Clostridium difficile Toxin CDT Induces Formation of Microtubule-Based Protrusions and Increases Adherence of Bacteria

Clostridium difficile causes antibiotic-associated diarrhea and pseudomembranous colitis by production of the Rho GTPase-glucosylating toxins A and B. Recently emerging hypervirulent Clostridium difficile strains additionally produce the binary ADP-ribosyltransferase toxin CDT (Clostridium difficile transferase), which ADP-ribosylates actin and inhibits actin polymerization. Thus far, the role of CDT as a virulence factor is not understood. Here we report by using time-lapse- and immunofluorescence microscopy that CDT and other binary actin-ADP-ribosylating toxins, including Clostridium botulinum C2 toxin and Clostridium perfringens iota toxin, induce redistribution of microtubules and formation of long (up to >150 µm) microtubule-based protrusions at the surface of intestinal epithelial cells. The toxins increase the length of decoration of microtubule plus-ends by EB1/3, CLIP-170 and CLIP-115 proteins and cause redistribution of the capture proteins CLASP2 and ACF7 from microtubules at the cell cortex into the cell interior. The CDT-induced microtubule protrusions form a dense meshwork at the cell surface, which wrap and embed bacterial cells, thereby largely increasing the adherence of Clostridia. The study describes a novel type of microtubule structure caused by less efficient microtubule capture and offers a new perspective for the pathogenetic role of CDT and other binary actin-ADP-ribosylating toxins in host–pathogen interactions.

CDT belongs to the family of binary actin-ADP-ribosylating toxins [12,13], which are produced by various pathogenic species of the genus Clostridium and Bacillus, including C. botulinum (C2 toxin), C. perfringens (iota toxin), C. spiroforme (C. spiroforme toxin, CST) and B. cereus (vegetative insecticidal proteins, VIP) [1,14]. These toxins consist of a biologically active, actin-modifying ADPribosyltransferase and a separated binding component, which is involved in binding and transport of the enzyme component into the cytosol of target cells.
In the cytosol, the enzymatic component of the toxins ADPribosylates G-actin at Arginine-177 [15,16], thereby blocking actin polymerization. Moreover, ADP-ribosylated actin acts as a capping protein at barbed ends of actin filaments to inhibit elongation and favoring depolymerization of F-actin [17,18].
Although the effects of ADP-ribosylating toxins on actin are well understood, their role as virulence factors is not clear. Here, we report that the actin-ADP-ribosylating toxins C. difficile CDT, C. perfringens iota toxin and C. botulinum C2 toxin not only affect the actin cytoskeleton but induce the formation of a novel type of microtubule structures, consisting of long microtubule-based protrusions on the surface of epithelial cells, leading to increased adherence of Clostridia.

Morphological changes of colon epithelial cells induced by actin-ADP-ribosylating toxins
To test the effects of the actin-ADP-ribosylating toxin CDT, we employed the human colon carcinoma cell line Caco-2. Treatment with the toxin caused shrinking and rounding-up of target cells. These morphological changes are typical for all binary actin-ADP-ribosylating toxins and depend on ADP-ribosylation of actin [12,19]. To get more information on toxin-induced morphological changes, we made use of time-lapse-microscopy ( Figure 1A). Surprisingly, we detected toxin-induced formation of cellular processes, occurring earlier than shrinking of the cell body. About 60-90 min after application of CDT (20 ng/ml CDTa and 40 ng/ml CDTb), the binary toxin induced formation of thin cell processes that extended from the surface of epithelial cells ( Figure 1A, Video S1). The structures were different in length, usually ranging from 5 -.150 mm with a diameter ranging from 0.05 to 0.5 mm. Neither the enzyme component nor the binding component of CDT alone had any effects on cell morphology and process formation, indicating the specificity of the toxin effect ( Figure S1A). Moreover, intoxication with a catalytic inactive mutant of CDTa (E430Q) [20] revealed no formation of protrusions ( Figure S1B). On the other hand, identical protrusions were caused by treatment of epithelial cells with other actin-ADPribosylating toxins like C. botulinum C2 toxin and C. perfringens iota toxin ( Figure S1C). In addition to Caco-2 cells, formation of toxininduced protrusions was observed in human colon carcinoma HT-29 cells (see below scanning electron microscopy, Figure 1C, right panel) as well as in primary intestinal epithelial cells isolated from the rat colon ( Figure S1D).
Quantification (for method see Figure S2) of toxin-induced formation of protrusions revealed that maximal effects occurred at medium concentrations of CDT (20 ng/ml CDTa and 40 ng/ml CDTb) ( Figure 1B). At high concentrations of CDT (150 ng/ml CDTa and 300 ng/ml CDTb) outgrowth of protrusions was rapid and occurred within 40 min but length of protrusions was reduced as compared to medium concentrations. At 5 ng/ml CDTa and 10 ng/ml CDTb, first protrusions were formed after ,120 min. Similar results were obtained with C2 toxin (data not shown).
Next we employed scanning electron microscopy. These studies showed that CDT-induced formation of protrusions occurred all over the cell surface of Caco-2 cells ( Figure 1C). Whereas with time lapse DIC microscopy formation of straight protrusions at rather early time points was observed, scanning electron microscopy at later time points (.3 h) revealed formation of a meshwork of cellular extensions. The protrusions were bended and exhibited contact with each others. Interestingly the shape of protrusions was often twisted in a corkscrew-like manner with small leaf-like extensions (arrows in Figure 1C, lower panel). These structures were clearly distinguishable from short microvilli observed on control cells by their shape and their variation in length ( Figure 1C) [21][22][23]. A similar formation of protrusions and a disappearance of microvilli were observed on polarized Caco-2 cells that were grown on filters for 2 weeks to ensure polarization ( Figure S3A). In these experiments the cell borders and the presence of tight junctions indicative for polarization were determined by occludin and ZO-2 stainings, respectively ( Figure  S3B+C).

