Identification of a conserved membrane localization domain within numerous large bacterial protein toxins

Abstract

Vibrio cholerae is the causative agent of the diarrheal disease cholera. Many virulence factors contribute to intestinal colonization and disease including the Multifunctional Autoprocessing RTX toxin (MARTX_Vc). The Rho-inactivation domain (RID) of MARTX_Vc is responsible for inactivating the Rho-family of small GTPases, which leads to depolymerization of the actin cytoskeleton. Based on a deletion analysis of RID to determine the minimal functional domain, we have identified a subdomain at the N terminus of RID that is homologous to the membrane targeting C1 domain of Pasteurella multocida toxin. A GFP fusion to this subdomain from RID colocalized with a plasma membrane marker when transiently expressed within HeLa cells and can be found in the membrane fraction following subcellular fractionation. This C1-like subdomain is present in multiple families of bacterial toxins, including all of the clostridial glucosyltransferase toxins and various MARTX toxins. GFP-fusions to these homologous domains are also membrane associated, indicating that this is a conserved membrane localization domain (MLD). We have identified three residues (Y23, S68, R70) as necessary for proper localization of one but not all MLDs. In addition, we found that substitution of the RID MLD with the MLDs from two different effector domains from the Vibrio vulnificus MARTX toxin restored RID activity, indicating that there is functional overlap between these MLDs. This study describes the initial recognition of a family of conserved plasma membrane-targeting domains found in multiple large bacterial toxins.

Bacterial pathogens use a wide array of secreted proteins to intoxicate mammalian cells and mediate disease. Many effectors target a specific organelle, necessitating strategies to traffic to certain subcellular locations after entry into the cytosol. For example, the cytolethal distending toxin active subunit, CdtB, uses a short N-terminal nuclear localization sequence attached to traffic to the nucleus (1). Similarly, many Type III secretion effectors contain distinct targeting signals: BteA uses a 130 aa N-terminal sequence to target lipid rafts (2); ExoS and YopE use internal 21 aa domains to associate with the host membrane (3); and ExoU uses a C-terminal sequence to localize to the membrane (4). For each of these effectors, subcellular localization is determined by a specific aa sequence separate from the catalytic site.

The multifunctional-autoprocessing RTX toxins (MARTX) are large bacterial toxins composed of multiple effector domains released by autoprocessing upon translocation into the host cell cytosol (5). The best characterized member of this family is the MARTX of Vibrio cholerae (MARTX_Vc), which is predicted to be 4545 aa in length (6, 7). MARTX_Vc carries three effector domains delivered by autoprocessing of which two function to disrupt the actin cytoskeleton. G-actin is covalently cross-linked into oligomers by the MARTX_Vc actin cross-linking domain (ACD) (8), whereas the Rho-inactivation domain (RID) reversibly inactivates host cell RhoGTPases (9). Transfection of cells with a plasmid expressing RID as a fusion with GFP induced cell rounding. In addition, when fused to the N-terminal portion of Anthrax toxin lethal factor (LF_N) and delivered to cells by Anthrax toxin protective antigen (PA) (10), this domain was sufficient to induce cell rounding and Rho-inactivation (9).

Proteins with significant structural similarities to MARTX_Vc are present in other pathogenic bacteria such as Vibrio vulnificus, Aeromonas hydrophila, Yersinia pestis, Clostridium difficile, Xenorhabdus nematophila, and Photorhabdus luminescens. However, primary sequence comparisons indicate each toxin possesses a distinct collection of effector domains (reviewed in ref. 5). The V. vulnificus MARTX toxin (MARTX_Vv) contains four putative effector domains of unknown function (DUF) and a putative RID but no ACD (5). Three of the DUFs are homologous to other bacterial toxin domains; DUF3 with the α/β-hydrolase family of enzymes, DUF4 with the P. luminescens makes caterpillars floppy toxins (11), and DUF5 with the C2 domain of Pasteurella multocida toxin (PMT) (5).

In this work, we identify a conserved peptide sequence in 16 different bacterial toxins produced by both Gram-positive and Gram-negative bacteria. Functional analysis of four of these domains reveals that these toxins use this sequence in a shared strategy for plasma membrane targeting, despite the varied mechanisms of host cell intoxication employed by each toxin.

