Plasma membrane association of three classes of bacterial toxins is mediated by a basic-hydrophobic motif

Summary

Plasma membrane targeting is essential for the proper function of many bacterial toxins. A conserved four helical bundle membrane localization domain (4HBM) was recently identified within three diverse families of toxins; clostridial glucosylating toxins, MARTX toxins, and Pasteurella multocida-like toxins. When expressed in tissue culture cells or in yeast, GFP-fusions to at least one 4HBM from each toxin family show significant peripheral membrane localization but with differing profiles. Both in vivo expression and in vitro binding studies reveal that the ability of these domains to localize to the plasma membrane and bind negatively charged phospholipids requires a basic-hydrophobic motif formed by the L1 and L3 loops. The different binding capacity of each 4HBM is defined by the hydrophobicity of an exposed residue within the motif. This study establishes that bacterial effectors utilize a normal host cell mechanism to locate the plasma membrane where they can then access their intracellular targets.

Introduction

Many cellular proteins that function at the inner face of the plasma membrane (PM) have specific structural elements that facilitate interaction with the lipid bilayer. A common strategy is to incorporate post-translational lipid modifications to anchor proteins to the hydrophobic lipid portion of the membrane bilayer (reviewed in (Levental et al., 2010)). As an alternate strategy, a protein can directly interact with the hydrophilic portion of phospholipids via electrostatic interactions dependent upon either bound calcium ions or clusters of surface-exposed positively charged amino acids (Roy et al., 2000, Mulgrew-Nesbitt et al., 2006). For example, the myristoylated alanine-rich protein kinase C substrate (MARCKS) and the membrane penetrating peptide Antp are well-characterized proteins that utilize multiple surface exposed basic residues for membrane binding (Zhang et al., 2003, Drin et al., 2001). Nuclear magnetic resonance studies of these proteins demonstrate that the incorporation of exposed hydrophobic residues within the basic patch greatly enhances membrane association because the hydrophobic side-chain can penetrate the lipid bilayer (hereafter referred to as a basic-hydrophobic motif) (Zhang et al., 2003, Zhang et al., 2005). Many proteins, including the RhoGTPase superfamily of signaling proteins, combine strategies by utilizing both lipid modifications and basic-hydrophobic motifs for membrane association (Heo et al., 2006).

Mimicry of normal cellular processes is a strategy frequently used by pathogens to manipulate host cells for the benefit of the infecting agent. Since numerous cellular components essential to host cell integrity and cellular signaling are PM-associated proteins, many bacterial effectors injected by Type 3 secretion or delivered from large toxins by autoprocessing to the cell cytosol must be redirected to the PM by a membrane localization domain (MLD) to access their intracellular targets. Structural studies of the Pasteurella multocida toxin (PMT) revealed an ~85 aa four-helical bundle termed the C1 domain that was subsequently shown to be both necessary and sufficient to localize the C2/C3 catalytic domain of PMT to the PM (Kitadokoro et al., 2007, Kamitani et al., 2010). Similarly, structural studies revealed a four-helical bundle at the extreme N-terminus of all of the clostridial glucosylating toxins (CGTs), including Clostridium difficile Toxin B (TcdB), Clostridium sordelli lethal toxin (TcsL), and Clostridium novyi alpha toxin (TcnA) (Reinert, 2005, Ziegler et al., 2008). These helical domains are structurally similar to the PMT C1 domain (Kitadokoro et al., 2007) and likewise are now recognized to have a role in localizing the glucosylating catalytic effector domains to the PM (Mesmin 2004, Geissler, 2010, and Fig. 1) after autoprocessing releases the effector to the cytosol (Egerer et al., 2010). All of the multifunctional autoprocessing RTX toxin (MARTX) Rho inactivation domains (RIDs) from various Vibrio species and other human and insect pathogens contain a similarly sized four-helical bundle. In addition, 9 effector domains of unknown function (DUF) that have limited homology to the C1/C2 domains of PMT (known as the PMT-like effectors) also possess a structurally similar domain. In Vibrio vulnificus, we will refer to this domain as DUF5 because it is the fifth effector domain of the MARTX_Vv toxin. The four-helical bundles from Vibrio cholerae RID (VcRID) and V. vulnificus DUF5 (VvDUF5) have been shown to localize to the PM (Geissler et al., 2010). In total, 24 different toxin effector domains have been identified as having MLDs that can be structurally modeled against the four helical bundle domains of PMT and the CGTs (Fig. S1).

Fig. 1.

Identification of residues involved in 4HBM localization in HeLa cells. HeLa cells were transiently transfected with plasmids expressing the indicated versions of MLD_VvDUF5-GFP (A), MLD_VcRID-GFP (B), MLD_TcdB-GFP (C), or MLD_TcsL-GFP (D). Electrostatic potential predictions from crystal structures (MLD_TcdB and MLD_TcsL) or modeled structures (MLD_VvDUF5 and MLD_VcRID) are shown at the left to indicate the expected location of each mutated residue (blue surfaces are positively charged; red surfaces are negatively charged). Transfected cells were incubated for 16–24 hrs, fixed, and imaged by confocal microscopy. Scale bars, 10 µm.