Binary toxins affect microtubules
Cell surface projections such as filopodia, microspikes or microvilli harbor actin filaments which are thought to be their key structural component [23]. Therefore, we applied TRITCphalloidin to study whether the toxin-induced protrusions contained F-actin. Whereas in control cells cortical filamentous actin was stained at cell borders (Figure 2A), treatment with CDT (50 ng/ml CDTa and 100 ng/ml CDTb) for 1 h induced redistribution of cortical actin filaments and formation of unstructured F-actin dots. After 2 h, the toxin caused severe destruction of the actin cytoskeleton. However, the protrusions induced by the actin-ADP-ribosylating toxins were not stained by TRITC-phalloidin. By contrast, indirect immunofluorescence with anti-a-tubulin antibodies revealed that the toxin-induced protrusions consisted of microtubules ( Figure 2A). One hour after addition of toxin, microtubules started to form bundles, grew along the cell cortex and/or crossed cell borders by forming protrusions at the cell surface. After 2 h of toxin treatment, long bundles of microtubules projected from the cell borders and the intracellular microtubule meshwork was redistributed. Additionally, by confocal microscopy of the cell surface of polarized Caco-2 cells the meshwork of microtubule-based protrusions was clearly visible ( Figure S3B+C). A magnification ( Figure S3D) of the surface identified the structures seen in scanning electron microscopy ( Figure S3A). Altogether, the microscopic studies indicated that the toxins mainly affected the microtubule network at the cell cortex. Accordingly, formation of protrusions was inhibited by addition of nocodazole (not shown).
To study the microtubule-based protrusions in more detail, we used antibodies specific for post-translational modifications of tubulin. In normal growth conditions, Caco-2 cells displayed very low amounts of detyrosinated a-tubulin (Glu-tubulin), which was not increased after disruption of the actin cytoskeleton by toxins ( Figure S4A). Also the amount of acetylated tubulin was not enriched after toxin treatment ( Figure S4B). As both detyrosinated tubulin and acetylated tubulin are markers for more stabilized microtubules [24,25], these results indicate that the microtubulebased protrusions mainly consist of dynamic microtubules. Accordingly, antibodies specific for tyrosinated a-tubulin stained microtubules in the protrusions ( Figure S4C).
To study whether formation of microtubule protrusions was accompanied and/or preceded by ADP-ribosylation of actin, we analyzed CDT-catalyzed modification of actin in Caco-2 cells in a time course by differential ADP-ribosylation ( Figure 2B). To this end, intact cells were treated with CDT for the indicated times. Thereafter, cells were lysed and cellular proteins were incubated with radioactively labeled NAD. 32 P-ADP-ribosylation of actin was induced by addition of the enzyme component C2I of C2 toxin, which also modifies actin at Arg177. One hour after

Author Summary
Clostridium difficile is responsible for ,20 percent of antibiotic-related cases of diarrhea and nearly all cases of pseudomembranous colitis. The pathogens produce two protein toxins (toxins A and B), which inactivate Rho-GTPases of host cells by glucosylation. Recently emerging hypervirulent strains of C. difficile release higher amounts of toxins A and B, are resistant towards fluoroquinolones and produce an additional protein toxin called C. difficile transferase (CDT). CDT is a binary toxin, which modifies Gactin by ADP-ribosylation, thereby inhibiting actin polymerization. So far the pathogenetic role of CDT is not clear. Here we studied the effects of CDT on human colon carcinoma cells and show that the toxin causes rearrangement of microtubules and formation of long cellular protrusions. The microtubule-based protrusions form a dense meshwork at the cell surface, which wrap and embed Clostridia, thereby increasing adherence of the pathogens. We observed similar effects with other members of the family of binary actin-ADP-ribosylating toxins like C. botulinum C2 toxin and C. perfringens iota toxin. Our findings show a novel type of microtubule structures induced by actin-ADP-ribosylating toxins and propose an important role of these toxins in hostpathogen interactions by their effects on adherence and colonization of Clostridia.
treatment of cells with CDT, the 32 P-ADP-ribosylation of actin in the cell lysate was reduced by ,30%. At this time point the first microtubule extensions were formed at the cell surface, suggesting that formation of protrusions and ADP-ribosylation of actin are concomitant processes.
To address the question, whether the formation of microtubule protrusions was a consequence of actin disruption or an actin independent effect, we tested the microfilament destabilizing drugs cytochalasin D and latrunculin B ( Figure S5A) [26]. Both cytochalasin D and latrunculin B induced microtubule protrusions; however, they induced protrusions less efficiently than CDT and rather high concentrations (e.g., 0.5-5 mM latrunculin B or cytochalasin D) were necessary to achieve similar effects on microtubules as CDT. The effects induced by cytochalasin D or latrunculin B started early after addition but the process of outgrowth of protrusions was less pronounced. When CDT was applied 10 min after latrunculin B, the typical strong effect of the ADP-ribosylating toxin on the formation of microtubule protrusions was still observed (data not shown).
Treatment of Caco-2 cells with jasplakinolide, which is known to stabilize F-actin and to enhance polymerization [27], delayed and reduced the formation of microtubule protrusions after treatment with CDT ( Figure S5B). By contrast, the Rho GTPase-inactivating toxins C. difficile toxin B and C. limosum C3 toxin, both of which alter the dynamics of the actin cytoskeleton, did not induce microtubule protrusions ( Figure S5C).

