Plasmids pXO1 and pXO2
The possibility that the genes for the dominant B.anthracis virulence factors might be extrachromosomal was suggested by the work of Louis Pasteur and Max Sterne, who each isolated variants of B. anthracis having reduced virulence. After some initial difficulties,
plasmids associated with toxin and capsule were discovered (Mikesell et al., 1983; Uchida et al., 1985; Green
sheared. Both these large plasmids have been sequenced. Plasmid pXO1 is 181.6 kb (Okinaka et al., 1999) and pXO2 is 96.2 kb (Pannucci et al., 2002). The replication origin of pXO2 is known to be similar to that of pAMβ1 (Tinsley et al., 2004), while that of pXO1 remains unknown.
The B. anthracis plasmids can each be selectively cured, pXO1 by repeated passage at 42°C, and pXO2 by growth in novobiocin. Comparison of plasmid-cured variants proved that pXO1 is needed for production of toxin and pXO2 for production of capsule (Thorne, 1985). Conjugal transfer of pXO1 to plasmid-cured strains of B. anthracis confirmed that all the genes necessary for toxin production are contained on the plasmid (Thorne, 1985; Heemskerk and Thorne, 1990).
These results explain the properties and the efficacy of the anthrax vaccines developed by Louis Pasteur and Max Sterne. It is now well established that B. anthracis strains must produce PAin order to induce protective immunity (Ivins and Welkos, 1988). Elimination of the pXO1 plasmid yields an avirulent strain, but one that does not induce immunity. In retrospect, it is evident
that Louis Pasteur’s attenuation of the virulence of B. anthracis cultures by growth at 42°C was due to partial curing of plasmid pXO1. The cultures he used to successfully immunize sheep probably contained a small number of virulent (pXO1+, pXO2+) bacteria with a larger number of avirulent (pXO1−, pXO2+) bacteria, with their efficacy due to the former, which would
induce antibodies to PA. While effective, these vaccines could cause infection if the fraction of virulent organisms was too high. Max Sterne’s important contribution was to analyze carefully the rare, spontaneous non-capsulated variants appearing on agar plates, and to show that they were greatly reduced in virulence (Sterne, 1937). These variants, now known to have lost the pXO2 plasmid, are effective animal vaccines and do not revert to virulence. The (pXO1+, pXO2−) “Ster ne” strain continues in use today as the preferred veterinary vaccine and as a convenient strain for laboratory studies because of its relative safety.
Genes specifying toxin structure and expression
The recognition that anthrax toxin was encoded on pXO1 facilitated the cloning and sequencing of the genes encoding PA (pagA), LF (lef), and EF (cya) (Vodkin and Leppla, 1983; Robertson and Leppla, 1986; Tippetts and Robertson, 1988). Each of the genes has a G+C content of about 30% (Table 18.1), similar to that of the B. anthracis genomic DNA (35% G+C). Upstream of
the ATG start codons in each of the genes is an appropriatelyappropriately located ribosome binding site, AAAGGAG for the PA and LF genes, and AAAGGAGGT for the EF gene. Each of the three proteins contains a typical bacillus signal peptide of 29–33 amino acids, with cleavage
occurring after an Ala or Gly. Following the stop codon of the PA gene is an inverted repeat that may act as a transcriptional stop; no similar structures are present in the LF or EF regions.
