RTX proteins: a highly diverse family secreted by a common mechanism

Repeats-in-toxin (RTX) exoproteins of Gram-negative bacteria form a steadily growing family of proteins with diverse biological functions. Their common feature is the unique mode of export across the bacterial envelope via the type I secretion system and the characteristic, typically nonapeptide, glycine- and aspartate-rich repeats binding Ca2+ ions. In this review, we summarize the current state of knowledge on the organization of rtx loci and on the biological and biochemical activities of therein encoded proteins. Applying several types of bioinformatic screens on the steadily growing set of sequenced bacterial genomes, over 1000 RTX family members were detected, with the biological functions of most of them remaining to be characterized. Activities of the so far characterized RTX family members are then discussed and classified according to functional categories, ranging from the historically first characterized pore-forming RTX leukotoxins, through the large multifunctional enzymatic toxins, bacteriocins, nodulation proteins, surface layer proteins, up to secreted hydrolytic enzymes exhibiting metalloprotease or lipase activities of industrial interest.


Introduction
With molecular cloning and DNA sequencing taking grounds in bacteriology labs, it has become clear since 1987 that a novel family of large secreted cytolytic toxins of Gramnegative pathogens emerged. Summarizing the similarities of the first five characterized determinants in a seminal Micro-Review in 1991, Rodney A. Welch first introduced the concept of the RTX family of proteins characterized by the presence of arrays of glycine-and aspartate-rich nonapetide repeats. He predicted that this was a broadly disseminated family, while at that time it was difficult to imagine how broad and variable it could be. RTX proteins are produced by a variety of Gram-negative bacteria and exhibit two common features. The first is the presence of repetitions of glycine-and aspartate-rich sequences, typically nonapeptides, which are located in the carboxy-terminal portion of the protein. These form numerous sites for the binding of Ca 21 ions and are at the origin of the historical name of the protein family, where RTX stands for repeats in toxin (Welch, 2001). The second key feature is the unique mode of secretion of RTX proteins via the type I secretion system (TISS). Protein translocation occurs through an oligomeric secretion channel spanning across the entire Gram-negative bacterial cell envelope (i.e. cytoplasmic membrane, periplasmic space and outer membrane). These dedicated ATP-binding cassette (ABC) transporter-based secretion apparati recognize uncleavable C-terminal secretion signals and mediate a single-step translocation of the RTX polypeptides from bacterial cytosol across both the inner and the outer bacterial membrane, directly into the extracellular space and without a periplasmic secretion intermediate. The rtx genes and genes needed for secretion are mostly located within a single larger rtx locus. RTX proteins represent a family of proteins that exhibit a wide range of activities and molecular masses from 40 to 4 600 kDa. A prominent and historically first described group of RTX proteins consists of toxins, mostly exhibiting a cytotoxic pore-forming activity, often first detected as a hemolytic halo surrounding bacterial colonies grown on blood agar plates (Goebel & Hedgpeth, 1982;Muller et al., 1983;Felmlee et al., 1985;Welch, 1991).
While the word 'toxin' is, for historical reasons, embodied in the name of the RTX protein family, a broad class of RTX proteins also comprises secreted proteases and lipases. These can act as synergistic virulence factors causing tissue damage and/or by eliciting the production of inflammatory mediators. Recently, a subgroup of very large RTX proteins (4 3200 amino acid residues) with multiple activities [multifunctional autoprocessing RTX toxins (MARTX)] was discovered. For example, the prototype Vibrio cholerae MARTX Vc was shown to cause rounding of epithelial cells by catalyzing covalent cross-linking of cellular actin. RTX proteins can further act as bacteriocins or contribute to defense against environmental aggression by forming protective bacterial surface (S)-layers. Some RTX proteins were also found to play a role in plant nodulation or Cyanobacteria motility, while the biological role of most RTX proteins remains unknown. Bioinformatic mining of the explosively growing database of bacterial genomes indicates that RTX proteins form a large and diverse family of proteins, with a broad spectrum of biological and biochemical activities.

RTX repeats
The requirement for calcium ions in RTX toxin activities was first documented for Escherichia coli a-hemolysin (Short & Kurtz, 1971) and Bordetella pertussis CyaA (Hanski & Farfel, 1985). Binding of calcium ions to the repeats of RTX toxins occurs only upon secretion, as the intracellular cytoplasmic calcium concentration in bacteria is quite low (o 100 nM) (Gangola & Rosen, 1987). The RTX protein needs to unfold or remains in a floppy conformation before translocation out of the cell through the TISS (Kenny et al., 1991;Koronakis et al., 2000). Calcium binding to the nonapeptide repeats in the C-terminal portions of these toxins then appears to promote folding and imposes adoption of a functional conformation on the secreted RTX proteins in the extracellular environment (Felmlee & Welch, 1988;Ludwig et al., 1988;Rose et al., 1995;Rhodes et al., 2001).
Analysis of the three-dimensional structure of Pseudomonas aeruginosa alkaline protease possessing six of the RTX motifs with a consensus sequence X-(L/I/F)-X-G-G-X-G-(N/D)-D, where X means any residue, revealed that the repeated sequences constitute a new type of calcium-bind-ing structure (Fig. 1), called a parallel b-helix or a parallel b-roll (Baumann et al., 1993). In this structure, the first six residues of each motif form a turn that binds calcium, and the remaining three residues build a short b-strand. The Fig. 1. The parallel b-roll structure of the RTX repeats. Top view (a) and side view (b) projection of the RTX repeats comprising residues Gly333 and Phe376 from the three-dimensional model of alkaline protease of Pseudomonas aeruginosa (Baumann et al., 1993). The protein backbone, short b-strands and Ca 21 ions are represented as grey ribbon, cyan arrows and yellow balls, respectively. (c) A detailed view of the calcium-binding site within the RTX repeats. Each nonapeptide motif forms two half-sites for Ca 21 binding, where each Ca 21 ion is bound in a six coordinate site between two consecutive turns. The first turn contributes the main chain carbonyls of Asn343 and Val345, and one carboxyl oxygen of Asn347. The second turn contributes the carbonyls of Gly360 and Leu362, as well as the carboxyl oxygen of Asp365. Residues whose side chains directly coordinate the Ca 21 ion are highlighted. The carbon atoms of the side chains are green, nitrogens are blue, oxygens are red. arrangement of consecutive turns and b-strands builds up a right-handed helix of parallel b-strands, where one turn of this helix consists in two consecutive nonapeptides. Calcium is then bound within two consecutive turns by the conserved aspartic acids.
The numbers of RTX repeats vary among RTX proteins from o 10 to 4 40. While RTX proteases and lipases typically have a single block of seven to eight RTX nonapeptide repeats, very extensive RTX repeat domains were found recently in the large MARTX or in putative RTX proteins encoded in some sequenced genomes. Somewhere in the middle between the extremes is the RTX domain of CyaA from B. pertussis. For example, this is organized in five successive blocks, containing about eight nonapeptide RTX motifs each, which are separated by linkers of variable lengths (Glaser et al., 1988a;Osicka et al., 2000). The repeat domain of CyaA was then shown to harbor a small number (three to five) of high-affinity (K d o 1 nM) and about 40 low-affinity (K d $0.5-0.8 mM) binding sites for Ca 21 ions (Rose et al., 1995;Rhodes et al., 2001).

TISS
Gram-negative bacteria have evolved several (I-VI) pathways for protein secretion beyond the outer membrane to the extracellular environment (for reviews, see Saier et al., 2008;Fronzes et al., 2009). The RTX proteins contain an $60-residue-long C-terminal secretion signal that is not processed during secretion (Gentschev et al., 1990;Stanley et al., 1991;Sebo & Ladant, 1993). This is recognized by the sec-independent TISS, which mediates the translocation of proteins directly from the cytoplasmic compartment into the extracellular space through a channel spanning the entire cell envelope (Fig. 2).
Type I secretion across the cell wall depends on three specific proteins: a polytopic inner membrane protein with a cytoplasmic ATPase domain operating as an ABC exporter, a membrane fusion protein (MFP) and an outer membrane protein (OMP). The MFP spans out from the inner membrane into the periplasm and contacts both the inner membrane ABC exporter and the OMP. The paradigm of the type I secretion pathway is based on the analysis of the mechanism of secretion of the E. coli a-hemolysin (HlyA). The Hly exporter was also shown to promote to some extent the secretion of a number of heterologous RTX proteins expressed in E. coli, including the CyaA of B. pertussis (Sebo & Ladant, 1993), LtkA of Aggregatibacter (formerly Actinobacillus, Haemophilus) actinomycetemcomitans (Lally et al., 1989), PaxA of Pasteurella aerogenes  or FrpA of Neisseria meningitidis (Thompson & Sparling, 1993).
The hlyCABD operon (Fig. 3a) codes for the toxin activation protein (HlyC), the hemolysin itself (HlyA), the ABC transporter (HlyB) and the MFP protein (HlyD) (Wagner et al., 1983). The outer membrane component (TolC), a multifunctional protein, is encoded outside of the Fig. 2. The schematic depiction of the TISS assembly operation. Upon recognition of a C-terminal secretion signal on a given RTX protein translocation substrate, the inner membrane complex formed by an energy-providing ABC transporter and a MFP contacts the trimeric OMP. A sealed channel-tunnel assembly spanning across the entire Gram-negative bacterial cell envelope is formed, through which the RTX protein is exported in a single step from the bacterial cytoplasm directly to the external bacterial surface, without transiting through the periplasmic space. While concentrations of Ca 21 ions are typically o 100 nM in bacterial cytoplasm, allowing for maintenance of an unfolded RTX domain, millimolar calcium concentrations are typically encountered in host extracellular space colonized by pathogenic bacteria. Loading of RTX repeats of the secreted protein by Ca 21 ions then promotes its folding and acquisition of biological activity.
hly operon on E. coli chromosome ) and is part of the mar-sox regulon (Aono et al., 1998). In some other species, however, the gene for a TolC homologue, such as cyaE of B. pertussis, is comprised in the rtx operon (Glaser et al., 1988b). TolC forms a trimeric export channel in the outer membrane and its presence plays a critical role in type I protein secretion (Delepelaire, 2004;Koronakis et al., 2004).
The structure of TolC was determined by X-ray crystallography (Koronakis et al., 2000). It has been proposed that the trimeric accessory protein HlyD forms a substratespecific complex with the inner membrane protein HlyB, which subsequently recognizes the C-terminal signal peptide of HlyA. Upon binding of HlyA, the HlyD trimer interacts with the trimeric TolC protein of the outer membrane, inducing its conformational change and export of HlyA (Andersen et al., 2001). This complex appears to be transient, with the inner membrane complex of HlyB, HlyD and TolC disengaging and reverting to a resting state once the substrate has been transported (Thanabalu et al., 1998). The energy necessary for the secretion process depends not only on ATP hydrolysis mediated by HlyB but also on the proton motive force on the inner membrane (Koronakis et al., , 1993(Koronakis et al., , 1995. However, the work on secretion of Serratia marcescens hemoprotein HasA and Erwinia chrysanthemi metalloproteases B and C indicated a slightly different order of events (Letoffe et al., 1996), where the ABC transporter and MFP would associate only upon substrate binding.
It is generally assumed that type I secretion involves the translocation of unfolded proteins (Young & Holland, 1999). Secretion of the HasA protein of S. marcescens was, indeed, shown to depend on the binding of the chaperone SecB. Moreover, HasA cannot be transported if allowed to fold in the cytoplasm (Delepelaire & Wandersman, 1998;Wolff et al., 2003). Intriguingly, contact with HlyD was suggested to affect directly or indirectly the folding of HlyA following or during its transit through the translocator (Pimenta et al., 2005).
An atypical TISS has been described for the large MARTX proteins in Vibrio. Here, the apparati consists of four proteins, an analogue of HlyB (RtxB), an analogue of HlyD (RtxD), a TolC-like protein and an additional ATP-binding protein RtxE, where both ABC exporter proteins, RtxB and RtxE, appear to be necessary for MARTX protein secretion (Boardman & Satchell, 2004;Lee et al., 2008).
A gene for a TolC-like OMP is also present in the protease and lipase secretion loci, but is absent from most other operons encoding the RTX toxin family TISS, except for B. pertussis (cya) or Moraxella bovis (mbx) loci (Glaser et al., 1988b;Angelos et al., 2003). A unique genetic organization of TISS for RTX proteins was revealed in N. meningitidis (Wooldridge et al., 2005). In contrast to the cistronic organization of the secretion genes for most other rtx operons, homologues of hlyD and tolC genes are flanked by genes normally associated with mobile genetic elements and do not form an operon with the hlyB gene. Furthermore, the three genes were shown to be expressed independently and mutation at either locus resulted in the inability to secrete the FrpC and FrpC-like (FrpC2) RTX proteins.
Initially, the gene encoding a TolC homologue had not been recognized in the genomes of A. actinomycetemcomitans HK1651 (http://www.genome.ou.edu/act.html) and Mannheimia haemolytica PHL213 (Gioia et al., 2006). Crosby & Kachlany (2007), however, found an ORF in A. actinomycetemcomitans that encodes the TdeA protein of predicted structural properties similar to TolC and showed that inactivation of the tdeA gene resulted in a mutant unable to secrete LtxA.