Binary toxins influence microtubule plus end tracking proteins
Our studies showed that CDT affects microtubule formation especially at the cell cortex. Since the dynamics and capturing of microtubules is mainly controlled by plus end-tracking proteins (+TIPs), we studied +TIP end-binding protein 1 (EB1), which is a marker of polymerizing microtubules [28,29]. Staining of EB1 in control cells revealed a typical pattern with short comet-like structures at the tip of microtubules ( Figure 3A) [30]. Measurement of comets yielded a length of ,2.04+/20.04 mm. Treatment with CDT increased the length of EB1-decorated tips of microtubules to 3.57+/20.08 mm (p,0.001) ( Figure 3C). Also the ends of most microtubules that crossed cell borders and formed protrusions were decorated with EB1. This indicated that microtubules, which projected from the cell surface into the protrusions, were still in their growth phase. Moreover, these findings suggested that toxin-induced formation of protrusions is at least partially generated by tubulin polymerization and rather not by a mechanism that involves sliding of microtubules. Toxininduced elongation of EB1 comet length was also observed in HT29 cells and in primary epithelial cells from the rat colon (data not shown).
Similar results as observed for EB1 were obtained for cytoplasmic linker protein 170 (CLIP-170), which is another +TIP member [31] ( Figure 3B). The length of microtubule tip decoration with the rescue factor CLIP-170 increased after the treatment with CDT ( Figure 3C). Microtubule protrusions, which extended the cell surface, were decorated with both EB1 as well as CLIP-170 after treatment. Additionally, CLIP-115 and EB3 displayed a similar increase in microtubule tip decoration ( Figure  S6) [32].
To analyze the consequences of increased +TIP-association on microtubule dynamics, a GFP-version of the +TIP EB3 was expressed and the cells were analyzed by fluorescence time-lapse microscopy with high temporal resolution ( Figure 4A+B, Video S2). Only cells, expressing low levels of the GFP construct, were selected for analysis. The elongation of +TIPs after CDT treatment was also evident in EB3-GFP-transfected cells. The length of GFP-positive comets before and after the treatment with CDT was similar to the endogenous EB1-comets in immunofluorescence. This reflects the similar behavior of the endogenous +TIP as compared to the expressed recombinant EB3-GFP protein. In control cells, +TIP comets faded as soon as they reached the cell cortex. By contrast, in CDT-treated cells the EBcomets crossed the cell cortex and formed initial protrusions. Often microtubules grew along already existing ones as seen by the appearance of EB-comets using the same track into the protrusions. By means of EB3-GFP the rate of polymerization of ,100 steadily growing microtubules was measured before and after 1 and 2 h of treatment with CDT ( Figure 4C). Surprisingly, after the treatment, the rate of polymerization was significantly reduced from 0.26+/20.006 mm/sec to 0.23+/20.004 mm/sec (p,0.001) after 1 h and to 0.17+/20.003 mm/sec (p,0.001) after 2 h. However, in CDT-treated cells, microtubules and EB3 were associated for a longer period of time than in controls, indicating that capture, pause and/or catastrophe of microtubules were decreased ( Figure 4D).

Binary toxins influence distribution of CLASP2 and ACF7
We addressed the question whether proteins, which are suggested to be involved in capture of microtubules, were affected by actin-depolymerizing toxins. Besides EB1, CLASPs (CLIPassociated proteins) appear to participate in capture processes [33,34]. As shown in Figure 5A, in controls CLASP2 is located at the cell cortex and binds to the tip of microtubules. After the treatment with actin-ADP-ribosylating toxin, CLASP2 translocated from the cell cortex and associated with filament structures, which were at least partially stained with anti-tubulin antibody ( Figure 5A). EB1 interacts with the large spectraplakin ACF7 (actin crosslinking family 7; also termed MACF1, microtubule and actin crosslinking factor 1) that functionally links microtubules to actin microfilaments [35]. After the treatment with CDT (or with C2 toxin) the distribution pattern of ACF7 changed. ACF7 was released from the cell cortex and associated with the microtubule lattice ( Figure 5B). Thus, our data indicate that the actin-ADPribosylating toxin interferes with the capture of microtubules at the cell cortex by causing redistribution of proteins involved in capture processes.
Because interaction of ACF7 with F-actin and microtubules appears to be controlled by the small GTPase RhoA and its effector mDia [35], we studied the effects of the Rho-inactivating C. difficile toxin B on the formation of CDT-induced microtubule protrusions ( Figure 6A). Treatment of Caco-2 cells with toxin B (300 ng/ml) for 2-4 h induced major morphological changes such as shrinking of the cell body. However, addition of CDT caused the same formation of protrusions as observed in the absence of toxin B. Also the pretreatment of cells with the Rho-inactivating toxin C3 ( Figure 6B), which was applied as a C2IN-C3 fusion protein together with the binding component C2II, did not affect CDT-induced outgrowth of protrusions. These findings suggest that Rho GTPases do not play a major role in the effects caused by CDT.

