Microvilli-like structures are associated with the internalization of virulent capsulated Neisseria meningitidis into vascular endothelial cells

The internalization of bacterial pathogens into non-phagocytic cells has been increasingly investigated and recognized as playing an essential role in bacteria—host-cell interactions. The first mode of entry is illustrated by Listeria and Yersinia, and it involves the formation of cell protrusions in tight contact with the bacterial surface. The second one is exemplified by Shigella and Salmonella and involves dramatic cytoskeletal rearrangements at the site of entry, with the formation of large membrane folds or ruffles around bacteria(Galan and Zhou, 2000;Nhieu and Sansonetti, 1999). Cytoskeletal modifications are known to require the activation of Rho family GTPases. Rho isoforms control the formation of actin stress fibers and focal adhesion plaques (Amano et al.,1997); Rac isoforms promote the formation of ruffling lamellipodia, and Cdc42 generates filopodia and microvilli structures(Gauthier-Rouviere et al.,1998; Hall, 1998). Whereas Shigella entry required Cdc42, Rac and Rho GTPases, invasion by Salmonella was only dependent upon Cdc42(Galan and Zhou, 2000;Nhieu and Sansonetti,1999).

Neisseria meningitidis (also referred to as meningococcus) is an extracellular pathogen that, once in the bloodstream, invades the meninges and the cerebrospinal fluid. A key event in N. meningitidis invasion is the interaction of bacteria with brain endothelial cells. Indeed, examination of post-mortem material has clearly shown a direct interaction of extracellular N. meningitidis with endothelial cells of both the choroid plexus and meningeal capillaries(Pron et al., 1997). Two meningococcal attributes are clearly essential for meningeal invasion: (i) the capsular polysaccharide, which allows bacterial survival in extracellular fluids, and (ii) type IV pili (TFP), which are multimeric structures essential for adhesion of virulent capsulated N. meningitidis to host cells(Nassif et al., 1999). This latter process seems to be due to a tip-located adhesin designated PilC(Rudel et al., 1995). In vitro studies using tight monolayers of human epithelial cells have shown that piliated capsulated meningococci can cross the monolayers by transcytosis(Merz et al., 1996;Pujol et al., 1997), thus suggesting that, in vivo, meningococci may use the transcellular route to cross the vascular endothelium to the meninges. For non-capsulated meningococci, as well as for the closely related pathogen N. gonorrhoeae, several mechanisms of bacterial internalization have been described, that involve interactions of cellular receptors with bacterial surface components, such as Opa and Opc outer membrane proteins(Bauer et al., 1999;Edwards et al., 2000;Meyer, 1998;Naumann et al., 1999;Virji et al., 1993;Virji et al., 1992;Virji et al., 1996). The role of these components in virulent capsulated strains of N. meningitidisis uncertain since non-piliated capsulated isolates are unable to efficiently interact with the cells. In addition, piliated capsulated N. meningitidis, which do not express Opa and Opc outer membrane proteins,can efficiently be internalized (Merz et al., 1996; Pujol et al.,1997; Pujol et al.,1999). Type IV pili thus remain the only component able to initiate the interaction of virulent capsulated N. meningitidis with human cells.

Pilus-mediated adhesion of capsulated strains was shown to lead to the formation of cortical plaques underneath bacterial colonies on the apical surface of epithelial cells. These structures result from the localized polymerization of cortical actin associated with the clustering of tyrosine phosphorylated and integral membrane proteins (ICAM-1, CD44, EGF receptor) as well as ezrin, a member of the ERM (ezrin-radixin-moesin) protein family(Merz et al., 1999;Merz and So, 1997). The role and the mechanisms by which these cytoskeletal modifications eventually lead to bacterial internalization remained mostly unknown. Recently we have observed in endothelial cells that ErbB2, a receptor-tyrosine kinase of the EGF receptor family, is specifically recruited underneath bacterial colonies following pilus-mediated adhesion. ErbB2 is activated by homodimer formation and is required for an efficient bacterial internalization, whereas ErbB2-associated signaling is not involved in bacteria-induced cytoskeletal modifications (Hoffmann et al.,2001).