Classes of RTX proteins RTX cytotoxins
Cytotoxic RTX proteins are produced by a broad range of Gram-negative pathogens (Table 1) including the genera Escherichia, Bordetella, Proteus, Morganella, Moraxella, Vibrio and Kingella, and members of the Pasteurellaceae family (Mannheimia, Pasteurella and Aggregatibacter). RTX cytotoxins can be roughly divided into two families: the conventional and broadly studied pore-forming leukotoxins and the more recently discovered very large MARTX. The best-studied prototype of the MARTX group is VcRtxA (or MARTX Vc ) from V. cholerae, an over 4500-amino-acidresidue-long protein, which causes depolymerization of F-actin stress fibers in a broad range of cell types (Fullner & Mekalanos, 2000).

Pore-forming RTX cytotoxins
Pore-forming RTX cytotoxins represent a unique class of bacterial proteins that share (1) the requirement for posttranslational activation through amide-linked fatty acylation of internal lysine residues; (2) possess a hydrophobic domain that was shown or is presumed to form cationselective pores in target cell membranes; (3) are exported by TISSs; and (4) upon secretion are activated for exerting biological activity by binding calcium ions within the acidic glycine-and aspartate-rich nonapeptide repeats.
Based on the most obvious and/or historically first characterized activity, RTX toxins were divided into hemo-lysins and leukotoxins. The so-called RTX 'hemolysins' were initially found to exhibit a limited target cell and species specificity, while the activity of leukotoxins was considered to be species and cell-type specific (Welch, 1991;Coote, 1992). For example, E. coli a-hemolysin (HlyA) appears to be rather promiscuous, exhibiting a well-detectable cytotoxic activity on a wide spectrum of cells from various species, including erythrocytes, granulocytes, monocytes, endothelial cells or renal epithelial cells from mice, ruminants and primates (Gadeberg & Orskov, 1984;Keane et al., 1987;Bhakdi et al., 1989Mobley et al., 1990;Suttorp et al., 1990;Crosby & Kachlany, 2007). In turn, the leukotoxins of A. actinomycetemcomitans (LtxA) and M. haemolytica (LktA) appear to be quite selective and cytotoxic only to a restricted group of cell types in a species-specific manner (Shewen & Wilkie, 1982;Taichman et al., 1984Taichman et al., , 1987Strathdee & Lo, 1989).
However, this traditional classification as 'hemolysins' (pore-forming cytolysins) and leukotoxins appears to be somewhat obsolete. Even the 'promiscuous' a-hemolysin of E. coli (HlyA) and the 'hemolytic' adenylate cyclase (AC) toxin (CyaA) of Bordetella were now found to preferentially bind and target leukocytes expressing the b 2 -integrins LFA-1 and Mac-1, respectively (Lally et al., 1997;Guermonprez et al., 2001). It appears more appropriate to assume that all pore-forming RTX toxins are primarily leukotoxins, with selectivity for leukocytes being, at least for some of them, eroded at supraphysiological toxin concentrations. The residual activity of the most potent cytolytic (pore-forming) leukotoxins, with a less narrow host spectrum, would then be readily detected as cytolytic activity towards a broader variety of cell types, including erythrocytes. In contrast to these 'hemolysins,' the true 'leukotoxins' would lack any obvious activity on other cell types other than leukocytes from a certain host species.

Post-translational activation of RTX cytotoxins by covalent fatty acylation
The cytolytic (pore-forming) RTX leukotoxins are synthesized as inactive protoxins that undergo post-translational activation before export from the toxin-producing bacteria. This consists in post-translational modification of e-amino groups of internal lysine residues within conserved acylation sites by covalent attachment of amide-linked fatty acyl residues (Fig. 3b). This reaction is catalyzed by the RtxC acyltransferases expressed together with the protoxins (Goebel & Hedgpeth, 1982;Barry et al., 1991;Sebo et al., 1991). The mechanism of this novel type of protein acylation was analyzed in substantial detail for the prototype RTX toxinactivating and acyl-ACP-dependent protein acyltransferase HlyC, which converts the E. coli proHlyA to mature a-hemolysin toxin HlyA (Issartel et al., 1991; , 1998Ludwig et al., 1996). HlyC uses the fatty acyl residues carried by acyl-ACP to form a covalent acyl-HlyC intermediate, which transfers the fatty acyl residues to the e-amino groups of Lys 564 and Lys 690 residues of proHlyA (Worsham et al., 2001(Worsham et al., , 2005. Several residues, including Ser 20 and His 23 , were identified as being potentially involved in the catalysis of acyl transfer by HlyC (Issartel et al., 1991;Trent et al., 1998Trent et al., , 1999a. Acyl-ACPs carrying various fatty acyl residues, including the palmitic (C16:0) and palmitoleic (C16:1) residues most common in E. coli, could be efficiently used in vitro as acyl donors for modification of HlyA, while acyl-CoA is not used as a substrate by HlyC (Issartel et al., 1991;Trent et al., 1998). In vivo, however, HlyC exhibits a high selectivity for C14:0 myristic acid, which was found to constitute about 68% of the acyl chains covalently linked to Lys 564 and Lys 690 of native HlyA (Lim et al., 2000). Surprisingly, the extremely rare odd carbonsaturated C15:0 and C17:0 fatty-acyl residues were found to constitute the rest of the in vivo acylation of HlyA from two different clinical E. coli isolates (Lim et al., 2000). The biological relevance of the use of odd-carbon acyl residues for activation of HlyA as well as the mechanism by which HlyC selects these acyl-ACP loaded by the extremely rare acyl residues remains to be clarified. The role of acylation in toxin activity was analyzed for B. pertussis CyaA, where the extent of fatty acylation in vivo was found to depend on the producing strain. Initially, the Bp-CyaA extracted from a Tohama I-type B. pertussis 338 was found to be monoacylated by a single palmitoylation at the Lys 983 residue only (Hackett et al., 1994). Further work confirmed that the acylation of Lys 983 was necessary and sufficient for the activation of CyaA (Basar et al., 2001;Masin et al., 2005). The CyaA sequence, however, comprises two characteristic acylation sites conserved in the RTX cytolysin family, suggesting that CyaA can also be acylated on a second lysine residue, Lys 860 . The recombinant r-Ec-CyaA toxin produced in the presence of CyaC in E. coli was, indeed, found to bear a second acylation at Lys 860 (Hackett et al., 1995). Moreover, recombinant r-Bp-CyaA protein overproduced by a B. pertussis 18323/pBN strain was also later found to be acylated on both Lys 860 and Lys 983 residues (Havlicek et al., 2001). The reduced specific hemolytic activity of r-Ec-CyaA was then attributed to the modification by mainly the unsaturated palmitoleic (cisD9 C16:1) fattyacyl groups when produced in E. coli, while exclusively saturated C16:0 palmityl residues were found attached to Bp-CyaA in B. pertussis (Havlicek et al., 2001). Furthermore, acylation of each of the Lys 860 or Lys 983 residues alone was necessary and sufficient for conferring CyaA a full capacity to tightly bind its a M b 2 integrin receptor (CD11b/CD18). The mutant CyaA-K983R, acylated only on Lys 860 , still exhibited a fairly high ($20%) cytotoxic activity towards murine monocytic cells expressing CD11b/CD18, when compared with CyaA-K860R mutant monoacylated on the Lys 983 residue alone, or to intact r-Ec-CyaA acylated on both Lys 860 and Lys 983 residues (Masin et al., 2005). Acylation of Lys 983 appears, in turn, to be absolutely essential for the residual cytolytic activity of CyaA on cells lacking CD11b/ CD18 (Basar et al., 1999(Basar et al., , 2001. Pore-forming RTX toxins require fatty acylation for all known cytotoxic activities. However, the exact role of the post-translational modification in the mechanism of action is not truly understood. The nonacylated proHlyA and proCyaA form pores in planar lipid bilayers with a much reduced propensity, but the formed pores have quite similar properties as the pores generated by acylated toxin (Ludwig et al., 1996;Masin et al., 2005). Both nonacylated proHlyA and proCyaA are also quite active in penetrating a naked liposome membrane (Soloaga et al., 1996;Masin et al., 2004), suggesting that the acyl residues are not essential for toxin penetration into the membrane lipid bilayer. Recent evidence indicated that fatty-acylation status and nature may modulate toxin oligomerization and is essential for productive binding of RTX toxins to target cell receptors, allowing the cytotoxic action to occur (Sun et al., 1999;El-Azami-El-Idrissi et al., 2003;Thumbikat et al., 2003;Masin et al., 2005).
Pore-forming activity and interaction with the cell membrane in the absence of a specific cell receptor The highly potent and less specific cytolytic RTX leukotoxins, such as E. coli HlyA or Bordetella CyaA, also exhibit a readily detectable activity on cells other than leukocytes. Their interaction with the target cell membrane devoid of a specific proteinaceous receptor appears to occur in two steps, starting with a reversible adsorption of the toxin that is sensitive to electrostatic forces, which is then followed by an irreversible membrane insertion (Bakas et al., 1996;Ostolaza et al., 1997). Adsorption of RTX toxins is detectable on both toxin-sensitive cells and on certain toxinresistant cells (Iwase et al., 1990). Studies with the isolated calcium-binding domain of HlyA revealed that this part of the protein may adsorb on the membrane in the early stages of HlyA-membrane interaction (Sanchez-Magraner et al., 2007). Recent results with CyaA, HlyA and LtxA showed that these toxins exhibit a weak lectin activity and recognize and bind the N-linked oligosaccharides of their b 2 integrin receptors (Morova et al., 2008). This raises the possibility that the initial unsaturable binding of RTX cytotoxins to various cells might occur through the recognition of glycosylated membrane components, such as glycoproteins and gangliosides.
Whether proteinaceous receptors are involved in binding of RTX toxins to cell types other than leukocytes remains an open question. For example, earlier dose-response binding assays indicated an upper limit of 4000 HlyA binding sites per erythrocyte, implying at least some degree of specificity (Bauer & Welch, 1996b). Eberspächer et al. (1989), however, did not observe any saturability of binding of HlyA to erythrocytes, suggesting that binding was receptor independent. These data are compatible with later observations of Cortajarena et al. (2001Cortajarena et al. ( , 2003, showing that HlyA can use the abundant glycophorin protein as a high-affinity receptor on erythrocytes. HlyA binding and action on erythrocytes was blocked by antibodies binding glycophorin, by a competing peptide comprising residues 914-936 of HlyA or upon glycophorin digestion with trypsin. Once HlyA has inserted into the cell membrane, it appears to undergo an irreversible conformational change (Moayeri & Welch, 1997), after which it cannot be recovered from the membrane without the use of detergents (Bhakdi et al., 1986). However, the mechanism of membrane insertion and pore formation by RTX toxins remains poorly understood. Several studies confirmed that the hydrophobic domain of the N-terminal half of the pore-forming RTX leukotoxins is critical for their ability to form transmembrane pores (Ludwig et al., 1987;Glaser et al., 1988a;Cruz et al., 1990;Osickova et al., 1999;Basler et al., 2007). Other studies also showed that the hydrophobic region of E. coli HlyA was responsible for the insertion of the toxin into the target membrane (Hyland et al., 2001;Schindel et al., 2001). Biophysical studies demonstrated that RTX toxins form cation-selective pores of a defined size and with short lifetimes of only a few seconds (Menestrina et al., 1987(Menestrina et al., , 1996Benz et al., 1989Benz et al., , 1994Szabo et al., 1994;Lear et al., 1995;Maier et al., 1996;Schmidt et al., 1996;Karakelian et al., 1998). At higher toxin concentrations, however, these pores may change subunit stoichiometry over time and aggregate into larger lesions in the cell membrane (Moayeri & Welch, 1994).
Whether pore formation by the RTX leukotoxins depends on the toxin oligomerization step remained a matter of controversy. HlyA has been recovered from target membranes only as a monomer (Menestrina et al., 1987;Eberspächer et al., 1989;Stanley et al., 1993). On the other hand, the dose-response analyses indicated that the lytic activity on target cells was a highly cooperative function of toxin concentration, suggesting that oligomerization was involved in RTX toxin pore formation (Cavalieri & Snyder, 1982;Simpson et al., 1988;Bhakdi et al., 1989;Taichman et al., 1991b;Betsou et al., 1993;Szabo et al., 1994;Bauer & Welch, 1996a;Gray et al., 1998;Osickova et al., 1999). Moreover, in vitro complementation within pairs of individually inactive deletion variants of E. coli HlyA or B. pertussis CyaA allowed to restore, at least in part, the hemolytic and cytotoxic activities, suggesting that two or more toxin molecules had aggregated to form a pore. This substantiated the view that oligomerization was involved in pore formation by HlyA or CyaA (Ludwig et al., 1993;Iwaki et al., 1995;Bejerano et al., 1999). Recent results obtained with CyaA revealed the presence of rather unstable (dynamic stoichiometry) CyaA oligomers in the erythrocyte membrane, revealing a correlation between oligomerization of CyaA mutants in the membrane and their pore-forming capacity (Vojtova-Vodolanova et al., 2009).