Formation of protrusions increases adherence of bacteria
To address the question whether the formation of the microtubule meshwork may affect adherence and/or colonization of bacteria, we incubated Caco-2 cells in the presence of CDT with C. difficile strain VPI 10463, which produces toxins A and B but not the actin-ADP-ribosylating toxin CDT. Cells were treated for 1 h with CDT, then C. difficile bacteria were added and the incubation was continued for 90 min. Thereafter, cell monolayers were washed and bacteria, which remained adherent to the cell  Figure 3B. After CDT treatment, both +TIPs have an increased microtubule tip association. +TIPs were measured by Metamorph software. Data are given +/2SEM (*** = p,0.001, n = 100). CDT concentrations were as above. doi:10.1371/journal.ppat.1000626.g003 surface, were determined by an antibody directed towards C. difficile surface proteins. As shown in Figure 7A, treatment with CDT largely increased the number of bacteria adherent to cells. About 6-fold higher numbers of bacteria were counted in the presence of CDT as compared to controls. Because cell adherence studies with C. difficile were not performed under anaerobic condition, which is important for development of the bacteria, we repeated the experiments under anaerobic conditions. For this purpose polarized Caco-2 cell mono-layers infected with bacteria were placed in an anaerobic incubation chamber for 4 h. This incubation under anaerobic conditions did not cause visible defects of the cell monolayer. Thereafter, cell monolayers were washed and removed from the culture dishes. The cell suspensions, containing bacteria, were replated on blood agar plates and incubated under anaerobic conditions. Colony forming units were determined after 2 days, resulting in an ,4.5 fold increase with CDT as compared to controls ( Figure 7B). Electron microscopic studies revealed that bacteria were surrounded and wrapped by protrusions formed in the presence of CDT ( Figure 7C).
In the above mentioned experiments, purified CDT was added to cell culture and then adherence of Clostridia was studied. Therefore, we wanted to know whether production of CDT by Clostridia immediately affects adherence and colonization of the pathogen. To this end, Caco-2 cells were co-cultured with CDTproducing C. difficile strain Nap1/027 under anaerobic conditions for 5 h followed by aerobic incubation for further 3 h. Time-lapse microscopy revealed the formation of microtubule-based protrusions after 5 h, which largely increased within the next 3 h ( Figure 7D). The production of CDT under these conditions was verified by testing the ADP-ribosylating activity in the culture supernatant (data not shown).
By contrast, a C. difficile strain (VPI 10463) that does not produce CDT was not able to induce protrusions. Moreover, addition of anti-iota B antibody, which is known to interact with the binding component of CDT [36], prevented the formation of protrusions caused by C. difficile Nap1/027, indicating that the formation of protrusions depended on the actin ADP-ribosylating toxin CDT ( Figure 7D). To verify the importance of CDT in the establishment of disease, we employed an in vivo model of antibiotic-associated diarrhea.
In general, mice are resistant against C. difficile infection and do not develop any signs of disease (i.e. gut inflammation, colitis). For a long time, attempts to establish a mouse model for C. difficilemediated antibiotic-associated diarrhea have yielded ill-reproducible results. Most likely, mouse-colony specific differences in gut flora composition have been responsible for the poor reproducibility. Only very recently, antibiotic treatment regimes have been developed which allow re-capitulating the hallmarks of antibioticfacilitated C. difficile colitis in conventional mice [37,38]. To minimize interference by components of the normal gut flora, we used gnotobiotic mice colonized with the Altered Schaedler Flora (ASF) [39] for C. difficile infection. Mice were treated with clindamycin (0.2 mg) 24 h previous to infection with the human pathogenic C. difficile strain Nap1/027 (10 7 CFU). To investigate the role of CDT in enteric infection, we neutralized CDT in Nap1/027-infected mice by oral treatment with CDT-neutralizing rabbit anti-iota antiserum. The other group of mice was treated with control serum. Notably, in contrast to mice treated with CDT-neutralizing anti-iota toxin antibody, mice treated with the control serum developed signs of diarrhea (liquid stool) at 24 h after infection. Mice were sacrificed 30 h after infection. Upon macroscopic inspection, a great part of the control mice exhibited signs of colitis and approximately half of the mice showed overt inflammation of the cecum (data not shown). As the key experimental readout, we analyzed intestinal colonization by Nap1/027. C. difficile colonized anti-iota toxin treated mice at significantly lower levels as compared to control mice ( Figure S7).
Next, we analyzed in vivo epithelial adherence of C. difficile by immunofluorescent staining with a specific C. difficile antiserum. In CDT-neutralizing anti-iota toxin serum treated mice immunohistochemical stainings of sections of colon and cecum, revealed a lower number of bacteria directly adjacent to the mucosal surface ( Figure S8A-C).
Altogether, our data suggest that CDT increases epithelial adherence in vitro and in vivo, thereby supporting bacterial colonization.