Production of both capsule (Meynell and Meynell, 1964) and PA (Gladstone, 1948) by B. anthracis is ependent on addition of bicarbonate or CO2. Early studies showed that stimulation of PA synthesis by bicarbonate requires the presence of a gene located on pXO1 (Bartkus and Leppla, 1989). This gene, atxA, was mapped by transposon mutagenesis and cloned and sequenced (Uchida et al., 1993; Koehler et al., 1994). The product of the atxA gene, a protein of 56 kDa, increases by at least 10-fold the transcription from a start site, P1, located at bp -58 relative to the start codon. Constitutive, low-level transcription initiates at another site, P2, at bp -26. Both these start sites are located in a potential 58- bp stem-loop structure (Welkos et al., 1988). Disruption of the atxA gene and complementation with an atxAexpressing plasmid proved that this regulator is needed for transcription of all three toxin genes, and it follows that B. anthracis strains lacking atxA are less virulent for mice (Uchida et al., 1993; Dai et al., 1995). Aregion of 111 bp upstream of the pag coding sequence is required for AtxAaction, but no evidence is available
suggesting that AtxA itself binds there (Dai et al., 1995) and no common sequences can be identified in the regions upstream of pag, cya, and lef to whichtranscriptional regulators might bind. More recent studies with microarrays showed that AtxA is a key global regulator of many plasmid and chromosomal genes (Bourgogne et al., 2003; Mignot et al., 2004). The regulation of capsule biosynthesis by atxA requires the presence of either of two additional regulators, acpA
and acpB, encoded on pXO2, and having limited sequence similarity to atxA (Drysdale et al., 2004). Toxin gene sequence homologies and variations PA is similar to toxins produced by some pathogenic Clostridia and Bacillus species (Barth et al., 2004). All these homologues are binary toxins, like the anthrax toxins, but the second, catalytic component has ADPribosylation
activity. The PAhomologues are listed in Table 18.2, and the properties of the proteins are discussed in a later section. The most recent and novel addition to this list of proteins are two identified in an unusual Bacillus cereus strain, G9241, that contains a plasmid closely similar to pXO1 (Hoffmaster et al., 2004). This plasmid encodes proteins having high sequence identify to those in B. anthracis (99.7, 99, and 96% amino acid identity to PA, LF, and EF, respectively). A second large plasmid in the same strain, pBC218, contains additional PA and LF homologues resembling the B. cereus binary toxins mentioned above. It is not yet known whether these toxins are active. Comparison of these proteins may provide new nsights into PA structure and function. This finding
reminds us that B. anthracis and B. cereus are closely related and may represent the visible and more prominent parts of a genetic continuum. Within the B. anthracis species, there is very little
sequence variation between toxin genes of different isolates. Only three amino acid residues were found to vary among 26 sequenced PAgenes (Price et al., 1999). It is of interest that the variable residues 536 and 571 are adjacent and on the surface of the protein, suggesting that these residues may be part of an epitope recognized by the immune system of animals and therefore
under selective pressure to mutate to escape antibody neutralization. The EF gene, cya (Escuyer et al., 1988;Robertson et al., 1988), has homology to several other adenylate cyclase genes of pathogenic bacteria (Ahuja et al., 2004), the best studied being that encoding the “invasive” adenylate cyclase of Bordetella pertussis (Glaser et al., 1988; Hanski and Coote, 1991). The homology to the B. pertussis cyclase occurs only in the regions known to comprise the catalytic domain. More recently, adenylate cyclases that may contribute to pathogenesis have been identified in Pseudomonas aeruginosa (Yahr et al., 1998) and in Yersinia species (Parkhill et al., 2001). The EF gene also has strong homology to the LF gene in the amino terminal region that is now recognized as involved in binding to PA. Beyond residue 250, LF has a very limited sequence homology to other metalloproteases.
Role of the toxin in virulence
A number of tools are now available for performing genetic modifications to B. anthracis (Thorne, 1993; Koehler, 2002a). These include transduction and conjugation (Battisti et al., 1985; Heemskerk and Thorne, 1990) and transformation by electroporation (Bartkus and Leppla, 1989). The elegant methods developed for use of Tn917 in Bacillus subtilis have been adapted to
B. anthracis (Heemskerk and Thorne, 1990). Methods for gene disruptions are also available (Dai et al., 1995). Aconjugational transfer system was used to transfer a mutated PA gene into the Sterne strain, replacing the resident PA gene (Cataldi et al., 1990). The latter method was employed to produce strains expressing every combination of the three components (Table
18.3). The only strain that retained some virulence was RP9, which makes PA and LF (Pezard et al., 1991; Pezard et al., 1993). This suggests that LF is the more important virulence factor, whereas EF provides a smaller contribution. However, this analysis needs to be repeated in a strain having pXO2 in addition to pXO1 because of the interaction of regulatory genes on the two plasmids (Bourgogne et al., 2003).
The toxin proteins collectively, or the recombinant PA and LF proteins individually, constitute more than 50% of the protein present in B. anthracis culture supernatants grown in the R or FAmedia noted above. This makes purification of the PAand LF proteins relatively easy once the proteins have been protected from proteases and concentrated. Recovery from culture supernatants
has been done by hydrophobic “salting out” onto agarose resins (Leppla, 1991a), but other methods can also be used. Effective purification steps include chromatography on anion exchange resins or hydroxyapatite. Detailed protocols for purification from B. anthracis are available (Quinn et al., 1988; Leppla, 988; Leppla, 1991a; Farchaus et al., 1998; Park and Leppla, 2000; Ramirez et al., 2002). Methods used for purification of the toxin proteins from E. coli have been more diverse. However, because the proteins lack disulfide bonds, they can be refolded from inclusion bodies in good yields. E. coli expression systems have the disadvantage that the products need to be analyzed for endotoxin contamination.