While their sequence suggests that all 24 share a similar targeting domain, the attached effectors are not similar and can be sorted into three separate families; the PMT-like family, which is typified by PMT but also includes the PMT-like effectors from nine MARTX toxins; the RID family, which includes the RID effectors from eight MARTX toxins; and the CGT family, which are expressed by four species of pathogenic Clostridium. The ubiquity of this domain among toxin effectors suggests that this is a highly conserved functional domain that we now term the four helical bundle family of MLDs (4HBM), in order to distinguish them from unrelated membrane localization sequences identified in other bacterial effector proteins (Rabin et al., 2003, Krall et al., 2004, French et al., 2009).

Preliminary comparative studies of lipid interactions by the 4HBMs have shown that TcsL and TcdB, as well as the C1 domain of PMT associate directly with negatively charged phospholipids (Mesmin et al., 2004, Kamitani et al., 2010). However, the specificity of phospholipid binding varied for each protein. In vivo, the relative proportion of 4HBM protein localized to the membrane upon overexpression in culture cells was found to range from 16 to 56%, suggesting that while the 4HBMs are conserved in numerous toxins, the domains may have different innate capacities to associate with membranes (Geissler et al., 2010). Such diversity could be anticipated given that only two of ~85 residues are identical in all 24 4HBMs (Fig. S1).

The conserved structure of these domains and the fact that these domains are essential for the full activity of three different functional toxin families emphasize the significance of this region to toxin function (Geissler et al., 2010, Kamitani et al., 2010, Mesmin et al., 2004). By contrast, the differences in sequence highlight that knowledge of one 4HBM may not be directly applicable to the function of another. The initial characterization of the 4HBMs from VcRID, VvDUF5, and PMT identified several highly conserved residues as necessary for membrane localization. However, most of these residues were predicted to have a structural role in helical packing and may not represent residues that directly interact with phospholipids (Geissler et al., 2010, Kamitani et al., 2010). Therefore, it is of pivotal importance to not only understand how these 4HBMs associate with the PM, but also to compare toxins from different groups. In this study, we demonstrate that the 4HBMs contain a basic-hydrophobic motif that is critical for localization, indicating that these toxins mimic a normal eukaryotic localization mechanism. We further find that differences in the composition of the basic-hydrophobic motif leads to altered lipid association of the individual 4HBMs, thereby accounting for observed differences in the efficiency and stability of PM localization among the different toxins.

Results

Basic residues in L1/L3 are involved in 4HBM-membrane interactions

In an effort to isolate residues required for direct association of the 4HBMs with the PM, electrostatic surface charge models based on the crystal structures of TcdB and TcsL (Ziegler et al., 2008) and the predicted structures of Vibrio cholerae RID (MLD_VcRID) and Vibrio vulnificus PMT-like domain DUF5 (MLD_VvDUF5) (Geissler et al., 2010) were inspected for surface exposed basic residues that might interact with negatively charged phospholipids within the membrane. Numerous positively charged regions were identified in each of the structural models (Fig. 1).

In a previous study, 58% of a GFP-fusion to MLD_VvDUF5 was shown to be membrane-associated when expressed in HeLa cells compared to 34% of MLD_VcRID and 16% of MLD_TcdB ((Geissler et al., 2010). Thus, MLD_VvDUF5 was selected for an initial detailed analysis (Fig. 1A). In addition to the previously tested R71A mutation, Ala was substituted for all of the exposed Lys or Arg residues on the MLD_VvDUF5-GFP expression vector, and cells expressing these constructs were then viewed by fluorescence microscopy for reduced GFP localization (Fig. 1A). Four of the MLD_VvDUF5 mutants tested (R12A, K14A, K38A, R40A) localized to the periphery similar to wild-type. By contrast, substitution of Ala for Lys18 in the L1 loop disrupted the proper localization of MLD_VvDUF5 and the GFP signal was dispersed throughout the cytoplasm. Mutation of R71A in the L3 loop also blocked PM localization (Fig. 1A), consistent with previous results (Geissler et al., 2010).

Sequence alignment shows that a Lys or Arg residue at position 18 or 19 is present in 23 out of 24 4HBMs, while all 24 4HBMs have Arg at position 71 (Fig. S1). To test the necessity of a basic residue at these positions in two other 4HBMs representing the other subfamilies, Ala was substituted for K18 or R71 on MLD_VcRID-GFP and R18 and R71 on MLD_TcdB-GFP and MLD_TcsL-GFP. Similar to MLD_VvDUF5 and what was previously observed for the R71A mutation on MLD_VcRID and MLD_PMT (Geissler et al., 2010, Kamitani et al., 2010), these Ala substitutions abrogated PM localization of MLD_VcRID-GFP, MLD_TcdB-GFP, and MLD_TcsL-GFP in HeLa cells and resulted in exclusively cytosolic fluorescence (Fig. 1B–D).

Improved localization of MLD_VvDUF5 is not due to additional basic residues

Inspection of structural models (Fig. 1) suggested that the improved localization of MLD_VvDUF5 compared to MLD_VcRID could be due to basic patches at additional surface exposed sites, thereby allowing for multiple membrane contact points. This is probably not true since disruption of surface exposed Lys and Arg residues distant from L1/L3 face did not affect the localization of MLD_VvDUF5 (Fig. 1). In an attempt to improve the localization of the more weakly localizing MLD_VcRID, additional positively charged potential contact points were introduced. With the exception of the A12R mutation, substitution of Lys or Arg for several non-basic residues on MLD_VcRID (L10K, D43R, L79K) actually reduced PM fluorescence compared to wild-type (Fig. 1B), possibly due to disruption of the helical structure.

