Electron Tomography and Simulation of Baculovirus Actin Comet Tails Support a Tethered Filament Model of Pathogen Propulsion

Electron tomography reveals the structural organization of actin comet tails generated by a baculovirus, providing an understanding of how this pathogen hijacks host machinery to propel itself between cells.


Introduction
The seminal finding of Tilney and Portnoy [1] that Listeria monocytogenes exploits the actin cytoskeleton of infected cells to invade neighboring cells opened a new chapter in motile processes based on actin. Major progress in understanding how L. monocytogenes uses actin to move came from the identification of the Arp2/3 complex [2] as the downstream target promoting actin polymerization [3] and from the subsequent elucidation of the minimal protein cocktail required for propulsion in vitro [4]. Essential in the in vitro motility mix was actin, the Arp2/3 complex, ADF/cofilin, an actin capping protein and an activator of the Arp2/3 complex, ActA on L. monocytogenes, or N-WASP on plastic beads [4,5]. Subsequent studies have revealed a growing list of bacterial and viral pathogens that exploit the actin-based motile machinery of infected cells by mimicking or recruiting N-WASP to activate the Arp2/3 complex [6,7] or other actin nucleators [8]. To gain more insight into the mechanism of propulsion, several efforts have been directed at establishing the structure of the actin comet tail induced by pathogens. Electron microscopy of plastic-embedded, negatively stained L. monocytogenes comet tails, or critical point dried tails on ActA-coated beads [1,[9][10][11] showed actin filaments more or less randomly oriented, but the high density of filaments precluded definition of their spatial organization by conventional 2D imaging.
Using electron tomography we recently showed that lamellipodia networks comprise subsets of actin filaments linked by branch junctions structurally homologous to those formed from the Arp2/ 3 complex and actin in vitro [12]. With electron tomography practical limits are set by the thickness of the sample and L. monocytogenes are more than twice as thick as lamellipodia, making them less suitable for structural analysis. The timely finding that a baculovirus species, Autographa californica multiple nucleopolyhedrovirus, just 50 nm in diameter, moves on an actin comet tail in infected cells and bears a minor capsid protein p78/83 that directly activates the Arp2/3 complex [13][14][15] offered an ideal object for resolving comet architecture. In the present report we provide the first structure, to our knowledge, of an actin comet tail driving a pathogen and use this new information to re-evaluate alternative models (reviewed in [16,17]) of pathogen propulsion by actin.

Baculoviruses Generate Actin Comets in Vertebrate Cells
Lepidopteran cells, the natural host of baculovirus, proved too thick for analysis by electron tomography. However, we found that baculovirus was readily taken up by thinner vertebrate cells of mouse (B16 melanoma, NIH 3T3 fibroblasts), fish (CAR fibroblasts), and human (HeLa) origin. Figure 1A and Movie S1 show a B16 melanoma cell expressing GFP-actin that was infected with baculovirus tagged with mCherry [14]. The actin comet tails formed were indistinguishable from those seen in the native host cells, moved with a velocity of up to 50 mm/min, and contained typical tail components ( Figures 1B and S1) [11]. From intensity measurements of tails in cells transfected with different combinations of GFP/mCherry-tagged proteins, VASP and ArpC5 colocalized with actin and capping protein b and cofilin trailed behind the actin label ( Figure 1B and Figure S1).

Electron Tomography of Intracellular Comet Tails Induced by Baculovirus
By applying cryo-electron tomography to intact cells and cytoskeletons, we previously showed that actin networks in lamellipodia are preserved using a cytoskeleton preparation procedure involving simultaneous extraction with Triton X-100 and fixation in glutaraldehyde [18]. This procedure leads to a marked improvement in filament contrast and resolution as compared to un-extracted cells. Using the same routine with baculovirus infected cells, we could show that the fluorescence intensity profile and length of actin comet tails were retained by this extraction/fixation method ( Figure S2).
The ultrastructural organization of baculovirus actin comet tails was resolved in cytoskeletons of infected cells using two complementary approaches involving embedment either in negative stain ( Figure 2) or in vitreous ice ( Figure 3). A section of an electron tomogram of a negatively stained baculovirus comet tail in the cytoplasm of a fish fibroblast cytoskeleton and the model derived by filament tracking are shown in Figure 2 (see also Movie S2). As shown, the comet tail filaments form a fishbone-like array generated through the frequent formation of branch junctions (highlighted in the insets in Figure 2A) with an average angle of