Discussion
Here we report on a novel morphological feature induced by binary actin-ADP-ribosylating toxins, which modify actin at Arg-177, and thereby inhibit F-actin formation and cause microfilament depolymerization. We show that C. difficile CDT, C. botulinum C2 toxin and C. perfringens iota toxin induce cellular protrusions in epithelial cells, which are thin, highly dynamic structures with   lengths from 5 to .150 mm. Immunofluorescence microscopy revealed that the protrusions harbored dynamic microtubules. Formation of protrusions occurred concomitantly with ADPribosylation of actin and depolymerization of microfilaments. However, complete depolymerization of the actin cytoskeleton was not necessary for the formation of protrusions. The microtubule structures were formed when only a small part of actin was ADPribosylated and cell morphology was not grossly altered. Moreover, major formation of microtubule protrusions was observed at medium concentration of the toxins. It appeared that at rather high concentrations of CDT, formation of microtubulebased protrusions occurred rapidly but their elongation ceased soon.
Treatment of cells with the actin-depolymerizing toxins not only induced formation of microtubule protrusions but also changed cellular microtubule structures. Increased bundling of microtubules was observed. Previously it was reported that the actin-ADPribosylating C2 toxin affects microtubule patterns in human leukemia cells by causing formation of microtubule bundles [40]. In addition actin filament-depolymerizing agents like cytochalasin and latrunculin, which directly interact with actin to interfere with actin polymerization, can induce microtubule-based cell surface extensions [41,42]. However, these compounds were much less effective than the binary actin-ADP-ribosylating toxins. On the other hand, we observed that the F-actin-stabilizing compound jasplakinolide caused delay in the action of the binary toxins. Filamentous actin is a poor substrate for ADP-ribosylation by binary toxins, which may explain the delay of the action of CDT in the presence of jasplakinolide. Moreover, treatment with jasplakinolide may delay the release of microtubule-capturing proteins from the cell cortex (see below).
Microtubule dynamics are regulated by a variety of microtubule associated proteins. A major player in the regulation of microtubule dynamics is EB1, which promotes microtubule polymerization by increasing the number of rescue events and decreasing the rate of depolymerization or the time in pause [29,43,44]. Since almost all other +TIPs additionally interact with EBs [45], we decided to study the influence of actin-ADPribosylating toxins on the distribution of EB1. We showed that toxin treatment increased the length of EB1 comets at the growing tips of microtubules. Additionally, by the use of live imaging we showed that EB3-comets were present for longer time periods after toxin treatment, indicative of longer growth phases of microtubules. Surprisingly, these EB1-comets revealed a decrease in microtubule polymerization speed.
Capture of growing microtubules at the cell cortex is an important process in the control of microtubule dynamics. Several proteins have been suggested to be involved in microtubule capture besides EB1 at least CLASP1/2, and ACF7 appear to play a role [33][34][35]46]. We observed that CDT treatment of Caco-2 cells caused redistribution of CLASP2 and ACF7 from the cell cortex and microtubule ends to the cell interior. These findings are in line with the notion that microfilaments are essential for the localization of these proteins at the cell cortex and depolymerization of actin causes redistribution of these proteins. The functions of ACF7 have been studied in ACF72/2 mouse embryonic fibroblasts, showing that deletion of ACF7 causes defects in microtubule capture with bending of microtubules at the cell cortex. We also observed bending of microtubules after CDT treatment; however, the main effect of the toxin was the formation of protrusions. This was not described by deletion of ACF7 [35]. Importantly, the deletion studies were performed in the presence of functional microfilaments, whereas in our experiments the polymerization of actin is inhibited by the toxins.
So far the functional link between microfilaments and the microtubule system is still enigmatic. Master regulators of the actin cytoskeleton belong to the family of Rho GTPases, including Rho, Rac and Cdc42 proteins [47][48][49]. These switch proteins control actin polymerization by regulating the activity of the formin proteins mDia [50], or Wasp and Scar/Wave proteins, respectively, which control actin polymerization by Arp2/3 complexes [51]. Recently many studies showed that Rho GTPases play also a major role in controlling microtubules [52]. For example, RhoA regulates microtubules via mDia [53,54]. This effect appears to be independent of the action of mDia on actin [55]. Rac reportedly controls microtubules by stathmin phosphorylation via DOCK7 [56] and also Cdc42 is implicated in regulation of the microtubule system [57]. Another example is IQGAP, which interacts with Rac and Cdc42, and was shown to associate with microtubules via CLIP-170 [58]. However, our studies with C. difficile toxin B exclude the involvement of Rho, Rac or Cdc42 in toxin-induced formation of microtubule protrusions. Treatment of cells with toxin B, which inhibits the GTPases [3], did not prevent subsequent induction of formation of protrusions by CDT. Notably in these studies, rather high concentrations of toxin B were used and the incubation was continued until major cellular shrinking was observed, indicating full action of toxin B. Recently, it has been reported that inhibition of myosin II, which was shown to be involved in epithelial cyst formation, causes depolymerization of F-actin bundles and formation of cellular processes, which resemble in size and microtubule content our toxin-induced protrusions [59] but do contain a large part of F-actin, indicating that these protrusions are different. Ivanov and coworkers reported that phospholipase C might be involved in the process formation [59]. Until now we could not prove an involvement of phospholipase C in the formation of the reported microtubulebased protrusions.
CDT-induced formation of protrusions increased the adherence of C. difficile at the cell surface of epithelial cells. Electron microscopical studies revealed that the protrusions formed a dense meshwork in which the Clostridia were caught, suggesting that CDT has major effects on the colonization of Clostridia. In fact it has been suggested that the toxin favors colonization of C. difficile bacteria [60]. Besides toxin A and B, which both glucosylate Rho GTPases, CDT is the third toxin produced by C. difficile. It is generally accepted that toxins A and B are the cause of pseudomembranous colitis induced by the pathogen [2][3][4]61]. However, recently identified hypervirulent strains of C. difficile (e.g., ribotype 027/Nap1) also produce CDT and the number of isolates, producing CDT, is increasing [7][8][9][10][11]. The role of the additional toxin CDT as a virulence factor was not clear so far. It has been reported that CDT-producing strains, which do not produce the glucosylating toxins A and B (A 2 B 2 CDT + -strains), did not cause disease but colonize in hamster after challenge with clindamycin. On the other hand, A + B + CDT 2 -strains can colonize and cause disease under the same conditions [60].
Our findings show that the toxin has unexpected effects on the morphology of intestinal cells, which directly affect the environment of the Clostridia. We show that CDT induces a ,5 fold increase in adherence of C. difficile bacteria under anaerobic conditions. Moreover, in vivo studies in mice were in line with these results. When mice were infected with a CDT-producing strain, we observed a 4 fold increase in epithelial adherence of bacteria in the large intestine as compared to mice treated with CDTneutralizing antiserum.
These data were corroborated by the finding that C. difficile colonization of the cecal content was significantly decreased when CDT was functionally neutralized in the gut. Considering the massive effects of CDT on the microtubule system and the formation of a meshwork of microtubule-based protrusions tightly covering bacteria at the surface of intestinal epithelial cells, we propose that at least one major effect of CDT is increased epithelial adherence and thereby optimization of colonization of the pathogen. It was shown that C. difficile binds to extracellular matrix proteins like fibronectin or collagen [62]. Binding to extracellular matrix proteins appears to occur at least in part via surface layer proteins (SLPs) of C. difficile [63]. Also flagellar proteins are implicated in adherence of C. difficile [64]. It remains to be studied, which bacterial surface structures are involved in interaction with these cellular extensions. It has been suggested that interaction of Clostridia with extracellular matrix proteins is not possible before major alteration of epithelial cells are induced by cytotoxins. Thus, in the early phase of infection CDT-induced formation of protrusions might be of special importance for adherence of Clostridia.
Taken together, our findings suggest an important role of microtubule-based protrusions in enhancing cell surface adherence and colonization by C. difficile. Hence our findings of the toxin effects, which are shared by all actin ADP-ribosylating toxins studied, have a major impact in understanding the role of these toxins produced by several Clostridia. Different pathogenic Clostridia might exploit this mechanism by the production of ADPribosylating toxins. C. botulinum type C and D produces in addition to the highly potent neurotoxins, the actin modifying C2 toxin. For this reason, C2 might be of relevance for botulism in waterfowl [65]. Therefore, our studies not only demonstrate an important functional connection between actin microfilaments and regulation of microtubules but also offer a new perspective in understanding the role of the binary actin-modifying toxins in infectious biology and host-pathogen interactions.