Therefore, the presence of basic residues at position 18/19 and 71 within the L1 and L3 loops is essential for proper membrane association of members of all three families of 4HBM and additional positively-charged regions are not responsible for the differences in PM binding between the different 4HBMs.

4HBMs associate with negatively charged phospholipids

Previous studies showed that the isolated 4HBM from PMT and the glucosyltransferase domains of TcdB and TcsL (with intact 4HBMs) all associate with artificial liposomes containing negatively charged phospholipids (Kamitani et al., 2010, Mesmin et al., 2004). To determine if this is a conserved feature amongst each of the three classes of 4HBMs and that liposome binding is 4HBM specific, five representative 4HBMs were purified as recombinant proteins from E. coli by affinity chromatography (Fig. 2A) and tested for their ability to co-sediment with neutral liposomes supplemented with either neutral or anionic phospholipids (Mesmin et al., 2004).

Fig. 2.

4BHM proteins bind negative phospholipids through ionic interactions. (A) The 4HBM proteins used in this work were purified by affinity chromatography (1 µg of each protein is shown). (B, C) The percentage of each 4HBM construct bound to neutral liposomes supplemented with the indicated phospholipid (B) or PS-supplemented liposomes in the presence of increasing amounts of NaCl (in mM) (C) was determined using ultracentrifugation followed by densitometry of Tricine-SDS-PAGE gels (mean +/− SEM, n ≥ 2 for each condition).

Fig. 2B shows that the 4HBMs from VcRID, VvDUF5, TcdB, and TcsL each bound to liposomes containing anionic phospholipids (phosphatidic acid (PA), phosphatidylglycerol (PG), phosphatidylinositol (PI), phosphatidylserine (PS), and phosphatidylinositol 4,5-bisphosphate (PIP2)) with varying efficiencies; however, none of the 4HBMs significantly bound to the zwitterionic phosphatidylcholine (PC) supplemented liposomes. This result indicates that a phospholipid with a negatively charged head group is required for the MLD-liposome association. Despite the requirement for anionic phospholipids in the liposomes, the valence of the phospholipid did not influence the binding of the VcRID, VvDUF5, or TcdB 4HBMs, as there was no significant difference in the average percentage of protein in the pellets of the monoanionic (PA, PG, PI, PS) versus the polyanionic (PIP2) phospholipid supplemented liposomes (P values >0.2). Using surface plasmon resonance, Kamitani et al. (2010) showed that PMT C1 also associated with anionic liposomes. Importantly, MLD_VvRID, which did not localize to the membrane in HeLa cells, did not significantly bind to any of the liposomes tested, whereas all of the 4HBMs that localized to the HeLa PM (Fig. 1) also associated with anionic liposomes (Fig. 2B).

The 4HBM-phospholipid association is a calcium independent electrostatic interaction

The requirement for a negatively charged membrane and the necessity of basic amino acids for proper localization suggest that the 4HBM association with the membrane may be mediated by electrostatic interactions. To test this, a member of each class of 4HBM was tested for its ability to associate with PS supplemented liposomes under conditions of increasing ionic strength. For all three 4HBMs tested, the percentage of protein bound to liposomes significantly decreased with increasing NaCl (Fig. 2C), indicating a direct correlation between the presence of free negative charges on the phospholipids and the ability of 4HBM to bind liposomes. Interestingly, both MLD_VvDUF5 and MLD_TcdB showed some resistance to increasing NaCl as there was no significant decrease in the amount of liposome-bound protein until concentrations of NaCl higher than 100 mM were added to the reactions (Fig. 2C), whereas the more weakly associated MLD_VcRID had reduced binding at 100 mM NaCl. These results suggest that, in addition to electrostatic interactions, another mechanism may be contributing to membrane binding.

One of these mechanisms is for membrane-associated proteins to utilize bound Ca²⁺ to mediate binding to negatively charged phospholipids (reviewed in (Stace et al., 2006)). However, Ca²⁺ is not necessary for the 4HBM-liposome association since addition of EDTA to liposome binding assays had no effect on the percentage of either MLD_VcRID or MLD_VvDUF5 found in the pellet fraction with either PS or PG liposomes (Fig. S2).

To follow up on our in vivo finding that basic residues in the L1 and L3 loops were specifically involved in phospholipid binding, liposome binding assays were performed with K18A and R71A mutants of MLD_VvDUF5 as well as a Y24A mutant that had been previously characterized as defective in membrane localization (Geissler et al., 2010). In line with their localization defects in HeLa cells, substitution of K18 or R71 for Ala significantly reduced the association of MLD_VvDUF5 with liposomes supplemented with PG or PS (Fig. 3B) revealing these residues are important for direct phospholipid interactions. However, changing the 100% conserved Y24 to Ala had no effect on liposome binding (Fig. 3B) even though this mutant was unable to localize GFP to the membrane when expressed in cells (Geissler et al., 2010). Since Y24 is oriented towards the inside of the helical bundle and does not project towards the surface, it likely ensures proper positioning of the L1/L3 surface exposed residues for in vivo localization (Fig. 3A), but had less of an effect in the in vitro assay. This structural role is supported by the previous findings that substitution of Phe for Y24 permitted normal localization (Geissler et al., 2010). Importantly, CD spectroscopy showed that none of the mutations significantly altered the overall structure of the proteins (Fig. 3C).

