Preferential invasion of mitotic cells by Salmonella reveals that cell surface cholesterol is maximal during metaphase

Cholesterol is a key metabolic precursor and a determinant of membrane fluidity and lipid lateral clustering (Maxfield and van Meer, 2010). Cell surface-exposed cholesterol is crucial for the cell binding and invasion of many viruses and pathogenic bacteria, including Salmonella (Garner et al., 2002; Lafont et al., 2002; Rawat et al., 2003). Salmonella causes typhoid fever and a large proportion of gastroenteritis cases in humans. To invade cells, Salmonella assembles a type 3 secretion system (T3SS), encoded in its pathogenicity island-1 (SPI-1). T3SS acts as a ‘molecular needle’ that translocates virulence proteins into the host cell to trigger entry of the bacterium (Galán, 2001). The binding of Salmonella to its host depends on the presence of cholesterol in the targeted membrane (Garner et al., 2002). As the total cholesterol levels double between G1 and G2 (Fielding et al., 1999), we investigated whether Salmonella enterica serovar Typhimurium (Salmonella) preferentially invades cells at specific stages of their cell cycle. We found that Salmonella invades mitotic cells preferentially because cell surface cholesterol is maximal when cells divide.

Using flow cytometry and confocal microscopy, we showed that Salmonella targeted mitotic cells more efficiently than cells in other phases of their cell cycle (Fig. 1A–E; supplementary material Fig. S1A–F and Movies 1, 2), consistent with a recent observation (Misselwitz et al., 2011). This was true for the different wild-type strains (12023, SL1344 and LT2) and cell lines tested (supplementary material Fig. S1A). HeLa cells have been widely used to study Salmonella entry but, as tumour cells can have a perturbed cholesterol homeostasis (Gerlier et al., 1982), we also used the diploid epithelial cell line RPE1. After an infection of 10 minutes at a multiplicity of infection of 100, 60% of G2 cells but 93% of mitotic cells (although representing only 1% of the total population) were infected by SL1344 (Fig. 1D,E; supplementary material Fig. S1E). Less than 20% of interphase but 60% of mitotic cells contained more than two bacteria per cell (Fig. 1F; supplementary material Fig. S1G). As a consequence, 26% of all intracellular bacteria in the sample were inside mitotic cells (supplementary material Fig. S1H). Mitotic cell preference was not due to the different shape (spherical) as it remained when the infection was performed on cells rounded after detachment (Fig. 1G).

The translocon protein SipB binds directly to cholesterol and mediates the cholesterol-dependent attachment of Salmonella to targeted cells (Hayward et al., 2005). A mutant with a functional T3SS but lacking effectors required for invasion (ΔsopE/E2/B) retained preferential binding to mitotic compared with interphase cells (Fig. 2A–D), but, as expected, was unable to invade. However, a mutant lacking a T3SS altogether (ΔprgH) or one lacking SipB did not preferentially bind to mitotic cells (Fig. 2A–D). To test whether other bacterial cell surface proteins could mediate the targeting to mitotic cells, we induced the binding and internalization of the ΔsipB mutant by expressing Yersinia invasin, which allows cell entry by a different mechanism than the one used by Salmonella (Aiastui et al., 2010). Expression of the invasin did not restore the preference of ΔsipB for mitotic cells (Fig. 2E), confirming that SipB mediates the targeting. Cholesterol depletion abolished the preferential targeting to mitotic cells (Fig. 2F–H; supplementary material Fig. S2). By contrast, cholesterol enrichment, resulting in comparable levels in interphase and control mitotic cells (Fig. 2F), significantly reduced the preference of Salmonella for mitotic cells (Fig. 2H). Thus, cholesterol and SipB mediate the preferential targeting of mitotic cells by Salmonella. Interestingly, Salmonella invades cycling but not quiescent stem cells in a SipB- and cholesterol-dependent manner (Kolb-Mäurer et al., 2002; Yu et al., 2009), giving a potential relevance for our findings to infections in vivo.