Author Summary
Several bacteria and viruses hijack the motile machinery of cells they invade to generate networks of actin filaments (comet tails) to propel themselves from one cell to another. A proper understanding of the mechanism of propulsion has so far been hampered by a lack of information about the structure of the machinery. Using electron tomography we present here the three-dimensional structure of actin comet tails propelling a baculovirus, the smallest pathogen known to recruit the actin nano-machinery. We show that baculovirus is propelled by a fishbone-like array of actin filaments constructed from subsets linked by branch junctions, with an average of four filaments pushing the virus by their fast polymerizing ends at any one time. Using a stochastic mathematical model we have simulated comet tail organization as well as the tracks adopted by baculovirus inside cells. The simulations support a model of baculovirus propulsion in which the actin filaments are continuously tethered to the virus surface as they grow, branch, and push. Since larger pathogens like Listeria, Shigella, and Vaccinia virus generate comet tails exhibiting the same general morphology and components as those of baculovirus, the basic mechanism of their propulsion is likely a scaled up version of the one described here. Overview of GFP-actin expressing B16 melanoma cell that was infected with baculoviruses tagged with mCherry. (B) Relative distribution of actin and associated proteins along baculovirus comet tails. Tail length averaged 3.861.8 mm (s.d., n = 111 individual tails) and was normalized to 1.0. Curves were fitted using a centered sixth order polynomial function. Shown are 95% confidence bands for each dataset. For details, see Figure S1. Bar Figure 2B in different colors, each linked by a separate group of branch junctions (red spots in Figure 2B,C). Cross-correlation analysis [19,20] of the negatively stained comet tail filaments with reference models of the actin helix ( Figure S3) showed that the fast polymerizing, plus ends were directed forwards. The plus ends of the filaments analyzed are highlighted by black spheres in Figure 2C.
The filament array belonging to the actin comets was mainly separated in Z from the actin filaments of the cytoplasm (see additionally Figures S4 and S5), which were typically longer and randomly oriented. Intermingling of the two filament populations also occurred, but it was relatively straightforward to assign filaments to the comets according to additional criteria: 1, termination within the comet core; and 2, extension of filaments from the core towards the virus.
To circumvent the partial collapse of comet tails in negatively stained preparations, we analyzed them also by cryo-electron tomography. Optimal contrast for filament tracking in actin comets was obtained by imaging them in cytoskeletons of infected cells within cytoplasmic regions that lay over holes in the supporting film ( Figure 3 and Movie S3). Branch junctions with the typical angle were readily identified (insets, Figure 3A and Movie S4) with the comet tail filaments oriented at an average angle of 50.9623.7u (s.d., n = 485) to the core axis. The number of filaments transecting a core region of the comet tail corresponding to the cross-section of the virus (highlighted in yellow in Figure 3B) at different axial positions is shown in Figure 3C. These data indicated an average number of filaments involved in pushing of 3.961.4 (s.d., n = 252, range [1][2][3][4][5][6][7][8]. Taken together, the data from three negative stain and three cryo-electron tomograms gave an average filament length of 121.1699.9 nm (s.d., n = 961) ( Figure 3D shows a projection of the 3D model of the comet tail together with (in translucent grey) the endogenous actin filaments of the cytoskeleton that were mainly located above and below the comet (see also Movie S3 and Figure S6).