Fig. 3.

Mutation of conserved residues affects MLD_VvDUF5 phospholipid binding but not the overall structure. (A) The modeled location and orientation of the residues within L1/L3 that were tested for liposome binding. (B) The indicated proteins were incubated with neutral liposomes supplemented with the indicated phospholipid (PG, PS) and subjected to ultracentrifugation. The relative percentage of liposome-bound mutant protein in relation to wild-type (WT) MLD_VvDUF5 was then determined using densitometry of Tricine-SDS-PAGE gels (mean +/− SEM, n ≥ 2 for each condition). (C) CD spectra of MLD_VvDUF5 wild-type and mutant proteins.

In order to monitor if a similar mechanism of binding is present in the more weakly localizing MLD_VcRID, residues previously shown to affect membrane localization in vivo were tested for liposome binding. Similar to their effects on HeLa cell localization (Fig. 1B and (Geissler et al., 2010)), mutation of K18, S69, or R71 on MLD_VcRID almost completely abrogated binding to liposomes (Fig. S3A). However, these decreases are possibly due, at least in part, to structural changes as CD spectroscopy indicated an alteration to the structure, although these mutants remained predominantly alpha helical in structure (Fig. S3B). This reveals that the 4HBMs also differ in structural integrity, representing another variation in an otherwise conserved domain.

4HBM localization is dependent on membrane charge

If interaction of the L1/L3 basic patch with negatively charged phospholipids is essential for 4HBM localization, it should be possible to disrupt localization by altering the in vivo membrane composition. To test this, transfected HeLa cells were treated with chemical agents that neutralize membrane surface charge.

Dibucaine inserts into the membrane and acts to flip the major negative component, PS, from the inner leaflet of the PM to the outer leaflet (Yeung et al., 2006). Incubation of HeLa cells expressing either MLD_VvDUF5-GFP or MLD_TcdB-GFP with dibucaine resulted in a decreased association of each 4HBM with the membrane, with <4% of cells expressing either fusion retaining PM localization similar to untreated cells (Fig. 4, compare treated to untreated). This result suggests that in HeLa cells, proper PS orientation is required for 4HBM localization.

Fig. 4.

Altering the charge of the PM abrogates 4HBM-GFP membrane localization. HeLa cells transiently expressing each GFP-fusion were left untreated (Unt) or treated with either 10 mM dibucaine for 20 minutes (Dib) or 10 µM ionomycin for 5 minutes (Iono) then imaged by confocal microscopy. In order to quantify these effects, the percentage of cells displaying GFP localization similar to typical untreated cells (marked with an *) was determined for each condition and indicated in the upper right portion of each panel (n > 20 for each construct and condition). Scale bars, 10 µm.

The calcium ionophore, ionomycin, raises the intracellular Ca²⁺ concentration, which reduces the overall negative charge of the membrane (Fairn et al., 2009). The percentage of cells displaying MLD_VvDUF5-GFP or MLD_TcdB-GFP PM localization similar to untreated cells following ionomycin treatment dropped 5-fold and 13-fold, respectively (Fig. 4), indicating that the charge of the PM is involved in 4HBM localization. Together, these results show that the proper localization of the 4HBMs is dependent on the composition and charge of the phospholipids in the HeLa cell PM.

4HBM localization in Saccharomyces cerevisiae

In order to further characterize the ability of the 4HBMs to target the eukaryotic PM (Geissler et al., 2010), a Saccharomyces cerevisiae model was established. 4HBMs from VcRID, VvRID, PMT, VvDUF5, TcdB, and TcsL were expressed in S. cerevisiae as fusions to GFP and fluorescence was monitored microscopically. Similar to localization in HeLa cells (Fig. 1A and (Geissler et al., 2010), MLD_VvDUF5 showed the most pronounced peripheral localization in yeast (Fig. 5). MLD_VcRID-GFP also localized predominately to the periphery of S. cerevisiae, whereas the GFP-fusions to TcdB, TcsL, and PMT 4HBMs showed less peripheral localization and instead were distributed throughout the cytoplasm (Fig. 5 and Fig. S4A). A GFP-fusion to a sixth representative 4HBM, MLD_VvRID, was not membrane associated in yeast (Fig. S4A), consistent with data from HeLa cells (Geissler et al., 2010) and liposome binding assays (Fig. 2B), further demonstrating the ineffectiveness of this particular 4HBM.

Fig. 5.

Deleting the genes for enzymes involved in phospholipid biosynthesis alters 4HBM-GFP localization in yeast. The indicated strains of S. cerevisiae were transformed with plasmids expressing either GFP (A), MLD_VvDUF5-GFP (B), or MLD_VcRID-GFP (C) and grown for ~16hrs in media containing galactose to induce expression of each 4HBM-GFP. Cells were then immobilized and imaged by confocal microscopy. Insets show representative individual cells digitally magnified 3× and the average fluorescence intensity distribution across individual cells (n=10 for each). Scale bars, 10 µm.