Binding through cell-specific receptors
Recently, it has become increasingly clear that even the more 'promiscuous' RTX leukotoxins bind leukocytes through specific proteinaceous receptors of the b 2 integrin family. For example, CyaA of B. pertussis was shown to use the a M b 2 (CD11b/CD18) integrin (known also as a complement receptor 3 or Mac-1) as a target-cell specific receptor (Guermonprez et al., 2001). HlyA of E. coli was also found to specifically bind leukocytes (Welch, 1991) and to interact with the b 2 integrin CD11a/CD18 at low concentrations (Lally et al., 1997). Other studies, however, indicated that the binding of HlyA to cells occurred in a nonsaturable manner and the toxin did not interact with a specific protein receptor on granulocytes (Valeva et al., 2005). These contradictory results await definitive clarification and it is possible that too low concentrations of HlyA were used in the latter study and saturation of the abundant CD11a receptor was not reached. Moreover, A. actinomycetemcomitans and M. haemolytica leukotoxins (LtxA and LktA, respectively) are specific for human and bovine leukocytes, respectively, and were also found to interact with CD11a/CD18 (Lally et al., 1997;Ambagala et al., 1999;Li et al., 1999). The initial interaction of RTX leukotoxins with b 2 integrins then appears to rely on the recognition of N-linked glycans, as revealed for A. actinomycetemcomitans LtxA, E. coli HlyA and B. pertussis CyaA (Morova et al., 2008).

Elevation of calcium concentration in target cells
Elevation and modulation of free cytosolic calcium concentrations are basic strategies of host cell manipulation by pathogens (TranVan Nhieu et al., 2004). Cytosolic calcium levels are tightly controlled and their modulation is part of most prominent cellular signalling pathways that regulate many cellular processes (Berridge et al., 1998(Berridge et al., , 2000. The initial observation that HlyA of E. coli was responsible for significant calcium influx into cells was made by Jorgensen et al. (1983). This was corroborated by demonstrating that pore-forming RTX toxins, such as M. haemolytica LktA, A. actinomycetemcomitans LtxA or B. pertussis CyaA, also caused unregulated calcium influx into target cells Sun et al., 1999;Fiser et al., 2007). It has even been proposed that cell killing by RTX toxins was due to unregulated calcium influx, which would initiate cytoskeletal destruction and cell lysis (Welch, 2001). This mechanism may underlie the action of A. actinomycetemcomitans LtxA that was reported to promote Ca 21 changes in T-cells and to initiate a series of events that involve the activation of calpain, talin cleavage, mobilization of b 2 integrin molecules into membrane lipid rafts and subsequent cell lysis (Fong et al., 2006). Other findings indicate that the calcium influx induced by sublytic doses of RTX toxins accounts for the induction of inflammatory responses (Hsuan et al., 1999;Uhlen et al., 2000;Cudd et al., 2003b).
There persists, nevertheless, a controversy on the mechanisms by which specifically the RTX family of toxins would promote calcium influx into cells. Uhlen et al. (2000) reported that sublytic doses of E. coli HlyA stimulated oscillatory calcium responses in renal epithelial cells through the activation of L-type voltage-gated calcium channels and the subsequent response of IP 3 receptor channels in the endoplasmic reticulum. Moreover, M. haemolytica LktA appears to increase the cytoplasmic Ca 21 concentration in bovine alveolar macrophages and neutrophils both by the influx of extracellular Ca 21 through voltage-gated channels (Ortiz-Carranza & Czuprynski, 1992;Hsuan et al., 1998;Cudd et al., 2003a) as well as by promoting the release of Ca 21 from stores in the endoplasmic reticulum (Cudd et al., 2003a). However, working with different concentrations of HlyA on different cell lines, others have reported that HlyA induces an increase in cytoplasmic [Ca 21 ] c by allowing a passive influx of Ca 21 into cells through toxin pores (Valeva et al., 2005;Koschinski et al., 2006). Moreover, Fiser et al. (2007) reported that B. pertussis CyaA caused an increase of [Ca 21 ] c in monocytic cells by a mechanism that is independent of its pore-forming activity or of Ca 21 release from intracellular stores and depends on membrane translocation of the N-terminal cell-invasive domain polypeptide, but not on its enzymatic activity. The translocating AC domain, as such, appears, indeed, to participate in the formation of a novel and transient Ca 21 influx path in the host cell membrane (Fiser et al., 2007).
These seemingly contradictory results may be reconciled in part by the observed positive feedback effect on initial cell membrane permeabilization due to toxin insertion. As shown for HlyA, ATP leakage and Ca 21 influx, accompanying membrane insertion of HlyA, induces the activation of purinergic receptors and pannexin channels that are permeable for monovalent cations and Ca 21 . This further potentiates the influx of extracellular Ca 21 and contributes to cell lysis (Skals et al., 2009).

Broadly cytolytic RTX leukotoxins (hemolysins)
RTX leukotoxins are typically produced by Gram-negative pathogens and commensals of respiratory, gastrointestinal or reproductive tracts or oral cavities of animals and humans. The characterized broadly cytolytic RTX leukotoxins, thus far classified as 'hemolysins', are listed in Table 1. These toxins exhibit a hemolytic activity in vitro that is revealed by cultivating corresponding bacteria on sheep blood agar plates. In vivo, these toxins induce the production of inflammatory mediators or display cytotoxic and cytolytic effects on host immune cells of different species, thus inducing necrosis, apoptosis and inflammation (Czuprynski & Welch, 1995;Welch, 2001;Frey & Kuhnert, 2002).

