Previous observation showed that S. enterica serovar Typhimurium (STM) is able to invade PEC from the apical side. Apical invasion is accompanied by massive remodeling of the actin cytoskeleton and formation of membrane ruffles. Here we investigated the fate of the host cell brush border during STM invasion. We infected PEC line MDCK with STM wild-type (WT) bacteria. Cells not associated with STM maintained the typical finger-like appearance of the brush border MV ( Figures 1A, B , detail i). For infected cells we observed following modification of the apical side. a) Cells in contact with STM with formation of massive membrane ruffles and loss of MV ( Figures 1A, B , detail ii). b) Cells with intracellular STM had neither ruffles nor brush border, but showed prominent reticular structures extending over the entire apical cell side ( Figures 1A, B , detail iii). Fluorescence microscopy (FM) of Lifeact-eGFP transfected MDCK cells was performed to visualize F-actin and atomic force microscopy (AFM) to profile surface topologies ( Figure 1C ). Correlation of FM and AFM modalities indicated that elevated reticular structures fully co-localized with F-actin. Thus, we termed this morphotype ‘reticular actin’. These observations define STM as a pathogen inducing MV effacement.

We set out to identify the effector protein(s) responsible for the remodeling of the apical F-actin cytoskeleton of PEC. Invasion of MDCK or C2BBe1 cells by STM WT or mutant strains lacking SPI1-T3SS effector proteins SopE, SopE2, SopA, SopB or SipA was quantified ( Figures 2A , S2A, C ). Lack of SopE reduced invasiveness about 3-fold (26.0 ± 5.3% vs 9.2 ± 2.1% for WT and ΔsopE, respectively) and reduced areas of membrane ruffles were determined (33.4 ± 9.7 µm² vs. 8.5 ± 4.4 µm² for WT and ΔsopE, respectively) ( Figures 2A, C ). Lack of SopE2, SipA, SopA, or SopB had similar small, or no effects on membrane ruffling and invasion of MDCK and C2BBe1 cells ( Figures 2A, D, E ; S2A, B, C, D, E ). Since previous molecular analyses of SPI1-T3SS effector proteins revealed partially redundant functions in host cell F-actin manipulation (Hardt et al., 1998b; Zhou et al., 1999; Stender et al., 2000; Zhou et al., 2001; Hernandez et al., 2004; Boyle et al., 2006; Patel and Galan, 2006), we deployed a reductionist approach to monitor the contribution of each effector protein to the MV effacement. A mutant strain termed STM Δ5 lacking sipA, sopA, sopB, sopE, and sopE2 was still able to adhere to the apical side of host cells, but neither induced membrane ruffling, nor MV effacement ( Figures 2E, F ). STM Δ5 was more than 10,000-fold reduced in invasion of PEC ( Figures 2B, E, F ), comparable to STM ΔinvG lacking a functional SPI1-T3SS. We next complemented STM Δ5 with low-copy number plasmids for expression of single effectors, and determined invasion ( Figure 2B ). Complementation with sipA, sopB, sopE, or sopE2, but not sopA, each resulted in polarized cell invasion significantly higher than that of Δ5. Complementation by sopE was sufficient to restore invasion of MDCK cells to 21.0% of WT invasion, as well as membrane ruffling ( Figures 2E, F ). Complementation with sopE2 resulted in only 1.2% of WT invasion (( Figure 2B ), and limited F-actin accumulations were associated with bacteria ( Figures 2E, F ). Cells infected with WT, ΔsopE2, or Δ5 + [sopE] strains showed loss of MV and the presence of reticular F-actin. Invasion phenotypes were similar in the human cell line C2BBe1 ( Figures S2C, D, E ), but far less pronounced in the non-polarized cell line HeLa ( Figures 2C, D, F, G ). Compared to polarized cells, the invasion of non-polarized cells by STM was much lower ( Figures 2A, B vs. Figures S2D, G ).

Altogether, these results demonstrate that SopE as effector protein with GEF function is sufficient to trigger invasion of STM, to induce MV effacement, and formation of reticular F-actin networks. Although the contribution to invasion is similar in polarized and non-polarized epithelial cell lines, the effects of SopE on apical surface morphology are unique for polarized epithelial cells.