The ERM proteins play a major structural and regulatory role in many of the morphogenic changes of the plasma membrane, including the formation of microvilli (Yonemura and Tsukita,1999). These proteins control the organization of the cortical cytoskeleton by acting as linkers between the plasma membrane and the actin cytoskeleton. ERM proteins interact through their amino-terminal domain with the cytoplasmic domain of transmembrane proteins (so-called ERM binding proteins), such as CD44 or ICAM-1, and interact with F-actin by their C-terminal domain. ERM proteins are in a cytosolic dormant form, in which the binding site for F-actin is masked by an intramolecular interaction between the N- and C-terminal domains (Mangeat et al., 1999). Activation of ERM seems to require the binding of phosphotidylinositol 4,5-biphosphate and the phosphorylation of a specific threonine residue in their C-terminal domain by Rho-kinase, an effector of Rho GTPase (Gautreau et al., 2000;Matsui et al., 1998;Matsui et al., 1999).

In this work, we show that the internalization of piliated capsulated meningococci in endothelial cells, a process associated with meningeal invasion, is triggered by the formation of microvilli-like membrane protrusions, but not ruffles, containing ezrin, moesin and the ERM-binding proteins CD44 and ICAM-1. The role of ezrin recruitment in actin polymerization was confirmed by expression of a dominant-negative form of ezrin. Furthermore, we provide evidence that both Rho and Cdc42, but not Rac1,are critical for actin cytoskeletal modifications induced by N. meningitidis and subsequent bacterial internalization, although ERM binding to the plasma membrane occurs through a Rho-GTPase-independent pathway. All together, these data indicate that the formation of microvilli-like membrane protrusions at the surface of human endothelial cells is critically involved in the internalization of piliated capsulated virulent N. meningitidis.

The strain used in this study, designated ROU (group W135, ET37), is a piliated capsulated isolate obtained from the cerebrospinal fluid of a 2-month-old infant and has been previously described(Pron et al., 1997). Strains were routinely grown on gonococcal base medium (GCB containing 4% medium base,1% agar, 0.4% glucose, 0.2 mg/ml thiamine, 0.0005%(Fe(NO₃)3.9H₂O, 0.01% L-glutamine) at 37°C under 5%CO₂. The PilC^- derivative of this isolate, which is incapable of pilus-mediated adhesion, has been previously described(Pron et al., 1997).

Human umbilical vein endothelial cells (HUVECs; PromoCell, Heidelberg,Germany) were used between passages 1 and 8 and grown in Endo-SFM supplemented with 10% heat-inactivated FCS and 2 mM L-Glutamine (Life-Technologies, Grand Island, USA), 1 ng/ml of bFGF (Boehringer-Mannheim, Meylan, France), 0.5 UI/ml of heparin and 1.25 μg/ml of endothelial cell growth supplement (Sigma,Saint Louis, USA). The Human bone marrow endothelial cell line (HBMECs)(Schweitzer et al., 2000) was kindly provided by B. Weksler (Weill Medical College of Cornell University,NY, USA). Cells were cultured in DMEM-glutamax (Life-Technologies, Grand Island, USA) supplemented with 10% heat inactivated FCS, 7 UI/ml of heparin,7.5 μg/ml of endothelial cell growth supplement and 10 mM Hepes. Cells were seeded at 5×10⁴ cells/cm² in a 24-well or 6-well culture plates and grown at 37°C in a humidified incubator under 5%CO₂ for 2 to 3 days.

The day before the infection, the culture medium was replaced by a serum-free medium (starvation medium). Approximately 10⁷ bacteria in culture medium were added to the cells and allowed to adhere for 15-30 minutes. The monolayers were then washed every hour and new medium was added to avoid reinfection from the supernatant. At the indicated times, monolayers were harvested and the number of adhesive bacteria determined.

The number of internalized bacteria was determined by a gentamicin protection assay. Cells grown in a 24-well or 6-well culture plate were infected as above. 4 and 7 hours after infection, cells were washed and 150μg/ml of gentamicin was added to each well to kill extracellular bacteria. After 1 hour of incubation at 37°C, cells were scraped off the plates, and the number of intracellular bacteria was then determined. The proportion of internalized bacteria was calculated as the ratio of those that were gentamicin-resistant to adherent bacteria.

Where indicated, cells were pretreated with the following compounds,diluted in starvation medium: 1 ng/ml of Clostridium difficile toxin B (ToxB), a non-selective inhibitor of the Rho family GTPases(Boquet, 1999), kindly provided by P. Boquet (INSERM, Nice, France) or 30 μM of Y27632, a selective inhibitor of Rho kinase (Maekawa et al.,1999), an effector of the Rho protein, a gift from Yoshitomi Pharmaceutical (Osaka, Japan). These treatments were initiated 16 hours or 1 hour respectively, prior to the addition of bacteria and were maintained during the course of the experiments.