Results

Deletion Analysis of MARTX_Vc RID.

The MARTX_Vc RID (RID_Vc) corresponds to aa 2552–3099 of the MARTX_Vc holotoxin (9). To establish the minimum functional size of RID_Vc, a genetic deletion analysis on the domain was performed. Based on a CLUSTAL W alignment of the putative RID sequences of MARTX toxins from V. cholerae, V. vulnificus, Vibrio anguillarum, X. nematophila, and Xenorhabdus bovienii (Fig. S1), we generated three N-terminal and three C-terminal truncations in the eukaryotic expression plasmid pRID-GFP (Fig. 1A). The truncations were chosen to remove regions of significant (>60%) sequence divergence among the homologs, because these regions are likely not required for RID function. Transfection of pRID_2650–3099-GFP, pRID_2721–3099-GFP, or pRID_2552–3085-GFP into HeLa cells caused cell rounding indistinguishable from pRID_2552–3099-GFP but further deletions from either terminus of RID_Vc abolished cell rounding (Fig. 1A and Fig. S2). Together, these data indicated that residues within aa 2721–3085 are necessary and sufficient for RID_Vc activity when expressed in mammalian cells as a GFP-fusion protein.

Fig. 1.

Deletion analysis of RID_Vc identifies a broadly conserved domain. (A) Portions of the N- and C-termini of RID_Vc were deleted from pRID-GFP or LF_N-RID. Each truncated RID_Vc was tested for its ability to cause cell rounding after transfection (GFP-fusions) or delivery to the cell with PA (LF_N-fusions). (B) Cartoon of the entire PMT toxin (boxed, PDB ID: 2EBF) with the PMT C1 domain (blue) overlaid with a model of MLD_VcRID (red). (C) Alignment of 19 putative MLDs from various bacterial toxins: see Table S1 for accession numbers and protein details. Shading shows aa conservation and residue description: green = hydrophobic; orange = polar; blue = basic; red = acidic. Arrows indicate aa mutated to either Ala or a similar residue on MLD_VcRID or MLD_VvDUF5.

In an attempt to confirm that cell rounding was due to RhoA inactivation by the shorter RID fragment, an LF_N-fusion to residues 2721–3099 of RID_Vc was constructed and the amount of active GTP-bound RhoA within HeLa cells was monitored following PA-mediated delivery of purified LF_N-RID_2721–3099. Unlike the nearly 100% cell rounding caused by PA+LF_N-RID_2552–3099, incubation with PA+LF_N-RID_2721–3099 did not round HeLa cells even when added at a 1200× excess over the amount sufficient for LF_N-RID_2552–3099 toxicity (Fig. 1A and Fig. S3A). These results suggested that either the LF_N fusion inhibits the functionality of RID_2721–3099 or that the N terminus of RID_Vc contains an essential element that is required for functional activity when RID_Vc is delivered to the cell as a protein but is dispensable when the protein is overexpressed within the cell.

Residues 2561–2645 of MARTX_Vc Target the Plasma Membrane.

To further characterize the N-terminal sequence of RID_Vc, the peptide sequence corresponding to aa 2552–2720 was analyzed using PSI-BLAST (12). This analysis showed that MARTX_Vc aa 2561–2645 shares homology with the N terminus of all known RIDs from other MARTX toxins as expected (Fig. S1), and also the N terminus of DUF5 from MARTX_Vv (aa 3591–3669, hereafter referred to as VvDUF5). The DUF5 effector domain is also present within MARTX toxins from Vibrio splendidus, A. hydrophila, P. luminescens, and X. nematophila (5, 13). The N terminus of DUF5 has been previously noted as sharing homology with the C1 domain of PMT (5). In turn, PMT C1 is known to share structural homology with the N terminus of C. difficile Toxin B (TcdB) and other closely related clostridial glucosyltransferase toxins (CGT) (14). A pair-wise alignment of the 19 identified sequences showed a high degree of similarity throughout the region (44–89%), indicating that these may be shared functional domains important for activity (Fig. 1C). All 19 of these domains were found just to the N-terminal side of three different classes of effector domains (RID, PMT, and CGT) within various large toxins.