F-actin is also involved in maintaining the tight junction organization of cells in epithelial tissues (Eaton et al., 1995; Gopalakrishnan et al., 2002). We hypothesized that the effect of SopE on the actin cytoskeleton might affect the barrier function of polarized epithelia. As indicators of epithelial barrier integrity, the transepithelial electrical resistance (TEER) and distribution of tight-junctions (TJ) protein ZO-1 were analyzed in response to STM infection of C2BBe1 cells ( Figure 3 ). Infection with STM WT resulted in loss of epithelial barrier function as indicated by loss of TEER, while STM ΔinvG had no effect on barrier function ( Figure 3A ). Mutant strains only lacking single effectors ΔsopE or ΔsopE2 showed similar reduction in TEER as WT infection. Infection with STM Δ5 + [sopE] resulted in TEER decrease similar to the WT strain, whereas infection by STM Δ5 + [sopE2] ( Figure 3B ) or other effectors (data not shown) did not result in epithelial barrier damage. The effects of various strains on TJ integrity as determined by continuity of ZO-1 distribution were in line with TEER reduction by various mutant strains ( Figure 3C ). This result indicates that SopE-mediated re-modelling of the F-actin cytoskeleton also affects other F-actin-dependent structures such as TJ.

Effacement of MV, membrane ruffling and reticular F-actin formation were observed as outcome of STM infection. However, endpoint observations only provide restricted information about the kinetics of F-actin remodeling. Moreover, it remains open if STM induces membrane ruffling and MV collapse simultaneously, or if MV effacement is prerequisite for subsequent membrane ruffling.

We performed live-cell imaging (LCI) of STM invasion of MDCK Lifeact-GFP cells with high temporal and spatial resolution using SDM ( Figure 4A ). Shortly after addition of the inoculum, WT bacteria reached the cell monolayer and adhered to cells. Only a subset of the adhesion events led to productive invasion processes, and another subset of WT bacteria did not induce any detectable alteration of the host cell actin ( Figure 4A , ‘futile’). About 90-120 s after adhesion, individual bacteria triggered ruffle formation ( Figure 4A , ‘invasive’).

The Lifeact-eGFP signal is directly proportional to amount of F-actin, thus can indicate MV collapse or effacement induced by STM. No alterations of MV were detected if WT adhesion did not result in ruffle induction, or if cells were invaded by single bacteria of STM ΔsopE. We observed effacement of the entire brush border for cells invaded by WT or Δ5 + [sopE]. For quantification of MV effacement, maximum intensity projection images from the entire cell infected by STM were analyzed in order to avoid missing signals due to Z-drift. We quantified lifetime and size of ruffles induced by WT and ΔsopE strains. Individual invasion events were located and ruffle dimensions were quantification from initiation to termination over ca. 45 min. This allowed definition of maximal ruffles sizes for individual invasion events. Analysis of the kinetics of invasion revealed radial extension of ruffles, initiating from the point of bacterial adhesion over an interval of 8 min. ( Figure 4B ). Although the duration of ruffles induced by both strains was the same, the area of ruffles induced by WT strain was larger than those triggered by ΔsopE strain. ( Figure 4B ). Time-lapse series revealed that the loss of MV occurred in parallel to the ruffle induction ( Figures 4A, C , Movie 1 ). MV effacement is a consequence of the F-actin depolymerization, which might be active or passive. To address this question, we quantified the signal decay from those zones rich in MV during infection. This quantification demonstrated that the depolymerization of F-actin occurred exponentially ( Figure 4D , insert), suggesting an active depolymerization process.

Further observations from LCI showed that the brush border architecture was not restored after ruffle retraction, instead reticular F-actin appeared 30 s before complete retraction of the ruffle and remained present over the entire imaging period ( Figure 4A and Movie 1 ). F-actin depolymerization during ruffle formation was restricted to the apical side of infected cells ( Figure 4C ), since the basolateral F-actin cytoskeleton remained unaltered throughout the entire acquisition period ( Figure S3C ).

