d) strongly suggests that the reduction in lesion size is due to the observed attenuation of virus-induced cell motility.

Fig. 4: VGF is required for lesion formation in vivo.
figure 4

a, Multiphoton microscopy of WR and ΔVGF lesions in mice ear pinnae 6 days post-infection. Infected cells (green), collagen and blood vessels are shown. b, Confocal imaging of WR and ΔVGF lesions in cross-section. Infected cells (green) and nuclei (blue) are shown. c,d, Quantification of lesion widths (c) and depths (d) from a and b, respectively. e, Model of VGF-mediated VACV-induced cell motility (refer to text for details). EV, enveloped virion. Representative data from two mice/virus in biological triplicates (a,b). Lines represent the means of 10–15 lesions per condition from 5 mice/virus (3 cross-sections, 2 frontal sections). An unpaired t-test was applied (****P < 0.0001, **P < 0.01) (c,d). See Supplementary Table 1 for exact statistics.

The cell-to-cell spread of poxvirus infection is a complex process relying on the production of extracellular virus, actin tail formation, cell motility and superinfection repulsion2,34,35. Blocking of any one of these reduces VACVs ability to spread cell to cell, results in a concomitant reduction in plaque size and attenuation in mice. That loss of each of these processes results in a diminution, but not complete ablation, suggests that each provides a relative contribution to VACVs ability to spread. Future studies should be aimed at determining this relative contribution, whether these processes are additive or cooperative and whether their contribution differs depending on whether virus spread is occurring in a cell monolayer (2D) or in an infected host tissue (3D).

Here, we show that the VACV EGF homologue, VGF, contributes to this process through activation of EGFR-mediated cell motility. As illustrated in Fig. 4e, our data indicate that, within the context of a VACV plaque, VGF expression is relegated to newly infected cells (purple) at the leading edge. The short window of VGF expression, from 2 to 8 hpi, in combination with ADAM10-mediated shedding seem to spatially and temporally restrict VGF paracrine activity. This invokes a model in which autocrine or paracrine activation of EGFR–MEK–ERK–FAK signalling promotes rapid and efficient motility of infected cells in the vicinity, including those infected for longer than 6 h and in the process of producing virus (blue cells). Thus, it reasons that VGF-mediated cell motility increases the efficiency of spread by assuring increased cell-to-cell contact between virus-producing cells and their more distant uninfected neighbours. This model is supported by the appearance of kinetically synchronized waves of infected cell motility away from the plaque origin. Our previous simulations of VACV plaque formation indicate that the average diameter of uninfected and infected cells is 43.5 ± 13.4 µm and 34.7 ± 14.6 µm, respectively36. Based on radial velocity measurements in WR plaques (0.3 µm min−1), infected cells can traverse on average 10 uninfected cells within 24 h. By providing new molecular understanding of VACV-induced cell migration, using virus mutants and inhibiting EGFR signalling at multiple levels, we show that infected cell motility is directly linked to the ability of VACV to spread. The induction of cell motility through hijacking of the EGFR–MAPK–FAK signalling axis by VGF is reminiscent of metastatic transformation37. Perhaps motile infected cells harbouring cell-associated virus rather than free virus particles38 act as mediators of tissue-to-tissue virus spread. This would explain the ability of EGFR-targeting chemotherapeutic compounds to prevent poxvirus spread in vitro29 and block VACV-mediated lethality in vivo39, thus encouraging the potential repurposing of other chemotherapeutics as antiviral agents.



HeLa (ATCC CCL-2) and BSC40 (from P. Traktman, Medical University of South Carolina, Charleston, SC, USA) cells were maintained in DMEM supplemented with 10% FBS, 2 mM GlutaMAX and penicillin–streptomycin at 37 °C and 5% CO2. BSC40 medium was supplemented with 100 μM non-essential amino acids and 1 mM sodium pyruvate (Thermo Fisher Scientific). HeLa (ATCC CCL-2) cells have been authenticated by the American Type Culture Collection (ATCC) and BSC40 cells have not been authenticated. Both cell lines were tested regularly and remained mycoplasma free throughout this study.

