Actin-based motility is sufficient for bacterial membrane protrusion formation and host cell uptake


*For correspondence. E-mail; Tel. (+1) 650–725–7968; Fax (+1) 650–723–6783.


Shigella flexneri replicates in the cytoplasm of host cells, where it nucleates host cell actin filaments at one pole of the bacterial cell to form a ‘comet tail’ that propels the bacterium through the host's cytoplasm. To determine whether the ability to move by actin-based motility is sufficient for subsequent formation of membrane-bound protrusions and intercellular spread, we conferred the ability to nucleate actin on a heterologous bacterium, Escherichia coli. Previous work has shown that IcsA (VirG), the molecule that is necessary and sufficient for actin nucleation and actin-based motility, is distributed in a unipolar fashion on the surface of S. flexneri. Maintenance of the unipolar distribution of IcsA depends on both the S. flexneri outer membrane protease IcsP (SopA) and the structure of the lipopolysaccharide (LPS) in the outer membrane. We co-expressed IcsA and IcsP in two strains of E. coli that differed in their LPS structures. The E. coli were engineered to invade host cells by expression of invasin from Yersinia pseudotuberculosis and to escape the phagosome by incubation in purified listeriolysin O (LLO) from Listeria monocytogenes. All E. coli strains expressing IcsA replicated in host cell cytoplasm and moved by actin-based motility. Actin-based motility alone was sufficient for the formation of membrane protrusions and uptake by recipient host cells. The presence of IcsP and an elaborate LPS structure combined to enhance the ability of E. coli to form protrusions at the same frequency as S. flexneri, quantitatively reconstituting this step in pathogen intercellular spread in a heterologous organism. The frequency of membrane protrusion formation across all strains tested correlates with the efficiency of unidirectional actin-based movement, but not with bacterial speed.


Shigella flexneri, the causative agent of bacillary dysentery, is a Gram-negative facultative intracellular bacterial pathogen that invades and spreads through epithelial cells lining the human large bowel (Formal et al., 1958; 1966). The bacterium directs its entry into non-phagocytic host cells by injecting effector molecules into the cytoplasm via a type III secretion apparatus encoded on the large (180–230 kbp) virulence plasmid that is required for full virulence (Parsot and Sansonetti, 1996). S. flexneri then rapidly escapes the initial phagosome by a type III protein secretion-dependent mechanism and enters the cytoplasm (Sansonetti et al., 1986; High et al., 1992). Within the host cell cytoplasm, the bacteria continue to grow and divide and move rapidly through the cell by nucleating actin filaments at one pole of the bacterium, leaving behind actin-rich ‘comet tails’ in the cytoplasm (Ogawa et al., 1968; Bernardini et al., 1989; Pal et al., 1989). Shigella spread from an infected cell to uninfected neighbours through membrane-bound protrusions (Kadurugamuwa et al., 1991; Prevost et al., 1992; Sansonetti et al., 1994). Bacterial actin-based motility is necessary for intercellular spread in tissue culture monolayers and for virulence in infected animals (Sansonetti et al., 1991).

Although actin-based motility is necessary for intercellular spread, it is not known whether the force generated by this mechanism is sufficient for protrusion formation and/or intercellular spread. Alternatively, S. flexneri may interact specifically with molecular receptors on the inside of the host cell plasma membrane. To date, four classes of S. flexneri mutants capable of primary cellular invasion in non-phagocytic host cells but incapable of intercellular spread have been reported. The first class has a defect in intracellular multiplication because of a mutation in a metabolic gene or in a gene involved in completing septation (Okada et al., 1991a; Suzuki et al., 1994; MacSiomoin et al., 1996; Hong et al., 1998). The second class consists of mutants affected in their ability to polymerize actin or to move by actin-based motility (Makino et al., 1986; Bernardini et al., 1989; Pal et al., 1989; Nakata et al., 1993; Rathman et al., 2000). A third class contains mutants that are unable to escape into the cytoplasm of the adjacent epithelial cell (Allaoui et al., 1992; Suzuki et al., 1994; Page et al., 1999; Schuch et al., 1999). A fourth class is defective for intercellular spread via unknown mechanisms (Rathman et al., 2000); these may include mutants unable to form membrane protrusions and/or mediate protrusion uptake into adjacent cells.

The virulence plasmid-encoded protein product of the icsA (virG) gene is necessary and sufficient for actin-based motility (Makino et al., 1986; Bernardini et al., 1989; Lett et al., 1989; Goldberg and Theriot, 1995; Kocks et al., 1995). IcsA, a 120 kDa outer membrane protein, is located at a single pole on the surface of the bacterium, at the junction with the actin tail (Goldberg et al., 1993). Polar localization of IcsA requires polar protein secretion through a pathway that is common to the four Gram-negative bacteria S. flexneri, Escherichia coli, Yersinia pseudotuberculosis and Salmonella typhimurium (Sandlin and Maurelli, 1999; Steinhauer et al., 1999; Robbins et al., 2001). Once IcsA is delivered to the bacterial pole, it diffuses slowly in the outer membrane to generate a steady-state gradient. The polar gradient of surface IcsA expression is sharpened by the presence of the outer membrane serine protease IcsP (SopA), also encoded on the virulence plasmid, which cleaves IcsA at the Arg-758–Arg-759 bond (Fukuda et al., 1995; Egile et al., 1997; Shere et al., 1997; Robbins et al., 2001).

Furthermore, variations in S. flexneri's lipopolysaccharide (LPS) molecules affect the polar distribution of surface IcsA, in that more elaborate LPS structures yield a sharper gradient of IcsA (Sandlin et al., 1995; 1996; Van den Bosch et al., 1997; Robbins et al., 2001). Thus, S. flexneri strains that contain mutations in LPS synthesis loci galU, rfa, rfb and rfc and the O-antigen chain length regulator rol are less efficient for actin-based motility and intercellular spread than wild-type strains (Okada et al., 1991a,b; Rajakumar et al., 1994; Sandlin et al., 1995; 1996; Hong and Payne, 1997; Van den Bosch et al., 1997; Rathman et al., 2000).