Infection with STM Δ5 induced no detectable alterations of the actin cytoskeleton (see Figure 2E ), whereas infection with STM Δ5 + [sopE] resulted in extensive ruffle formation ( Figure 4A ). Although the membrane ruffling was similar for STM Δ5 + [sopE] and WT, the invasion by Δ5 + [sopE] often resulted in extensive spikes decorating the ruffle and F-actin protrusions ( Figure 4A ). Furthermore, cells invaded by STM Δ5 + [sopE] showed effacement of the entire brush border, while no alteration of MV was detected if cells were invaded by single bacteria of the ΔsopE strain ( Figure 4A ). In contrast to the effacement caused by STM Δ5 + [sopE] and WT STM, cells infected by the ΔsopE strain did not show any appreciable decay of F-actin signal from MV ( Figure 4D ).

Our results demonstrate that MV effacement and ruffle formation were simultaneously caused by STM during SPI1-T3SS-mediated invasion. In addition, MV were not restored after complete engulfment of STM, but reticular F-actin was newly polymerized. This also suggests that the increased F-actin polymerization at the ruffle led to consumption of the G-actin pool. Once this pool is exhausted, MV retract by enhanced depolymerization, as suggested by the exponential decay of the F-actin signal. The reduced F-actin polymerization activity observed during infection by the ΔsopE strain would not require consumption of the G-actin pool. In turn, MV integrity is not affected.

To further test if membrane ruffling triggered by STM consumes the cytoplasmic G-actin and finally concludes in MV depolymerization, two inhibitors of F-actin polymerization were applied to invasion and LCI analyses. Latrunculin B (Lat B) binds G-actin and prevents polymerization, whereas Cytochalasin D (Cyt D) blocks F-actin polymerization by binding to the barbed ends of F-actin (Pollard and Mooseker, 1981), avoids F-actin depolymerization (Morris and Tannenbaum, 1980), and the interaction with cofilin at the pointed ends (Shoji et al., 2012) ( Figure 5A ). At concentrations of 10 µM, both inhibitors blocked the invasion by interrupting ruffle formation ( Figure 5B ). However, Lat B inhibited invasion more efficient than Cyt D ( Figure 5C ), since Lat B sequesters G-actin. Titration experiments showed that at inhibitor concentrations of 1 µM, invasion of Lat B-inhibited cells was 1.44-fold lower than of Cyt D-inhibited cells. Time-lapse series of WT infection in presence of 1 µM of either inhibitor supported this observation ( Movie 2 ). Lat B-treated cells showed full disruption of F-actin cytoskeleton, and aggregation of actin at the apical side. However, ruffles were not fully formed by STM-triggered F-actin polymerization. In contrast, Cyt D-treated cells maintained MV and membrane ruffling proceeded slower and remained for more than 60 min without retraction. In DMSO-treated controls, the ruffle formation terminated in less than 20 min ( Figure 5D ). The distinct effects of the inhibitors indicate that STM-induced membrane ruffle formation deploys G-actin from sources other than F-actin in MV. If depolymerization of these structures is blocked by Cyt D, ruffle formation is dependent on the availability of the limited pool of cytoplasmic G-actin.

To follow the fate of the G-actin pool during STM infection, staining was performed with Phalloidin or DNase I to label F-actin or G-actin, respectively. At 25 min. p.i. G-actin in STM-infected cells was weakly stained by DNase I ( Figure 5E ). Most of the G-actin signal localized at ruffles, the origin of ruffles, and basolateral sides, but no G-actin was observed at the apical side. In contrast, non-infected cells presented homogenous distribution of G-actin with the exception of the nucleus. Taken together, the data support that membrane ruffles initiate by consumption of G-actin from the cytoplasmic pool. Moreover, these results show that only the apical F-actin cytoskeleton was affected by STM infection ( Figures 4 , S3 ).

We set out to identify regulators of F-actin required for the STM-induced cytoskeletal remodeling that could explain MV effacement during STM invasion. In accordance with previous reports, we observed that cofilin, IQGAP1 and cortactin were recruited to ruffles triggered during invasion of MDCK cells by STM ( Figure S4 ). Since all of these proteins are controlled by MAPK and Src kinases, we investigated the effect of pharmacological inhibitors PP1 (for Src kinase) and PD98059 (for ERK1/2 kinase) on STM invasion. Presence of 150 µM PD98059 or 50µM PP1 only mildly diminished invasion compared to cells treated with solvent DMSO ( Figure S4B, C ). These results indicate that although proteins involved in the Src and ERK1/2 kinase pathways are recruited to the ruffle formation in PEC, their function is secondary.