Architecture of Baculovirus Actin Comets Formed in Vitro
By exposing the p78/83 capsid protein of budded baculoviruses from cell supernatants using detergent ( Figure 4A,B) we could reconstitute comet tail formation in vitro ( [4], Figures 4 and 5). In the motility cocktail (actin, Arp2/3 complex, gelsolin, ADF/ cofilin, and profilin) filament length was influenced most strongly by variations in gelsolin concentration ( Figure 4F,G). Addition of VASP to the protein cocktail substantially increased the frequency of particles bearing actin comets. In all cases actin filaments were nucleated at one end of the virus particle and with appropriate concentrations of cofactors, the in vitro comet tails resembled closely those observed in vivo. Cryo-electron tomography was used to determine the three-dimensional organization of the in vitro comet tails ( Figure 5 and Movie S5). As for the in vivo comet tails, we observed a fishbone-like array of filaments linked by branch junctions (inset and red spheres, Figure 5) with four to five filaments in close contact with the rear of the virus ( Figure 5C).
Cross-correlation analysis [19] of individual filaments in the cryotomograms ( Figure S7) showed that the plus ends were oriented towards the virus, consistent with the analysis of the in situ comet tails embedded in a negative stain.

Mathematical Model of Baculovirus Propulsion
Using a quasi-static approach in a two-dimensional model (see Text S1) we sought to mimic the intracellular movement of baculovirus and the structural features of the comet tails. We assumed several forces acting on the virus: Cytoplasmic friction, Brownian forces, and the forces exerted by the actin filaments. In the model, actin filaments are simulated as stiff, immobile rods, which can push the virus due to polymerization. We initially considered three variations of the model ( Figure 6A-C). In the ''tethered'' case ( Figure 6A) actin filaments proximal to the virus surface are continuously tethered, both during pushing and when they lag behind. When lagging behind they exert a pulling force described by a spring connection ( Figure 6A, 2). In the ''tethered during branching'' case ( Figure 6B) tethering to the virus occurs only during branching events ( Figure 6B, 2) and is described by a spring connection to the Arp2/3 complex. In the ''untethered'' case ( Figure 6C) no tethering takes place between the filament plus ends or the branch points with the virus surface. In the latter two cases, the untethered, lagging filaments (Figures 6B,C and 7) may convert into pushing or capped filaments. The primary difference between the three cases is then the degree of pulling forces exerted on the virus by the filament network. In each case, Arp2/3dependent branches (marked in red in Figure 6A-C) are induced by the p78/83 nucleation promoting factor at the rear of the pathogen. Filaments that move laterally off the rear of the virus continue to polymerize ( Figure 6A-C, 4) until they are capped ( Figure 6A-C, 5), and those at the rear of the tail de-polymerize from their minus ends ( Figure 6A-C, 6).
To compare the electron microscope data with the simulations, we quantified (1) the angles that filaments subtended with the axis of the comet tails ( Figure 6D,E), (2) the length distribution of the filaments ( Figure 6K), and (3) the number of filaments joined together in subsets ( Figure 2B). By an appropriate choice of input parameters (Table S1) we could mimic the measured parameters only for the ''tethered'' case ( Figure 6F,J,K; summarized in Tables S2 and S3). In the ''tethered'' situation, the distribution of angles subtended to the trajectory (52.6626.8u, s.d., n = 49,321) compared closely to the value obtained from three cryo-tomograms (50.9623.7u, s.d., n = 485), whereas the ''tethered during branching'' and ''untethered'' models produced tails with filaments emanating at significantly higher angles (64.8642.6u, s.d., n = 58,895, and 59.7641.1u, s.d., n = 51,168, respectively; Figure 6F,G,H,J). Pushing the virus along a straight path tends to be dynamically unstable due to its elongated shape [21]. To obtain in the model a persistent and relatively straight movement, an average of 3-4 filaments evenly distributed over the rear of the virus were required at any one time. Due to statistical variation, this condition was not satisfied by a single subset of filaments. Instead, regions on the virus rear, where filaments became sparse, required the engagement of new mother filaments, through de novo nucleation or recruitment from the cytoplasm that then initiated new filament subsets. In the ''tethered'' case, the simulations generated an average of 11 filaments per subset, comparable to the value determined from the tomograms, whereas for the ''untethered'' case a new mother filament was required after every 2-3 branching events, leading to an average of 3 filaments per subset.
The straightness of tracks adopted by the virus in cells was compared to the simulations by measuring the angular divergence of 200 nm segments along the comet tail axis (Figure 7). Experimental data were derived from 19 comet tails imaged by conventional 2D electron microscopy ( Figure 7C,D). Again, the range of observed turning angles fitted best the values obtained with the ''tethered''  Figure 7A,F and Movies S1 and S6), whereas tracks were highly tortuous in the ''untethered'' simulations ( Figure 7B,F and Movie S6) as also previously reported for simulated trajectories of L. monocytogenes [22]. Values of the angular divergence of trajectories from tails acquired by transmission electron microscopy (22.0624.4u, s.d., n = 255 angles in 19 individual tails) and ''tethered'' simulations (17.4614.8u, s.d., n = 1,449) were significantly smaller than those from ''untethered'' simulations (51.2638.0u, s.d., n = 2,681) and those from simulations for the case of ''tethered during branching'' (44.4633.2u, s.d., n = 2,337) ( Figure 7F, Table S3).
Although the electron tomography data indicated branching only towards the virus surface, we also considered a ''tethered'' model in which branching could also occur away from the virus, leading to a subpopulation of filaments unable to engage in pushing. Surprisingly, simulations of this ''tethered random branching'' case produced relatively straight tracks ( Figure 6I) with normal deviations (24.5620.9, s.d., n = 1,430, Figure 7F) but failed to reproduce the typical fishbone-like array of actin filaments. This was reflected in significantly higher filament to trajectory angles as well as in a 40% decrease in the number of filaments per subset. In addition the number of productive,