PMT C1 was previously shown to localize GFP to the plasma membrane of 293T cells when ectopically expressed (14). To test if the C1-related domain from RID_Vc also facilitates membrane localization, aa 2561–2645 were transiently expressed in HeLa cells as a fusion to either the N or C terminus of GFP. A plasmid expressing the myristylation site and the polybasic membrane targeting sequence present within the first 15 aa of c-Src fused to mCherry (S15-mCherry) was used as a marker for the plasma membrane (15–18). This portion of c-Src mediates a direct association with host cell membrane components and has been shown in macrophages to colocalize with a biomarker for phosphatidylserine (PS) (15, 19, 20). When the plasmids expressing the N-terminal GFP-fusion (2561-2645-GFP) and S15-mCherry were cotransfected into HeLa cells, an overlap of both fluorescence signals at the cell periphery was observed, particularly compared to GFP alone where there was very little overlap with mCherry (Fig. 2A). Both GFP and 2561–2645-GFP showed fluorescence within the nucleus due to passive transport across the nuclear membrane; however, GFP alone was rarely observed at the membrane colocalized with S15-mCherry (Fig. 2A and Table 1). Transfection of a plasmid expressing MARTX_Vc aa 2561–2645 attached to the C terminus of GFP (GFP-2561-2645) did not display any distinct localization pattern within HeLa cells or overlap with S15-mCherry and was indistinguishable from GFP alone by microscopy (Table 1).

Fig. 2.

RID_Vc aa2561-2645 is sufficient to drive GFP to the HeLa cell plasma membrane. (A) HeLa cells transfected with a plasma membrane marker (S15-mCherry) and either GFP or MARTX_Vc aa2561-2645-GFP (MLD-GFP) were imaged by deconvolution microscopy. Arrow shows overlap at the plasma membrane of mCherry and GFP signals. (B) Representative anti-GFP immunoblot of three separate experiments following membrane fractionation of HeLa cells transfected with the indicated constructs; S, soluble; M, membrane. (C) The average percentage of signal (+/− the standard deviation) in the membrane fraction following fractionation was determined by densitometry of immunoblots for each MLD-GFP construct (n = 3). Tabulated raw densitometry measurements are found in Table S2. A Student's t test was employed to determine the statistical significance of the difference between the amount of GFP versus each GFP-fusion in the membrane fractions; *, significantly different from GFP alone (P < 0.05).

Table 1.

Localization of each GFP fusion to HeLa cell–cell junctions after ectopic expression

MLD construct	Bright junctions*	Total junctions	% Bright junctions
GFP	3	40	8%
GFP-VcRID	2	31	6%
VcRID-GFP	34	38	89%
VvRID-GFP	7	31	23%
VvDUF5-GFP	43	43	100%
TcdB-GFP	19	19	100%
VcRID-GFP mutants†
Y23F	0	21	0%
Y23A	2	29	7%
S68T	2	21	10%
S68A	1	25	4%
R70K	4	49	8%
R70A	3	40	8%
ΔH1-2	1	15	7%
VvDUF5-GFP mutants†
Y23F	27	27	100%
Y23A	0	34	0%
S62T	11	71	15%
S62A	7	56	13%
R64K	0	36	0%

* HeLa cell–cell junctions were scored for fluorescence intensity across a line bisecting the cell–cell junction. Bright junctions exhibited a single central peak of fluorescence intensity (Fig. 3B and SI Materials and Methods). Data are the aggregate counts of the total junctions scored for at least three independent experiments.

^†Representative images for site-directed mutants can be viewed in Fig. S5.

In support of the microscopy, membrane fractionation showed 33 ± 8% of the 2561–2645-GFP fusion protein (MLD-GFP) was present in the membrane fraction, whereas GFP-2561–2645 (GFP-MLD) was rarely at the membrane and was not significantly different in localization from GFP alone (Fig. 2 B and C and Table S2). Immunoblotting with anti-GFP antibody confirmed stable fusion proteins were produced for all fusions tested and thus failure to localize to the membrane was not due to protein degradation. Altogether, these data support that aa 2561–2645 of MARTX_Vc is a peptide sequence required for plasma membrane targeting and thus represents a bacterial toxin membrane localization domain (MLD).

MLD Is a Functionally Conserved Domain Found in Many Bacterial Toxins.