Intercellular spread of S. flexneri occurs when bacteria produce a membrane protrusion into a neighbouring cell. Transmission electron micrographs have captured a few intermediate steps that show membrane-bound protrusions of S. flexneri invading neighbouring cells (Kadurugamuwa et al., 1991; Sansonetti et al., 1994; Schuch et al., 1999). No protein has yet been identified in S. flexneri that is necessary for mediating the intermediate step in cell-to-cell spread or protrusion uptake.

This study was undertaken to determine whether the force generated by actin-based motility is sufficient to form membrane protrusions and induce their uptake by neighbouring uninfected cells in the absence of additional virulence factors. We expressed IcsA in two strains of E. coli that differed in their LPS structure and in the presence or absence of IcsP (and thus differed in the relative polarity of bacterial surface IcsA expression). We found that actin-based motility was sufficient for membrane protrusion formation and that protrusions into neighbouring cells were pinched off in a host cell-mediated endogenous process consistent with the previously described process of paracytophagy (Robbins et al., 1999). The efficiency of this reconstituted intercellular spread is highly correlated with the unipolarity of IcsA on the bacterial surface and with unidirectional actin-based motility.


Escherichia coli expressing IcsA replicates in the cytoplasm and forms polymerized actin tails in HeLa cells

We initially tested whether E. coli expressing the S. flexneri protein IcsA would move by actin-based motility in the cytoplasm of host cells. We expressed IcsA in two strains of E. coli that differed in their LPS structure and were deficient for an outer membrane protease (OmpT) that cleaves IcsA non-specifically. These E. coli strains also expressed the Y. pseudotuberculosis invasion protein, invasin, which is sufficient for pathogen-mediated endocytosis in non-phagocytic host cells (Isberg and Falkow, 1985). In order to test whether E. coli expressing IcsA can polymerize actin in host cells, the bacteria must gain access to the cytoplasm. To achieve this, E. coli bacteria were incubated with purified six-His-tagged listeriolysin O (HisLLO), a pore-forming haemolysin from Listeria monocytogenes that is necessary and sufficient for its escape from the host vacuolar compartment (Bielecki et al., 1990). This purified protein is capable of rescuing an LLO-deficient mutant in trans (Gedde et al., 2000).

HeLa cells were infected with the engineered E. coli strains, and the escape of internalized E. coli from the host primary vacuoles was evaluated microscopically. Two criteria indicated that the engineered E. coli could escape and replicate in the cytoplasm: (i) many bacteria exhibited polymerized actin at their surface, which is only available in the cytosolic compartment (Fig. 1); and (ii) the number of bacteria present per host cell increased over time (Fig. 2). The E. coli that had entered the cytosol replicated during the 5 h assay with a doubling time of ≈ 1 h (Fig. 2), which is slightly slower than the 40 min doubling time of wild-type S. flexneri in the host cytosol (Sansonetti et al., 1986). E. coli that had not been treated with HisLLO did not polymerize actin and did not replicate inside host cells (not shown).

Figure 1.

E. coli expressing icsA form actin tails in the cytosol of HeLa cells. HeLa cells were infected with bacteria for 5 h, fixed, and stained with rhodamine–phalloidin to show actin filaments. The first column is the fluorescence image, the second column the phase-contrast image, and the third column is the overlay of the fluorescence and phase-contrast images. Each row represents a different strain.

A–C. S. flexneri 2457T.

D–F. E. coli K-12 expressing icsA (DMM1) (arrows point to a bacterium with a uniform actin cloud).

G–I. E. coli O8 containing icsA and icsP (DMM4) (arrowheads point to a bacterium with an asymmetric actin cloud).

Scale bars = 2 µm. See Supplementary material.

Figure 2.

E. coli expressing icsA replicate in the cytosol of HeLa cells. HeLa cells were infected with DMM1, E. coli K-12 containing icsA, fixed at indicated times and stained with rhodamine–phalloidin. The number of bacteria that had polymerized actin at their surface and were therefore in the cytosol were counted per infected cell. The average number of cytosolic bacteria per host cell was calculated by repeating the experiment three times (N = 40 for 2 h; N = 100 for 3 h; N = 63 for 4 h; N = 105 for 5 h; N is the total number of infected cells counted. Error bars are SD).

IcsA unipolarity is necessary for efficient actin-based motility of S. flexneri and E. coli in HeLa cells

Wild-type S. flexneri expresses IcsA at one pole of the bacterium and thus forms polar actin tails in the cytosol of host cells (Ogawa et al., 1968; Bernardini et al., 1989; Pal et al., 1989; Goldberg et al., 1993; Goldberg and Theriot, 1995). Several factors contribute to the polar localization of IcsA, including the rate of diffusion of outer membrane IcsA, which is affected by the O-antigen polysaccharide chain length of the LPS molecules (Sandlin et al., 1995; 1996; Sandlin and Maurelli, 1999; Robbins et al., 2001), and the specific cleavage of IcsA in the outer membrane by the protease IcsP (SopA) (d'Hauteville et al., 1996; Egile et al., 1997; Steinhauer et al., 1999). We wished to examine the relative contributions of these two factors to comet tail formation by E. coli in HeLa cells.

As reported previously (Sandlin et al., 1996), wild-type S. flexneri polymerized actin at one pole of the bacterium, whereas the S. flexneri LPS mutant, BS520 (which does not make any O-antigen), polymerized actin in a less polar fashion than wild-type bacteria. We quantified this difference by calculating the ratio of bacteria with comet tails (moving bacteria) to the total number of actin-associated bacteria. By this criterion, wild-type S. flexneri formed actin tails fourfold more frequently than the S. flexneri LPS mutant BS520 (Fig. 3).

Figure 3.