Discussion
The resolution achievable by electron tomography is dependent on a number of factors, not least the thickness of the specimen. Baculovirus is at least 20-fold smaller than L. monocytogenes, with a diameter only six times the thickness of an actin filament, making it an ideal object for electron tomography. For cryo-electron tomography, we found that filament tracking and polarity analysis was only possible in energy-filtered tomograms obtained by imaging comet tails in cytoskeletons over holes in perforated films, to reduce background noise. As we show, useful complementary structural information was also provided from electron tomography of negatively stained samples due to higher image contrast and filament detail. Taking into account these different factors we have been able to provide the first 3D structure, to our knowledge, of an actin comet tail propelling a pathogen. Our demonstration of a minimal actin filament cassette for propulsion provides a new framework for analyzing and modeling the underlying mechanisms of actin force production by Arp2/3-complex-dependent actin assemblies.
The actin-based propulsion of loads (from L. monocytogenes to beads) has already attracted the interest of several groups [16,17,23,24]. In the case of the baculovirus tail with its limited number of filaments, we can exclude the squeezing mechanism proposed in the active gel model [23]. Mogilner and Oster [25] and Dickinson and Purich [26] provide useful alternative models of how actin filaments may push at the pathogen surface disregarding actin tail architecture. Alberts and Odell's [27] comprehensive in silico model simulates details of the actin filament organization but without knowledge of the actual structure. We chose a tractable modeling approach (see Text S1) using a minimal number of parameters (Table S1) to define the essential set of mechanisms required to simulate the observed filament organization and the motion of the virus in vivo.
By comparing alternative mathematical simulations with our experimental data, we may propose key mechanistic features of comet tail propulsion. In particular, comet tail organization and the trajectories of the virus were best explained by assuming continuous attachment of actin filament to the virus surface, supporting the hypothesis that tethering, possibly by VASP ( Figure 1B) [28], which is present in the baculovirus tail, and pulling forces between the tail and the pathogen [17,29] are needed to stabilize viral movement. A critical parameter influencing the modeling outcome was Brownian motion (see Text S1). Without it, there was little difference between the simulations for tethered and un-tethered filaments. Since the virus must experience external jostling forces in the cytoplasm, we consider the model featuring Brownian motion and tethering to mimic more closely the situation in vivo. An interesting feature of the model was the prediction that nucleation of filaments on the virus is required not only to initiate movement, but also intermittently to correct sharp turns and a stochastic, regional paucity of pushing filaments. As a consequence, sequential subsets of branched filaments are created, consistent with the tomography data and reminiscent of the organization of actin filaments in lamellipodia [12]. The fishbone-like array of filaments in the comet tail could only be simulated if branching was biased toward the virus surface with the result that all new filaments are involved in pushing. The viral nucleation promoting factor, p78/83, therefore seems to restrict branching from tethered mother filaments in a biased way, through conformational constraints, or possibly as a result of local bending of actin filaments at the filament-virus interface [30]. The delay of capping protein incorporation in the comet tail corresponding to around 1 s ( Figure 1B) is consistent with the observed filament length distribution ranging from 10 nm to 800 nm (Figures 2, 3, and 6K). The later incorporation of cofilin ( Figure 1B) is likewise consistent with its preferential binding to ADP-F-actin [31].
Fishbone-like patterns of actin have been described in comet tails induced in cells by vaccinia virus [32] as well as by L. monocytogenes in in vitro motility assays [5] at low gelsolin concentrations. The principles of organization of actin filaments in the baculovirus actin comet tail shown here thus likely apply generally to comets formed on pathogens that hijack the Arp2/3 complex machinery for propulsion and invasion [6].