As MLD_VvDUF5 was the most effective for localization in yeast, residues essential for localization were tested by substituting either Ala or conservative residues for several 100% conserved amino acids on MLD_VvDUF5 that were required for localization in HeLa cells (Geissler et al., 2010). Localization of MLD_VvDUF5-GFP to the periphery of yeast, as in HeLa cells, was reduced by mutation of K18, R71, and the two other highly conserved residues, Y24 and S69 (Fig. S4B). Altogether, these data show that GFP-fusions expressed in S. cerevisiae can be used as a model to study the localization of the strongly localizing 4HBM proteins (MLD_VvDUF5 and MLD_VcRID).

Altering the yeast membrane composition decreases 4HBM localization

Development of the S. cerevisiae model facilitated using genetics to address host requirements for the 4HBM localization. S. cerevisiae strains deleted for genes involved in phospholipid biosynthesis were transformed with plasmids expressing unfused GFP, MLD_VvDUF5-GFP, or MLD_VcRID-GFP and the fluorescence distribution across each cell was then measured. Neither CGT-family 4HBM nor MLD_PMT was tested due to their poor localization in wild-type yeast.

In S. cerevisiae, cho1 encodes for phosphatidylserine synthase, which is responsible for generating PS, the major negative phospholipid component (~30%) of the yeast PM (Zinser et al., 1991). Deletion of cho1 almost completely abrogated peripheral fluorescence of both MLD_VvDUF5-GFP and MLD_VcRID-GFP, resulting in significant differences to the average localization patterns compared to the wild-type strains (Fig. 5; P values ≤0.0002) and the localization patterns of GFP, MLD_VvDUF5-GFP, and MLD_VcRID-GFP in the cho1 strain were not significantly different from each other (P values >0.5). Deletion of the gene encoding inositol 1-phosphate synthase (ino1), which generates precursors for all inositol-containing phospholipids (Stein et al., 2002) had no effect on the localization patterns of either MLD_VvDUF5-GFP or MLD_VcRID-GFP (Fig. 5; P values >0.1). These data highlight the importance of the role of PS in generating the negative membrane charge required for proper 4HBM peripheral localization in yeast.

An exposed hydrophobic residue in L1 enhances membrane association

Thus far we have established that the 4HBMs utilize surface exposed basic residues to recognize anionic phospholipids in the PM. Although this is a common mechanism of membrane association utilized by numerous prokaryotic and eukaryotic proteins, many proteins augment PM association through inclusion of a hydrophobic post-translational modification (e.g. myristoylation) and/or a hydrophobic residue(s) into the positive surface, thereby creating a basic-hydrophobic motif. The crystal structures of TcdB, TcnA, and TcsL and models of MLD_VvDUF5 and MLD_VcRID show that each L1/L3 region contains 2–3 Arg or Lys residues, and that these surround a surface exposed hydrophobic residue at position 16 or 17 (Fig. 6A and Fig. S1). Importantly, the hydrophobicity of the residues found at these positions varies greatly with a Phe residue in MLD_VvDUF5, TcdB, and TcsL; a Leu residue in TcnA; and a Thr residue in MLD_VcRID (Fig. 6A). Despite >89% aa identity with MLD_VcRID, MLD_VvRID lacks a hydrophobic residue at this position and instead encodes a more hydrophilic Asn (Fig. S1).

Fig. 6.

An exposed hydrophobic residue within Loop1 influences the 4HBM-membrane interaction. (A) Amino acid and structure alignments of the L1 region. K/R are highlighted in blue, F are highlighted in green, I/L are highlighted in yellow, and N are highlighted in magenta. Residues in positions 16/17 are depicted as sticks in the structure alignment and are colored as in the left panel. (B) S. cerevisiae transformed with either wild-type or mutant 4HBM-GFP constructs were grown for ~16 hrs in media containing galactose to induce expression. Insets show representative individual cells magnified 3×. (C) The average fluorescence intensity distribution across individual cells for each condition (n=10 for each construct). (D) HeLa cells transiently expressing each GFP-fusion were fixed and imaged by confocal microscopy ~20hrs following transfection. Arrows highlight areas of clear PM fluorescence. (E) HeLa cells transfected with the indicated constructs were separated into soluble and membrane fractions and the average percentage signal (+/− the standard deviation) in the membrane fraction was determined by densitometry of immunoblots (mean +/− SEM, n ≥ 3). (F) Liposome binding assays were conducted with each mutant protein using PS-supplemented liposomes. The relative binding of each mutant protein in relation to its corresponding wild-type is presented (mean +/− SEM, n ≥ 3). Scale bars, 10 µm.

In order to test the importance of a strongly hydrophobic residue at this position, a codon for Phe was substituted for N16 on the non-localizing MLD_VvRID-GFP. This substitution significantly increased the peripheral localization of MLD_VvRID in both S. cerevisiae and HeLa cells, resulting in a localization profile that appeared more similar to cells expressing MLD_VvDUF5-GFP than those expressing wild-type MLD_VvRID-GFP (Fig. 6B–D). Fractionation of membranes from transfected HeLa cells showed that the N16F substitution significantly increased the membrane association of MLD_VvRID (Fig. 6E; P=0.0371).

Conversely, F16 in MLD_VvDUF5 was mutated to Asn to mimic MLD_VvRID. When MLD_VvDUF5-GFP F16N was expressed in yeast or HeLa cells, there was a clear increase in cytosolic fluorescence and a concomitant decrease in peripheral fluorescence compared to wild-type MLD_VvDUF5-GFP (Fig. 6B–D). In support of these results, HeLa cell fractionations showed a significant decrease in the amount of MLD_VvDUF5 F16N–GFP in the membrane compared to wild-type (Fig. 6E; (P=0.0078)).