Inhibitors and antibodies

Salirasib (5 μM; SML1166, Sigma), sorafenib (5 μM; 8705) and U0126 (20 μM; 9903; Cell Signaling Technology), gefitinib (10 μM; Iressa, S1025, Selleck-Chem), GI254023X (20 μM; 3995) and PF573228 (10 μM; 3239, Tocris) and GW280264X (20 μM; AOB3632, Aobious) were all used at the indicated concentrations. Monoclonal anti-B5 (VMC-20) was a kind gift of G. H. Cohen and R. J. Eisenberg (University of Pennsylvania, Philadelphia, PA, USA)40. The phospho-EGFR antibody sampler kit (9922, containing 4267, 3777 and 2237), phospho-ERK1/2 pathway sampler kit (9911, containing 9427, 9154, 4370 and 11989), FAK antibody sampler kit (9330, containing 3281, 3284, 8556 and 13009), α-tubulin DM1A antibody (3873) and anti-rabbit and anti-mouse IgG horseradish peroxidase-linked antibodies (7074 and 7076) were purchased from Cell Signaling Technology and used at 1:1,000. Anti-ADAM10 (ab124695) and anti-ADAM17 (ab2051) were purchased from Abcam and used at 1:1,000. IRDye-coupled secondary antibodies were purchased from Licor and used at 1:10,000. Alexa Fluor-conjugated secondary antibodies and phalloidin were purchased from Invitrogen/Thermo Fisher Scientific and used at 1:1,000 or 1:100, respectively.

Construction of recombinant viruses

Recombinant VACVs were generated using homologous recombination as previously described41. Briefly, VACV-infected BSC40 cells were transfected with linearized plasmid 4 hpi and harvested 48 hpi. Plaques were selected by fluorescence through four rounds. Final plaques were sequenced to confirm the correct insertion of the construct. The VGF double knockout virus, vSC20, was a generous gift from B. Moss (National Institute of Allergy and Infectious Diseases (NIAID); National Institutes of Health (NIH), Bethesda, MD, USA)18. WR lacZ NP-SIINFEKL-eGFP (WR NP-eGFP) and vSC20 lacZ NP-SIINFEKL-eGFP (ΔVGF NP-eGFP) viruses were constructed by inserting pNP-SIINFEKL-eGFP21 into the TK locus of WR or vSC20. The WR ΔF11 NP-eGFP and ΔVGF/ΔF11 NP-eGFP viruses were generated by replacing amino acids 1–320 of the F11L locus with mCherry under the control of a late VACV promoter.

Mature virion/EEV 24-h yield

Confluent BSC40 cells in 6-well plates were infected with virus at multiplicity of infection (MOI) 10 and fed with 1 ml full medium. At 24 hpi, the supernatant containing EEVs was collected and cleared of cells by 2× centrifugation at 400g for 10 min. For mature virions, cells were collected by scraping, centrifugation and resuspension in 100 μl 1 mM Tris (pH 9.0), prior to 3× freeze–thaw. The plaque forming units/millilitre (p.f.u. per ml) were determined by crystal violet staining of plaques, 48 hpi after serial dilution on confluent monolayers of BSC40 cells.

CEV/actin tail analysis

BSC40 cells were seeded on fibronectin-coated coverslips. Cells were infected with WR or mutant viruses at MOI 5. For inhibitor treatment, cells were infected with WR mCherry-A4 virus and inhibitors were added at 1 hpi. At 10 hpi, cells were fixed with methanol-free formaldehyde for 20 min before two PBS washes. Fixed cells were blocked with 2% BSA in PBS for 30 min before incubation with anti-B5 (VMC-20; 1:10,000) in PBS (1% BSA) for 1.5 h. After three PBS washes, cells were incubated with anti-mouse Alexa 647 secondary antibody (1:1,000) for 1 h. Cells were then permeabilized with 0.1% Triton in PBS for 20 min prior to staining with phalloidin-Alexa 488 (1:100) and Hoechst (1:10,000). Samples were mounted with Immu-Mount and stacks were acquired using a Leica TCS SPE confocal microscope. CEVs were counted for each cell using the spot detection function of Imaris (version 7.6.5, Bitplane). Identical settings were used for all samples (0.6-μm spot size, quality above 40). Actin tails were counted manually.

Plaque assays

Confluent BSC40 monolayers were infected with 100 p.f.u. After 1 h, cells were fed with DMEM without supplements. For inhibitor experiments, compounds were added at 22 hpi. For assays with GI254023X, the drug was added at 1 hpi and topped up every 12 h. Plaque diameters were measured manually using Fiji. One hundred plaques per condition or experiment were measured.

Live-cell imaging of VACV plaques

BSC40 cells infected with VACV were imaged from 24–48 hpi in an environmental chamber maintained at 37 °C and 5% CO2. Two different time-lapse microscope settings were used: a Zeiss Axiovert 200 M with a Hamamatsu Orca AG camera and a Nikon Ti inverted microscope with a Nikon DS-Qi2 high-sensitivity scientific CMOS camera. A ×4 objective was used for the Zeiss Axiovert and a ×10 objective was used for the Nikon Ti. For each condition, 10–15 plaques were imaged every 10 min. For plaques next to wounds, each well was scratched at 22 hpi with a 200-μl pipette tip and the well was washed twice with DMEM before imaging as above.