Anti-human CD44 mAb was purchased from R&D (Abington, UK), anti-ICAM-1,anti-VE-cadherin mAbs were from Serotec (Oxford, England), anti-paxillin mAb was from Transduction Laboratories (Lexington, USA) and anti-human vinculin mAb from Sigma (Saint Louis, USA). Phalloidin-alexa488 (Molecular Probes,Eugene, Or) was used to stain the actin cytoskeleton (F-actin). ERM proteins were detected using selective rabbit polyclonal antisera kindly provided by P. Mangeat (CNRS, Montpellier, France). mAb 26C4 specifically raised against RhoA(Lang et al., 1993) is a gift from J. Bertoglio (INSERM, Chatenay-Malabry, France). In infected monolayers,bacteria were stained using either ethidium bromide or a rabbit polyclonal antibody directed against the meningococcus strain Rou. This antibody was generated as follows: an overnight culture of Rou was either fixed with 2.5%paraformaldehyde or sonicated. These preparations were mixed with Freund's complete adjuvant (1:1), and 1 ml was used for inoculation of a New Zealand white rabbit. The rabbit was boosted with the same mixture three times every 2 weeks. Microinjected proteins were detected by immunostaining using antimyc 9E10 mAb from Biomol (Plymouth, USA) or anti-VSVG P5D4 mAb from Roche(Indianapolis, USA).

HUVECs were plated at a density of 5×10⁴cells/cm² onto 12-mm diameter acid-washed glass coverslips coated overnight with fibronectin (10 μg/ml). Cells were infected as described above, fixed with 2.5% paraformaldehyde in PBS for 20 minutes, neutralized with 0.1 M glycine in PBS for 5 minutes, permeabilized for 1 minute with 0.5%Triton X100 in PBS and then saturated for 20 minutes with PBS containing 0.2%gelatin before incubation for 30 minutes with the primary antibodies diluted in PBS/gelatin. Cells were then washed with PBS and incubated for 30 minutes with anti-rabbit or anti-mouse secondary antibodies conjugated with Cy3 or Cy5 and immunoabsorbed against human, bovine, rabbit or mouse serum proteins(Jackson Immunoresearch, West Grove, USA). Finally, cells were washed three times in PBS and mounted in moviol (Sigma, Saint Louis, USA) before analysis with a Zeiss LSM 510 confocal microscope. The pinhole of each PMT was adjusted separately in order to take XY optical sections with the same step. Sequential scannings were also performed in order to confirm the absence of any interference between signals detected at different wavelengths. Series were subjected to an orthogonal projection in order to make a two-dimensional reconstruction of the area under study.

Microinjections were performed using an Eppendorf Transjector 5246 coupled to an Eppendorf Micromanipulator 5179. HUVECs were seeded onto glass coverslips as described for the immunofluorescence protocol. About 50 cells per well were microinjected into the nucleus with the eukaryotic expression vector. The vector used was pRK5-myc, encoding the dominant-negative forms of Rac or Cdc42, RacN¹⁷ or Cdc42N¹⁷(Nobes and Hall, 1999). These plasmids were kindly provided by G.Tran Vhan Nieu (Institut Pasteur, Paris). pRK5-myc-RacN¹⁷ was microinjected at 117 μg/ml 18 hours before cell infection and pRK5-myc-Cdc42N¹⁷ at 133 μg/ml 5 hours before cell infection. Eukaryotic expression vectors encoding either the N- terminal domain of human ezrin (amino acids 1-309), a dominant-negative form of ezrin(pcB6-ezrin-Nter) or the full-length ezrin (pCB6-Ezrin) were obtained from A. Gautreau (Institut Curie, Paris). They were microinjected into the nuclei of HUVECs at a concentration of 120 μg/ml 2 hours prior infection of the monolayers with N. meningitidis.

The recombinant C3-transferase from Clostridium botulinum(Boquet, 1999), a Rho-selective inhibitor generously provided by M. Popoff (Institut Pasteur, Paris) was injected into the cytoplasm at 200 ng/ml with 2.5 mg/ml rhodamine-labelled dextran diluted in distilled water, 2 hours before infection by N. meningitidis as described above. Control injections carried out with rhodamine-labelled dextran did not produce any significant effect on cell morphology or actin organization.