To test if conserved C1-like MLDs from other toxins likewise function in membrane targeting, HeLa cells were transfected with plasmids expressing the putative MLDs from VvRID (MLD_VvRID), VvDUF5 (MLD_VvDUF5), and TcdB (MLD_TcdB) fused to GFP. Similar to MLD_VcRID-GFP, cells ectopically expressing the MLD_VvDUF5 and MLD_TcdB fusions displayed clear membrane localization of GFP by epifluorescence microscopy (Fig. 3A) and were identified in the membrane by immunoblotting with anti-GFP antibody (Fig. 3 B and C). Fluorescence intensity distribution plotting for MLD_VcRID, MLD_VvDUF5, and MLD_TcdB showed maximum GFP-signal at cell–cell junctions compared to GFP alone, which showed nearly uniform fluorescence throughout the cell with maximum fluorescence occurring randomly (Fig. 3D and Fig. S4). This biased localization suggests that these MLDs may preferentially target cell–cell junctions (Table 1 and Fig. S4). Intriguingly, even with 88% aa identity and 93% similarity with MLD_VcRID, MLD_VvRID showed only weak membrane localization and very little junction targeting (Fig. 3, Fig. S4, and Table 1). By contrast, MLD_VvDUF5-GFP showed an apparent higher affinity for the plasma membrane (Fig. 3 C and D) and was also localized to vesicles throughout the cells (arrows in Fig. 3A). Despite its clear association with the plasma membrane by epifluorescence microscopy (Fig. 3 A and B and Table 1), the percentage of MLD_TcdB found in the membrane upon fractionation was less than MLD_VcRID-GFP or MLD_VvDUF5-GFP but was statistically greater than GFP alone (Fig. 3 C and D). These data indicate that although each MLD has a function in targeting proteins to the plasma membrane, the relative amount of protein localized, the stability of each protein at the membrane upon fractionation, the specificity for cell–cell junctions, and the pattern of recycling varied among the four MLDs tested.

Fig. 3.

MLD_VcRID homologs localize to the HeLa cell plasma membrane and at cell–cell junctions. (A) HeLa cells were transfected with the indicated MLD-GFP plasmid then fixed and imaged by epifluorescence microscopy 24 h later. (B) Representative anti-GFP immunoblot following membrane fractionation of HeLa cells transfected with the indicated constructs; S, soluble; M, membrane. (C) Densitometry was used to determine the average percentage of signal (+/− the standard deviation) (n ≥ 3) in the membrane fraction for the indicated construct as was done in Fig. 2C. Tabulated raw densitometry measurements are found in Table S3. (D) The average fluorescence intensity measurement of cell–cell junctions (white lines in A) was determined for each construct using ImageJ (SI Materials and Methods) (n ≥ 2 junctions from 3 independent transfection experiments).

Structural Requirements for Membrane Localization.

Analysis of the PMT (PDB ID: 2EBF), TcdB (PDB ID: 2BVL), and TcsL (PDB ID: 2VL8) structural models and a predicted model of MLD_VcRID (Fig. 1B) shows the conserved MLD regions consist of four α-helices (designated H1-H4) connected by three short loop regions (designated L1-L3) (Fig. 1 B and C). To further test the structural requirements for proper membrane targeting, the first 30 aa of MLD_VcRID, corresponding to H1 and most of H2, were deleted from MLD_VcRID-GFP and the cellular location was monitored following transfection. This deletion prevented MLD_VcRID-GFP localization to cell–cell junctions (Table 1), suggesting that all four helices are required for proper membrane targeting.

Fig. 1C shows that despite high divergence of sequence among the various MLDs, nine residues are conserved in at least 90% of the homologs with only three residues 100% identical. According to the PMT, TcdB, and TcsL structural models, Leu8, Ala12, Leu30, Leu54, Tyr61, and Ala73 (numbered according to their location on MLD_VcRID) lie buried within the hydrophobic core formed by bundling of the four helices. These residues likely have conserved function to maintain structural integrity. The three 100% identical residues, Tyr23, Ser68, and Arg70, are near or within loops L1 and L3 and are oriented near each other on the same end of the helical bundle. Electrostatic potential predictions of the known and modeled MLD structures show that the interface formed by L1 and L3 appears to form a positively charged surface possibly involved in either protein or membrane interactions (Fig. 4).

Fig. 4.