Single-cell tracking

To obtain the single infected cell trajectories, live-cell time-lapse movies of plaque formation were analysed using TrackMate version v3.5.1 or lower42. To detect cells, the Laplacian of Gaussian filter was applied, and estimated blob diameter and threshold were fitted for the optimal detection of cells in each data set. For tracking, the linear assignment problem tracker was used with a maximum frame-to-frame linking of 30 μm, maximum gap-closing distance of 30 μm and maximum gap-closing frame gap of 1.

Measurement of radial velocity and directional migration efficiency in plaques and single-cell experiments

To determine the radial velocity of cells within plaques, the coordinate system zero was defined at the plaque origin. Coordinates of the plaque origin were obtained using the algorithm described in ref. 43. The coordinates of each cell track were then converted from a Cartesian to a polar coordinates system. The radial component of each cell track was then used to compute an average radial velocity of the cells within a plaque using equation (1).

$$RV{\mathrm{ = }}\frac{{d < \Delta \rho > }}{{dt}},$$

where RV is radial velocity, ∆ρ is the maximum radial component of trajectory, d/dt is a temporal derivative and t is time from experiment start.

Following the radial velocity measurement, the directional migration efficiency of infected cells within plaques was determined using equation (2).

$$DME{\mathrm{ = }}\omega _{\Delta \rho }\left( {1 - \omega _{\Delta \theta }} \right),$$

where DME is the directional migration efficiency, ωρ is the minmax normalized radial velocity and ωθ is the maximum range of the normalized angular polar component of each track relative to the origin. Values were averaged to obtain a representative value for each plaque.

To measure radial velocity and directional migration efficiency in single-cell experiments, live-cell, time-lapse phase-contrast images were collected. Images were processed by pixel classification using a Random Forest44 machine learning algorithm in Weka software45 to ensure compatibility with TrackMate42. Similar to cell tracking in plaques, TrackMate with a spot size parameter of 80 pixels was used. The radial velocity and directional migration efficiency of single-cell tracks were computed using equations (1) and (2), respectively. To overcome undersampling bias in radial velocity and directional migration efficiency measurements associated with downscaling from plaques to single cells, we performed a Monte-Carlo-based bootstrapping46 resampling of the experimental data with 100,000 permutations. Reciprocal hypothesis testing was performed using permutation tests.

Vector field analysis of directional cell motility

To determine the general directional tendency of motile infected cells, the spatiotemporal tensor of live-cell, time-lapse tracks of plaque formation were fitted to a vector field. For this, the vector field k-means clustering algorithm47 was applied to the trajectory data. To ensure background-to-signal separation, before application of the algorithm, the cell tracking data were appended with synthetic background trajectories of constant radial velocity, distance and direction.

VGF antibody production

Anti-VGF was produced by GenScript USA Inc. The peptide DSGNAIETTSPEITC, previously used by Chang et al.14, corresponding to residues 1–14 of the cleaved VGF including an additional cysteine at the carboxy terminus, was conjugated to KLH. The peptide–KLH conjugate was used to immunize one rabbit, and the anti-VGF antibody was affinity purified after three immunizations.

Expression and purification of recombinant VGF/EGF

The sequence of cleaved VGF was amplified from VACV genomic DNA and inserted into the pQE30 vector, resulting in 6×His-VGF. The sequence of fully cleaved EGF was codon optimized for expression in bacteria, ordered as gblock from IDT and inserted into the pQE30 vector using Gibson cloning, resulting in 6×His-EGF. Transformed XL1 blue bacteria were inoculated and grown overnight with antibiotics. LB medium (500 ml) was inoculated with the cultures and grown at 30 °C. When the absorbance at 600 nm (OD600) reached 0.4–0.6, gene expression was induced with 1 mM isopropyl-β-d-thiogalactoside (IPTG). After 4 h, cells were harvested by centrifugation at 4,000 r.p.m. for 15 min at 4 °C. Cell pellets were resuspended in 30 ml suspension buffer (500 mM NaCl, 50 mM Na2HPO4, 10 mM imidazole and 0.1% Tween-20, pH 8.0) and sonicated on ice (15 pulses of 15 s). Crude extracts were filtered through a 0.22-μm filter. Protein was purified on Qiagen Ni-NTA agarose columns. Briefly, columns were washed with five column volumes of suspension buffer, followed by a 3-ml elution with 125 mM imidazole and a 12-ml elution with 250 mM imidazole. Fractions of 1 ml were collected and analysed by SDS–PAGE. The most concentrated fractions were pooled and dialysed overnight in suspension buffer without imidazole, using a membrane with a molecular weight cut-off of 3.5 kDa. Samples were then adjusted to 25% glycerol, aliquoted and stored at −80 °C. A bicinchoninic acid (BCA) assay was performed to determine protein concentration.