HlyA of E. coli
Among the best-characterized RTX 'hemolysins' is HlyA, the single polypeptide (107 kDa) a-hemolysin secreted by uropathogenic as well as many commensal fecal isolates of E. coli (Fig. 3). The N-terminal 200 amino acid hydrophobic domain of HlyA is predicted to contain nine amphipathic a-helices (Hyland et al., 2001), while the C-terminal Ca 21 -binding domain contains 11-17 of the glycine-and aspartate-rich nonapeptide b-strand repeats (depending on the stringency of the criterium for consensus motif conservation). It is assumed that membrane interaction of HlyA occurs mainly through the amphipathic a-helical domain. However, it has been proposed recently that both major domains of HlyA are directly involved in the membrane interaction of HlyA, the Ca 21binding domain being responsible for the early stages of HlyA docking to the target membrane (Sanchez-Magraner et al., 2007). Similarly, Masin et al. (2004) showed that both the pore-forming and the acylation domain of CyaA are required for membrane interaction. Cortajarena et al. (2001) observed that a short sequence from the C-terminal domain (amino acid 914-936) was the main HlyA segment binding the a-glycophorin on erythrocytes.
A homologous EhxA protein is produced by the enterohemorrhagic E. coli O157:H7 from a gene located on a 90 kbp plasmid (pO157). EhxA exhibits 61% identity to HlyA, but displays a more narrow target cell specificity and binds erythrocytes less efficiently (Bauer & Welch, 1996b;Stanley et al., 1998), exhibiting virtually no activity against human leukocytes (Bauer & Welch, 1996b).
The MbxA secreted by M. bovis is implicated in the pathogenesis of infectious bovine keratoconjunctivitis (Brown et al., 1998), where it promotes corneal epithelial cell damage (Rogers et al., 1987;Beard & Moore, 1994). The mbxCABD genes form a classical rtx operon that is absent from nonhemolytic strains (Angelos et al., 2003) and is part of a mobile genetic element designated as a pathogenicity island of M. bovis (Hess & Angelos, 2006).
Studies involving apx deletion mutants and trans-complementation experiments show that ApxIIA is essential in the pathogenesis of porcine pleuropneumonia and that the combination of ApxIA and ApxIIA, secreted by certain serotypes of Actinobacillus pleuropneumoniae, accounts for the severe course of the disease with a fatal outcome (Reimer et al., 1995). The ApxIIIA protein (Rycroft et al., 1991;Frey et al., 1993a, b) as well as PaxA of P. aerogenes  are nonhemolytic on erythrocytes, but show a significant CAMP reaction, a cohemolytic reaction dependent on the sphingomyelinase from b-hemolytic Staphylococcus aureus (Frey et al., 1994;Kuhnert et al., 2000). ApxIIIA was shown to be highly cytotoxic against macrophages and was previously called the macrophage toxin (Lally et al., 1989;MacDonald & Rycroft, 1992). The cytotoxic activity of PaxA, also produced by Pasteurella mairi, has not yet been analyzed in detail (Frey & Kuhnert, 2002), but all P. aerogenes and P. mairi isolated from aborted feta or neonatal septicemia of pigs produced PaxA, while strains devoid of the paxA gene are isolated as opportunistic pathogens or commensals . In general, there are little molecular and functional data available on the role in virulence and host or target cell specificity of most of the RTX toxins from animal pathogens.
LktA plays a role in the pathogenesis of bovine and ovine pneumonic pasteurellosis (Jeyaseelan et al., 2002), whereas A. actinomycetemcomitans LtxA is the main virulence factor of localized aggressive periodontitis in humans (Henderson et al., 2003;Fine et al., 2006;Haubek et al., 2006). These toxins can inhibit the mobility, chemotaxis and respiratory burst of neutrophils, release inflammatory mediators from granulocytes or macrophages, prevent phagocytosis by polymorphonuclear cells and disrupt the phagosome, thus allowing bacterial invasion of the phagocyte. Altogether, these effects strongly reduce the immune response of the host. The A. actinomycetemcomitans LtxA was further reported to interact with Cu, Zn superoxide dismutase, which may protect both the bacteria and the LtxA from reactive oxygen species produced by host inflammatory cells (Balashova et al., 2007).
Cytolytic RTX leukotoxins ( Table 1) are proteins of a typical molecular mass of 100-120 kDa that possess similar structural and functional domains and are encoded within similar rtx operons as the RTX 'hemolysins' (Fig. 3) (Chang et al., 1987;Lo et al., 1987;Lally et al., 1989). LtxA shares 40-50% amino acid homology with the E. coli HlyA and M. haemolytica LktA. Similarly, both the nonacylated proLktA and proLtxA were in turn able to bind target cells without increasing the intracellular calcium concentration or inducing cytolysis (Sun et al., 1999;Highlander et al., 2000;Thumbikat et al., 2003;Balashova et al., 2009). The proLktA can be acylated to mature LktA by E. coli HlyC, Bordetella CyaC and Actinobacillus LtxC (Lally et al., 1994;Westrop et al., 1997). Substitution of Lys-554 to Thr or Cys reduced the lytic activity of LktA against bovine lymphocytes by only 40%, indicating the presence of a second functionally redundant acylation site (Pellett & Welch, 1996). Recently, two internal lysine residues of A. actinomycetemcomitans LtxA (Lys-562 and Lys-687) have been identified as targets for covalent fatty acid modification by the ltxC gene product (Balashova et al., 2009). Interestingly, LtxA is the only RTX toxin that has a basic pI of 8.9. Also, the RTX domains of both LktA and LtxA are shorter than that of HlyA and contain only seven and eight calcium-binding nonameric repeats, respectively Kuhnert et al., 1997). Their hydrophobic domain in the N-terminal part was proposed as the region that spans the target cell membrane and deletions in this region impair the cytolytic and pore-forming activities (Cruz et al., 1990). While the region of LtxA required for the recognition of human target cells appears to span the glycine-rich repeats (Lally et al., 1994), the segment determining specificity for ruminant cells has been associated with the N-terminal portion of LktA (Forestier & Welch, 1991;Thumbikat et al., 2003).
A difference in the release of LtxA into the culture medium and retention on the cell surface was observed between nonadherent (smooth) and adherent (rough) strains of A. actinomycetemcomitans Lally et al., 1991a;Ohta et al., 1991;Berthold et al., 1992;Kachlany et al., 2000) and LtxA release into the medium could be related to mutations in the tad gene involved in the tight nonspecific adherence of A. actinomycetemcomitans (Kachlany et al., 2001). Recently, A. actinomycetemcomitans LtxA was also found to be released within outer-membrane vesicles (Kato et al., 2002;Diaz et al., 2006). These are likely to deliver LtxA to host cells (Demuth et al., 2003), similar to the delivery of E. coli heat-labile enterotoxin into mammalian cells (Kesty et al., 2004).
The quite narrow host range of leukotoxins appears to be due to cell-specific binding through the b 2 integrins. The requirement for b 2 integrins in cytotoxic action was first observed for M. haemolytica LktA, which was not cytotoxic to neutrophils isolated from animals suffering from a bovine leukocyte adhesion deficiency. This deficiency is due to a single point mutation (D128G) in the highly conserved extracellular region of the CD18 subunit, which results in reduced membrane expression of the b 2 integrin LFA-1 (Shuster et al., 1992;Nagahata, 2004). In fact, the b 2 integrins share a common b 2 subunit, CD18, which is combined with either one of the unique a chains, aL (CD11a), aM (CD11b), aX (CD11c) or aD (CD11d). So far, two of these integrins have been found to bind RTX leukotoxins: a L b 2 integrin, known as lymphocyte functionassociated antigen 1 (LFA-1, CD11a/CD18), was shown to bind HlyA, LktA and LtxA, and the a M b 2 integrin binds Bordetella CyaA, respectively (Lally et al., 1997). Transfection of an LFA-1-expressing construct rendered an otherwise resistant K562 cell line sensitive to killing by LtxA (Lally et al., 1997). Recently, Kieba et al. (2007) explained the species selectivity of LtxA by showing that it recognizes the N-terminal 128 amino acid residues of human CD11a that are missing in the murine LFA-1 counterpart. However, the cysteine-rich tandem repeats of the human CD18 subunit were also reported to interact with LtxA (Dileepan et al., 2007a).
The CD18 subunit of bovine b 2 integrins was, in turn, identified as the receptor for M. haemolytica LktA (Ambagala et al., 1999;Li et al., 1999;Dassanayake et al., 2007). Expression of either bovine LFA-1 or chimeric LFA-1 (bovine CD18/murine CD11a) rendered the LktA-resistant cells susceptible to LktA (Deshpande et al., 2002;Dileepan et al., 2005), with the magnitude of LktA-induced cytotoxicity correlating with the levels of LFA-1 expression on the target cell surface (Jeyaseelan et al., 2000;Leite et al., 2000). It remained, however, controversial as to which part of the CD18 subunit LktA binds, as Gopinath et al. (2005) mapped it within amino acids 1-291, while Dileepan et al. (2005Dileepan et al. ( , 2007b found it between residues 541 and 581 of the extracellular region of bovine CD18. Nevertheless, Shanthalingam & Srikumaran (2009) have recently shown that the LktA-binding site is formed by amino acids 5-17 of CD18, which, surprisingly, comprise most of the amino acids of the signal peptide that remains intact on mature CD18 molecules on the cell surface.
Leukotoxin activities against target cells are highly dose dependent. At sublytic concentrations, leukotoxins induce an increase in [Ca 21 ] c in cells and activate neutrophils and mononuclear cells to undergo oxidative burst and degranulation (Czuprynski et al., 1991;Maheswaran et al., 1992;Stevens & Czuprynski, 1995;Balashova et al., 2009). This is accompanied by the production of several lipid mediators and proinflammatory cytokines from human macrophages (Johansson et al., 2000;Kelk et al., 2005Kelk et al., , 2008. Pulmonary mast cells were shown to release histamine in response to LktA (Adusu et al., 1994) and LktA has also been shown to inhibit bovine lymphocyte blastogenesis induced by concanavalin A and pokeweed mitogen (Majury & Shewen, 1991). Mannheimia haemolytica LktA also affects the adhesion of platelets, which gives rise to fibrin deposits in lung alveoli of cattle with pasteurellosis (Cheryk et al., 1998;Nyarko et al., 1998). As the concentration of leukotoxins increases, target cells are stimulated to undergo apoptosis, involving LFA-1 signalling through protein kinase C and adverse effects on the mitochondria (Kato et al., 1995;Korostoff et al., 2000;Kelk et al., 2003;Atapattu & Czuprynski, 2005). At higher leukotoxin concentrations, the apoptotic mechanisms are exceeded and necrosis occurs due to pore formation. LktA-mediated cell permeabilization was shown to cause a rapid leakage of the intracellular potassium and cell swelling (Clinkenbeard et al., 1989a;Clinkenbeard & Upton, 1991). In vitro studies showed that LtxA formed voltage-gated ion channels of large conductance in the planar lipid bilayer, with an approximate functional diameter of 0.96 nm (Iwase et al., 1990;Lear et al., 1995). A pore size of about 1.2 nm was, in turn, deduced for LktA from osmotic protection experiments with raffinose (Clinkenbeard et al., 1989b). In addition, metabolites from phospholipase C activation (arachidonic acid) appear to contribute to LktA-induced cytolysis significantly (Jeyaseelan et al., 2001).
CyaA primarily targets and paralyzes with high efficacy the leukocytes expressing the a M b 2 integrin (CD11b/CD18). With reduced efficiency, however, the toxin can also penetrate and deliver the AC enzyme into a variety of cells lacking the CD11b/CD18 (Hanski, 1989;Bellalou et al., 1990). Unlike most other enzymatically active toxins, which penetrate into cell cytosol from endosomes, several reports showed that the AC is delivered directly across the cytoplasmic membrane of cells without the need for receptormediated endocytosis of the toxin (Gentile et al., 1988;Gordon et al., 1988;Guermonprez et al., 1999;Schlecht et al., 2004;Basler et al., 2006). In target cytosol, the N-terminal AC domain of CyaA binds intracellular calmodulin, whereupon its specific enzymatic activity is increased by $10 000-fold and catalyzes uncontrolled conversion of cellular ATP to cAMP, a key second messenger signalling molecule (Wolff et al., 1980;Confer & Eaton, 1982). This subverts the signalling of protein kinase A and essentially instantaneously ablates the bactericidal functions of phagocytes, such as oxidative burst and phagocytosis, and induces the secretion of immunomodulatory cytokines (Vojtova et al., 2006). Recently, the crystal structure of the AC domain of CyaA in the complex with the C-terminal fragment of calmodulin was resolved by Guo et al. (2005).
The main segment of CyaA required for binding to the a M b 2 integrin was located in the glycine-rich repeat region between residues 1166 and 1281 of CyaA (El-Azami- El-Idrissi et al., 2003). However, cooperation and structural integrity of all domains of the 'hemolysin' moiety of CyaA appear to be critical for membrane insertion and translocation of the N-terminal AC enzyme domain into cell cytosol (Bellalou et al., 1990;Iwaki et al., 1995). Translocation itself, but not the mere insertion of CyaA into the cytoplasmic membrane of cells, is driven by a negative membrane potential (Otero et al., 1995). Fiser and colleagues showed that the translocating AC polypeptide inserted into the cell membrane participates in the formation of a novel type of membrane path for Ca 21 influx into monocytic cells. The latest results from our laboratory show that calcium influx induces talin cleavage by calpain and enables mobilization of CyaA with CD11b/CD18 into lipid rafts, where the choles-terol-enriched lipid environment supports the translocation of the AC domain across membrane (Bumba et al., 2010). Moreover, translocation of the AC domain and oligomerization into cation-selective pores appear to represent two independent and parallel/competing activities of the membrane-inserted form of CyaA. Either activity can be upmodulated at the expense of the other by specific substitutions of key glutamate residues forming pairs in the predicted transmembrane segments between residues 500 and 700 of CyaA. These were found to play a critical role in cell binding, formation of cation-selective pores and translocation of the AC domain into the cells (Osickova et al., 1999(Osickova et al., , 2010Basler et al., 2007;Fiser et al., 2007;Vojtova-Vodolanova et al., 2009). The 'hemolysin' pores formed by CyaA have a diameter of only about 0.6-0.8 nm and the specific 'hemolytic' activity of CyaA is relatively low, compared with HlyA for example (Bellalou et al., 1990). It appears, nevertheless, to synergize with the invasive AC enzyme activity in maximizing the overall cytotoxic potency of the toxin on CD11b 1 cells (Basler et al., 2006;Hewlett et al., 2006).
The role of CyaA in the interaction of Bordetella with cells of the respiratory epithelia and in the modulation of the host immune response through the induction of proinflammatory cytokine secretion remains poorly explored. Recent work has suggested that CyaA activity may account for the induction of IL-6 in tracheal epithelia colonized by B. pertussis (Bassinet et al., 2004). CyaA contributes to numerous pathological effects in the murine model of lung infection, such as efficient pulmonary colonization, induction of histopathological lesions in lungs, recruitment of inflammatory leukocytes and induction of lethality (Weiss et al., 1984;Gueirard et al., 1998). Low concentrations (1-5 ng mL À1 ) of CyaA have recently been shown to effectively inhibit complement-mediated opsonophagocytosis, which is a crucial defense mechanism of naive unimmunized hosts (Kamanova et al., 2008). In fact, CyaAmediated intoxication by cAMP was also found to inhibit the phagocytosis of B. pertussis cells via Fc receptors by neutrophils (Weingart & Weiss, 2000). CyaA activity further causes a loss of chemotactic and oxidative burst capacities required for the bactericidal activity of leukocytes (Friedman et al., 1987). Moreover, the toxin can induce macrophage apoptosis (Khelef et al., 1993;Khelef & Guiso, 1995) by a mechanism involving disruption of the membrane potential of mitochondria (Bachelet et al., 2002). In many Bordetella isolates, the CyaA protein remains attached to the bacterial surface following secretion, due to interaction with the filamentous hemagglutinin (Zaretzky et al., 2002). Most of the cell-attached CyaA appears, however, to be aggregated and unable to act as a 'contact weapon', because only the newly secreted CyaA was found to be capable of penetrating target cells and increasing the intracellular cAMP levels (Gray et al., 2004).