The frequency of actin tails produced by E. coli increases in the presence of icsP or O8 side-chains on the LPS. Host cells were infected for 4 h, fixed and stained with rhodamine–phalloidin. The frequency of actin tails was calculated as the number of bacteria with actin tails divided by the total number of bacteria with polymerized actin at their surface (clouds + tails). S. flexneri 2457T (N = 52, n = 391); S. flexneri BS520 (N = 45, n = 541); S. flexneri BS520 soaked in HisLLO (N = 80, n = 590); E. coli K-12 expressing icsA, DMM1 (N = 111, n = 771); E. coli K-12 expressing icsA and icsP, DMM2 (N = 100, n = 719); E. coli O8 expressing icsA, DMM3 (N = 60, n = 236); E. coli O8 containing icsA and icsP, DMM4 (N = 90, n = 568). N is number of infected cells; n is the number of bacteria. The average and standard errors were calculated from three different experiments.

An E. coli K-12 strain expressing icsA secretes IcsA at one pole of the bacterium, but the absence of repeating O side-chains on the LPS allows for a relatively rapid diffusion of IcsA in the outer membrane in a fashion analogous to that in the S. flexneri LPS mutant strain BS520 (Goldberg and Theriot, 1995; Robbins et al., 2001). In contrast, IcsA localization in the E. coli strain expressing LPS of the O8 serotype (ATM379) remains unipolar (Sandlin and Maurelli, 1999). Comparing the percentage of tails formed by E. coli K-12 expressing IcsA (DMM1) versus ATM379 expressing IcsA (DMM3), we found that the E. coli with the O8 antigen (DMM3) formed tails at a frequency that was moderately, although significantly, greater than that of their E. coli K-12 counterparts (Fig. 3; P = 0.01 for DMM1 versus DMM3). Thus, LPS elaboration enhances tail formation in both S. flexneri and E. coli, although this effect is more pronounced in S. flexneri.

The addition of the outer membrane protease, IcsP, has been shown to sharpen the polar gradient of surface IcsA in both S. flexneri (Steinhauer et al., 1999) and the Gram-negative bacterium Y. pseudotuberculosis (Robbins et al., 2001). The isogenic E. coli K-12 strain containing icsP as well as icsA (DMM2) had a larger percentage of bacteria with polymerized actin at one pole or on one side of the bacterial surface than did DMM1. This increase in the number of bacteria with a polarized polymerization of actin also correlated with a reproducibly significant increase (2.5–threefold; P = 0.02 at 4 h) in the frequency of actin tails for the E. coli K-12 strain containing icsA and icsP (DMM2 versus DMM1; Fig. 3). In the serotype O8 background, the addition of icsP in the strain DMM4 did not significantly increase the frequency of tail formation (DMM4 versus DMM3). However, the O8 strain containing icsP as well as icsA (DMM4) did form actin tails more frequently than the relatively non-polar E. coli K-12 strain DMM1 (Fig. 3; P = 0.05 DMM4 versus DMM1). Thus, although the relative contributions to tail formation of LPS structure and protease activity are strain dependent, the efficiency of actin tail formation by engineered E. coli in the cytosol of host cells is clearly affected by the unipolar localization of surface IcsA.

Escherichia coli with unipolar surface distribution of IcsA move with higher average speeds

The speed of bacteria that are moving through the cytosol by actin-based motility can be measured using video microscopy. We compared the average speeds of the E. coli and S. flexneri strains constructed (bacteria moving unidirectionally were used to obtain the average velocities). Wild-type S. flexneri moved with an average speed of 0.19 µm s−1, which is significantly faster than the S. flexneri LPS mutant, BS520, which displays a less polar distribution of surface IcsA (Table 2; twofold faster, P < 0.0001). The E. coli K-12 strain containing icsA (DMM1) did not move significantly faster than BS520, but the addition of icsP (DMM2) increased the average speed moderately, yet significantly (1.2-fold faster; P = 0.03). The E. coli K-12 strain containing icsA and icsP (DMM2) had an average speed that was not significantly different from that of wild-type S. flexneri(Table 2).

Table 2.  Average speeds and movement classes of S. flexneri and E. coli strains expressing IcsA.
Average speed
± SD (µm s−1)
NPercentage moving
spinning on axis
moving in circles
2457T (S. flexneri)2a  0.19 ± 0.05161000025
BS520 (S. flexneri)2a  0.09 ± 0.0316NDNDND 
DMM5 (S. flexneri)5+ 0.29 ± 0.1221NDNDND 
DMM1 (E. coli)K-12+ 0.13 ± 0.083053182966
DMM2 (E. coli)K-12++0.16 ± 0.0752701713100
DMM3 (E. coli)O8+ 0.05 ± 0.021068191331
DMM4 (E. coli)O8++0.17 ± 0.08108215421

Interestingly, the addition of icsP caused an increase in the percentage of bacteria that were moving in a direction perpendicular to their long axis (7% for DMM1 versus 41% for DMM2). More importantly, the addition of icsP not only increased the average speed, but also resulted in a significant increase in the number of bacteria that were moving in a unidirectional manner regardless of whether the bacteria were moving in the same direction as or perpendicular to their long axis (Table 2; 53% for DMM1 versus 70% for DMM2). Thus, the increase in the frequency of tails observed for the E. coli K-12 strain containing icsP(Fig. 3) correlates with an increase in the average speed and in the percentage of bacteria moving unidirectionally.

The E. coli serotype O8 strain containing icsA (DMM3) moved at an average speed that was not significantly different from that of BS520 or DMM1 but, when icsP was added, the average speed of this strain, DMM4, was increased 3.5-fold (P < 0.01) and was very similar to the speed for wild-type S. flexneri(Table 2). This increase in speed was accompanied by an increase in the percentage of moving bacteria that were travelling unidirectionally (Table 2). Thus, there is a correlation between IcsP expression and both increased speed and unidirectional movement in both E. coli strain backgrounds.

Finally, we wanted to address whether additional unidentified plasmid-encoded virulence factors could be involved in efficient S. flexneri actin-based motility. We expressed icsA in a strain of S. flexneri cured of the virulence plasmid (DMM5) and introduced it into host cytoplasm as described above for E. coli strains. We found that the average speed of DMM5 was ≈ 30% faster than the average velocity of wild-type S. flexneri(Table 2, P < 0.01). DMM5 differs from wild-type S. flexneri both in lacking icsP and other virulence plasmid-encoded genes and in having more copies of icsA (25–30 copies per DMM5 cell versus one or two copies per wild-type S. flexneri cell). Overall, we conclude that the speed of both S. flexneri and E. coli in HeLa cells is affected by the co-ordination of both LPS and IcsP, and that the spatial concentration of surface IcsA probably plays a major role in determining the average speed.