Virus Preparation
The supernatant from Sf9 cells infected with either wild-type or mCherry-tagged baculovirus [14] was harvested after 3-5 d, precleared at 2,500 g for 5 min, filtered through a membrane with 0.45 mm pore size, then centrifuged at 18,000 g for 1 h and resuspended in 150 ml of baculovirus buffer (10 mM HEPES, 0.5 M KCl, pH 7.4) to avoid clustering of the viruses. For in vitro assays, de-enveloped viruses were prepared by adding 2% Triton X-100 (Fluka) to the virus suspension and incubating the mixture on a shaker for 15 min at 25uC. The suspension was then centrifuged at 500 rpm to remove large debris and the supernatant centrifuged at 20,000 g for 1 h. The resulting pellet was resuspended in 50 ml baculovirus buffer supplemented with 1% BSA and used as stock in the motility assay.  (1). The virus surface is continuously tethered to the barbed ends (A, 2) or to the Arp2/3 complex during branching (B, 2) or not at all (C). Tethering is modeled by a spring connection. Branches (3) are initiated by Arp2/3 complex (red) recruited to the plus ends of actin filaments at the virus surface. Elongation of filaments that can no longer push (4) continues until they are capped (5). Filaments at the rear of the tail become de-branched and depolymerize from their minus ends (6). In (B) and (C) filaments lagging behind are not tethered (7). In Vitro Motility Assay The in vitro assay was basically performed as described [4]. Briefly, ADF (3.7 mM), Profilin (2.5 mM), Gelsolin (25-200 nM), Arp2/3 (75 nM), G-Actin (7.6 mM), VASP (100 nM), and the purified, de-enveloped baculovirus were mixed in X-Buffer (10 mM HEPES, 100 mM KCl, 1 mM MgCl 2 , 0.1 mM CaCl 2 , pH 7.8) supplemented with 1% BSA, 2 mM ATP, 4 mM MgCl 2 , and 6.7 mM DTT. The motility assay was incubated for different times as a drop on a grid in a wet chamber at room temperature.