Use of CyaA in research and vaccine applications
Because Gram-negative bacteria generally do not express calmodulin homologues and the AC domain exhibits only residual enzyme activity in the absence of calmodulin, the AC domain of CyaA could elegantly be used as a reporter enzyme for tracing protein translocation into the eukaryotic cell cytosol (Sory & Cornelis, 1994). The AC (Cya) reporter protein is fused to an effector protein secreted through the type III secretion pathway (TTSS), such as the Yersinia Yop proteins. While the fusion exhibits only very low AC enzyme activity in the bacterial cell and/or the culture supernatants, once it is injected through the TTSS pathway into a eukaryotic host cell, the AC enzyme is activated 4 1000-fold by host calmodulin and catalyzes the rapid conversion of ATP to cAMP. Use of the AC (Cya) fusion reporter has now become a standard technique for the demonstration of contact-dependent direct translocation of TTSS effector proteins into animal and plant host cells by a number of Gram-negative species.
Another original application makes use of the fact that the residual activity of the AC domain in the absence of calmodulin requires the physical interaction of the T25 and T18 fragments of the AC domain. This could be well exploited to develop a bacterial two-hybrid system for the detection of protein-protein interactions (Karimova et al., 1998). When the T25 and T18 fragments are individually fused to peptides or proteins that are able to bind each other, the interaction of the chimeric polypeptides brings together the T25 and T18 fragments, resulting in the restoration of a residual capacity of the enzyme to convert intracellular ATP to cAMP, even in the absence of eukaryotic calmodulin. This, in turn, can be monitored in a rather sensitive manner using E. coli hosts lacking the endogenous AC activity as indicator strains for the two-hybrid screening, because even very low levels of cAMP produced by the reconstituted AC enzyme will allow transcription of the genes involved in lactose and maltose catabolism in E. coli.
In another application, the genetically detoxified CyaA, unable to increase cAMP levels, could be exploited for highly efficient in vivo delivery of foreign T-cell epitopes into the major histocompatibility complex class I-and II-dependent antigen presentation pathways of CD11b 1 dendritic cells (Osicka et al., 2000;Loucka et al., 2002;Schlecht et al., 2004). This allowed the use of CyaA for antigen delivery and induction of strong Th1-polarized and epitope-specific CD8 1 cytotoxic T-cell responses, effective in prophylactic vaccination against viruses and in immunotherapy of certain tumors Fayolle et al., 1996Fayolle et al., , 1999Fayolle et al., , 2001Saron et al., 1997;Dadaglio et al., 2000;Loucka et al., 2002;Preville et al., 2005;Mackova et al., 2006). Phase I/II human clinical trials aimed at exploring this exciting application of CyaA for cervical cancer and melanoma immunotherapy are currently in preparation.

MARTX
A quite different and recently discovered division of the RTX-toxin family is a group of very large toxins that differ from all previously known RTX proteins by the molecular structure and rtx gene cluster organization (Table 1, Fig. 3). These MARTX have thus far been identified in several different Vibrio species (Vibrio sp. RC385, V. cholerae, Vibrio splendidus, Vibrio anguillarum, Vibrio vulnificus) and are also present in Aeromonas hydrophila, Yersinia enterocolitica and Yersinia kristensenii, Proteus mirabilis, Photorhabdus luminescens and Photorhabdus asymbiotica (Supporting Information, Tables S1 and S2) (Satchell, 2007;Li et al., 2008). The best-studied prototype of MARTX is the V. cholerae RtxA (VcRtxA), renamed to MARTX Vc (Satchell, 2007).
All rtxA-like MARTX genes encode proteins that range from 3212 to 5206 amino acid residues in length. In contrast to other toxins, the C-terminal repeats of MARTX proteins exhibit an 18-residue-long consensus sequence X(V/I)XXGXXNX(V/I)XXGDGXDX. These share a common G-7X-GXXN central motif, instead of the typical nonapeptide repeat (Lin et al., 1999;Satchell, 2007). Moreover, MARTX proteins possess additional N-terminal repeats, which fall in two classes. The first has a 20-residue consensus sequence GXXG(N/D)(L/I)(T/S)FXGAG(A/G) XNX(L/I)X(RH) and the second has a 19-residue consensus T(K/H)VGDGX(S/T)VAVMXGXAN(I/V)X. Altogether, the glycine-rich repeats represent about 25% of the sequence of MARTX Vc and were proposed to bind the eukaryotic cell surface and to facilitate the translocation of a central $1700 amino acid portion of the MARTX Vc to the target cell cytosol. These central regions of MARTX are composed of differing activity domains that, upon entry into the eukaryotic cell, may exert different cytotoxic activities (Boardman & Satchell, 2004;Satchell, 2007). However, except for the central domains of MARTX Vc and part of V. vulnificus RtxA (MARTX Vv ), the biological activity and function of these large RTX proteins remain unknown. Little is known about the post-translational modifications of MARTX proteins. An rtxC gene similar to the hlyC acyltransferase gene required for maturation of E. coli a-hemolysin is located in the same operon with the rtxA gene (Lin et al., 1999). The recent data indicate that rtxC is not necessary for the MARTX toxin function in V. cholerae, but the MARTX toxin without rtxC activation showed reduced actin cross-linking activity (Cheong et al., 2010), while a deletion of rtxC had no effect on the virulence of V. vulnificus Liu et al., 2007). These results suggest that acylation may not be essential for all MARTX toxins (Satchell, 2007).
Analysis of the available genomic structures revealed that the martx loci typically consist of two divergent operons (rtxHCA and rtxBDE). The toxin gene is found as the third gene downstream of the rtxC homologue (a putative acyl transferase) and a conserved hypothetical gene (rtxH) of unknown function, which is found only in the martx gene clusters. The divergent operon contains three more genes that encode homologues of the ATP-binding transporter protein RtxB, the MFP RtxD and a second ATPase, RtxE (Boardman & Satchell, 2004). The latter also appears to be involved in MARTX secretion by TISS, as rtxE gene disruption in V. vulnificus and V. anguillarum blocked the secretion of MARTX Vv (MARTX Va , respectively) and induced a significant reduction in bacterial cytotoxic activity against epithelial cells in vitro (Lee et al., 2008;Li et al., 2008). The necessary homologues of the OMP tolC are then found outside the rtx loci. This atypical four-component TISS seems to be a conserved feature across the entire MARTX family (Boardman & Satchell, 2004).
The production of MARTX TISS components is regulated by the growth phase. The repressor regulating rtxBDE expression is encoded outside the rtx locus and is not directly linked to quorum sensing, while V. cholerae may apparently couple the regulation of the rtx locus to the detection of stress (Boardman et al., 2007). The rtxH, rtxC and rtxA genes are coordinately expressed on a single mRNA (Boardman et al., 2007).
The prototype rtxA gene was found in both clinical and environmental isolates of V. cholerae, but not in the O1 classical biotypes (Lin et al., 1999;Chow et al., 2001). The deduced MARTX Vc protein is 4545 residues long, with a predicted molecular mass of 4 485 kDa (Lin et al., 1999). In contrast to pore-forming RTX leukotoxins, the MARTX Vc does not appear to disrupt membrane integrity or cause cell death. Rather, MARTX Vc activity contributes to the severity of acute inflammatory responses in the pathology of cholera by inducing alteration of permeability of the paracellular tight junctions. This results from the capacity of MARTX Vc to induce cell rounding and depolymerization of the actin cytoskeleton in a broad range of cell types, and yet the cells remain viable (Cordero et al., 2006). Concurrent with actin stress fiber disassembly, actin monomers are covalently cross-linked into dimers, trimers and higher multimers by the actin cross-linking domain (ACD), which utilizes Gactin as a substrate and hydrolyzes one molecule of ATP per cross-linking event (Kudryashov et al., 2008a). The ACD of MARTX Vc , located between residues 1963 and 2375, catalyzes a unique reaction consisting of the formation of an intermolecular iso-peptide bond between the g-carboxyl group of glutamic acid residue 270 and the e-amino group of lysine residue 50 of actin (Kudryashov et al., 2008b). Contrary to expectation, however, deletion of the ACD did not ablate the cell-rounding activity of MARTX Vc , revealing that the large toxin carried a second cell-rounding activity (Lin et al., 1999;Fullner & Mekalanos, 2000;Fullner et al., 2001Fullner et al., , 2002Sheahan et al., 2004). This targets the regulation of the small Rho GTPases, Rho, Rac and Cdc42, rather than the Rho GTPase proteins directly, and a 548-residue-long Rho inactivation domain (RID) of MARTX Vc was recently found to inactivate the Rho GTPases by a mechanism distinct from other Rho-modifying bacterial toxins (Sheahan & Satchell, 2007).
The MARTX Vc toxin was found to insert into the host cell plasma membrane and is supposed to directly translocate the ACD to the cytosol of cells in a way involving insertion of the N-and C-terminal repeat regions into the cell membrane. Inside the cell, the ACD is released into cell cytosol (Sheahan et al., 2004) through self-processing catalyzed by a conserved cysteine protease domain (CPD), which cleaves MARTX Vc between residues L 3428 and A 3429 . Three additional CPD-dependent processing sites were identified, all at leucine residues delimiting the ACD, RID and a/b junction domains (Prochazkova et al., 2009;Shen et al., 2009). Autoprocessing of MARTX Vc thereupon results in the release of individual activity domains into target cytosol. Autoprocessing activity of the CPD is induced by binding inositol hexakisphosphate (InsP 6 ), which is abundant at the inner surface of the cell membrane (Prochazkova & Satchell, 2008), and the CPD activation mechanism was recently characterized in molecular detail (Prochazkova et al., 2009). Because InsP 6 is exclusive to eukaryotes and is present at cytosolic concentrations 4 10 mM, the evolution of a proteolytic biosensor responding to InsP 6 appears to be an ingenious strategy for assuring that the functional activation of a secreted toxin occurs only once it has reached the host cell cytosol (Lupardus et al., 2008).
The sequence of the V. vulnificus RTX toxin (VvRtxA, MARTX Vv ) was reported in 2003 (Chen et al., 2003), showing that it consists of 5206 residues and has a predicted molecular mass of 556 kDa . The deduced primary amino acid sequence of MARTX Vv is $80-90% identical throughout most regions to that of MARTX Vc . However, no ACD is present in MARTX Vv , which also does not cause actin cross-linking (Sheahan et al., 2004) and posesses only the Rho-inactivating activity associated with an RID (Sheahan & Satchell, 2007). This is followed by a CPD located towards the C-end of MARTX Vv (Prochazkova & Satchell, 2008).
Unlike MARTX Vc , the MARTX Vv may be able to disrupt membranes by the predicted segments homologous to those of pore-forming RTX toxins. This would go well with the difference in the virulence of the pathogens, where V. vulnificus is exceedingly more destructive and cytolytic, as compared to V. cholerae . Indeed, MARTX Vv appears to be crucial for V. vulnificus virulence and cytotoxicity Lee et al., 2007;Liu et al., 2007).
The RtxA of V. anguillarum appears to play a major role in the virulence of the agent causing vibriosis in fish. The rtx operon encodes a second hemolysin gene cluster in V. anguillarum M93Sm, which also has a hemolysin gene, vah1. While Vah1 causes cell vacuolation, the MARTX Va causes cell rounding. Analysis of the MARTX Va sequence reveals that the protein does not contain an ACD, but a homologue of the RID was identified. The contribution of the RID to cell rounding needs to be investigated further. Single mutations in vah1 or rtxA attenuate the cytotoxicity of V. anguillarum M93Sm and a vah11rtxA double mutant is no longer cytotoxic (Li et al., 2008).
The genes for MARTX homologues of P. luminescens are clustered in two chromosomal regions and are tandemly organized. There are four loci containing intact rtxA genes, while four are disrupted by frameshifts or insertion sequences. The organization of genes encoding the RTX secretion system is identical to that of V. cholerae (Duchaud et al., 2003). The function of these genes in pathogenesis was not analyzed, but the actin cross-linking domain was not found in MARTX sequences from Photorhabdus, which appear to have evolved unique cellular activities through the acquisition of a different genetic material (Sheahan et al., 2004).