Escherichia coli can be engineered to form membrane protrusions as efficiently as wild-type S. flexneri

The ability of S. flexneri to form membrane protrusions is dependent on actin-based motility. So far, we have shown that all E. coli strains expressing icsA, with or without icsP, were able to form actin tails in the cytosol of HeLa cells and move at varying speeds. We next addressed whether the force generated by actin-based motility would allow an E. coli bacterium that came in contact with the plasma membrane to push its way out into a membrane protrusion. We observed by live video microscopy that all E. coli strains containing icsA were in fact able to form protrusions in HeLa cells expressing green fluorescent protein (GFP)–actin. The morphology of the protrusions formed by the E. coli strains was indistinguishable from protrusions formed by wild-type S. flexneri, as determined by transmission electron microscopy (Fig. 4A) and fluorescence microscopy staining for polymerized actin with phalloidin (data not shown). This result indicated that genes other than icsA carried on the S. flexneri virulence plasmid were not necessary for protrusion formation. We confirmed this by infecting HeLa cells with the plasmid-cured S. flexneri strain containing icsA in multicopy (DMM5) and observing the formation of protrusions in live video microscopy and in fixed monolayers (data not shown).

Figure 4.

The frequency of membrane protrusions produced by E. coli increases to wild-type Shigella levels in the presence of icsP and O8 side-chains on the LPS.

A. HeLa cells infected with E. coli K-12 containing icsA and icsP, DMM2, for 5 h were fixed and processed for transmission electron microscopy. The membrane protrusions were morphologically identical to those reported for wild-type Shigella (Kadurugamuwa et al., 1991). Scale bar = 1 µm.

B. HeLa cells on coverslips were infected for 4 h with individual strains of Shigella or E. coli followed by fixation and staining with rhodamine–phalloidin. The frequency of protrusions was calculated by dividing the number of protrusions by the sum of the number of protrusions plus the number of actin tails. Each bar represents the average of three different experiments; the standard errors were calculated for the three experiments (N > 60 and n > 100 for all strains). N is the number of infected cells scored, and n is the number of bacteria with actin polymerized at the surface.

C. The frequency of protrusions correlates with the percentage of bacteria moving unidirectionally. Percentage unidirectional values from Table 2 for 2457T, BS520, DMM1, DMM2, DMM3 and DMM4 were plotted against the protrusion frequencies from (B). The line shown is best fit by least squares. The correlation coefficient, R2, is 0.88.

We quantified the frequency of membrane protrusions present at various times during the course of infection with the E. coli strains and compared the frequencies of protrusion formation by our E. coli strains with the frequencies for wild-type S. flexneri and BS520. At all time points, the frequency of protrusions formed by wild-type S. flexneri was significantly higher than that in the S. flexneri LPS mutant, BS520, and the E. coli K-12 strain containing icsA alone (DMM1; 17- to 30-fold; P < 0.01 for both strains compared with 2457T). The E. coli strains that had increased frequencies of actin tail formation, DMM2 and DMM3, produced significantly more membrane protrusions at 4 and 5 h after infection compared with the S. flexneri LPS mutant (BS520) and the E. coli K-12 strain containing icsA (DMM1). Thus, the presence of icsP, as in strain DMM2, or the presence of an increased number of O side-chains, as in strain DMM3, increased the frequencies of protrusions to 50% of the wild-type S. flexneri frequency. Finally, we combined the presence of the outer membrane protease, IcsP, and the increased O side-chains in E. coli serotype O8 and measured the protrusion frequency of this strain (DMM4) at 4 h. Remarkably, the frequency of protrusions for DMM4 was essentially the same as that for wild-type S. flexneri(Fig. 4B).

We were surprised to find a poor correlation between speed and the ability to form protrusions. For example, DMM3 formed protrusions 3.6-fold more frequently than DMM1 (Fig. 4C; P < 0.03), yet DMM3 moved 2.6-fold more slowly than DMM1. Likewise, although DMM2 moved with only a slightly faster average speed than DMM1, DMM2 formed protrusions at a 6.2-fold higher frequency than DMM1 (Fig. 4B; P < 0.0001). However, we found a strong correlation between the ability to form protrusions and the percentage of E. coli that are moving unidirectionally (Fig. 4C), indicating that consistent direction rather than rapid speed is most critical for a cytoplasmic bacterium moving by actin-based motility to distort the plasma membrane and form protrusions.

Actin-based motility is sufficient for uptake by adjacent HeLa cells

Shigella flexneri and the unrelated Gram-positive pathogen L. monocytogenes both spread from cell to cell by an actin-based motility-dependent mechanism. The first step in intercellular spread is the formation of a membrane-bound protrusion into an adjacent cell. Intercellular protrusions must therefore distend two membranes: one membrane from the donor cell and the other from the recipient cell (Tilney and Portnoy, 1989; Parsot and Sansonetti, 1996; Schuch et al., 1999). The donor cell protrusion is then taken up by the recipient cell, leaving the bacterium enclosed in a double-membrane vacuole. Cell-to-cell spread is completed when the bacteria escape from the double-membrane vacuole into the cytosol of a secondary cell. Bacteria in the secondary cell cytosol continue to grow, nucleate actin and again form protrusions. The type III protein secretion genes contained on the virulence plasmid are required for S. flexneri to escape the secondary vacuole (Page et al., 1999; Schuch et al., 1999). As we did not provide E. coli with any of the S. flexneri genes necessary for secondary vacuole lysis, and the HisLLO initially provided on the surface of bacteria to allow for escape from the primary vacuole was either rapidly diluted or degraded (Geoffroy et al., 1987; Portnoy et al., 1992; Decatur and Portnoy, 2000; Gedde et al., 2000), we expected to find E. coli that had protruded into adjacent cells to be surrounded by a double membrane.