The preferential targeting of mitotic cells suggested they could have more cholesterol at their surface. Short (1 minute) incubation of live cells with the fluorescent cholesterol-binding compound filipin stained the plasma membrane without being significantly internalized (Fig. 3A–C; supplementary material Fig. S3). Because cholesterol (and thus filipin) can flip across the bilayer (Garg et al., 2011), we also used a fluorescent ester of polyethylene glycol-cholesterol (fPEG-cholesterol; Fig. 3D–F), which partitions in cholesterol-rich domains. Because of its size, PEG cannot flip across the plasma membrane (Madenspacher et al., 2010; Sato et al., 2004) and thus quantitatively accounts for the cell surface-exposed, endogenous cholesterol. In addition, fPEG-cholesterol can be fixed and thus coupled with phospho-Histone H3 (pH 3) labelling to identify mitotic cells by flow cytometry. However, pH 3 labels all phases within mitosis (prophase to cytokinesis) (Li et al., 2005), and because we observed two populations (Fig. 3F), we also used 3D confocal microscopy. It showed that the strongest signals corresponded to metaphase cells and, thus, that plasma membrane and cell surface exposed cholesterol were maximal during this stage (Fig. 3G–J).

The large differences between G2 and metaphase are unlikely to be due to changes in cholesterol synthesis; therefore, we investigated its transbilayer repartition. Studies in neurons, red blood cells and epithelial cells have determined that plasma membrane cholesterol is mainly (70–80%) located at the cytoplasmic leaflet (Mondal et al., 2009; Schroeder et al., 1991; Wood et al., 2011). Interestingly, phosphatidylserine and phosphatidylethanolamine are mainly located at the cytoplasmic leaflet during interphase but are transiently exposed at the cleavage furrow during cytokinesis (Emoto et al., 2005). We assessed cholesterol asymmetry using an established assay that takes advantage of the fluorescence of the natural derivative of cholesterol DHE and of its fast quenching by TNBS (Mondal et al., 2009). As TNBS is cell impermeable, it only quenches DHE located at the outer leaflet of the plasma membrane. Fast 3D stacks of images of DHE located at the plasma membrane (∼1 minute after loading) were acquired before and after TNBS application. In interphase cells, TNBS quenched ∼20% of DHE signals (Fig. 3K). As >90% signals were quenched in permeabilized cells, we concluded that ∼80% of DHE molecules were protected and thus located in the inner leaflet of the plasma membrane, in agreement with the literature (Mondal et al., 2009). In mitotic cells, however, TNBS quenched ∼45% of the DHE signals, indicating a partitioning closer to 50∶50 between both leaflets (Fig. 3K,L). This suggests that the increase of cell surface-exposed cholesterol we observed with fPEG-cholesterol on metaphase cells was due to a loss of transbilayer asymmetry. The decreased quenching in mitotic cells was not due to the differences in starting intensities as on giant unilamellar vesicles (GUV) DHE signals were linear across a range of concentrations encompassing those used in cells (Fig. 3M). TNBS quenched ∼50% of signals at the highest DHE concentration, as expected for an equal partitioning of the probe between the two leaflets. Thus, we concluded that the asymmetrical distribution of plasma membrane cholesterol ceases during mitosis, exposing more at the outer surface, thereby explaining why dividing cells are more susceptible to Salmonella invasion. Interestingly, this was not observed for phosphatidylserine during metaphase, which remained mostly at the inner leaflet (Fig. 3J), indicative of a cholesterol-specific property.