Western blot

For each sample, confluent HeLa cells (10-cm dish) were maintained without serum overnight prior to infection or stimulation. To harvest, cells were transferred to ice, washed with ice-cold PBS and 150 μl lysis buffer (25 mM Tris-HCl, pH 7.5, 100 mM NaCl, 1% Triton X-100, 0.5% NP-40, 1 mM EDTA and 1 mM EGTA) with protease/phosphatase inhibitor (5872, NEB) was added. Cells were harvested by scraping and lysates were incubated on ice for 20 min prior to centrifugation at 20,000g (4 °C). Samples were boiled in reducing sample buffer, run on 4–12% Bis-Tris gels and transferred to nitrocellulose. Membranes were incubated with primary antibodies (1:1,000) overnight in either 5% BSA or 5% milk in TBST according to the manufacturer’s recommendations. IRDye-coupled secondary antibodies were used for detection on a Licor Odyssey or horseradish peroxidase-coupled secondary antibodies with horseradish peroxidase substrate (Merck Millipore) were used for detection on a GE Healthcare ImageQuant.

EGFR PathScan activation array

Confluent HeLa cells were starved overnight prior to infection with WR or ΔVGF at MOI 5 for 6 h or treatment with EGF or VGF (2.5 μg ml−1) for 1 h. Subsequently, cells were put on ice, washed with cold PBS and lysed with 200 μl lysis buffer (9803, Cell Signaling). Supernatants were collected by scraping, cleared by centrifugation and the EGFR PathScan activation array (12622, Cell Signaling) was coated with a 1:2 dilution of lysate in array buffer. The array was performed according to the manufacturer’s protocol and a GE Healthcare ImageQuant was used for detection. For quantification, images were masked to retain only spot-related intensities within the image. The total intensity per spot was quantified and compared to array controls.

Supernatant/cell transfer experiment

For supernatant/cell transfer experiments, HeLa cells were kept in serum-free media overnight, then infected in the presence or absence of indicated inhibitors. At 6 hpi, supernatants were removed and cleared of cells by centrifugation (2 × 300g). Cells were washed, gently scraped in 1 ml medium and separated by gentle pipetting. Cells and supernatant were then diluted in fresh serum-free medium as indicated and added to starved HeLa cells for 1 h. For cell transfer, GI254023X was added during activation to prevent further cleavage of VGF. Cells were harvested for western blotting as described above. Input supernatant was concentrated using a 10-kDa cut-off Amicon centrifugal filter before analysis. Input supernatant and cell samples were adjusted to equal volumes and immunoblot analyses were performed.

Short interfering RNA silencing of ADAM10/ADAM17

HeLa cells were reversed transfected with scrambled, ADAM10 or ADAM17 short interfering RNA (siRNA) at a final concentration of 20 nM. Seventy-two-hour post-transfection cells were infected with WR at MOI 2 for 4 h and supernatant transfer experiments were performed as described above. ON-TARGETplus SMARTpool siRNA for human ADAM10 (L-004503-00-0005) and ADAM17 (L-003453-00-0005) were purchased from Dharmacon. The sequences for ADAM10 siRNA were CAUCUGACCCUAAACCAAA, CAAGGGAAGGAAUAUGUAA, GAACUAUGGGUCUCAUGUA and CGAGAGAGUUAUCAAAUGG. The sequences for ADAM17 siRNA were GAAGAACACGUGUAAAUUA, GCACAAAGAAUUAUGGUAA, UAUGGGAACUCUUGGAUUA and GGAAAUAUGUCAUGUAUCC. AllStars negative control siRNA was purchased from Qiagen.

VGF immunofluorescence

WR and ΔVGF plaques were fixed at 36 h with 4% formaldehyde and stained with anti-VGF (1:500), anti-rabbit Alexa 594 secondary antibody (1:1,000) and Hoechst (1:10,000). Samples were imaged with a ×10 objective on the Nikon TI microscope.