HeLa cells infected with the E. coli K-12 strain containing icsA and icsP for 5 h were examined by transmission electron microscopy. Primary infected cells were observed at low magnifications and were found to contain many cytoplasmic bacteria as expected as a result of replication in the cytoplasm (Fig. 5A). Adjacent cells were examined at higher magnifications for E. coli surrounded by double membranes. We often observed E. coli that had protruded into adjacent cells and were surrounded by the remnants of the donor cell membrane containing polymerized actin, which was then surrounded by the recipient cells' membrane (Fig. 5B).

Figure 5.

Double membranes surround E. coli protrusions into adjacent cells.

A. At low magnification (3000×, bar is 5 µm), HeLa cells infected with E. coli K-12 containing icsA and icsP, DMM2, in the cytoplasm for 5 h were detected frequently.

B. At higher magnification (10 000×, bar is 1 µm), protrusions from primary infected cells into adjacent cells could be observed to be surrounded by two membranes (short arrows); one membrane was from the primary infected cell, and the second membrane was from the plasma membrane being forced into the recipient cell. Several bacteria are in the process of forming a membrane protrusion into a neighbouring cell (B, arrowhead) or into the nucleus (A, asterisk).

Recently, L. monocytogenes cell-to-cell spread was observed by live video microscopy in MDCK cell monolayers (Robbins et al., 1999). These observations showed that a very specific sequence of events occurred, which we were able to replicate with wild-type S. flexneri. Like Listeria, membrane protrusions in recipient cells were initially elongated and ovoid in shape (Fig. 6A, 0 min). Subsequently, the protrusion appeared to shrink and then ‘snap’ to a roughly spherical morphology. At this point, the protrusion appeared to have been released from tension, the proposed point at which the recipient cell pinched off the protrusion (Fig. 6A, 45 min). The secondary vacuole continued to shrink and, within ≈ 20 min, S. flexneri lysed the vacuole, and the GFP–actin that was co-localized with the bacterium disappeared (Fig. 6A, 67 min). An engineered E. coli strain, DMM2, was likewise able to protrude into the adjacent HeLa cell that was not expressing GFP–actin, and the shape of this protrusion was indistinguishable from that of S. flexneri intercellular protrusions (Fig. 6B, 0 min). There was a ‘snap’ followed by a morphological change to a roughly spherical shape similar to that observed for S. flexneri(Fig. 6B, 29 min). However, the GFP–actin remained associated with the bacterium in the recipient cell, and the secondary vacuole did not lyse (Fig. 6B, 63 min). Thus, E. coli moving by actin-based motility generated enough force to spread from cell to cell, and the subsequent uptake of the protrusion in the recipient cell is not dependent on S. flexneri-specific protein(s), although vacuolar lysis is.

Figure 6.

Uptake of E. coli membrane protrusions occurs with kinetics similar to S. flexneri protrusion uptake. HeLa cell infections were followed by live time lapse video microscopy, and the time points for each fluorescence (top) and phase-contrast or DIC frame (bottom) are indicated. An S. flexneri 2457T protrusion (A) and a DMM2 protrusion (B) extending from a GFP–actin-expressing cell into a cell expressing undetectable levels of GFP–actin were followed over time. The protrusion was originally oblong and clearly connected to the donor cell by a slender, actin-containing stalk (A, 0 min; B, 0 min). The protrusion ‘snapped’ (A, 10 min; B, 11 min) and collapsed to a roughly spherical shape (A, 45 min; B, 29 min). The GFP–actin localized with the S. flexneri bacterium (black arrows in A) disappeared by 67 min, whereas the GFP–actin associated with E. coli remained and did not change shape (B, 63 min). Scale bars = 2 µm. The frames shown are representative of three observations of cell-to-cell spread by S. flexneri and four observations using E. coli. See Supplementary material.


Metabolic requirements for bacterial replication in cytoplasm

The replication of S. flexneri in the host cell cytosol is required for its pathogenesis (Sizemore et al., 1995). The mammalian cytoplasm is nutrient limited and strongly reducing (Keese et al., 1999) compared with normal bacterial growth media, so it might be expected that cytoplasmic bacterial pathogens have specific metabolic adaptations (Goebel and Kuhn, 2000). However, the Gram-positive non-pathogenic bacterium Bacillus subtilis can replicate in the cytoplasm of macrophages if the gene encoding listeriolysin O (hlyA) is provided to enable escape from the phagosome (Bielecki et al., 1990), indicating that host cell cytoplasm may be permissive for bacterial growth, at least for some Gram-positive organisms.

Our results show that the cytoplasm of mammalian endothelial and epithelial cells is also permissive for replication of some Gram-negative organisms. The four species of Shigella are closely related to E. coli, sharing > 90% sequence identity by DNA–DNA reassociation analysis (Brenner et al., 1969). When Shigella spp. evolved from E. coli to become pathogens, they not only acquired virulence genes on a plasmid and in the chromosome but also shed chromosomal genes via deletions referred to as ‘black holes’ (Maurelli et al., 1998). If there were a metabolic adaptation allowing S. flexneri but not E. coli to replicate in cytoplasm, it could in principle result from (i) acquisition of a gene or genes on the virulence plasmid; (ii) acquisition of a gene or genes on the chromosome; or (iii) loss of specific genes into ‘black holes’ on the chromosome. Previous work has shown that E. coli K-12 hybrids carrying both the 220 kb virulence plasmid and the purE-linked kcpA chromosomal locus from S. flexneri can replicate intracellularly (Pal et al., 1989), demonstrating that no chromosomal alterations apart from kcpA are required and, therefore, adaptations (ii) and (iii) are unlikely. We have extended this finding by demonstrating that the S. flexneri virulence plasmid is also not required to confer the ability to perform cytoplasmic replication on E. coli, ruling out adaptation (i). In other experiments, we have found that the Gram-negative bacterium Y. pseudotuberculosis can also replicate in the cytosol of HeLa cells and MDCK epithelial cells (unpublished data), suggesting that this metabolic permissiveness may extend to many Gram-negative species.