As final proof of the importance of this hydrophobic residue in defining the membrane binding capacity as part of basic-hydrophobic motif, the ability of these mutant proteins to associate with PS-supplemented liposomes in vitro was assessed. Fig. 6F shows that substitution of Phe for N16 on MLD_VvRID significantly increased the percentage of liposome-bound protein (P=0.0002), whereas the Asn for F16 mutation on MLD_VvDUF5 significantly decreased the association (P=0.0023).

Altogether, these data show that the presence of an exposed hydrophobic amino acid within the basic L1/L3 region is critical for proper peripheral membrane localization, thus demonstrating that the 4HBMs utilize a basic-hydrophobic motif for targeting effector domains to the PM.

Discussion

We have previously used protein structure prediction and modeling to identify a surface exposed positively charged region present within the 4HBMs. Initial characterization of members within this conserved family of MLDs suggested that they might associate with a specific negatively charged membrane phospholipid (Geissler et al., 2010, Mesmin et al., 2004, Kamitani et al., 2010). In this work, we tested this hypothesis using targeted mutagenesis coupled with in vivo localization studies and in vitro liposome binding assays designed to mimic the host cell membrane (Mesmin et al., 2004). Altogether, our findings show that the ability of a given 4HBM to associate with the membrane is dependent on the concerted action of an exposed highly hydrophobic residue at the top of L1 (position 16/17) and electrostatic interactions between basic residues surrounding L1/L3 with negatively charged membrane phospholipids. These residues form a basic-hydrophobic motif within the 4HBM at the alternate side of their cognate effector proteins, which mediates their interaction with the overall negative charge of the PM rather than with specific phospholipids.

Differences in the composition of the basic-hydrophobic motif can explain both similarities and observed variations in membrane binding between the different 4HBMs (Fig. 7). Importantly, all of the 4HBMs shared K/R18, which is expected to provide the main positively charged side chain necessary for each of the 4HBMs to associate with negatively charged phospholipids. All of the 4HBMs also have an R71 residue that might interact with negatively charged phospholipids as well. However, its interior position within the known and modeled 4HBM crystal structures suggests that R71 plays a predominately structural role in properly exposing both K/R18 and the hydrophobic residue to the membrane, although R71 most likely further contributes to the overall positive charge of the L1/L3 face.

Fig. 7.

Model for 4HBM plasma membrane association through a basic-hydrophobic domain. Panels show how two representative 4HBMs mediate their association with the host cell PM. The modeled structures of MLD_VvDUF5 (left panel) and MLD_VvRID (right) are depicted as electrostatic surfaces over ribbon structures, with residues affecting localization labeled (blue surfaces are positively charged; red surfaces are negatively charged). Comparing the panels shows the difference in electrostatic interactions (arrows, green=attractive, red=repulsive) between the polybasic surface of each 4HBM and the anionic phospholipids within the membrane (red). Applying our data to the mechanism of membrane association of other basic-hydrophobic domain-containing proteins suggests that the exposed Phe within L1 of MLD_VvDUF5 (or other hydrophobic containing 4HBMs) inserts into the lipid bilayer and may act as a membrane anchor (left panel). Conversely, the right panel depicts the basis for the weaker membrane association of 4HBMs such as MLD_VvRID that lack an exposed hydrophobic residue and/or multiple basic residues surrounding L1.

In addition to Phe17, K18, and R71, MLD_VvDUF5 has three other positively charged residues surrounding L1/L3 and thereby displays a discreet PM localization with minimal residual protein in the cytosol, which is particularly visible when expressed in yeast (Fig. 5B and 6B). The CGT 4HBMs have only two additional basic residues along with Phe17 in this region and thereby have consistently displayed decreased PM association compared to MLD_VvDUF5. The RID-like 4HBMs have even fewer Arg/Lys and lack a truly hydrophobic residue in L1, and thus show even less success at membrane localization. In this way, the toxin effectors are mimicking a mechanism used by eukaryotic, bacterial, and viral proteins to localize to the PM. For the MARCKS peptide, it has been shown that substituting Ala for the five Phe residues within the basic-hydrophobic motif decreases PS-liposome binding 10–100-fold (Wang et al., 2002, Arbuzova et al., 2000). These Phe have been shown to draw the MARCKS peptide closer to the membrane, thereby increasing the electrostatic interactions between the exposed basic residues (Zhang et al., 2003). Similar to MARCKS, a surface exposed Phe residue within a basic region on Bacillus subtilis DivIVA likely plays a comparable role in membrane association, as substitution of F17 for Ala significantly reduced DivIVA liposome binding (Oliva et al., 2010). Likewise, HIV-1 Gag uses electrostatic interactions and a myristoyl moiety to mediate its association with negatively charged phospholipids in the membrane (Dalton et al., 2007). Therefore, our results using various bacterial toxin domains further highlight the importance of an exposed hydrophobic residue(s) to the electrostatic interactions involved in protein-membrane binding as well as the conservation of this mechanism for PM association. These differences in PM association may then impact how the various toxins function. In particular, how the Phe substitution in VvRID might augment activity or its absence from VvDUF5 might reduce activity. Unfortunately, the molecular mechanisms of these two effectors are not known so there is no ability to quantitatively assess how these substitutions might affect the activity of the individual effectors or the MARTX_Vv holotoxin that carries both effectors.