The L1-L3 interface of each MLD displays a positive charge. (A) Cartoon representation of MLD_PMT for orientation of the diagrams in B. (B) Electrostatic potential predictions for two faces of the crystallized MLDs (PMT, TcdB, and TcsL) and the modeled MLDs (VvDUF5, VcRID, and VvRID). All of the predictions indicate a net positive charge (blue) at the interface of L1 and L3 although the remainder of the helical bundles contain large patches of negative charge (red).

The potential importance of the three 100% identical residues for membrane localization was tested by altering each residue on MLD_VcRID and MLD_VvDUF5 to either an Ala or to a closely conserved residue (Tyr-Phe, Ser-Thr, Arg-Lys). Epifluorescence and confocal microscopy showed that substitution of any of these three residues on MLD_VcRID abrogated GFP localization to cell–cell junctions or to the membrane (Table 1 and Fig. S5 B–G). Similarly, substitution of the homologous residues on MLD_VvDUF5-GFP also abolished membrane localization (Table 1 and Fig. S5 H–M). However, MLD_VvDUF5-GFP localization was decreased by mutation of Tyr23 to Phe (Fig. S5J). These results indicate that although plasma membrane localization is a conserved function of these domains, only the Ser and Arg residues within L3 are absolutely essential, suggesting that other requirements and specificities for localization are determined by nonconserved residues.

Functional Complementation Between the MLDs.

The variance of requirements for conserved residues and the extent of sequence divergence of exposed residues of the various MLDs suggest that each MLD may be uniquely paired with its cognate toxin. To test if one MLD can functionally complement another, chimeric toxins were generated in which MLD_VcRID on LF_N-RID_Vc was replaced with either MLD_VvRID or MLD_VvDUF5, resulting in LF_N-MLD_VvRID-RID_Vc or LF_N-MLD_VvDUF5-RID_Vc, respectively. When applied along with PA, either chimera rounded HeLa cells indistinguishably from LF_N-RID_Vc after incubation for 3 h. (Fig. S3A).

A time course of cell rounding was then performed, comparing the percentage of HeLa cells that were round following incubation with varying concentrations of LF_N-MLD_VcRID-RID_Vc, LF_N-MLD_VvRID-RID_Vc, LF_N-MLD_VvDUF5-RID_Vc, or LF_N-ΔMLD-RID_Vc. Both LF_N-MLD_VvRID-RID_Vc and LF_N-MLD_VvDUF5-RID_Vc were able to fully intoxicate cells similar to LF_N-RID_Vc at concentrations above 0.1 nM and the time required for cell rounding was not altered (Fig. S3B). Therefore, despite the differences in membrane localization of the individual MLD constructs when fused to GFP, replacing MLD_VcRID with another MLD had a less than 10-fold effect on the dynamics of LF_N-RID_Vc-mediated cell rounding when delivered to HeLa cells using PA.

Discussion

The ability of a bacterial toxin to identify the location of its cellular target is essential for toxic activity. In this work, we describe the presence of a conserved peptide sequence within 16 large bacterial protein toxins capable of targeting known and putative effector domains within these toxins to the plasma membrane. Interestingly, the MLD is found associated with three distinct effectors: RID, typified by the RID of MARTX_Vc and also found in six other MARTX toxins; PMT C2, typified by PMT and present as DUF5 in five MARTX toxins; and CGT, typified by C. difficile TcdB and found in four other related toxins. Each of these effectors is known to specifically target proteins present at the membrane. The PMT C-terminal complex deamidates heterotrimeric G proteins via the C3 domain attached to C1/C2 (21), CGTs glucosylate Rho family GTPases (22), and RID inactivates Rho family GTPases although potentially by an indirect mechanism (9). This domain appears to be unique to these toxins and is different from the known membrane localization sequences of T3SS effectors (BteA, ExoS, YopE, ExoU) (2–4).