Polar IcsA is sufficient for unidirectional actin-based motility in living host cells

In addition to intracellular replication, S. flexneri must also spread from cell to cell to be fully pathogenic (Sansonetti et al., 1991). The ability to move by actin-based motility is required for intercellular spread of S. flexneri (Bernardini et al., 1989). Previous reports demonstrating that IcsA from S. flexneri is sufficient to confer actin-based motility on non-pathogenic E. coli have relied exclusively on the reconstitution of motility in concentrated cytoplasmic extracts or in mixtures of pure host proteins, but have not demonstrated motility inside living host cells (Goldberg and Theriot, 1995; Kocks et al., 1995; Loisel et al., 1999). Similar studies have demonstrated that L. monocytogenes ActA protein is sufficient to confer actin-based motility on non-pathogenic Gram-positive species (Kocks et al., 1995; Smith et al., 1995) or on inert polystyrene microspheres (Cameron et al., 1999). This reliance on cytoplasmic extracts as a surrogate for host cells has come about simply because of the technical difficulty of introducing non-pathogenic bacteria into the cytoplasmic compartment of living cells. However, it is clear that studies of actin-based motility in cytoplasmic extracts are not always predictive of events in the host cell cytosol. For example, wild-type S. flexneri is perfectly capable of actin-based motility in host cells, but cannot perform actin-based motility in cytoplasmic extracts under any conditions tested, including conditions that support actin-based motility of E. coli expressing IcsA (Goldberg and Theriot, 1995). Conversely, polystyrene microspheres coated with purified ActA perform normal actin-based motility in cytoplasmic extracts (Cameron et al., 1999), but are never observed to move after introduction into living epithelial cells (J. R. Robbins and J. A. Theriot, unpublished observation).

In this report, we have overcome previous technical limitations using a two-step method to introduce non-pathogenic bacteria into cytoplasm: induction of pathogen-directed endocytosis by expression of invasin from Y. pseudotuberculosis followed by vacuolar escape mediated by listeriolysin O from L. monocytogenes. Using this method, we have confirmed that IcsA expression is indeed sufficient to confer actin-based motility on E. coli in living host cells, although we found significant differences among strains with respect to the efficiency of movement initiation. Previous work on both S. flexneri and L. monocytogenes actin-based motility has shown that cytoplasmic bacteria initially nucleate an actin-rich ‘cloud’, which surrounds all or most of the bacterium but is not associated with movement. Movement is initiated when the cloud becomes polarized (Bernardini et al., 1989; Tilney and Portnoy, 1989; Kadurugamuwa et al., 1991; Prevost et al., 1992). We have found that efficient initiation of movement, as measured by the tail–cloud ratio, is enhanced on E. coli expressing IcsA when either (i) core LPS structure is elaborated to include multiple copies of an O antigen; or (ii) the protease IcsP is co-expressed with IcsA, although these two modifications do not appear to act additively (Fig. 3). Both these alterations in the outer membrane also serve to sharpen the gradient of IcsA expression, either in S. flexneri (Sandlin et al., 1995; 1996; Sandlin and Maurelli, 1999; Steinhauer et al., 1999) or in heterologous species including E. coli and Y. pseudotuberculosis (Robbins et al., 2001). Taken together, these findings indicate that tightly regulated polar expression of IcsA is required for the efficient conversion of actin clouds into actin tails and for the initiation of actin-based movement. This conclusion also appears to hold true for ActA from L. monocytogenes; either non-pathogenic bacteria (Smith et al., 1995) or polystyrene beads (Cameron et al., 1999) with a polarized ActA distribution initiate movement at a greatly enhanced frequency compared with similar particles with a non-polarized distribution.

We also examined possible correlations between two other quantitative parameters describing actin-based motility and IcsA polarity as governed by both IcsP expression and LPS structure. These other parameters are speed and movement directionality. Surprisingly, speed does not correlate well with polarity, with IcsP expression or with LPS structure. For example, on S. flexneri, removal of O antigen is associated with a twofold decrease in speed (2457T versus BS530), whereas on E. coli, the removal of O antigen is associated with a 2.5-fold increase in speed in the absence of IcsP (DMM3 versus DMM1) or no change in speed in the presence of IcsP (DMM4 versus DMM2) (Table 2). Likewise, the removal of IcsP can be associated with either a substantial speed increase (2457T versus DMM5 and DMM3 versus DMM4) or, in another strain, with only a slight speed increase (DMM1 versus DMM2). We do not yet know what governs bacterial speed.

Another parameter of actin-based motility is the direction of movement, for example whether the bacteria move in circular versus unidirectional tracks. Wild-type S. flexneri exhibited exclusively unidirectional movement, whereas the various E. coli strains all showed some tendency to move in circles or, in extreme cases, to spin like propellers on a central axis (Table 2). Among the E. coli strains, we observed increases in the frequency of unidirectional movement associated with both the introduction of IcsP (DMM1 versus DMM2 and DMM3 versus DMM4) and the addition of O antigens onto LPS (DMM1 versus DMM3 and DMM2 versus DMM4). For movement directionality, the contributions of IcsP and elaborate LPS appear to be additive, so that the strain with both (DMM4) is significantly more unidirectional than any other E. coli strain.

Unidirectional movement is necessary and sufficient for membrane protrusion formation and uptake

Actin-based motility through the cytoplasm brings bacteria to the host cell surface, where they may have the opportunity to form a protrusion and be taken up by a neighbouring cell. One intriguing possibility is that pathogenic bacteria relying on this method of intercellular spread express virulence factors on their surface that interact with the cytoplasmic face of the host cell plasma membrane and mediate protrusion formation, analogous to the way many pathogenic bacteria invade cells by expressing virulence factors that bind to receptors on the extracellular face of the host cell plasma membrane [e.g. Yersinia invasin binds to β1-integrins (Isberg and Leong, 1990), and L. monocytogenes internalin binds to E-cadherin (Mengaud et al., 1996)]. An alternate possibility is that no molecular recognition is involved but, rather, that the force of actin-based motility alone is sufficient to deform the plasma membrane in protrusion formation drastically.