Transwell migration assay

BSC40 and HeLa cells were seeded in the top well of 6.5-mm transwells with 8-μm pore size polyester membrane inserts (Corning). The bottom well was filled with 650 μl serum-free DMEM and cells were incubated overnight. Cells in the top chamber were infected with ΔVGF NP-eGFP (MOI 1) and inserts were transferred into wells containing 650 μl serum-free DMEM, 5% FBS, 2 μg ml−1 VGF, uninfected cells or WR-infected cells (MOI 1). After 20 h, the top side of the transwell inserts were swabbed with a cotton bud to remove remaining cells. Wells were then fixed with 4% formaldehyde and stained with Hoechst to visualize cells that had migrated through the transwell membrane. The membrane was excised with a scalpel and mounted in Vectashield. Membranes were imaged using the ×10 objective of the Nikon Ti inverted microscope. Cells were counted using the spot detection function of the Fiji plugin TrackMate.

Single-cell infection, sorting and imaging

HeLa cells were left uninfected or were infected with WR LacZ NP SIINFEKL (MOI 1). At 1 hpi, cells were washed with PBS and detached from plates with 200 µl of 0.25% trypsin per well. For sorting, 400 µl of medium (5% FCS) was added and individual cells were sorted into single wells of a collagen-coated 384-well plate using the scatter of a 488-nm laser of a BD FACSAriaTM III (Becton Dickinson). Individual wells were imaged at 37 °C and 5% CO2 on an inverted microscope (TI-E Eclipse, Nikon) using a ×10 DIC, plan apo objective and a ×0.7 TV adapter in front of an Orca flash 4.0 camera (Hamamatsu). The field of view encompassed one well and images were acquired every 10 min for 27 h.

Inoculation of mice

Pathogen-free C57BL/6 and albino B6(Cg)-Tyrc-2J/J mice (used for multiphoton imaging) were acquired from Taconic or from The Jackson Laboratory. Adult female mice (6–12-weeks old) were used in all experiments. Mice were housed in pathogen-free conditions (including murine norovirus, murine parvovirus and murine hepatitis virus) and maintained on standard rodent chow and water supplied ad libitum. All animal experiments were conducted in accordance with the Animal Welfare Act and the recommendations in the Guide for the Care and Use of Laboratory Animals of the NIH. NIAID animal facilities have full accreditation from the Association for Assessment and Accreditation of Laboratory Animal Care and are PHS-assured (assurance number: A4149-01). All animal procedures were approved by the NIAID Animal Care and Use Committee. For epicutaneous infection, approximately 20 µl rVV stock solution (1 × 108 p.f.u. per ml) was placed on the ear skin and gently poked five times per ear with a bifurcated needle (similar to the human vaccination protocol). Virus stocks were grown and titred in-house.

Intravital multiphoton microscopy imaging

Multiphoton microscopy imaging was performed as described48. Briefly, images were acquired on an upright Leica SP5 MP microscope (Leica Microsystems) equipped with two Mai Tai Ti:Sapphire lasers (Spectra Physics) and 10-Watt pumps or on an inverted Leica TCS SP8 MP microscope (Leica Microsystems). Ears were immobilized on an imaging platform and bathed in warm saline. All images were acquired using a ×20 objective (NA: 1.00). Emitted fluorescence was collected with a four-channel non-descanned detector. For blue/green channels, wavelength separation was accomplished with a dichroic mirror at 495 nm, followed by emission filters of 460/50-nm bandpass and 525/50-nm bandpass. Second harmonic generation of collagen in the dermis is shown in blue according to convention. Tile scans of eight fields of view were collected, covering approximately 2.8 mm of the ear. Sequences of image stacks were transformed into maximum intensity projections using Imaris software (Bitplane).

Confocal imaging of frozen sections

Ears were fixed in periodate-lysine-paraformaldehyde overnight as reported49, cryoprotected in 15% sucrose, embedded in OCT medium (Electron Microscopy Sciences) and frozen in dry-ice-cooled isopentane. Cross-sections (18 µm) were cut on a Leica cryostat (Leica Microsystems), blocked with 5% goat or donkey serum, then stained with a combination of DAPI (Perkin Elmer), CD11b-Alexa 647 (clone M1/70, eBioscience) and VGF peptide antibody, followed by anti-rabbit Alexa 568 (Invitrogen). Images were acquired on an inverted SP5 confocal microscope (Leica Microsystems) using identical PMT (photomultiplier tube) and laser settings. Lesion depth was scored manually using Imaris software.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Code availability

All custom codes used in this study will be made available upon reasonable request.