Expression and purification of protein toxins
The components of C2 toxin (C2I and C2II) [66] and of iota toxin (Ia and Ib) [67] were purified as described.
The components of CDT (CDTa and CDTb) were purified as recombinant glutathione S-transferase proteins. GST-Proteins were expressed in E. coli TG1 pACYC IRL10 cells (CDTa) and in E. coli BL21 DE3 Rosetta cells (CDTb). The proteins were purified by affinity chromatography with glutathione-Sepharose 4B, according to the manufacturer's instructions. Glutathione Stransferase was cleaved off by thrombin (3.25 NIH units/mL of bead suspension). The inactivation of thrombin was performed by the addition of benzamidine beads. CDTb was activated by 0.2 mg of trypsin/mg of protein for 30 min at 37uC.
All binding components used (CDTb, C2II and Ib) mentioned in the text were used as protease-activated proteins according to [66,67].
Purification of the Rho-inactivating C3 toxin, which was used as a cell permeable fusion toxin, has been described [68].

Construction of CDTb expression plasmid
pGEX-CDTa construction was described previously [20]. For construction of pGEX2T-CDTb, cdtb (2631 bp) was amplified by PCR, using genomic DNA from C. difficile strain 196 (gift from M. Popoff, Institute Pasteur, Paris) as a template. Used primers: 59CDTb-BglII GGG GGG AGA TCT ACC ATG AAA ATA CAA ATG AGG AAT AAA AAG G and 39CDTb-EcoRI GGG GGG GAA TTC CTA ATC AAC ACT AAG AAC TAA TAA CTC. The PCR-product was cloned into the pGEX-2T-vector.

Antibodies and fluorescent dyes
Mouse anti-EB1 and anti-EB3 antibodies were from BD Biosciences (Franklin Lakes, NY), mouse monoclonal anti-atubulin, anti-acetylated tubulin antibody were from Sigma (St. Louis, MO), mouse anti-detyrosinated tubulin antibody was from Millipore (Billerica, MA). Anti-tyrosinated tubulin antibody YL1/ 2 was used as described before [70]. Mouse anti-MACF1/ACF7 antibody from Abnova (Heidelberg, Germany). Rabbit anti-MACF1/ACF7 antibody and anti-occludin antibody was from Santa Cruz (Santa Cruz, CA). Anti-ZO-2 antibody was from Cell Singnalling Technology (Beverly, MA). Anti-CLIP-170 antibody [31], anti-CLIP-115 antibody [32] and anti-CLASP2 antibody [34] were a gift from Dr. Niels Galjart. As secondary antibodies Alexa568-and Alexa488-conjugated anti-rabbit, anti-mouse and anti-rat antibodies (Invitrogen, Karlsruhe, Germany) were used. Nuclei were stained by DAPI-Prolong Gold and filamentous actin was stained using phalloidin-TRITC (Invitrogen, Karlsruhe, Germany). Antisera against C. difficile 630 were from Dr. Neil Fairweather (Imperial College, London). Bacterial culture C. difficile strain VPI 10463 was cultured in Brain Heart Infusion medium (BD, Franklin Lakes, NY) at 37uC under anaerobic conditions (Anaerocult A, Merck, Darmstadt, Germany). For quantification of bacterial adherence, 100 ml of a C. difficile over night culture (OD 1.2; 2.5610 8 bacteria/ml) were added to a 3 cm dish of polarized Caco-2 monolayer, containing 2 ml cell culture medium with 15 mM HEPES-Buffer (Biochrom, Berlin, Germany). The cells were incubated under anaerobic conditions for 4 h. After incubation the cells were washed, removed and plated on blood agar plates. After 2 days colony forming units (CFUs) were counted. Adherence of Clostridia after CDT treatment was quantified as percent of adherence on control cell monolayer.

ADP-ribosylation assay
ADP-ribosylation was performed at different time points after toxin treatment. 30 mg of whole-cell lysate were used. The subsequent ADP-ribosylation reaction was performed for 30 min at 30uC with 300 ng of C2I in a buffer, containing 20 mM Tris-HCl (pH 7.5), 1 mM EDTA, 1 mM dithiothreitol, 5 mM MgCl 2 , and Complete protease inhibitor (Roche, Basel, Switzerland)(according to the manufacturer's manual). The reaction was stopped by boiling the sample with SDS-sample buffer. The samples were run on SDS-PAGE, and [ 32 P]ADP-ribosylated proteins were detected by autoradiography with a PhosphorImager (Amersham Biosciences, Freiburg, Germany).