Class II myosins are involved in the invasion of STM (Hanisch et al., 2011), thus invasion experiments in presence of class II myosin inhibitor (-) blebbistatin were performed. In contrast to previously published data (Hanisch et al., 2010), invasion of STM was affected by S (-) blebbistatin, since we observed a 3-fold reduction in invasion of MDCK cells ( Figure 6A ). Furthermore, STM infection of cells treated with S (-) blebbistatin led to induction of very large ruffles at 1 h p.i. ( Figures 6B, C ). Quantification of ruffle areas demonstrated a two-fold increment in ruffles size for S (-) blebbistatin-treated cells ( Figure 6C ).

Interestingly, during LCI membrane vesicle formation at the apical side of cells treated with blebbistatin was observed ( Figure 6B ). The formation of vesicles at the apical side of the cells recalled the function of myosin 1a. Myosin 1a controls the tension in MV and formation of vesicles at the brush border, containing EPS8 L3, Ezrin, and sugar transporters (Mcconnell et al., 2009; Nambiar et al., 2009). Cells transfected with myosin 1a-eGFP were used to test the involvement of myosin 1a in invasion by STM. We observed the recruitment of myosin 1a-eGFP to ridges of ruffles triggered by STM, whereas β-actin-RFP was recruited to central areas of ruffles ( Figure 6D and Movie 3 ). To test if myosin 1a actually contributes to STM invasion, C2BBe1 cells were stably transfected by a lentiviral construct encoding shRNA for knockdown (k/d) of myosin 1a (Myo1a k/d). In accordance to previous results (Tyska et al., 2005), the apical side of Myo1a k/o cells released large amounts of vesicles ( Figure S4D ). The invasiveness of STM in Myo1a k/d cells was reduced to 82.8 ± 0.1% ( Figure 6E ). In spite of the reduced effect of the knockdown of myosin 1a, ruffle size was also altered in absence of myosin 1a and MV, although with vesicles, were still present in cells infected by STM ( Figures 6G, H ). These results demonstrate that class I and class II myosins contribute to the correct formation of ruffles triggered by STM. Abrogation of the function of these myosins impedes STM entry into host cells.

The exponential decay of the F-actin observed during MV effacement indicates an active depolymerization process. Since contribution of cofilin is unlikely due to its absence in the normal brush border architecture (Ashworth et al., 2004), we investigated further candidates. Villin regulates the F-actin polymerization in MV by severing and capping F-actin, concluding in G-actin release (Bretscher and Weber, 1980; Ferrary et al., 1999; Ubelmann et al., 2013). Thus, the functions of villin may also be required for invasion of polarized cells by STM. We generated C2BBe1 cells with lentiviral transfection of shRNA for villin k/d (VIL1 k/d). These cells showed a normal distribution and structure of MV ( Figure 6G , Figure S4D ) similar to previous reports (Ferrary et al., 1999; Revenu et al., 2012). Invasion of VIL1 k/d cells was reduced to 58.7 ± 1.2% compared to parental C2BBe1 cells ( Figure 6F ). Further examination demonstrated that the size of STM-induced ruffles was highly decreased in contrast to those observed in parental cells. Membrane ruffles did not show large extensions as observed for infection with WT ( Figures 6G, H ). In VIL1 k/d cells, MV were intact and invading bacteria surrounded by small ruffles, similar to the invasion phenotype of the ΔsopE strain (see Figure 2 ). These results show that MV are only effaced in presence of villin, which is probably associated with its severing activity.

Our results indicate that SopE dominantly controls the F-actin polymerization in host cells ( Figure 2 , Figure 4 ). Furthermore, we observed that F-actin depolymerization in MV is exponential and related to villin. However, the connection between SopE and villin is in this case unknown. Previous data already demonstrated that isoforms of phospholipase C-γ (PLCγ), PLCβ and PLA interact with Rac1 (Peppelenbosch et al., 1995; Athman et al., 2003; Jezyk et al., 2006). Thus, we inhibited PLCγ with pharmacological agent U73122, and determined the effect on invasion of the Δ5 + [sopE] strain. Invasion rates after treatment with U73122 of C2BBe1 and VIL1 k/d cells were 25% lower compared to that of the WT strain ( Figure 6I ). But in VIL1 k/d cells inhibited with U73122, no further inhibition of the STM invasion was detected. Therefore, SopE requires the function of PLCγ. If PLCγ is not activated, probably villin is not active. Similarly, if villin is absent, inhibition of PLCγ is futile, since there is not binding partner for PLCγ. Once villin is activated, depolymerization of F-actin culminates in MV collapse.