The relative binding capacity of the different 4HBMs for the membrane is consistent with the principles of the interfacial hydrophobicity of amino acids determined by Wimley and White (Wimley et al., 1996). Applying their hydrophobicity scale to our findings offers an additional explanation for the strong association of MLD_VvDUF5 and the apparent lack of membrane binding for MLD_VvRID. Their data shows that Phe, which is present at position 16/17 on many 4HBMs including MLD_VvDUF5 and MLD_TcdB, is the second most favorable residue for membrane insertion. By contrast, Asn (as present in the non-localizer MLD_VvRID) contributes unfavorably to membrane insertion, thereby removing an integral component of the 4HBM-membrane interaction (Fig. 7, compare panels). The overall effect of a strongly hydrophobic residue at this position would be predicted to contribute more favorably to bilayer insertion and should ultimately result in enhanced membrane association by decreasing the distance between the membrane and the positively charges surrounding L1. Consistent with this, substitution of Asn16 with a Phe on MLD_VvRID restored the ability of this domain to localize to the PM.

Asn from MLD_VvRID is not the only unique residue at this position (Fig. S1). The 4HBM from Clostridium perfringens lethal toxin (TpeL) has an extended L1 and contains a unique Pro adjacent to the conserved Lys19, and accordingly appears very unfavorable for insertion. In the 4HBM from the Candidatus Regiella insecticola PMT-like effector, the L1 loop is also extended and has a very favorable Tyr adjacent to Arg19. Equally divergent, MLD_PMT and MLD_TcdA have a favorable Ile at position 16/17 and only two positively charged residues surrounding the L1/L3 interface (Fig. S1). Demonstrating that Ile is a favorable residue at this position, MLD_PMT specifically interacts with negative phospholipids in vitro (Kamitani et al., 2010) and displays obvious PM localization in tissue culture cells (Kitadokoro et al., 2007, Kamitani et al., 2010). However, MLD_PMT-GFP did not localize to the periphery of yeast cells, perhaps due to this irregular basic-hydrophobic motif (Fig. S4). Therefore, these modifications within L1 show that there are no defined residue requirements within the L1/L3 membrane interaction face, as long as those present promote either membrane insertion or phospholipid binding. Our findings support a role for the composition of L1 in providing flexibility or specificity for the membrane association for each 4HBM and may explain why the individual 4HBMs interact differently with phospholipids.

Unlike the classic eukaryotic mechanisms of membrane targeting following myristoylation, palmitylation, or Ca²⁺ binding, we have shown that 4HBM membrane association is independent of host cell modifications. Further, 4HBM localization does not depend on binding to other proteins or to lipid rafts, as proposed for the localization of the Type III secretion system effectors (French et al., 2009). 4HBM-membrane binding can therefore occur immediately following autoprocessing of each effector from the holotoxin (reviewed in (Egerer et al., 2010)), without relying on further processing or activation by host factors. However, analogous to the MARCKS-membrane interactions (Kim et al., 1994), the catalytic domains attached to each 4HBM likely provide additional spatial specificity and augment membrane association upon target protein binding. Thus, despite utilizing common mechanisms for membrane targeting, there are many different nuances in the manner in which these 4HBMs associate with the host membrane. How these differences in 4HBM-membrane interactions will translate to the fundamental function of each toxin and the interaction with its cognate target remains to be fully understood but can be addressed when the catalytic mechanism of each effector has been solved.

Experimental procedures

Cell lines, yeast strains, reagents, and DNA sequencing

HeLa cells (ATCC) were cultured at 37°C with 5% CO₂ in DMEM containing 10% FBS, 50u/ml penicillin, and 50µg/ml streptomycin. Strains of Saccharomyces cerevisiae (Open Biosystems) were grown in YPD supplemented with 1 mM ethanolamine (cho1) or 2 mg/L myo-inositol (ino1) as necessary. All reagents and chemicals were purchased from Sigma or MP Biomedicals. Restriction enzymes were purchased from New England Biolabs. Plasmid DNA was isolated using Econospin spin columns (Epoch biolabs) and sequenced in the Northwestern University Genomics core facility.

Construction of TcsL and PMT MLD-GFP and mutants

Pfx50 and specific primers (Table S1) containing flanking NheI and BamHI sites were used to amplify the DNA encoding the amino acids for TcsL 1–83 (from pMQ101 (Voth et al., 2004)) or PMT 589–668 (from P. multocida BAA-428 genomic DNA, ATCC). These products were digested with NheI-BamHI and ligated with same-cut pEGFP-N3 (Clontech) to construct pMLD_TcsL-GFP or pMLD_PMT-GFP. Point mutations on pMLD_VcRID-GFP, pMLD_VvRID-GFP, pMLD_VvDUF5-GFP, pMLD_TcdB-GFP, and pMLD_TcsL-GFP were generated using the Quikchange XL II site-directed mutagenesis kit (Stratagene) or by Genscript (Piscataway, NJ). All mutations were verified by sequencing.