Escherichia coli K-12 hybrids containing more than 200 S. flexneri genes on the large virulence plasmid are able to spread from cell to cell, as detected by plaque formation (Pal et al., 1989). Thus, it is possible that the S. flexneri virulence plasmid encodes for a specific intracellular plasma membrane receptor that triggers protrusion formation. It has also been observed that variations in S. flexneri's LPS structure can affect its ability to spread from cell to cell (Sandlin et al., 1995; 1996; Van den Bosch et al., 1997; Rathman et al., 2000), although it could not be determined from these experiments whether the LPS were only affecting the polar distribution of IcsA and thus decreasing the efficiency of unidirectional movement, or whether the LPS also plays yet another role in protrusion formation or recognition at the cytoplasmic face of the host cell plasma membrane.

We have found that all E. coli strains tested, including those that express only a single S. flexneri gene (icsA), are capable of forming protrusions at some frequency (Fig. 4B). Furthermore, a strain of S. flexneri cured of its virulence plasmid but with icsA restored on a high-copy plasmid is also capable of forming protrusions. Thus, neither normal S. flexneri LPS nor any additional S. flexneri-specific factor is necessary for protrusion formation. Most strikingly, we found a significant strong correlation between the frequency of protrusion formation in all strains examined and the frequency of unidirectional movement. For the most unidirectional E. coli species (DMM4 with both IcsP and elaborate LPS), the frequency of protrusion formation was restored to the frequency of protrusion formation for wild-type S. flexneri(Fig. 4B). Therefore, we can quantitatively reconstitute protrusion formation in a non-pathogenic E. coli, lending strong support to the hypothesis that unidirectional actin-based motility alone is necessary and sufficient for protrusion formation.

It remains formally possible that IcsA itself interacts with a plasma membrane receptor during protrusion formation. However, in our most efficient protrusion-forming strains, IcsA protein localization is strongly polarized, with significant expression only on the back end of the bacterium, at the junction with the comet tail. To interact with the host plasma membrane, IcsA would need to be localized at the front end as well, where contact is first made. As the correlation we observe between polarity and protrusion formation strongly favours back-end localization, we tend to rule out the possibility that IcsA interacts directly with the membrane. As stated earlier, S. flexneri and E. coli are very closely related and, thus, it is also formally possible that both bacterial strains encode a common protein or molecule that is somehow involved in a signalling event at the host cell membrane that allows for protrusion formation. However, such a hypothetical event would not be dependent on a molecule that is specifically adapted for a pathogen with a cytoplasmic lifestyle.

Intriguingly, persistence of unidirectional movement appears to be more important in protrusion formation than motility parallel to the long bacterial axis. We found that many individual cells of E. coli K-12 expressing both icsA and icsP (DMM2) moved in a direction that was perpendicular to their longitudinal axis, but still unidirectionally. These ‘sideways’ individuals were still able to form membrane protrusions as observed by live video microscopy (data not shown). Protrusion formation was not correlated with average speed at all. Thus, to make plasma membrane protrusions, bacteria need not be moving particularly fast, nor need they be moving parallel to their long axes, but they must be moving in a determined and persistent manner.

Furthermore, we have shown that E. coli membrane protrusions into adjacent cells are taken up or ‘pinched off’ with kinetics that appear to be similar to those of wild-type S. flexneri and L. monocytogenes (Robbins et al., 1999). Although we were not able to determine the efficiency of spread for our E. coli constructs because of their inability to lyse the secondary vacuole, we did determine that the uptake by neighbouring cells was not a rare event and was easily detectable by video microscopy. Thus, the uptake of protrusions by adjacent cells is mediated by actin-based motility and is in no way dependent on additional S. flexneri virulence factors. The endogenous process of ‘paracytophagy’, in which epithelial cells internalize bits of their neighbours, is sufficient to explain the uptake of a protrusion by the recipient cell.

Experimental procedures

Bacterial strains and growth conditions

Bacterial strains used in this study are listed in Table 1. The wild-type S. flexneri strain 2457T and the isogenic mutant S. flexneri strains BS520 and BS176 (Sandlin et al., 1996) were grown in trypticase soy broth overnight, diluted 1:100 and grown for 180 min at 37°C with agitation for log phase bacteria. E. coli strains were grown overnight in Luria broth, diluted 1:100 and grown for 180 min at 37°C with agitation for log phase bacteria. Antibiotics were added at the following concentrations when appropriate: ampicillin, 100 µg ml−1; kanamycin, 30 µg ml−1; chloramphenicol, 30 µg ml−1.

Table 1.  Bacterial strains and plasmids.
Strain or plasmid nameDescription of relevant traitsSource
S. flexneri 2457TWild-type S. flexneri, serotype 2aFormal et al. (1958)
BS520S. flexneri 2457T rfaL1::TnphoASandlin et al. (1996)
BS176S. flexneri plasmidBernardini et al. (1989)
DMM5BS176 containing icsA on pHS3199 and invasin on pinvThis study
MBG264E. coli K12 FaraD139 lacU169 rpsL thi pyrC4Goldberg and Theriot (1995)
gyrA thyA his flaD ompT::kan 
DMM1MBG264 pHS3199 pinvThis study
DMM2MBG264 pHS3199 pDMM2This study
ATM379E. coli O8 thr-1 leuB6Δ(gpt-proA)66 argE3 thi-1Sandlin and Maurelli (1999)
rfbO8+lacY1ara14 galK2 xyl-5 mtl-1 mgl-51 
rpsL31 KdgK51 supE44 ompT::kan 
DMM3ATM379 pHS3199 pinvThis study
DMM4ATM379 pHS3199 pDMM2This study
 pHS3199IcsA cloned into pBR3222 colE1 ori, Aprd'Hauteville et al. (1996)
 pRI203inv encoded on 4kb BamHI insert in pBR325Isberg and Falkow (1985)
 pinvinv BamHI insert from pRI203 put into pACYC184, CmrThis study
 pACYC184Vector with P15A ori, CmrNEB
 picsP1icsP/sopA cloned into pACYC184, CmrRobbins et al. (2001)
 pDMM2inv BamHI insert from pRI203 put into picsP1. CmrThis study

Plasmid constructs

Plasmid DNA isolation, restriction enzyme digests, ligations and electroporations were performed essentially as described previously (Maniatis et al., 1989). icsP was cloned into pACYC184 to create picsP1 as described previously (Robbins et al., 2001). The 4000 bp BamHI fragment from pRI203, containing the gene coding for invasin from Y. pseudotuberculosis, inv, was gel purified (Qiagen kit) and inserted into the unique BamHI site within picsP1 to create pDMM2 or into pACYC184 to create pinv.