Cadherin domain proteins
A particularly novel class of RTX proteins of Vibrio is represented by RtxL1 and RtxL2, which are characterized by the presence of more than one cadherin domain and were first identified in V. cholerae N16961 (Chatterjee et al., 2008). The rtxL1 and rtxL2 genes are arranged in tandem, which is different from the arrangement of the V. cholerae RTX gene cluster (Lin et al., 1999). RtxL1 and RtxL2 belong to the RTX family of hemolysin/leukotoxins, exhibiting hemolytic activity on human erythrocytes, but also appear to play a role in adherence and biofilm formation by V. cholerae N16961.
Both rtxL1 and rtxL2 genes are expressed in all V. cholerae isolates belonging to O1 (strains N16961, O395 and 569B), O139 (strain SG24) and non-O1-nonO139 (strains VCE232 and VCE309) serovars under in vitro conditions and appear to play a role in virulence in a mouse model (Chatterjee et al., 2008).
Another member of the RTX-toxin group with cadherin domains was found as a novel RTX-like hemagglutinin (FrhA) of the V. cholerae O1 classical strain O395, the O1 El Tor strain A1552 and the O1 El Tor strain P27459 (Syed et al., 2009). FrhA expression is positively regulated by the flagellar regulatory hierarchy. It mediates adherence to chitin and epithelial cells, enhances biofilm formation and is involved in intestinal colonization in infant mice.

RTX proteases
The RTX proteases belong to microbial zinc metalloproteases (Hooper, 1994;Miyoshi & Shinoda, 2000) and form a group of approximately 50 kDa proteolytic enzymes secreted by a variety of pathogens. These proteases consist of an N-terminal proteolytic domain and a C-terminal calciumbinding RTX domain and are synthesized as zymogens that are activated by processing upon secretion.
The RTX proteases belong to the subgroup of metzincin metalloendopeptidases  that contains an extended zinc-binding motif HEXXHXXGXXH, of which the three histidine residues are involved in binding the catalytically essential zinc ion, while the glutamic acid residue was postulated to take part in the catalytic activity. In addition, the metzincins all share a conserved methionine, which is located on a turn near the catalytic site, some 40-60 residues towards the C-terminus (the Met-turn).

Genetic organization of RTX protease loci
The RTX protease locus of P. aeruginosa alkaline protease AprA was first identified by Duong in 1992(Duong et al., 1992. It consists of five ORFs, with aprD, aprE and aprF encoding the TISS proteins, the structural aprA gene for the protease and aprI encoding a protease inhibitor (inh).
The genetic organization of RTX protease operons varies. In Pseudomonas fluorescens CY091 (Liao & McCallus, 1998), the protease gene is located upstream of the inh and aprDEF transporter genes. The same organization of the locus was shown in P. luminescens (Bowen et al., 2003), starting with the structural prtA gene for the protease, and followed by a gene encoding a putative protease inhibitor, inh and then the three members of an associated TISS, prtB, prtC and prtD (Fig. 4). The genetic organization of RTX protease operons appears to depend on the number of secreted proteases. For example, E. chrysanthemi, a phytopathogenic enterobacterium, secretes four RTX proteases, where the first gene in the operon is the structural gene for prtG (Ghigo & Wandersman, 1992). This is followed by a gene encoding a putative protease inhibitor, inh, and then the three genes encoding TISS components, prtD, prtE and prtF. Structural genes for the other three RTX proteases prtB, prtC and prtA are adjacent to and belong to independent transcription units Dahler et al., 1990;Ghigo & Wandersman, 1992).
In some bacterial species, the RTX protease operon is combined with the genes for RTX lipases or other proteins. For example, in P. fluorescens B52, an aprX-lipA operon contains the protease (aprX) and the lipase (lipA) genes encoded at opposite ends of a contiguous set of genes (Woods et al., 2001). This also includes the protease inhibitor, TISS genes and two autotransporter genes (aprXinh-aprDEF-prtAB-lipA), constituting an operon (Ahn et al., 1999).

Structure-function relationships of RTX proteases
The typical size of RTX proteases is around 480 amino acid residues and 50 kDa, with the theoretical isoelectric point of most RTX proteases being about 4.0-4.6.  first showed that the extracellular proteases B and C of E. chrysanthemi were synthesized as inactive zymogens proB and proC that were activated by autoprocessing of 16 or 18 amino acid residues ($2 kDa) from the N-termini in the external medium containing divalent cations. As other zinc metalloproteases, the RTX proteases are inhibited by the general metalloprotease inhibitor EDTA, as well as by o-phenanthroline, a specific zinc metalloprotease inhibitor (Zhang et al., 1999).
Three-dimensional structures of the alkaline proteases of P. aeruginosa (AprA) and PrtSM of S. marcescens were the first RTX protein structures resolved (Baumann et al., 1993;Baumann, 1994). Both proteases have a very similar twodomain structure. The N-terminal part is the proteolytic domain with a folding topology very similar to astacin, the archetypical metzincin protease (Bode et al., 1996). The C-terminal RTX domain consists of a 21-strand b sandwich. Within this domain, a 'parallel b roll' structure was first described, in which successive b strands are wound in a righthanded spiral and Ca 21 ions are bound within the turns between strands by a repeated GGXGXD sequence, thus setting a paradigm for the RTX portions of the whole protein family.

RTX protease inhibitors
The RTX protease inhibitor from E. chrysanthemi (Inh) was shown to be synthesized as a 12 kDa polypeptide having a signal peptide of 19 residues, which is cleaved off during Secdependent secretion into the periplasm (Letoffe et al., 1989). The mature 10 kDa inhibitor is entirely located in the periplasm of E. chrysanthemi and its presumed physiological role consists in protecting periplasmic proteins against proteases that might potentially leak out from the TISS channel assembly. Inh forms a 1 : 1 complex with proteases A, B and C from E. chrysanthemi, AprA from P. aeruginosa and PrtSM from S. marcescens (Letoffe et al., 1989). In the crystal structure of the complex of S. marcescens PrtSM with Inh from E. chrysanthemi (Baumann et al., 1995), the Inh was found to fold into a compact eight-stranded antiparallel b-barrel structure, interacting with the protease via five N-terminal residues that insert into the active site cleft of PrtSM.
Similar inhibitors have been characterized for other RTX proteases, such as P. aeruginosa (Feltzer et al., 2000), S. marcescens (Kim et al., 1995) and Photorhabdus (Valens et al., 2002;Bowen et al., 2003). The SmaPI inhibitor of S. marcescens, however, shows a very high protease specificity, while the protease inhibitor of P. aeruginosa (APRin) exhibits a significantly higher inhibitory activity (K D of 4 pM) compared with the inhibitors of E. chrysanthemi and S. marcescens (K D values from 1 to 10 mM).

Biological activity of RTX proteases
Pseudomonas aeruginosa AprA was shown to hydrolyze fibrin and fibrinogen, with specific activities similar to plasmin (Shibuya et al., 1991). AprA also exhibits an anticoagulant activity in human plasma, which was attributed to its direct fibrinogenolytic function. This may account, at least in part, for the most characteristic pathologic feature of the P. aeruginosa septicemia, which consists in hemorrhagic lesions lacking thrombi (Fetzer et al., 1967). Purified P. aeruginosa AprA also readily cleaves soluble laminin (Heck et al., 1986), which suggests its direct role (together with elastase) in both tissue invasion and hemorrhagic tissue necrosis in P. aeruginosa infections. Further, AprA also degrades human g-interferon and inhibits its biological activity (Horvat & Parmely, 1988). Moreover, the effects of P. aeruginosa AprA on serum complement and on the isolated components C1q and C3 were also described (Hong & Ghebrehiwet, 1992). Because both C1q and physiological fragments of C3 (C3b, iC3b, and C3dg) are important opsonins, degradation of these molecules by Pseudomonas enzymes may facilitate the survival and proliferation of the organism in plasma.
The RTX protease from P. aeruginosa (Kreger & Gray, 1978) also appears to participate in liquefactive necrosis of the cornea through the digestion of the proteoglycan extracellular matrix, a major structural component of the cornea.
The ZapA protease, secreted by the urinary tract pathogen P. mirabilis, was found to cleave immunoglobulin G (IgG) (Loomes et al., 1993) and both IgA1 and IgA2 (Wassif et al., 1995), potentially conferring protection against opsonization of the bacterium during urinary tract infections.
The role of PrtA in the virulence of the bacterial entomopathogen Photorhabdus awaits clarification. The bacterium lives in symbiosis with nematodes that invade insects. Following entry into the insect, the bacteria are released from the nematode gut into the open blood system of the insect. Here, they secrete factors that kill the host and digest host tissues into nutrients for the replicating bacteria and nematodes. The secreted RTX protease PrtA of Photorhabdus, however, was not shown to be any major virulence factor (Bowen et al., 2003) and may play an alternative role in the host bioconversion.
Caulobacter crescentus synthesizes another unusual type of RTX protease (Umelo-Njaka et al., 2002) called Sap (S-layer-associated protease). The N-terminal half of Sap exhibits significant similarity to other RTX proteases (e.g. AprA of P. aeruginosa), including the characteristic RTX repeat sequences, while the C-terminal half of Sap exhibits a significant similarity to the N-terminal region of the S-layer protein RsaA. The hypothesis is that Sap evolved by combining the catalytic portion of an RTX protease with an S-layerlike protein, perhaps to associate with nascent S-layer monomers to scan for modifications. Despite no clearly identifiable type I secretion signal, Sap still appears to be secreted by a TISS (Ford et al., 2007).
Lipases from P. fluorescens show remarkable sequence similarity to that of S. marcescens, with an identity of about 65% over virtually the entire length of the sequence (Li et al., 1995). The S. marcescens SM6 and Pseudomonas sp. MIS38 enzymes are 613 and 617 residues long, respectively, and compared with P. fluorescens lipase, which is 476 amino acids in length, appear to bear an extra domain consisting of 138 and 135 amino acid residues inserted between residues Asn 405 and Thr 406 of the P. fluorescens lipase backbone (Amada et al., 2000).
Lipase production by P. fluorescens B52 was shown to be repressed by iron and is regulated by temperature. Optimal lipase production occurs well below the optimal growth temperature (Woods et al., 2001). Lipase production is also regulated by the homologue of the E. coli EnvZ-OmpR twocomponent osmoregulatory system and its secretion is reduced by NaCl (McCarthy et al., 2004). The lipase production by P. brassicacearum was, in turn, shown to be under the control of phase variation (McCarthy et al., 2004).
In S. marcescens, the lipA gene is not linked to TISS component genes for LipB (ABC protein), LipC (MFP) and LipD (OMP), which, besides secreting LipA, can also promote the secretion of the metalloprotease, PrtA (Nakahama et al., 1986), and of the S-layer protein homologue SlaA (Kawai et al., 1998). In P. fluorescens SIK W1, the RTX lipase gene tliA is situated downstream of the ABC exporter genes tliDEF, with genes prtA and inh, for an RTX protease and a protease inhibitor, being located upstream of tliDEF (Ahn et al., 1999). In P. fluorescens strain no. 33, lipA is clustered with an aprA gene for an alkaline protease, aprDEF genes for the TISS and pspA and pspB genes for two homologues of Serratia serine proteases (Kawai et al., 1999). In the P. fluorescens B52, lipA is situated downstream of the aprXinh-aprDEF-prtAB operon (Woods et al., 2001). The organizations of the various gene clusters are depicted in Fig. 5.