Construction of plasmids for MLD expression in yeast

Pfx50 and specific primers (Table S1) containing at least 30 bp homology to pYC2/NT A (Invitrogen) were used to amplify the DNA encoding the amino acids for the 4HBM-GFP fusions using the pEGFP-MLD constructs as template. S. cerevisiae gap repair cloning of these PCR products was then used to generate the pYC-MLD-GFP plasmids (Ma, 1987). Site-directed mutations on the various pYC-MLD-GFP plasmids were generated gap repair mediated cloning of either DNA from the pEGFP-MLD mutant construct or by incorporation of the desired mutation within a primer (Table S1).

Construction and purification of His-4HBM proteins

Plasmids expressing 6xHis-tagged 4HBM proteins were constructed using yeast gap repair cloning into pYCpet (Geissler et al., 2010). Pfx50 polymerase and specific primers (Table S1) containing homology to pYCpet were used to amplify DNA encoding each 4HBM using pEGFP-MLD plasmids as templates. Individual PCR products were then transformed into S. cerevisiae along with linearized pYCpet, generating pYCpet-4HBM-6xHis plasmids. Once confirmed by sequencing, these plasmids were transformed into E. coli BL21 DE3 and then proteins were purified as His-LF_N-RID in (Sheahan et al., 2007). Notably, the 6His-tag attached to the C-terminus would be associated with the opposite side of the 4HBM proteins than the L1/L3 face and is not expected to contribute to or interfere with membrane interactions.

Liposome binding assays

Liposomes were generated based on a neutral backbone; 30% PE, 20% cholesterol, 20% PC, with the remaining 30% either PC, the monovalent phospholipids (PS, PA, PG, PI), or the multivalent phospholipid PIP₂ (5% PI(4,5)P₂ + 25% PC) (Mesmin et al., 2004). Briefly, lipid chloroform stock solutions (Avanti Polar Lipids) were mixed and dried under nitrogen gas then by vacuum. The resulting lipid film was resuspended to 10mM final lipid concentration in 20mM HEPES, pH 7.5, 100mM KCl then sonicated. The resulting liposomes were monitored for size consistency by microscopy. Liposome binding assays were performed essentially as in (Mesmin et al., 2004), except that for each assay 3 µM of the indicated protein was mixed with 3 mM liposomes and incubated in 20 mM HEPES, pH 7.5, 100 mM KCl supplemented with 1 mM MgCl₂ and 1 mM CaCl₂ (and NaCl at the indicated concentrations) for 5 minutes at 37°C. Each sample was then centrifuged at 200,000xg for 2 hours at 25°C. The supernatants were removed and the pellets resuspended in an equal volume of buffer. Samples were then subjected to 16% Tricine-SDS PAGE followed by staining and densitometry with National Institutes of Health ImageJ 1.40g. All of the differences between samples were compared with an unpaired Student’s t-test using Prism 4.0c (Graphpad).

Circular dichroism

The indicated proteins were analyzed at 0.01 mg/mL (VcRID, VvRID, VvDUF5, and mutants) or 0.02mg/mL (TcdB and TcsL) in 10 mM KH₂PO₄, 50 mM Na₂SO₄, pH 7.5 at 23°C in 0.1 cm quartz cuvettes. Spectra were recorded from 190 – 250 nm using a J-815 circular dichroism (CD) spectrophotometer and Spectra Manager software (Jasco). Secondary structure assignments were determined using the CDPro software with the CONTIN analysis method (Colorado State University, Fort Collins, CO).

Membrane fractionation and immunoblotting

Transfected HeLa cells were separated into membrane and cytosolic fractions and processed for western blotting using GFP-specific antiserum conjugated to horseradish peroxidase (Miltenyi Biotec) as described (Geissler et al., 2010).

Microscopy

Live yeast cells were immobilized in 1% agarose in SC-media following overnight growth induction then imaged by confocal microscopy using an LSM510 Meta (Zeiss). The ability of each 4HBM-GFP fusion to localize to the yeast PM was determined by measuring the fluorescence intensity distribution across individual cells using NIH ImageJ similar to (Geissler et al., 2010). The average distributions for each 4HBM-GFP construct were determined from 10 separate cells from each image shown.

HeLa cells were grown to ~60% confluence on acid-washed coverslips in 12-well dishes then transiently transfected and processed for microscopy as in (Geissler et al., 2010). Slides were imaged as above and images were compiled using Adobe Photoshop CS3 Extended. For chemical modification of the PM, 24 hrs after HeLa cells grown on coverslips were transfected, spent media in each well was replaced with either fresh DMEM+FBS (dibucaine) or 20 mM HEPES, pH 7.4, 140 mM NaCl, 5 mM KCl, 1 mM MgCl₂, 2 mM CaCl₂ (ionomycin, untreated) (Yeung et al., 2006). Ionomycin (10 µM final) or dibucaine (10 mM final) was then added and incubated for 5 minutes (ionomycin) or 20 minutes (dibucaine) at 37°C with 5% CO₂, then cells were fixed, processed, and imaged as above.

Structure homology modeling and alignments

The GENO3D server (http://geno3d-pbil.ibcp.fr) was previously used to generate 3D structure models MLD_VcRID, MLD_VvDUF5, and MLD_VvRID (Combet et al., 2002). Structure alignments to the MLDs from TcdB, TcsL, and TcnA and electrostatic potential predictions were performed using MacPymol: PyMOL Enhanced for Mac OS X (Delano Scientific). Sequence alignments were generated using Jalview 2.6.1 (Waterhouse et al., 2009).