Loss of cholesterol asymmetry induces a concomitant decrease at the inner leaflet (Fig. 3N), consistent with the decrease in the cholesterol-binding protein caveolin-1 at the plasma membrane during metaphase (Boucrot et al., 2011). To mitigate the decrease in inner leaflet levels of cholesterol during mitosis, we enriched cells with cholesterol complexes. Addition of cholesterol perturbed cell compaction and rounding causing metaphase cells to become significantly bigger (Fig. 4A; supplementary material Movies 3, 4). The formation of the rigid cortical actin network that drives mitotic compaction is mediated by RhoA (Maddox and Burridge, 2003) and activated ERM (Ezrin, Radixin, Moesin) proteins that bridge actin filaments to the plasma membrane (Kunda et al., 2008). Cholesterol addition did not affect recruitment of Rho GTPases to the plasma membrane during mitosis, but strongly decreased the recruitment of activated ERM proteins (Fig. 4B,C; supplementary material Fig. S4A,B,and Movies 5, 6), potentially explaining the defect in mitotic cell rounding caused by cholesterol addition. As ERM proteins are not known to bind directly to cholesterol but do bind to phosphatidylinositol-4,5-bis-phosphate [PtdIns(4,5)P₂] (Roch et al., 2010), a decrease in cholesterol in the inner leaflet during mitosis might favour ERM proteins recruitment by supporting higher levels of PtdIns(4,5)P₂. Indeed, cholesterol enrichment led to a decrease in PtdIns(4,5)P₂ levels during metaphase (Fig. 4D), in agreement with a previous report (Chun et al., 2010). Clarification of the mechanisms involved will help our understanding of mitotic cell rounding as well as the propensity of dividing cells to be preferentially targeted by viruses and bacteria.

HeLa (ECACC 93021013) and hTERT-RPE1 (ATCC CRL-4000) cells were grown at 37°C in 5% CO₂ in DMEM (Gibco) or DMEM∶F12 HAM 0.25% (w/v) sodium bicarbonate (Sigma), respectively, supplemented with 1 mM glutamine (Sigma) and 10% fetal calf serum. Cholesterol levels were decreased and increased using 15 µM methyl-beta-cyclodextrin (MβCD) and 16 mg/ml water-soluble cholesterol, respectively (both Sigma).

Antibodies used were: rabbit anti-phospho-HistoneH3 and anti-phospho-ERM, FITC-Annexin V (both Cell Signaling Technology), mouse anti-MPM-2 (Millipore) and anti-α-tubulin (Sigma), goat anti-Salmonella (CSA-1) (KPL), mouse anti-RhoA 26C4, rabbit anti-Cdc42 P1 (both Santa Cruz Technologies), mouse anti-Rac1 (BD Biosciences), mouse anti-PIP2 2C1 (AbCam). Antibody specificity was verified on cells overexpressing the targeted proteins. Secondary antibodies (Alexa488-, Alexa555- or Alexa647-conjugated), phalloidin-rhodamine and wheat germ agglutinin Texas-Red-x conjugate were from Invitrogen. DRAQ5 (Biostatus), filipin III, dehydroergosterol (DHE), (2,4,6-trinitrobenzene sulfonic acid) TNBS, staurosporine, 1-butanol and propidium iodide were from Sigma. Synthesis and characterization of fPEG-cholesterol was described earlier (Madenspacher et al., 2010).

Wild-type S. Typhimurium strains were 12023, SL1344 and LT2. SL1344 mutants were: ΔsopE/sopE2/sopB/sptP (called ΔsopE/E2/B), ΔprgH (both from Dr Unsworth; Imperial College London) and ΔsipB (generated using λ red recombinase method; 5′-CGGAGACAGAGCAGCACAGTGAACAAGAAAAGGAATAATTGTGTAGGCTGGAGCTGCTTCG-3′ and 5′-GCGGCGGGATTTATTCCCACATTACTAATTAACATATTTTCATATGAATATCCTCCTTAG-3′). Strains harboured pFPV25.1 plasmid carrying gfpmut3A (rpsM constitutive promoter), or pRI203 plasmid, encoding for the Yersinia pseudotuberculosis invasion region, as indicated.

Cells grown overnight were incubated with sub-cultured bacteria (OD₆₀₀ 1.8) at MOI 100, incubated in Earle’s Balanced Salt Solution (Gibco), at 37°C for 10 minutes, washed and fixed [3% paraformaldehyde (PFA) for 20 minutes]. In some experiments, cells were detached using 0.025% Trypsin-EDTA (Sigma) just before infection.