For scanning electron microscopy, cells were cultured and infected as described above, then fixed with 2.5% glutaraldehyde solution in 0.1 M cacodylate buffer pH 7.3 Preparations were then coated with gold palladium after critical-point drying. The examination was performed on a JEOL 840A at the Centre Inter-Universitaire de Microscopie Electronique (Paris,France).

For transmission electron microscopy, cells were grown on transwells as described above and were fixed overnight at 4°C with a 1:1 mixture of 2.5%glutaraldehyde and 2.5% paraformaldehyde in cacodylate sucrose buffer (0.1 M cacodylate, 0.1 M sucrose, 5 mM CaCl₂, 5 mM MgCl₂, pH 7.2). Monolayers were then stained for 1 hour in a solution of 1%OsO₄ and placed for 1 hour in 1% uranyl acetate. After dehydration in a graded series of alcohols, cells were embedded with polyester filter in Epon. Thin sections were obtained by using an Ultracut ultramicrotome and analyzed with JEOL-100CX electron microscope.

In situ observations have demonstrated that meningococcal interactions with endothelial cells are a key event in the crossing of brain microvessels in the meninges and choroid plexuses (Pron et al., 1997). We therefore analyzed the interaction of N. meningitidis with human endothelial cells of both macrovascular and microvascular origin (HUVECs and HBMECs, respectively). The strain used in this work was a piliated capsulated isolate of ROU, obtained from the cerebrospinal fluid of a patient with fulminant meningococcal infection. Scanning electron microscopies are shown inFig. 1. As described for the interaction of N. meningitidis with epithelial cells(Pujol et al., 1997;Pujol et al., 1999), a two-step process was observed with bacteria interacting with endothelial monolayers. Following the initial attachment, bacteria proliferate and form small colonies, a step which is referred to as `localized adhesion'(Fig. 1A). Bacteria then disperse on the apical surface of the cells, forming a single monolayer covering the cells, a step referred to as `diffuse adhesion'(Fig. 1C). As expected, neither an interaction nor bacterial proliferation was observed with a PilC^- capsulated derivative of ROU, which is incapable of pilus-mediated adhesion (data not shown).

Prior to bacterial infection, no membrane protrusion was observed on endothelial cells, whereas during localized adhesion of N. meningitidis, small protrusions of the cellular membranes were induced beneath and around bacteria of most colonies(Fig. 1A,B). The formation of these membrane protrusions was transient and disappeared during the stage of diffuse adhesion (Fig. 1C,D). At this later time point, even though bacteria remained highly adhesive to the endothelial cell surface, no intimate adhesion with pedestal formation was observed with endothelial cells, in contrast to our previous observations with infected epithelial cells (Pujol et al.,1999). Moreover, as shown inFig. 1, when interacting with endothelial cells, meningococci remained mostly extracellular. However, in some cases, cellular protrusions can be seen surrounding and engulfing meningococci (Fig. 1B). Transmission electron microscopy analysis(Fig. 2A-D) confirmed that bacteria can be found surrounded by such membrane protrusions, leading to the formation of large vacuoles containing meningococci. These observations strongly suggest that these protrusions are responsible for meningococcal internalization

To confirm that bacterial internalization is correlated with the formation of these cellular protrusions induced during the localized adhesion of N. meningitidis, gentamicin protection assays were performed at 4 hours(localized adhesion) and 8 hours (diffuse adhesion) after infection of the endothelial monolayers. As shown in Fig. 3, whereas the number of adhering bacteria dramatically increased between 4 hours and 8 hours, owing to extracellular proliferation, the number of internalized bacteria remained relatively unchanged during the same time,thus indicating that bacterial internalization predominantly takes place during the localized adhesion, when cellular protrusions are observed. Taken together, these observations suggest that piliated capsulated meningococci induce the transient formation of cellular protrusions during the initial adhesion phase of N. meningitidis infection and that these cytoskeletal modifications are associated with the internalization of a small fraction of adhesive bacteria. Ezrin is required for the assembly of the microvilli-like cellular protrusions

In order to gain insight into the molecular mechanisms responsible for the formation of the cellular protrusions induced by pilus-mediated adhesion of N. meningitidis, we identified components involved in their organization. Since the cortical actin-based cytoskeleton plays a crucial role in local membrane reorganization, we assessed, by immunofluorescence analysis,whether actin was polymerized in the membrane protrusions induced by N. meningitidis. As shown in Fig. 4A, localized polymerization of cortical actin was observed underneath bacterial colonies. Moreover, xz section analysis of the actin staining (Fig. 4D) revealed that cortical actin was indeed polymerized in the membrane protrusions.