Functional analysis of representatives of each class of MLD identified in this study suggested that, although they show similar ability to localize GFP to the plasma membrane and MLDs can substitute for another in a chimeric toxin, the question that arises is whether an MLD is functional because it has specificity for lipids in membranes or because it targets host proteins that are present at the membrane. Previous studies support a mechanism of direct lipid association. Clostridium sordellii TcsL has been shown to bind liposomes enriched for PS (23), an anionic lipid preferentially associated with the cytosolic leaflet of the plasma membrane (19). Both binding to liposomes and Rac-glucosylation activity of TcsL were stimulated by PS-enriched liposomes, dependent upon the first 18 aa of the toxin, corresponding to H1 and L1 of the MLD. However, TcdB showed less selectivity for PS and was active also in liposomes enriched for phosphatidylglycerol. In our study, TcdB showed a weaker, but significant association with membranes such that localization observed in vivo following ectopic expression may have been dissociated from the membrane during sample preparation. These data indicate that although the MLDs may directly target lipids, they do so with varying affinity.

Further evidence supporting direct lipid association comes from analysis of the crystal structures of PMT, TcdB, and TcsL and structural modeling of MLD_VcRID and MLD_VvDUF5. The three 100% identical residues identified as essential for proper localization of MLD_VcRID and MLD_VvDUF5 are all found near each other, with their side chains directed inward rather than exposed to solvent. Our data suggest that the Tyr, Ser, and Arg residues may be directly involved in L1-L3 interactions required to maintain the overall structure of the four-helical bundle. The folding of MLD_VvDUF5 may be less dependent on the Tyr residue itself but rather the size of the residue in its position, thereby allowing it to be substituted for Phe with little effect on membrane association. The defect in localization of the Ser68 and Arg70 mutants in both MLD_VcRID and MLD_VvDUF5, even when changed to Thr or Lys, further emphasizes that the interactions between L1 and L3 may be essential to proper localization. Assessment of the electrostatic potential of this region indicates that all of the MLDs with a known structure are positively charged at this end. Three-dimensional models constructed for MLD_VcRID, MLD_VvRID, and MLD_VvDUF5 also suggest these MLDs are positively charged at the L1/L3 face (Fig. 4B). Thus, similar to the polybasic N terminus of Src, the positive surface of these MLDs may directly interact with the negatively charged acid lipids such as PS, at the plasma membrane (16, 19). As a further example, annexin, which is also known to specifically localize to PS lipids, is enriched in tight junctions suggesting there is an enrichment of these lipids at this site (24), providing justification for the observation that MLD localization is enhanced at cell–cell junctions.

An alternative model to explain the membrane localization is that the MLDs bind host membrane proteins. Because the different classes of MLDs show apparent distinct affinity for membrane, it is possible they target different proteins. Because the genetic requirement for localization may be limited to Ser68 and Arg70, it is possible that binding involves diverse residues at the L1/L3 loops or structural elements elsewhere on the helical bundle. The interface between the four helices includes a large hydrophobic patch. Similar patches in helical bundles have been shown to participate in protein–protein interactions, for example the ability of focal adhesion kinase to bind paxillin (25). Alternatively, differential affinity or binding to different proteins may depend upon the divergent L2-H3 region of each MLD. The amino acid composition of this region is highly variable among the homologs, with many of the MLDs lacking up to six residues from this region alone (Fig. 1C). The composition of the MLD in the L2-H3 region is more conserved within the subgroups of MLDs, an analysis that would suggest that interactions are effector related. However, this concept is contradicted by the finding that MLD_VvDUF5 and MLD_VvRID were able to complement the activity of MLD_VcRID, suggesting that the MLDs are not matched to their cognate toxin. Thus, although it is possible that protein–protein interactions are important for function at the membrane, the proteins targeted are either similarly localized themselves or the interaction is with different faces of the same target protein.

Interactions between the toxin activity domains and the target membrane protein certainly also contribute to binding specificity. Our data shows that RID_2721–3085 (without an MLD) was functional and able to round cells, but only when produced at high levels within the cell following transfection. Similarly, purified TcsLΔ18, although not toxic to cells after microinjection, was able to glucosylate soluble Rac in vitro (23). The toxic activity of PMT was also significantly decreased, but not completely abolished, when the C1 domain was deleted (14, 26). We believe that the high amount of RID_2721–3099-GFP produced in the cell after transfection was sufficient to locate its cellular binding partner, despite not being directly targeted to the plasma membrane. However, the necessity of the individual MLDs for functionality of each full-length toxin remains to be determined.