Salmonella enterica sv. Typhimurium strain SL1344 was used as wild-type (WT) strain and mutant strains were isogenic to either WT. Bacterial strains used in this study are listed in Table 1 . In order to obtain mutant strains of sopA, sopB and sopE2 in the strain background of SL1344, gene replacements by an aph cassette were generated in NCTC 12023 by Red-mediated recombination (Datsenko and Wanner, 2000) using pKD13 as template and primers specified in Table S1 . Subsequently, P22 transduction according to standard methods (Maloy et al., 1996) was used to move the mutations to SL1344. Plasmids used for complementation of mutations, or expression of fluorescent proteins are listed in Table 2 and construction is described in Suppl. Materials.

Cell lines were cultured at cells were incubated at 37°C in a humidified atmosphere containing 5% CO₂. For invasion assays and microscopy analyses, MDCK clone Pf were used as standard cell culture model, kindly provided by the Nephrology department of the University Hospital Erlangen. Confluent monolayers in 25 cm²-cell culture flasks were seeded each week in a new 25 cm² flask with MEM supplemented with 1 x non-essential amino acids (PAA, Germany), 10% inactivated fetal calf serum (FCS, Sigma, Germany) and 1 x Glutamax (PAA, Germany). Cell line C2BBe1, a derivate of CaCo2 cell line (ATCC CRL-2102) was used as human polarized epithelial cell line to analyses of STM interactions. These cells were cultured in DMEM high glucose without pyruvate (PAA, Germany), containing Glutamax, 10% FCS and 2.5 µg x ml^-1 holo-transferrin (Sigma-Aldrich, Germany). For cultivation, media were supplemented with penicillin/streptomycin (PAA, Germany). Medium was changed every third day. Medium are replaced by antibiotic-free medium at least one day prior infection. Cells were seeded at 10⁵ cells per 12 mm polycarbonate filter insert (0.4 µm pore size, Millipore, Germany). The trans-epithelial electrical resistance (TEER) was measured every third day with a platinum electrode and an Ohmmeter EVOM (World Precision Instruments, USA). Cells were cultured until a TEER of 500-700 Ω per well was observed, usually for 10 to 15 d. TEER was determined prior and after infection by STM as indicator to epithelial barrier integrity. Generation of knock-down cell lines is described in Supplementary Materials and Methods.

Five days prior to infection, MDCK cells were seeded at 1 x 10⁵ cells per well in 24-well plates. At least 4 h before infection, the medium was substituted by medium without antibiotics. Bacterial strains were precultured in LB overnight at 37°C with continuous aeration in glass test tubes in a roller drum. Overnight cultures were diluted 1:31 in fresh LB and cultured for 3.5 h as above. Cultures were adjusted to OD₆₀₀ = 0.2 in PBS and a master mix was prepared in MEM. Cells were infected in triplicates with each strain from the master mix at multiplicity of infection (MOI) of 5. After invasion for 25 min, non-internalized bacteria were removed by washing three times with PBS. Thereafter, fresh medium containing 100 µg x ml^-1 gentamicin was added for 1 h. Finally, infected cells were washed five times with PBS and lysed with 0.5% deoxycholic acid for 10 min. Lysates were diluted and platted onto Mueller-Hinton agar (BD, Germany) plates with an Eddy Jet spiral platting instrument (IUL Instruments, Barcelona). Plates were incubated at 37°C overnight and numbers of colonies were counted.

For infection of monolayers in transwell filters, bacterial strains were added at MOI 50 and TEER was recorded each 15 min for the first hour, thereafter each 30 min until 2.5 h. Finally, cells were washed thrice with pre-warmed PBS and fixed with methanol at -20°C overnight.