Cell Culture
B16 mouse melanoma cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM) (Sigma-Aldrich) with 10% fetal calf serum (Thermo Fisher Scientific), 1% penicillin/ streptomycin, and 1% L-glutamine. B16 cells were transfected as subconfluent monolayer cultures in 30 mm Petri dishes using 2 mg of DNA, 100 ml Optimem (Invitrogen), and 6 ml Fugene (Roche). Before addition to the cells overnight, the transfection mix was incubated for 20 min. The next day, the medium was changed and the transfected cells plated on coverslips coated with 25 mg/ml laminin (Sigma) for fluorescence microscopy. Sf9 cells were kept in Grace's insect medium with 10% fetal bovine serum, 1% penicillin/streptomycin, and 1% L-glutamine. Goldfish fin fibroblasts (line CAR, No. CCL71, ATCC) were maintained in Basal Eagle's Medium (Sigma-Aldrich) supplemented with 1% nonessential amino acids, 2.5% HEPES (Invitrogen), 1% penicillin/ streptomycin, 1% L-glutamine, and 12% fetal bovine serum (Thermo Fisher Scientific) at 27uC. For transient transfection, subconfluent monolayer cultures on 30 mm Petri dishes were used. The transfection mixture was prepared as follows: 2 mg of DNA and 12 ml of Superfect lipofection agent (Qiagen) were mixed in 200 ml Optimem (Invitrogen). After 15 min incubation at room temperature, a further 5% serum-containing medium with transfection mixture was added to the cells for 5 h. The cells were then washed and returned to normal medium. Transfected cells were replated after 24 h on 15 mm coverslips coated with human fibronectin (Roche) at a concentration of 50 mg/ml.
For virus uptake 150 ml baculovirus stock was added to 2 ml of a subconfluent cell culture growing on coated coverslips or electron microscopy grids and incubated at 27uC or 37uC depending on the cell type, and imaging was performed after 1-3 h.

Live Cell Imaging and Analysis
Coverslips carrying baculovirus-infected cells were mounted in an open chamber on a heating platform (Harvard Instruments) in prewarmed medium. Intracellular virus-induced tails were usually observed after about 20 min postinfection. The samples were observed on an inverted Zeiss Observer epifluorescence microscope equipped with a Perkin Elmer UltraView spinning disc system (ProSync2). For kymograph analysis, Fiji together with a MatLab-based script were used. Statistical analysis was performed using Prism. Intensity values of tail components were normalized to tail length based on the linear relationship between tail length and speed ( Figure S1E).