We then analyzed the distribution of various proteins known to link the actin cytoskeleton and membrane components. Paxillin(Fig. 4B) and vinculin (data not shown) were not recruited to the site of polymerization of cortical actin beneath bacterial colonies but remained highly enriched at the tip of actin stress fibers in focal adhesions, indicating that cytoskeletal modifications induced by N. meningitidis differed from focal adhesions or focal complexes that are seen in ruffles (Yamada and Geiger, 1997; Zamir et al., 1999). On the other hand, using specific antibodies, ezrin(Fig. 4C) and moesin (not shown) were found to be recruited underneath bacterial colonies to the site of actin polymerization together with the ERM-binding proteins CD44(Fig. 4C) and ICAM-1 (not shown), whereas radixin, another member of the ERM family, remained localized at cell junctions (not shown). In addition, xz section analysis of ezrin distribution in infected cells showed a massive recruitment just beneath the plasma membrane at the tip of the cellular protrusions(Fig. 4D), reminiscent of ezrin distribution in epithelial microvilli(Yonemura and Tsukita,1999).

The ERM N-terminal domain interacts with ERM-binding proteins whereas the C-terminal domain binds to F-actin. Indeed, the N-terminal domain (amino acids 1-309) of ezrin can compete with ezrin for interaction with ERM-binding proteins, thus behaving as a dominant-negative form of ezrin. In order to assess the actual function of ezrin in the assembly of the bacteria-induced cellular protrusions, we expressed its N-terminal domain in endothelial cells before N. meningitidis infection. As shownFig. 5A, expression of this dominant-negative form of ezrin not only inhibited the recruitment of endogenous ezrin but also prevented polymerization of actin underneath bacterial colonies. This observation was further confirmed by cell counting(Fig. 5B). It should be pointed out that microinjection of full-length ezrin had no effect on actin polymerization as shown in Fig. 5C, thus eliminating the possibility that the lack of actin polymerization observed in Fig. 5B was a consequence of microinjection. These data clearly indicate that ezrin is playing a pivotal role in the cytoskeletal modifications induced by pilus-mediated adhesion of N. meningitidis.

To investigate the putative role of Rho family GTPases in the organization of membrane protrusions induced by N. meningitidis, endothelial cell monolayers were incubated with Clostridium difficile toxin B (Tox B),which non-selectively inhibits all members of the Rho family(Boquet, 1999). The number of bacterial colonies inducing cortical actin polymerization and/or ezrin recruitment was measured (Fig. 6A and 6B). Tox B induced a dramatic decrease in the number of colonies associated with cortical actin polymerization(Fig. 6B). However, the number of bacterial colonies recruiting ezrin was not affected(Fig. 6A), although, as shown in Fig. 6C, ezrin recruitment was less pronounced than that observed with no ToxB treatment when actin was polymerized. We then conclude that Rho family GTPases are involved in cortical actin polymerization but not in the initial ezrin and ezrin-binding protein recruitment induced by N. meningitidis underneath bacterial colonies.

We next determined which GTPases of the Rho family were involved in the cytoskeletal modifications induced by N. meningitidis. For that purpose, we used a monoclonal antibody specifically raised against RhoA, 26C4(Lang et al., 1997). Immunofluorescence analysis of RhoA distribution in infected cells demonstrated an enrichment of this molecule in the microvilli-like structures (Fig. 7A). Furthermore, the role of Rho was further analyzed using (i) Clostridium botulinum C3 transferase, a selective inhibitor of Rho,and (ii) Y27632, a selective inhibitor of Rho kinase, a Rho effector(Maekawa et al., 1999). Under the conditions used, Y27632 induced a loss of actin stress fibers, without inducing cell rounding, and prevented actin polymerization underneath bacterial colonies (Fig. 7B). Indeed, the number of bacterial colonies able to induce cortical actin polymerization was significantly lower in cells incubated with Y27632 than in control cells (Fig. 8). Similar results were obtained after microinjection of C3 transferase(Fig. 8). All together, these data demonstrate the role of Rho and of its effector Rho kinase in mediating cortical actin polymerization induced by N. meningitidis.