Selection of HeLa cells stably expressing GFP–actin and cell culture

HeLa cells (ATCC CCL2) were transfected with pEGFP-actin (Clontech) using the lipofectamine protocol from Gibco BRL. Clones were selected in 400 µg ml−1 G418 (geneticin; Gibco BRL). Stably expressing clones were selected on the FACSorter and passaged in the presence of G418. Low-passage aliquots of drug-resistant HeLa clones were frozen and stored in liquid nitrogen: clones were passaged in DMEM containing 10% FCS (Gibco BRL) and 200 µg ml−1 G418 for a maximum of 4–6 weeks.

HeLa cell infections

Bacterial strains were grown to log phase as described previously. E. coli or plasmid-cured S. flexneri strains (1 × 108) were spun down at 5000 r.p.m. for 10 min, resuspended in 600 µl of phosphate buffered saline (PBS), pH 7.0 diluted 1:5 (buffer A) and incubated in 10 mM nickel sulphate, pH 7.0, for 10 min at room temperature. The bacteria were spun down, resuspended in 600 µl of buffer A and incubated in 4 µg of purified six-HIS-tagged LLO for 10 min at room temperature (Gedde et al., 2000). The bacteria were spun down, resuspended in 600 µl of buffer A, and 40 µl was added to each well of a six-well dish containing 106 HeLa cells (5 × 105 GFP–actin-expressing HeLa cells plus 5 × 105 HeLa cells not expressing GFP–actin for observing intercellular spread in live video microscopy) and a 25 mm no.1 round glass coverslip. The infections were synchronized by centrifuging for 10 min at 1000 r.p.m., incubated for 1 h at 37°C, 5% CO2 before washing with 3 ml of PBS. DMEM containing 10 µg ml−1 gentamicin was added, and the infections were then incubated for 4, 5 and 6 h at 37°C, 5% CO2.

Fluorescence video microscopy and analysis

Infected cells grown on glass coverslips were observed 3–8 h after infection after mounting on a stage whose temperature was maintained at 37°C by a circulating water bath. Cells were overlaid with phenol red-free DMEM with 10% FBS buffered with 20 mM HEPES (pH 7.3) and covered with a thin layer of silicone DC-200 fluid (Serva) to prevent evaporation. Observations were performed on a Nikon Diaphot-300 inverted microscope equipped with either phase-contrast or DIC and epifluorescence optics. Time-lapse video microscopy was acquired with a cooled CCD camera (NDE/CCD; Princeton Instruments) and metamorph (Universal Imaging) software. Phase-contrast/fluorescence or DIC/fluorescence image pairs were recorded every 10–30 s, with 50 ms exposures.

The tracking feature within metamorph software was used to obtain distances and velocities. The tracks taken at 10 s intervals were resampled at 30 s intervals in excel 4.0 to obtain the average speed per bacterium. The averages and standard deviations for each bacterial strain are reported (Table 2).

Phalloidin staining

HeLa cells (5 × 105) were plated on coverslips in six-well tissue culture dishes. Cell were infected and fixed with 4% paraformaldehyde in PBS at the indicated time points. Fixed cells were then extracted with CSK buffer, washed with blocking buffer and incubated for 10 min with rhodamine–phalloidin (Molecular Probes; 1 unit per coverslip) in blocking buffer, washed and processed for microscopy (Robbins et al., 1999). Imaging was performed on an Axioplan 2 microscope (Zeiss) or an inverted Nikon Diaphot 300, and images collected with a Micromax:512BFT cooled CCD camera (Princeton Instruments) connected to an Optiplex computer (Dell) using metamorph software (Universal Imaging).

Transmission electron microscopy

HeLa cells (1 × 106) were seeded onto 25 mm round coverslips and allowed to adhere overnight. Cells were infected as described above. At 5 h after infection, the cells were fixed with 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.2) for 30 min on ice. Monolayers were then post-fixed with 1% osmium tetroxide in 0.1 M phosphate buffer for 20 min and stained with 1% aqueous uranyl acetate for 15 min. Samples were washed with water before dehydration through a graded series of ethanol washes. The samples were infiltrated with Poly/Bed 812 for 1 h. Samples were embedded by placing gelatin capsules filled with resin on top of the coverslip. The polymerization process was allowed to proceed for 16 h at 60°C before removal of the coverslip. Serial sections were cut and examined with a Phillips transmission electron microscope model CM12.

Statistical analysis

Statistical analysis was performed using the Student's two-tailed t-test for independent means.


We thank Jennifer Robbins for scientific input and experimental advice throughout the course of this work. We gratefully acknowledge Daniel Portnoy and Margaret Gedde for communication of unpublished results and the HisLLO clone. We thank Anthony Maurelli and Marcia Goldberg for sending bacterial strains. We also thank Lisa Cameron and Jennifer Robbins for providing purified HisLLO, Susanne Rafelski and Fred Soo for help in velocity analyses, Nafisa Ghori for assistance in transmission electron microscopy, and Stanley Falkow and Igor Brodsky for stimulating discussions and helpful comments on the manuscript. This work was supported by National Institutes of Health grant AI36929 and a Fellowship from the David and Lucile Packard Foundation (to J.A.T). D.M.M was supported by NIH training grant 5T32AI07328 (to the Department of Microbiology and Immunology).

Supplementary material

The following material is available from 143/cmi143sm.htm.

Fig. S1. Phalloidin fluorescence, phase-contrast and overlay images for all strains listed in Table 2.

AVI files showing cell-to-cell spread in S. flexneri and E. coli.