For imaging, bacterial strains harboring p3589 for constitutive expression of mCherry were used. 1 x 10⁵ cells were seeded on cover slips as for invasion assays. Bacteria were diluted in MEM to an OD₆₀₀ of 0.2 and applied for infection at MOI 50. Infections were made in duplicate at least three times independently. After 15, 25, or 60 min, the infection was stopped by washing cells with PBS four times. For fixation, 3% PFA in PBS was added and cells were incubated for 1 h at 37°C. After fixation or any subsequent incubation with reagents or antibodies, cells were washed thrice with PBS at 37°C. Fixed cells were permeabilized by incubation for 15 min at 37°C with 0.5% Triton X-100 (Sigma-Aldrich, Germany) in blocking solution consisting of 2% BSA (Biomol, Germany), 2.0% Goat serum (Gibco, Germany) in PBS in a humid chamber. Antibodies were diluted in blocking solution and incubations were performed at 37°C in a humid chamber. Rabbit anti Ezrin was diluted 1:500 and incubated with the samples for 1 h. Alexa488-conjugated Phalloidin (Invitrogen, Germany) was added in 1:200 dilution and incubated 45 min at 37°C. Cover glasses were then mounted on glass slides with Fluoroprep (Biomerieux, France), sealed with Entellan (Merck, Germany) and kept in the dark at 4°C. C2BBe1 cells were also used for microscopy analysis of tight junction integrity. After fixation with cold methanol overnight at -20°C, tight junctions were stained with rabbit anti ZO-1 at a dilution of 1:200 for 2 h. Alexa488-conjugated anti-rabbit secondary antibody was applied 1:500 for 1 h. All washing steps were performed as for MDCK cells. Finally, filter inserts were recovered and mounted between a cover slip and glass slide as described above.

Fixed samples were observed with a SP5 II confocal laser-scanning microscope (CLSM, Leica Microsystems Wetzlar, Germany). Images were acquired using a 100x objective with a numerical aperture of 1.51 and 1 Airy unit for the pinhole, the pixel size of the images was 70.85 x 70.85 nm. The 488 nm argon laser line was used for Alexa488-conjugated antibodies and eGFP. The HeNe 543 nm laser line was used for excitation of Alexa568-conjugated antibodies and mCherry or mTagRFP. Images were acquired with Leica Acquisition Software V. 2.3.6 and further processed with Imaris V. 7.6.1 (Bitplane, Switzerland) and FIJI (Max-Planck Institute for Cell Biology, Dresden, Germany). LCI of MDCK cells expressing Lifeact-eGFP, Myo1a-eGFP, β-actin-RFP, IQGAP1-eGFP, Cortactin-mCherry or cofilin-RFP and mCherry-, GFP-, or mTagRFP-expressing STM strains was performed using a CellObserver microscopy system (Zeiss, Germany) equipped with Yokogawa spinning disc unit. Images were acquired for 120 min starting shortly after infection at maximal speed with intervals of 100 to 200 ms, recording Z-stacks with distances of 0.30 – 0.35 µm between Z-planes. A water immersion objective with a numerical aperture of 1.333 was used. Acquisition was performed with either a cooled CCD camera (CoolSNAP HQ², Photometrics) with a chip of 1,040 x 1,392 pixels for high spatial resolution, or an EM-CCD camera (Evolve, Photometrics) with a chip of 512 x 512 pixels for high sensitivity. Acquisition and processing of time-lapse images was performed with AxioVision 4.8.2 or ZEN 2011 software (Zeiss). Images from CLSM or SDM were deconvolved with Huygens V.4.2 using a theoretical point spread function. Bleaching and Z-drift were also corrected with Huygens.

AFM measurements were conducted using the NanoWizard II AFM system (JPK Instruments AG, Berlin, Germany). High-resolution surface images were acquired by operating the AFM under ambient conditions in soft contact mode using silicon nitride AFM probes with a nominal force constant of 0.06 N/m (SiNi, Budget Sensors, Wetzlar, Germany). Samples were prepared as described above. For each sample, topographic overview images with a 90 x 90 µm scan area were taken before zoom-ins were generated. All images were polynomial fitted and unsharpened mask filtered using JPK data processing software (JPK Instruments AG). 3D projections of height profiles are shown, tilted 12° in X direction.

To correlate surface structures recorded by AFM with the cytoskeleton, epifluorescence images of Lifeact-eGFP cells were acquired and aligned with AFM images by matching landmarks observed in both images using the transform tool in Adobe Photoshop.