The ability of the chimeric LF_N toxins to fully mimic LF_N-RID_Vc activity suggests that the enzymatic activity of RID_Vc could be the limiting factor in the dynamics of cell rounding. Accordingly, despite a possibility of MLD_VvDUF5 to enhance the membrane association of RID, cell rounding did not occur more rapidly. Furthermore, although MLD_VvRID did not significantly increase GFP membrane targeting, a sufficient amount of LF_N-MLD_VvRID-RID_Vc required for RID-mediated cell rounding may have been properly localized to the membrane, thereby allowing intoxication to proceed indifferently from LF_N-RID_Vc.

Discovery of this conserved MLD advances our understanding of the biology of the MARTX toxins as a whole as well as the strategies many pathogenic Vibrios and emerging pathogens use to intoxicate host cells. Because this MLD is present in multiple families of bacterial toxins suggests a conserved mechanism for targeting protein toxins to the mammalian cell plasma membrane.

Materials and Methods

Construction of the Rho-Inactivation Domain (RID) Deletion Plasmids.

Fragments of V. cholerae rtxA encoding for the aa segments of RID indicated in Fig. 1A were amplified by PCR and inserted into pEGFP-N3 (Clontech) to yield the RID-GFP plasmids. Alignments were performed using either CLUSTAL W (27) and highlighted using TEXTSHADE through the Biology Workbench 3.2 server (http://workbench.sdsc.edu) or KAlign (http://www.ebi.ac.uk/Tools/kalign/) (28) and manually fitted to match secondary structure predictions.

Construction of Membrane Localization Domain GFP-Fusions.

The DNA corresponding to aa 2561–2645 of V. cholerae rtxA [according to the Lin sequence (7)], 2377–2461 and 3591–3669 of V. vulnificus rtxA, and 1–83 of C. difficile tcdB, were amplified by PCR and inserted into either pEGFP-N3 or pEGFP-C3 (Clontech). Point mutations on pMLD_VcRID-GFP and pMLD_VvDUF5-GFP were generated using the Quikchange XL II site-directed mutagenesis kit (Stratagene).

Purification and Application of LF_N-Fusion Proteins.

PA, LF_N-RID_Vc, LF_N-ΔMLD-RID_Vc, LF_N-MLD_VvRID-RID_Vc, and LF_N-MLD_VvDUF5-RID_Vc were all purified as in Sheahan and Satchell (9) using an ÄKTA Purifier. Purity of each protein was assessed by SDS/PAGE. Intoxications were performed as in (29) with either PBS or 28 nmol PA and the indicated amount of LF_N-fusion protein added to HeLa cells.

Microscopy.

HeLa cells were grown to approximately 70% confluence on acid-washed coverslips in 12-well dishes then transiently transfected using FuGene HD (Roche) following the manufacturer's protocol. Coverslips were washed with PBS and fixed in 3.7% formaldehyde in 100 mM Pipes pH 7.4 then mounted onto glass slides (18). Slides were imaged for epifluorescence (Fig. 3, and Figs. S2 and S3) with an inverted Leica DMIRE2 microscope with a CCD camera, for deconvolution (Fig. 2) with a Deltavision microscope and software (Applied Precision) (18), or for confocal microscopy (Fig. S5) with a Zeiss LSM510-Meta. Images were then processed and compiled using Adobe Photoshop CS3 Extended. Details for cell–cell junction quantification can be found in SI Text and Fig. S4.

Membrane Fractionation and Western Blotting.

Transfected HeLa cells grown in 6-well dishes were divided into cytosolic and membrane fractions as described previously (9). Equivalent volumes of each fraction were separated by 15% SDS/PAGE and transferred to nitrocellulose membranes. GFP and GFP-fusions were detected with GFP-specific antiserum conjugated to horseradish peroxidase (Miltenyi Biotec). Membranes were developed with Supersignal West Pico chemiluminescent substrate (Pierce). Band intensities were determined using National Institutes of Health ImageJ 1.40g. Raw data for densitometry measurements can be found in Tables S2 and S3.

Molecular Modeling.

The MLD sequences from VcRID, VvDUF5, and VvRID were input into the GENO3D server (http://geno3d-pbil.ibcp.fr) for 3D structure modeling (30). The structures of TcdB, PMT, and TcsL were used as target templates for all models. Structure alignments and electrostatic potential predictions were performed using MacPymol (Delano Scientific).