We next investigated whether the related proteins, Rac1 and Cdc42, also affected N. meningitidis-induced actin polymerization. HUVECs were microinjected with plasmids encoding Rac1N¹⁷ and Cdc42N¹⁷, the dominant-negative forms of Rac1 and Cdc42,respectively. As shown in Fig. 8, the number of bacterial colonies with polymerized actin was significantly reduced in endothelial cells microinjected with Cdc42N¹⁷, whereas microinjection of Rac1N¹⁷ had no effect. Fig. 7C shows that microinjection of Cdc42N¹⁷ induced a dramatic increase in the formation of stress fibers, a cytoskeletal organization controlled by Rho(Hall, 1998), without inducing polymerization of cortical actin underneath bacterial colonies (see xz section in Fig. 7C). This increase in the formation of stress fibers in cells microinjected with a dominant-negative form of Cdc42 indicates that the effect of Cdc42 N¹⁷ on actin cytoskeleton is not due to an inhibition of Rho activity.Fig. 7D confirmed that microinjection of the dominant-negative form of Rac1 had no effect on cortical actin polymerization induced by N. meningitidis, even though microinjected cells were retracted and had no stress fibers, as expected from Rac1 inhibition (Hall, 1998). Taken together, these data demonstrate that the cytoskeletal modifications induced by N. meningitidis required the activation of both Cdc42 and Rho but not of Rac1.

To confirm that bacterial internalization is promoted by the induction of these cytoskeleton changes, we analyzed the effect of ToxB and Y27632 on bacterial adhesion and entry into endothelial cells. As shown inFig. 9, bacterial entry into cells treated by either of these two inhibitors was significantly decreased,whereas N. meningitidis adhesion was unaffected. These data demonstrate that actin cytoskeleton reorganization induced by both Cdc42 and Rho, via Rho kinase activation, are crucial for the internalization of N. meningitidis into endothelial cells. They suggest that in vivo activation of both Cdc42 and Rho during pilus-mediated adhesion onto brain endothelial cells may be involved in meningeal invasion by N. meningitidis.

Few bacterial pathogens are capable of crossing the blood brain barrier. It is therefore likely that these neuroinvasive bacteria have developed sophisticated tools that enable them to do so, either by a paracellular or a transcellular route. In the case of Streptococcus pneumoniae,trafficking towards the cerebral tissue seems to occur by a transendothelial migration induced by the interaction between bacterial phosporylcholine and the endothelial receptor of platelet-activating factor(Cundell et al., 1996;Cundell et al., 1995;Ring et al., 1998). Previously reported in vivo data have clearly demonstrated that an important step in meningeal invasion by N. meningitidis is its interaction with brain endothelial cells. In this study, we present data that indicate that meningococci are internalized into vascular endothelial cells by triggering multiple signaling pathways within the host cells. We provide evidence that the cytoskeletal modifications associated with the internalization of N. meningitidis into endothelial cells are associated with the formation of cellular protrusions, which are reminiscent of epithelial microvilli in terms of shape and ezrin content. These structures are probably responsible for bacterial engulfment and internalization. On the basis of the capacity of an ezrin dominant-negative form to block actin polymerization, our data suggest that these events occur in a two-step process: first, the recruitment of ezrin and moesin underneath bacterial colonies; second, the polymerization of cortical actin. Moreover, the present study shows that the first step is not dependent on Rho GTPases, whereas the second step requires the activation of Rho, and its downstream effector Rho kinase, as well as Cdc42, but not Rac1.

The precise mechanism by which pilus-mediated adhesion is recruiting ezrin is still unknown. ERM proteins have been shown to be present in the cytoplasm as oligomers, which can be recruited to the cytoplasmic membrane where threonine phosphorylation in the C-terminal domain is required for the transition to monomers. Only monomers can crosslink actin filaments to the plasma membrane to form microvilli in epithelial cells(Gautreau et al., 2000). The N-terminal domain is responsible for membrane targeting via interaction with ezrin-binding proteins, such as CD44 or ICAM-1. Indeed, CD44 has been proposed to be involved in the formation of microvilli by concentrating activated ERM proteins at the plasma membrane (Yonemura et al., 1998; Yonemura and Tsukita, 1999). We therefore hypothesize that, during pilus-mediated adhesion of N. meningitidis, the recruitment of ezrin-binding proteins might similarly target ezrin and moesin to the plasma membrane, underneath bacterial colonies, where they would be activated and would trigger actin polymerization.