RTX bacteriocins
Bacteriocins are structurally and functionally diverse bacterial toxins inhibiting the growth of other bacterial strains (Jacob et al., 1973). Plasmid-borne genes for bacteriocins belonging to the RTX protein family (Table 3) were found in Gram-negative plant endosymbionts and pathogens, such as Rhizobium leguminosarum, Bradyrhizobium elkanii, Xylella fastidiosa, Xanthomonas oryzae or Agrobacterium tumefaciens, respectively (Oresnik et al., 1999;Simpson et al., 2000;Venter et al., 2001;Watson et al., 2001;Cherif et al., 2006;Sugawara et al., 2007). The mechanism of action of these putative bacterial toxins remains to be characterized. Nodulation competition experiments with mutants lacking the RTX bacteriocin activity indicated a role in the competitiveness of R. leguminosarum 248 (Oresnik et al., 1999).

S-layer RTX proteins
S-layer proteins form regularly arranged two-dimensional crystalline arrays covering the entire outer surface of a broad spectrum of bacteria and archaea. These are often composed of a single protein or glycoprotein species of 40-200 kDa, which is endowed with the ability to assemble on the supporting envelope layer, thus representing one of the simplest self-assembly systems (Sleytr & Beveridge, 1999;Sleytr et al., 2007).
Several RTX proteins have been identified among S-layer proteins of pathogenic as well as nonpathogenic bacteria and cyanobacteria (see the section on RTX proteins involved in the motility of Cyanobacteria for more details). S-layer proteins have an acidic pH, lack cysteines and are produced in large quantities (10-12% of cell protein). They usually possess two structurally different domains, with the N-terminal part typically showing similarity to other S-layer proteins and the C-terminal part containing the characteristic RTX structures and an uncleaved TISS signal, respectively. The RTX repeats of S-layer proteins were shown to bind calcium, which was proposed to mediate proper Slayer crystallization.
The best-studied S-layer RTX protein is the 98 kDa RsaA that forms a hexagonal lattice on the cell surface of C. crescentus. Six copies of RsaA form a ring-like subunit that interconnects with other subunits to form a twodimensional array consisting of approximately 40 000 RsaA units and with a porosity predicted to exclude molecules larger than $17 kDa . Some RsaA monomers are anchored to the outer bacterial membrane via interaction with smooth lipopolysaccharide, while the others are presumed to remain surface-associated by interaction with already attached RsaA monomers (Bingle et al., 1997a;Awram & Smit, 2001;Ford et al., 2007). Calcium appears to mediate proper RsaA crystallization and perturbations in RsaA at or near the RTX repeats result in the shedding of RsaA (Walker et al., 1994;Beveridge et al., 1997;Nomellini et al., 1997;Ford et al., 2007). It is presumed that a key function of the S-layer is to form a selective porosity barrier protecting the bacterium from a variety of predatorial assaults in the complex environments of bacterial biofilms. The S-layer appears to provide a barrier protecting against attack by the predatory Bdellovibrio-like bacterium (Koval & Hynes, 1991).
The N-terminal part of the 98 kDa RsaA protein shows the highest homology score to Campylobacter fetus S-layer protein, while its C-terminal portion contains five RTX repeats (Gilchrist et al., 1992). RsaA is produced constitutively to make up 10-12% of the total cell proteins, whereby it largely exceeds the secretion levels of any other type of RTX proteins. The ABC and MFP components of the TISS complex are encoded by rsaD and rsaE genes, located immediately downstream of the rsaA gene (Awram & Smit, 1998), while the rsaF a gene encodes one of two alternative OMP components of the RsaA secretion machinery and is located several kilobases downstream of the rsaA gene. The rsaF b gene encoding a second OMP component is completely unlinked. This is the only known example of a TISS that can utilize either of the two OMPs for the secretion of the same protein. Both OMPs, however, appear to be needed to handle the large amounts of RsaA produced, as neither one of the OMPs alone could induce wild-type secretion levels of RsaA (Toporowski et al., 2004).
Because of high secretion levels, cell surface location and geometrical packing of the RsaA protein, and thanks to the ease of genetic manipulation, the Caulobacter S-layer system has been exploited for biotechnology applications. These comprise the secretion of large quantities of 'passenger' proteins of economic and research interest into culture media (Bingle et al., 2000), or surface presentation of heterologous protein inserts on the the S-layer, such as S-layer-mediated display of the IgG-binding domain of streptococcal protein G (Bingle et al., 1997b;Umelo-Njaka et al., 2001;Nomellini et al., 2007).
Three other S-layer RTX proteins, Crs (1361 residues), CsxA (1123 residues) and CsxB (1238 residues), have also been identified in Campylobacter rectus, a Gram-negative bacterium associated with several forms of human periodontal disease (Miyamoto et al., 1998;Wang et al., 1998;Braun et al., 1999;LaGier & Threadgill, 2008). These proteins appear to be the potential virulence factors involved in the evasion of host defense, such as phagocytic uptake and bactericidal activity of serum (Okuda et al., 1997;Thompson, 2002).
SlaA, another S-layer family RTX protein of 1002 residues partially similar to the C. crescentus RsaA protein, was found in S. marcescens (Kawai et al., 1998). The 101 kDa protein appears to be exported by the Lip TISS system and a sequence dissimilarity in the N-terminal regions of SlaA has been observed among different strains of S. marcescens, which may be related to the antigenic variation of S. marcescens (Kawai et al., 1998).

RTX proteins involved in the motility of Cyanobacteria
RTX proteins appear to be abundant in cyanobacteria, while the function of most of them remains elusive. Quite unexpectedly, however, some of the described cyanobacterial RTX proteins have been shown to be involved in cell motility, including SwmA of Synechococcus sp. strain WH8102, oscillin of Phormidium uncinatum or hemolysinlike protein Sll1951 of Synechocystis sp. strain PCC 6803. The RTX protein SwmA was suggested to be part of the S-layer and to be required for the swimming motility of the marine unicellular cyanobacterium Synechococcus sp. strain WH8102 (Brahamsha, 1996;McCarren et al., 2005). Swimming Synechococcus strains are observed to rotate around their longitudinal axis, as they translocate at speeds of up to 25 mm s À1 and once fortuitously attached to a microscope slide, they rotate around the point of attachment. Intriguingly, cells with an insertionally inactivated swmA gene lack the Slayer and are nonmotile, and yet still rotate around the point of attachment. Thus, SwmA is somehow required for the generation of thrust, but not torque. However, the mechanism by which the SwmA and the S-layer function in motility remains elusive (Brahamsha, 1996;McCarren et al., 2005).
SwmA is a 130 kDa polypeptide that appears to be glycosylated and contains glycine-and aspartate-rich repeats. It is assumed that RTX repeats of SwmA function in calcium ion bridging to the outer membrane and thereby mediate the anchoring of the S-layer. Treatment of Synechoccocus sp. strain WH8102 with the chelator EDTA, indeed, removes the outer membrane and solubilizes SwmA (Brahamsha, 1996;McCarren et al., 2005).
The surface fibrils on the top of P. uncinatum S-layer consist of a single rod-shaped RTX protein of 646 residues with 46 repeats, the oscillin. Its structure appears to favor gliding, a relatively slow and smooth surface-associated translocation (Hoiczyk & Baumeister, 1997). As proposed by Hoiczyk and colleagues, the highly glycosylated surface of oscillin fibrils possesses ideal physicochemical properties for the temporary adhesion of the slime necessary for the generation of thrust and the helical arrangement of oscillin fibrils might guide the rotation of the P. uncinatum filament (Hoiczyk & Baumeister, 1997;Hoiczyk, 2000).
The hemolysin-like RTX protein Sll1951 of 1741 residues produced by a unicellular freshwater cyanobacterium Synechocystis appears to be related to the elimination of motility, although the mechanism of its action remains largely unknown (Sakiyama et al., 2006).

Nodulation RTX proteins
A 30 kDa RTX protein, NodO, was identified in R. leguminosarum bv. viciae and was shown to play a role in pea and vetch nodulation (Economou et al., 1990). The exact mechanism of the action of NodO in the process of the formation of nitrogen-fixing nodules on legume roots remains to be clarified. NodO was shown to form cationselective channels in planar lipid bilayers (Sutton et al., 1994), and two hypotheses were proposed on how this might enhance nodulation. One assumes that NodO pores in root cell membrane facilitate the passage of lipooligosac-charide nodulation factors. Another possibility could be the synergy of signalling resulting from cation fluxes through NodO channels across the plasma membrane, such as Ca 21 entry into root cells, thus amplifying the response induced by lipooligosaccharide nodulation factors (Sutton et al., 1994). The pores formed by NodO in planar lipid bilayers are relatively large (4 2 nm) and remain stably open for rather long, but are not voltage gated.
The prsDE genes for the TISS used by NodO appear to be conserved in all members of the Rhizobiaceae tested, even though these strains do not contain a nodO gene. The TISS, however, appears to be required for the secretion of several other Ca 21 -binding proteins that are involved in the formation and the nitrogen-fixing capacity of nodules induced by R. leguminosarum bv. viciae (Finnie et al., 1997). It also mediates the secretion of non-RTX glycanases PlyA and PlyB involved in the processing of Rhizobium exopolysaccharide (Finnie et al., 1998).

RTX proteins of unknown biological function
Initially, two partially homologous proteins, FrpA and FrpC, possessing the characteristic carboxy-proximal repetitions of the RTX nonapeptide motif, were discovered in N. meningitidis, a commensal of the human nasopharynx that occasionally causes invasive meningococcal disease (Thompson et al., 1993a, b). Biological activity of the Frp (Fe-regulated protein) proteins remains unknown. However, their secretion under iron-limited growth conditions, which mimic the condition in body fluids, and the elevated titers of antibodies against Frp proteins found in convalescent sera of patients from meningococcal disease, suggest a potential role of the FrpC-like proteins in meningococcal carriage or virulence .
Meningococci appear to carry a whole polymorphic family of frpC-like genes that code for proteins sharing large portions of identical sequence and varying in the number of C-terminal RTX repeat blocks and/or in insertions/deletions in the N-terminal portions. For example, the 122 kDa FrpA harbors 13 copies of nonapeptide repeats, while the 198 kDa FrpC has 43 copies. The N-terminal 293 residues of FrpA and the 407 N-terminal residues of FrpC do not exhibit any sequence homology, apparently due to the foreign DNA inserted at the 5 0 -end of the frpA gene (Thompson et al., 1993b;Parkhill et al., 2000;Tettelin et al., 2000).
The prototype FrpC protein of N. meningitidis FAM20 or MC58 strains is an 1829-residue-long protein with an amino-terminal portion of 876 residues and a carboxyterminal RTX moiety of 953 residues. The N-terminal portion of FrpC (residues 1-414) does not exhibit significant homology to any known proteins. However, it binds with very high affinity the FrpD lipoprotein, which is expressed from a gene located immediately upstream of the frpC gene in a predicted iron-regulated frpDC operon (Prochazkova et al., 2005). Recently, a unique 'clip-and-link' selfprocessing module (SPM) located between residues 414 and 657 of FrpC was characterized. It exhibits a high degree of sequence homology to segments of several RTX proteins of unknown function that are encoded in genomes of plant and animal pathogens (Osicka et al., 2004). Upon binding of calcium ions, the SPM produces a novel type of autocatalytic cleavage of the peptide bond between residues Asp 414 and Pro 415 of FrpC. Moreover, the newly generated N-terminal fragment of FrpC can be covalently linked to another protein molecule by a novel type of Asp-Lys isopeptide bond, which forms between the carboxyl group of the C-terminal Asp 414 residue of the thus generated fragment and the e-amino group of an internal lysine of another protein molecule (Osicka et al., 2004). This defines a novel class of autoprocessing RTX proteins of unknown biological function, as the same type of calcium-dependent processing and cross-linking activity was also demonstrated for the purified ApxIVA protein of A. pleuropneumoniae (Osicka et al., 2004).
ApxIVA is specific to the species of A. pleuropneumoniae (Schaller et al., 1999) and is genetically quite distant from other known RTX toxins. It is produced by A. pleuropneumoniae only during infections in vivo. When expressed as a recombinant protein in E. coli, it shows slight hemolytic activity and a distinct cohemolytic (CAMP) reaction. The apxIV determinant lacks the activator C gene and the type I secretion genes B and D, which are found in loci for the production of additional Apx toxins (ApxIA-ApxIIIA). However, upstream of apxIVA, an ORF can be found that is necessary for the hemolytic and CAMP activity of ApxIVA (Schaller et al., 1999;Frey & Kuhnert, 2002). ApxIVA was shown to be required for the full virulence of A. pleuropneumoniae, although its mechanism of action remains unclear (Liu et al., 2009). Recently, a system for single-step affinity chromatography purification of untagged recombinant proteins based on the SPM of N. meningitidis was developed (Sadilkova et al., 2008). The N-terminus of the SPM is fused to a target protein of interest and the C-terminus to an affinity tag. Upon loading of cell lysate and binding of the fusion protein to an affinity matrix, contaminating proteins are washed away and sitespecific cleavage of the Asp-Pro bond linking the target protein to the self-excising module is induced by the addition of calcium ions. This results in the release of the target protein with only a single aspartic acid residue added at the Cterminus, while the self-excising affinity module remains trapped on the affinity matrix (Sadilkova et al., 2008).