Immunostaining, measurement of +TIP length
Cells were washed with warmed PBS, fixed in ice-cold methanol, containing 1 mM EGTA (left out for phalloidin-TRITC staining), postfixed with 4% formaldehyde, and permeabilized with 0.15% Triton X-100 in PBS. Cells were blocked by 1% BSA and 0.05% Tween 20 in PBS. Incubation with first antibody was performed over night at 4uC in 1:200, 1:300 or 1:1000 dilutions in blocking buffer dependent on the antibodies. Then, samples were washed with 0.05% Tween 20 in PBS and incubated with 1:200 dilutions of the secondary antibodies for 1.5 h at room temperature. Thereafter, cells were washed again and embedded with Prolong Gold. Fixed samples were analyzed by fluorescence microscopy, using an upright Zeiss Observer microscope system (Carl Zeiss GmbH, Jena, Germany) equipped with an apotom. For the measurement of +TIP length, 100 +TIPs in each group were analyzed with Metamorph software (Universal Imaging, Downingtown, PA).

Live cell imaging and quantification of formation of protrusions and +TIP dynamics
For live cell imaging cells were observed in a chamber that provided a humidified atmosphere (6.5% CO2 and 9% O2) at 37uC on a Zeiss Axiovert 200 M inverted microscope (Carl Zeiss GmbH, Jena, Germany). Specific fluorescence illumination was generated by a monochromator (Cairn Optoscan; Faversham Kent, UK). Images of 16-bit depth were collected with a digital camera (Coolsnap HQ; Roper Scientific, Tucson, AZ) driven by Metamorph imaging software (Universal Imaging, Downingtown, PA). For the observation of formation of microtubule protrusions time-lapse series, lasting 2-8 h, were acquired with 90 sec intervals. For quantification of formation of microtubule protrusions, the lengths of all processes were summated every 15 min and normalized by the respective section of cell perimeter (an example of quantification of protrusion formation is given in Figure S2).
For the observation of +TIP dynamics cells were transiently transfected with an EB3-GFP expression vector [71] (kind gift of Dr. N. Galjart) and imaged 24-48 h after transfection. Only cells with low expression levels were employed. Time-lapse series with 2-4 min duration were acquired with a 2 sec interval. Control cells were imaged and after treatment for 1 and 2 h, the same cells were imaged again. All quantifications were performed by the Metamorph software.
For the quantification of microtubule polymerization speed, 100 steadily moving +TIPs for each group were tracked for 10 sec. The rate of polymerization was calculated by the covered distance during this time. For the quantification of the time in growth, 30 microtubules with EB3-GFP association at tips of microtubules were recorded from their appearance until fading.
In live cell imaging experiments photo-bleaching and photostress were reduced by minimal aperture of the monochromator exit slit and the microscope was equipped with high speed shutters for fluorescence and transmitted light illumination.

Preparation of primary colon epithelial cells
The colon of a Wistar rat was excised and transferred to icechilled PBS, containing 50 mg/ml penicillin/streptomycin (Biochrom, Berlin, Germany). The colon was washed three times. Ions were chelated by 1 mM EDTA and 1 mM EGTA for 1 h at 21uC. To release the epithelial cells from the lamina propria, the tube was vigorously shaken. The piece of tissue was removed and the liberated epithelial cells were pelleted at 50 g for 3 min. The cells were washed in growth DMEM medium (supplemented with 15% FCS, 1% Na-pyruvate, 1% non essential amino acids and 50 mg/ ml penicillin/streptomycin). Subsequently, cells were plated on collagen-coated dishes or coverslips. Experiments were performed after 1-2 days in vitro.

Scanning electron microscopy
For scanning electron microscopy of cells the cultivation medium was replaced by ice-cold cacodylate buffer (0.1 M cacodylate, 0.01 M CaCl 2 , 0.01 M MgCl 2 , 0.09 M saccharose, pH 6.9) and prefixed by adding paraformaldehyde to a final concentration of 5%, after 10 min glutaraldehyde was added to a final concentration of 2%. Cells were dehydrated by incubation in a graded series of acetone in 10% steps to 100%, critical-point dried with liquid CO 2 (CPD030, Bal-Tec, Liechtenstein) and covered with a thin gold film for conductivity by sputter coating (SCD040, Balzers, Liechtenstein or Cressington 108 auto, Watford, UK). Samples were examined in a field emission scanning electron microscope DSM982 Gemini at an acceleration voltage of 5 kV or in a LEO 435VP (Carl Zeiss SMT, Oberkochen, Germany) at an acceleration voltage of 10 kV.

Animals
Gnotobiotic mice were created by colonizing germfree Swiss Webster mice with a donor mouse, associated with the Altered Schaedler Flora (ASF), according to the protocol published on the Taconic webpage (Taconic, Hudson, NY, www.taconic.com/ library). Gnotobiotic mice were maintained under barrier conditions in individually ventilated cages with autoclaved chow and autoclaved, acidified water. The experiments were performed on 2 independent occasions with a total of 10 Nap1/027 infected mice treated with anti-iota CDT-neutralizing serum, 11 Nap1/027 infected mice treated with control serum and 4 control mice. 2 of the control mice were also treated with control serum to exclude a serum induced inflammation.