Several studies have shown that Rho stimulates the activity of ezrin and moesin (i) by inducing their recruitment to the plasma membrane, (ii) by promoting the formation of phosphotidylinositol 4,5-biphosphate, which is responsible for an activating conformational change in ERM proteins, and(iii), although still a matter of debate(Matsui et al., 1998;Matsui et al., 1999), by activating Rho kinase, which can phosphorylate ERM proteins in their C-terminal domain. Surprisingly, we observed in the present study that ToxB, a non-selective inhibitor of Rho family GTPases, did not affect the recruitment of ERM proteins or ERM-binding proteins underneath bacterial colonies. Further investigation will aim at identifying the alternative signaling pathways involved in ERM protein recruitment promoted by the pilus-mediated adhesion of N. meningitidis.

Our data using specific inhibitors of the Rho family GTPases demonstrate the crucial role of both Rho and Cdc42 in N. meningitidis-induced actin polymerization and bacterial entry. However, their respective contribution still remains unknown, as does the identity of Cdc42 targets. Considering the role of Rho kinase in actin polymerization induced by N. meningitidis, it is tempting to speculate that the role of Rho activation is to promote ezrin phosphorylation following its recruitment to the plasma membrane at the site of bacterial attachment. Since a role for Cdc42 in the formation of microvilli-like structures has previously been reported in epithelial cells (Gauthier-Rouviere et al.,1998), we hypothesize here that it might be responsible for actin nucleation leading to the formation of the microvilli-like membrane protrusions in endothelial cells.

Interestingly, the membrane protrusions induced by N. meningitidison the endothelial surface significantly differ from those induced by other known invasive bacteria. Cytoskeletal modifications leading to the formation of ruffles have been shown to be responsible for the internalization of bacterial pathogens such as Shigella, Salmonella and, more recently, Neisseria gonorrhoeae (Edwards et al., 2000; Galan and Zhou,2000; Nhieu and Sansonetti,1999). In the case of Shigella, ezrin recruitment was shown to be dependent on bacteria-induced actin polymerization(Skoudy et al., 1999). Moreover, Shigella-induced protrusions are enriched in paxillin and vinculin, two proteins usually localized at focal adhesions, which are not recruited in N. meningitidis-induced membrane protrusions. Indeed,the lack of ruffle formation reported in this study in response to N. meningitidis adhesion was correlated with our observation that Rac1,which generally controls ruffle formation(Hall, 1998), is not involved in bacteria-induced actin cytoskeleton modifications in endothelial cells.

Despite the lack of the type III secretion system, which is involved in Shigella and Salmonella internalization, virulent capsulated N. meningitidis are internalized into host cells. This internalization is likely to promote their subsequent transcytosis through cellular barriers in vivo. As previously mentioned, type IV pili are the only bacterial components identified so far involved in the formation of the microvilli-like membrane protrusions induced by N. meningitidis. That pilusmediated adhesion has been shown to be important for meningeal invasion(Pron et al., 1997) is consistent with the role of these cytoskeletal modifications in the crossing of the cerebral vascular endothelium. It is, however, possible that additional meningococcal components are involved in the signaling events leading to the formation of the membrane protrusions observed here. Recently, a role for the lipooligosaccharide (LOS) in the actin cytoskeleton rearrangement induced by piliated Neisseria gonorrhoeae has been shown(Song et al., 2000). Gonococcal strains expressing LOS lacking the lacto-N-neotetraose did not promote actin polymerization and were less invasive. However, we did not observe here such inhibition of actin polymerization with a mutant of the ROU strain lacking the LOS core (data not shown).

The present study thus provides new insights into the molecular mechanisms used by N. meningtidis to enter endothelial cells, a critical step in the pathogenesis of this bacteria for its trafficking through the cerebral vasculature. Further studies on the meninogoccal components and their cellular receptors involved in this interaction should provide essential clues to the understanding of the molecular mechanisms of N. meningitidisinfection of the central nervous system.

We thank Patrice Boquet, Guy Tran Van Nhieu, Paul Mangeat, Alexis Gautreau,and Michel Popoff for fruitful discussions and for providing us with materials, Joanne Dove for careful reading of the manuscript. X.N.'s laboratory is supported by INSERM, and the Université RenéDescartes ParisV. This work was supported in part by a special grant from the`Programme de Recherche Fondamentale en Microbiologie et Maladies Infectieuses'.