Electron Microscopy
For negative staining electron microscopy, cells were grown on Formvar-coated 200 mesh hexagonal nickel grids (Agar Scientific) or 135 mesh NHF15-A gold Finder grids (Maxtaform) in standard medium and allowed to spread overnight. After spreading, the samples were washed with cytoskeleton buffer (10 mM MES buffer, 150 mM NaCl, 5 mM EGTA, 5 mM glucose, 5 mM MgCl 2 , pH 6.8), fixed and extracted with 0.5% Triton X-100 and 0.25% glutaraldehyde in cytoskeleton buffer for 1 min, and kept in 2% glutaraldehyde with 1 mg/ml phalloidin in cytoskeleton buffer at 4uC. For routine inspection in an 80 kV transmission electron microscope (Morgagni, FEI), grids were stained with 70 ml 2% SST including 1 mg/ml phalloidin. For electron tomography, grids were stained with 70 ml 6%-8% sodium silicotungstate (SST) including 1 mg/ml phalloidin. For cryo-electron tomography, virus-infected cells were plated in growth medium onto Quantifoil R1/4, R1.2/1.3, and R2/2 perforated carbon films on 200 mesh Au grids, and allowed to spread for 6 h. The cells were fixed as for negative staining and the grids subsequently transferred to forceps in a grid-plunging device (Leica EM GP), supplemented with 4 ml of medium and 10 nm BSA-saturated colloidal gold [12]. During mounting and blotting, the grid was held in a chamber with controlled humidity and temperature (95% and 22uC) and was blotted automatically for 1.5-2 s with humidified Whatman No. 1 filter paper applied to the backside of the grid to avoid any contact with the cells. Samples were frozen by plunging into liquid ethane cooled to 80 K by liquid nitrogen. In vitro polymerized actin tails were prepared for electron microcopy as follows. For negative staining the incubation mix was fixed after different times on the 200 mesh Au grids by injecting 2% glutaraldehyde (final concentration 0.5%) for 4 min and stained with 70 ml 2% sodium silicotungstate (SST). For cryo-electron microscopy, samples were incubated on Quantifoil R3.5/1 perforated carbon films on 200 mesh Au grids for 45 min, rinsed with 70 ml X-Buffer, followed by 70 ml X-Buffer with 1:10 BSA-colloid gold, and without fixation or fixed as above, frozen as described above with blotting times between 1.2 and 1.6 s. Tilt series were acquired on an FEI Tecnai F30 Helium (Polara) microscope, operated at 300 kV, and cooled to approximately 80 K. Automated acquisition of tilt series was driven by SerialEM 3.x. Normally, the tilt range was 260u to +60u using the Saxton tilt scheme based on 1u increments for negative stain and 2u increments for cryo-samples from 0u tilt, at a nominal defocus value of 25 mm to 27 mm for negative stain and 28 mm to 212 mm for cryo-samples. For negative stain samples, two tilt series around orthogonal axes were recorded on a Gatan UltraScan 4000 CCD camera at on-camera magnifications with filaments tethered, untethered, tethered during branching, and tethered with random branching. Measured versus tethered n.s. (p.0.9999); measured versus tethered during branching **** (p,0.0001); measured versus untethered **** (p,0.0001); measured versus tethered random branching n.s. (p = 0.0905); tethered versus tethered during branching **** (p,0.0001); tethered versus untethered **** (p,0.0001); tethered versus tethered random branching **** (p,0.0001). Data nonparametric, by Kruskal-Wallis test. Bars, (A-C) 500 nm, (D, E) 100 nm. doi:10.1371/journal.pbio.1001765.g007 typically from 23,0006 to 59,0006. For frozen hydrated samples, zero-loss images were recorded on a Gatan MSC 2 k camera on a GIF 2002 using a slit width of 15 eV. The total electron dose for the cryo-tilt series was less than 130 electrons per Å 2 , and the primary on-screen magnification was from 20,0006 to 29,0006.

Image Processing and Analysis
Re-projections from the tilt series with gold particles as fiducials for alignment and contrast transfer function corrections were performed using IMOD software [38]. A typical tomogram comprised a z-stack of 60-120 sections of 0.4-1.5 nm each. The polarity of actin filaments in tomograms of negatively stained and frozen samples was determined using a filament straightening protocol and cross-correlation analysis as described elsewhere [20]. Filaments were manually tracked using IMOD, as previously described [12].
The identity of branch junctions was based on filament tracking analysis, performed by three investigators independently. Filaments were tracked individually and branch junctions identified as blunt intersections of filament pairs subtending an acute angle in the range of 60-90u, as deduced for branch junctions in lamellipodia [12]. An example of the tracking analysis in the cryo-tomograms is shown in Movie S4.

Mathematical Modeling
Mathematical modeling was performed using MatLab software. Details of the model are provided in Text S1.  Figure 6D). The projections serve to illustrate the spatial segregation between the comet tail filaments (green) and the filaments of the host cell (grey). Movie S3 Cryo-electron tomogram of comet tail in vivo. Scan of electron tomogram of comet tail in vitreous ice from Figure 3A and the model derived by manual tracking in Figure 3B,D. Actin filaments, first in green, are highlighted as subsets in different colors in the middle of the movie. Branch points are in red. Actin filaments belonging to the cytoskeleton of the cell are indicated in grey, as is also a microtubule crossing the field. At the end of the movie, the pushing core of the tail is highlighted in yellow.

(MOV)
Movie S4 Example of the filament tracking procedure in cryoelectron tomograms. Movie shows Z-scan through same tomogram as Movie S3, but with the manual tracking of several filament segments in the second part. Red spheres indicate the branch junctions and blue spheres individual tracking points. (MOV) Movie S5 Cryo-electron tomogram of comet tail assembled in vitro. Scan of electron tomogram of in vitro comet tail in vitreous ice from Figure 5A   Text S1 Details of mathematical simulation. (DOCX)