Identification of novel RTX proteins
Currently, almost 1000 prokaryotic genomes are fully sequenced and 4 2000 genome sequencing projects are in progress. We downloaded 840 fully sequenced bacterial genomes available in February 2009 and the putative RTX proteins were identified using three different methods: (1) pattern search; (2) Hidden Markov Model (HMM) search by HMMER 2.3.2 (Eddy, 1998); and (3) RPS-BLAST (Marchler-Bauer et al., 2002).
For the identification of RTX proteins by pattern search, we tested the Prosite pattern PS00330 (DX[L/I] XXXXGXDX[L/I]XGGXXXD) (Hulo et al., 2006), the pattern proposed for the AC toxin GGXG(N/D)DX(L/I/F) (Bauche et al., 2006), and a universal short pattern for the calcium-binding site GGXGXD. Matches to a database of all bacterial proteins were identified by the program preg, which is a part of the EMBOSS package (Rice et al., 2000). We analyzed 4 2.75 million sequences of a total length of 865 million amino acid residues. As expected, the longer the tested pattern was, the fewer the proteins that contained a positive hit. In total, 2598 hits to the Prosite pattern were found in 773 proteins. Shorter patterns were identified 5337 and 34 027 times, respectively, in 1210 and 20 963 proteins, respectively. To assess how many hits could be expected at random, we calculated an average amino acid composition of the bacterial protein database and generated a database of proteins of a random sequence, but with the same average composition. The average protein composition was calculated by pepstats and random sequences were generated by makeprotseq from the EMBOSS package (Rice et al., 2000). The overall size of the randomized database was equal to the size of the database containing all bacterial proteins. The randomized database was tested for the presence of the aforementioned patterns and randomization and testing was repeated 100 times. On average, only 1.6 hits to the Prosite pattern were found in the random database. Not surprisingly, the other two shorter patterns were more frequent, with 373 and 20 504 hits on average for GGXG(N/D)DX(L/ I/F) and GGXGXD motifs, respectively. The specificity of shorter patterns in identifying RTX proteins could be improved by searching for multiple hits in a single protein.
For the purpose of this review, we only considered proteins containing a hit to the Prosite pattern as being a putative RTX protein.
The alternative approach to identification of calciumbinding sites common to RTX proteins was to use the HMM-based model for hemolysin-type calcium-binding repeats, the PF00353 from Pfam database release 23.0 (Finn et al., 2006), which is an 18-residue-long model of two calcium-binding sites. We used HMMER 2.3.2 (Eddy, 1998) for the identification of proteins containing these calciumbinding sites. One thousand and thirteen proteins exhibited significant homology to the PF00353 Pfam model. On average, only 0.1 hits were found in the above-described random database. Thus, this approach appeared to be highly specific and generated virtually no false-positive hits.
The last used approach was to predict RTX proteins with RPS-BLAST (Marchler-Bauer et al., 2002). We searched the bacterial protein database for homology to the 510-aminoacid-long COG2931, position-specific scoring matrix (PSSM) that represents RTX toxins and the related Ca 21binding proteins. In contrast to Pfam models, there was no cutoff e-value defined for PSSM; therefore, we tested e-value cutoffs lower from 10, 1, 0.1 and 0.01. An e-value cutoff equal to or higher than 1 was not specific enough and 4 15 proteins were identified even in the random sequence database. Using an e-value of 0.1 as a cutoff, 4 400 proteins were identified in the bacterial database, compared with only 1.25 in the random sequence database. We therefore considered a cutoff of e-value equal to or o 0.1 as being sufficiently stringent for the purpose of this review.

Predicted RTX proteins
Bacteria harboring RTX protein genes were mostly Gammaproteobacteria (100 strains), Alphaproteobacteria (56 strains) and Betaproteobacteria (32 strains). Cyanobacteria were represented 25 times and RTX proteins appear, indeed, to be most abundant in cyanobacteria, with Trichodesmium erythraeum IMS101 bearing genes for 35 putative RTX proteins and up to 28 RTX proteins being encoded in the Acaryochloris marina MBIC11017 genome. Surprisingly, putative RTX protein genes were also detected in seven Gram-positive bacteria genomes, with six Actinobacteria and one Firmicutes (Streptococcus sanguinis SK36), respectively. None of the Actinobacteria genomes, however, harbored any significant homologues of TISS components. Interestingly, the S. sanguinis SK36 genome contains an operon encoding an RTX protein, a HlyB-like protein and a HlyD-like protein. The TolC protein was not identified. A TolC homologue would, however, not be needed for Type I RTX secretion in a Gram-positive bacterium, which lacks the outer membrane. Whether the RTX protein and TISS components are really produced and functional in S. sanguinis remains to be determined.
RTX proteins identified by the aforementioned approaches vary significantly in length and range from o 100 residues (protein NP_437763 of Sinorhizobium meliloti) to 36 800 residues (protein YP_378930 of Chlorobium chloro-chromatii). Ninety percent of the putative RTX proteins fit to the range of 300-5000 residues, with an average of about 1600 residues. With the exception of three proteins, all RTX proteins appear to be acidic, with a theoretical pI in the range from 3.2 to 4.9 for 90% of the putative RTX proteins (Fig. 6).
Independent of the primary annotation, the prediction of functions of the putative RTX proteins was attempted. We used RPS-BLAST ver 2.2.20 (Altschul et al., 1997) to detect significant homologies of putative RTX proteins to PSSMs in the CDD database version 2.16 (Marchler-Bauer et al., 2002), which contains almost 32 000 domain models from Pfam, COG, SMART and KOG databases. Based on detected homologies to the CDD database, we were able to assign putative functions to about 30% of the identified RTX proteins (Table S1). The largest group would consist of proteases (112) and adhesins (113), the latter falling into two major classes of cadherins (52) and vWA domain Fig. 6. Comparison of size and pI distribution of the bulk of bacterial proteins and of the sum of characterized and putative RTX proteins. Molecular weight (MW) and pI was calculated for the bulk of bacterial proteins (about 2.75 mil., shown in black) and for the identified and predicted RTX proteins (1024 proteins, shown as red crosses) using the program pepstat, a part of the EMBOSS package (Rice et al., 2000). MW was log base 10 transformed (y-axis) and plotted against calculated pI (x-axis) in statistical package R version 2.9.0 (http://www.r-project.org/). proteins, respectively (46). A further 20 lipases and 16 peroxidases were predicted.
About 20 enzymes were predicted to be involved in the degradation of saccharides and polysaccharides as glycosyl hydrolases, polysaccharide hydrolases and endoglucanases. About 20 putative RTX proteins might be involved in the degradation of DNA as endonucleases, nucleotidases and phosphodiesterases. A further four might act as phytases, one as a sulfatase, three as cyclophilin-like peptidylprolyl cis-trans isomerases and five as phosphatases, respectively.
Interestingly, three b-lactamases and one putative lysozyme were also predicted among the computationally detected putative RTX proteins. A further nine putative RTX proteins appear to exhibit weak homology to Hedgehog/ Intein domain proteins, suggesting that they might undergo autoprocessing and splicing.

Conclusions
Being first recognized as a group of pore-forming bacterial leukotoxins two decades ago (Welch, 1991), the RTX family nowadays comprises a particularly broad range of exoproteins that play important roles in the colonization of various habitats and hosts by Gram-negative bacteria (Fig. 7). Bioinformatic mining of the growing database of bacterial genomes indicates that a large spectrum of biological and biochemical activities of RTX proteins still remains to be characterized. The unique and central overarching feature of RTX proteins is the presence of variable numbers of the calcium-binding glycine-and aspartate-rich C-terminal repeats. These separate the unprocessed C-terminal secretion signals from the specific bodies of individual proteins. By being intrinsically unstructured at the very low calcium concentrations found within bacterial cytoplasm, the repeats maintain an unfolded or a loosely folded state of the RTX protein that enables recognition of its secretion signal by a dedicated type I secretion ATPase assembly. This function of RTX repeats allows a single-step passage of extremely large proteins through the entire Gram-negative cell envelope. Binding of extracellular calcium ions to repeats emerging at the bacterial surface then 'turns on' structuration of the repeats, pulling the protein out of the Fig. 7. Use of a simple RTX building unit in generation of complex biological functionalities. By maintaining an unfolded state which allows a posttranslational secretion from the calcium-depleted cytoplasm, and by promoting protein folding upon the binding of calcium ions (yellow balls) in the extracellular environment, the C-terminal assemblies of glycine-and aspartate-rich nonapeptide RTX repeat units first assist in the translocation of even very large RTX proteins. These transit across the entire Gram-negative bacterial cell envelope in a single step mediated by the dedicated the type I secretion machinery recognizing unprocessed C-terminal secretion signals. Proteins using this secretion pathway exhibit a very broad range of biological functions in colonizing diverse host environments. RTX proteins were found to exert activities like structural proteins involved in protective S-layer formation and motility of bacteria, in colonization of root nodules of plants by symbiotic bacteria, serving as bacteriocins on other bacteria, exerting hydrolase activities, or playing a prominent role as essential colonization and virulence factors of bacteria in animal hosts, respectively. Besides of a large group of pore-forming leukotoxins a particular sophistication of function is observed for the very large (thousands of residues long) RTX toxins consisting of multiple domains exhibiting enzymatic and cytotoxic activities (MARTX). cells and driving folding of the rest of the protein. Moreover, the RTX repeats themselves support an amazingly vast array of biological activities, be it a role in the formation of Slayers, bacterial adherence/motility or host-receptor interaction and membrane penetration of RTX proteins. Hence, through a simple variation of number and block arrangement of a small repeat unit, an elegant construction of sophisticated biological functions is achieved.
during Bordetella pertussis infection in vivo. Infect Immun 66: Table S1. Predicted RTX proteins identified by bioinformatic screening of sequence databases. Table S2. Bacterial genomes containing rtx genes, as identified by sequence database screening.
Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.