Cells infected or treated as described above were permeabilized (0.1% saponin), stained with primary and secondary antibodies, mounted using Aqua polymount (Polysciences Inc.) and imaged using a confocal laser-scanning microscope (LSM 510 or 710, Zeiss). In some experiments, cells infected with EGFP-expressing Salmonella were labelled without permeabilization with CSA-1 to label extracellular bacteria. Total fluorescence (sum of the Integrated Density of each images in 3D stacks) of cholesterol, Rho GTPases and pERM was quantified using ImageJ. Volumes of metaphase cells were measured using Huygens software (Scientific Volume Imaging) and visualized (3D rendering) using Volocity 5.0 (PerkinElmer).

Cells on dishes infected as described above were permeabilized (0.08% Triton X-100 for 10 minutes) and stained with primary and secondary antibodies. Data were collected on FACS Calibur™ (BD Biosciences) and analysed with FlowJo 7.6 (TreeStar).

Detached cells (flow cytometry) or cells grown on glass-bottom dishes (microscopy) were successively incubated in imaging buffer (DMEM without Phenol Red, 5% FBS, 15 µM HEPES) containing 5 µM DRAQ5 (10 minutes, 37°C) and 5 µg/ml filipin for 1 minute (to limit the staining to the plasma membrane) before measuring each samples live.

Cells grown on dishes or coverslips were stained successively with 5 µM DRAQ5 (10 minutes, 37°C) and with 1 µg/ml fPEG-cholesterol for 20 seconds before fixation with 4% PFA. Cells were then immunostained and prepared for flow cytometry or confocal microscopy.

Live-cell 3D stack imaging was performed on a spinning-disk confocal microscope (Eclipse TE-2000, Nikon; UltraVIEW VoX, Perkin-Elmer) controlled by Volocity 5.0. Flow cytometry and fixed-cell microscopy were performed on instruments described above.

Quenching of DHE by TNBS was performed as described by Mondal et al. (Mondal et al., 2009). Each MatTek dishes were placed onto the microscope stage (37°C chamber) and a field of view containing mitotic cells selected (brightfield). Cells were incubated for 1 minute in imaging medium containing 0.5 mM DHE-MβCD complexes (DHE in ethanol dried under argon and dissolved in buffer containing MβCD and DHE at 8∶1 mol∶mol ratio, sonicated and incubated overnight at 37°C under gentle agitation), washed once and rapidly imaged (3D stacks) using an epifluorescence microscope equipped with 350/50 and 420LP filters (Chroma). TNBS (10 mM) was added and another set of images taken after 1 minute. In some experiments, the cells were permeabilized using 40 mg/ml digitonin (Sigma).

GUVs were produced as described previously (Meinecke et al., 2013) using a lipid mixture (25% phosphatidylcholine, 15% phosphatidylethanolamine, 2% phosphatidylinositol, 8% phosphatidylserine, 10% sphingomyelin, 40% cholesterol) and imaged as described above.

Results shown are mean ± standard error of the mean (s.e.m.). Statistical testing was performed using Student’s t-test (continuous data, two groups), chi-square test (binomial data) or one-way ANOVA and Dunnett’s test (continuous data, at least three groups), as appropriate. ns, not significant; *P<0.05; **P<0.01; ***P<0.001. n, number of cells, was >30,000 and >50 cells per FACS and microscopy experiment, respectively; N, number of experiments, was greater than three for each quantification.

We thank Francisco García-del Portillo (Centro Nacional de Biotecnologia) and Kate Unsworth for providing reagents; we thank Richard Hayward, Mair Thomas and António Ferreira for comments on the manuscript; and the members of the Holden and Boucrot labs, especially Kieran McGourty, for discussions. E.B. thanks Harvey McMahon (Cambridge) for kind access to reagents and equipment.