Infection of mice
For histology, intestinal tissue was cryo-embedded in Tissue Tek OCT (Sakura) and flash frozen. For immunofluorescent staining, the tissue was fixed in 4% PFA over night followed by over night equilibration in 20% sucrose and cryo-embedding. Cryosections (7 mm) of PFA-fixed and cryo-embedded tissue from cecum and colon were mounted on glass slides and air dried for 2 h at room temperature prior to immunostaining. Sections were fixed again in 4% paraformaldehyde for 5 min, washed and blocked with 1% BSA in PBS. The primary antibody was detected with an Alexa 488 labeled anti-rabbit antibody. The sections were incubated over night with a rabbit anti-C. difficile serum. Actin was stained by TRITC-phalloidin and the nuclei by DAPI. For bacterial quantification, bacteria adjacent to the epithelium, stained with anti-C.difficile serum, within 3 mm of the epithelium were counted.
If bacterial aggregations were attached to the epithelium all bacterial cells in the aggregation were counted. This was done for $15 optical fields per group. The number of C. difficile per mm mucosa surface was determined.
For bacteriology, cecal content was removed aseptically and homogenized in 4uC PBS. The bacterial loads in cecal content were determined by plating on C. difficile selective agar plates (Oxoid, Basingstoke, UK) at different dilutions. The minimal detectable value was 10 CFU/g.

Statistics
Student's t test was used when two groups of parametric data with normal distribution had to be compared. The Mann-Whitney U test was applied for processed data without normal distribution. Statistic evaluation was performed with the Sigma Stat software (Jandel Scientific, Corte Madera, CA) or with Prism 4 (GraphPad, La Jolla, CA). P values ,0.05 were considered statistically significant. (B) Series of time-lapse images of cells, which were treated with 500 nM jasplakinolide (jas) for 30 min. Then 20 ng/ml CDTa and 40 ng/ml CDTb were added and the formation of protrusions was monitored. Scale bar represents 20 mm. Actin stabilization can delay and decrease the formation of protrusions. (C) DIC timelapse microscopy of Caco-2 cells. Subconfluent Caco-2 cells were treated with 300 ng/ml toxin B (Tox B) or 300 ng/ml C3 fusion toxin (C3) and 500 ng/ml C2II (to deliver the C3-fusion toxin into the cells) for the indicated times (h). The toxins were added to the cell culture medium. Incubation with toxins was continued until major morphological changes occurred. No formation of protrusions was observed under these conditions. Scale bar represents 20 mm. Found at: doi:10.1371/journal.ppat.1000626.s005 (0.87 MB PDF) Figure S6 Influence of CDT on EB3 and CLIP-115 proteins. Subconfluent Caco-2 cells were treated with 20 ng/ml CDTa and 40 ng/ml CDTb and fixed after 1 h. Indirect immunofluorescence pictures of EB3 (red) by using anti-EB3 antibody and CLIP-115 (green) by using anti-CLIP-115 antibody in Caco-2 cells are shown. The nucleus was stained by DAPI (blue). Scale bar represents 20 mm. The increase of EB3 and CLIP-115 comet length after toxin treatment is seen. Found at: doi:10.1371/journal.ppat.1000626.s006 (0.84 MB PDF) Figure S7 CDT increases bacterial loads of C. difficile in cecal content in vivo. Mice were infected by gavage with Nap1/027 (10 7 CFU) and subsequently treated with control serum or CDTneutralizing anti-iota toxin serum. Cecal content was aseptically removed and homogenized. Bacterial loads in the cecal content from uninfected untreated mice, Nap1/027 (10 7 CFU) infected CDT-neutralizing anti-iota toxin serum treated mice and Nap1/ 027 infected control antiserum treated mice were determined by plating on C. difficile selective agar plates at different dilutions. The minimal detectable value was 10 CFU/g. The experiments were performed on 2 independent occasions with a total of 10 Nap1/ 027 infected mice treated with anti-iota CDT-neutralizing serum, 11 Nap1/027 infected mice treated with control serum and 4 control mice. Boxes indicate 25th and 75th percentiles, black bars indicate medians, and whiskers indicate data ranges. Y-axis is scaled logarithmically (log 10 ). * indicates p#0.05. Found at: doi:10.1371/journal.ppat.1000626.s007 (0.03 MB PDF) Figure S8 CDT increases adherence of C. difficile in the cecum and colon in vivo. (A) Mice were infected with Nap1/027 (10 7 CFU) and subsequently treated with control serum or CDTneutralizing anti-iota toxin serum. Cryosections (7 mm) of PFAfixed and cryo-embedded tissue from cecum and colon were immunostained for C. difficile. Actin was stained by TRITCphalloidin and the nuclei by DAPI. M marks the mucosa and L marks the lumen of the intestine. White boxes are magnified. White arrows mark bacteria or bacterial aggregations adjacent to the epithelium. Calibration bar represents 20 mm. (B) Quantification of C. difficile directly adjacent to the epithelium in the cecum. For bacterial quantification, bacteria adjacent to the epithelium, stained with anti-C. difficile serum, within 3 mm of the epithelium were counted. If bacterial aggregations were attached to the epithelium all bacterial cells in the aggregation were counted. This was done for $15 optical fields per group. The number of C. difficile per mm mucosal surface was determined. Nap1/027 infected mice treated with control serum were set 100%. (C) Quantification of C. difficile directly adjacent to the epithelium in the colon. The number of C. difficile per mm mucosa was determined as in B. Found at: doi:10.1371/journal.ppat.1000626.s008 (0.85 MB PDF) Video S1 DIC time-lapse microscopy movie of Caco-2 cells. Subconfluent Caco-2 cells were treated with 20 ng/ml CDTa and 40 ng/ml CDTb. The toxin was added to the cell culture medium. Thereafter, a picture was taken every 90 sec. After ,50 min the formation of toxin-induced protrusions started. The maximum of protrusion formation was observed after ,4 h. The incubation time (h:min) is indicated in the movie. Scale bar represents 20 mm.