Structural features common to the three toxin components
All three of the anthrax toxin proteins are similar in size and charge (Table 18.1). Especially notable is that all three proteins lack cysteine. This has proved advantageous for structure function analyses, because any cysteine added by mutagenesis for subsequent chemical modification is unique (Nassi et al., 2002; Mourez et al., 2003). It was noted some years ago that extracellular
bacterial proteins generally have a low cysteine content (Pollack and Richmond, 1962), and this generalization appears to hold for a number of other secreted bacterial toxins. Perhaps most striking is the absence of cysteines in the B. pertussis adenylate cyclase, a protein of 1706 residues (Glaser et al., 1988).Overview of toxin binding and internalization by cells
Data on toxin structure combined with studies on interaction with cells to be discussed below has led to a model of toxin uptake depicted in Figure 18.1 (Petosa et al., 1997; Moayeri and Leppla, 2004). PAbinds to cell surface receptors and is cleaved by cell surface proteases, principally furin, with release of the 20-kDa N-terminal fragment. PA63 then oligomerizes and also binds LF or EF. The complex is internalized by endocytosis, and acidification of the vesicle causes insertion of
the PA63 heptamer into the endosomal membrane to produce a channel through which LF or EF translocate to the cytosol. Details of the individual steps are discussed in later sections.PA structure and function PAcrystal structure
The structure of PA was solved by X-ray diffraction (Petosa et al., 1997). PAis a long, flat protein that is rich in β-sheet structure (Figure 18.2). Four domains are distinguished that correspond to functional regions previously defined by analysis of large fragments produced
by trypsin and chymotrypsin. Domain 1 (aa 1-258) contains two tightly bound calcium ions and a
large flexible loop (aa 162–175) that includes the sequence 164RKKR167 that is cleaved during proteolytic activation. Domain 2 (aa 259–487) contains several very long β strands and forms the core of the membrane-inserted channel. It also has a large flexible loop (aa 303–319) implicated in oligomer formation and membrane insertion. Domain 3 (aa 488–595) has no clearly defined function. Domain 4 is loosely associated with the other three domains and is involved in receptor binding.PAproteolytic activation, nicking, oligomerization,and binding of LF and EF
The two flexible loops mentioned above contain sites uniquely sensitive to proteolytic cleavage, as was recognized even before the aa sequence was known. The sequence 164RKKR167 is extremely sensitive to cleavage by trypsin, clostripain, and other proteases that recognize
basic residues. PAin solution at 1 mg mL−1 is completely cleaved in 30 minutes when treated with 0.1 μg mL−1 of trypsin (Leppla et al., 1988; Singh et al., 1989). The fragments of 20 and 63 kilodaltons (designated PA20 and PA63, respectively) do not easily dissociate. However, at pH 7.5, concentrated solutions of the nicked PAincubated for several hours form a precipitate, which can be shown by SDS gel electrophoresis to contain PA63. Chromatography of trypsin-treated PA
on the MonoQ anion exchange resin (Pharmacia) at pH 9.0 yields an early peak of PA20 and a later peak containing PA63. PA63 purified on MonoQ resin remains soluble indefinitely if kept at pH 9.0. Analysis by nondenaturing gel electrophoresis and gel filtratio n chromatography showed this material to be a large oligomer of the 63-kDa peptide (Leppla et al., 1988; Singh et al., 1994). This oligomer is extremely stable. By transmission electron microscopy after negative staining,
the oligomer was shown to be a heptamer (Milne et al., 1994). This species was crystallized and its structure determined by X-ray diffraction, which confirmed the heptameric nature of the oligomer (Petosa et al., 1997). More recently, a higher resolution structure of the heptamer bound to receptor was determined (Lacy et al., 2004a). This showed that the loop including residues 303–309 stabilizes the heptamer by closely associating with neighboring PA63 monomers in the
heptamer. The functional importance of the trypsin-sensitive site at residues 164–167 became evident when it was noted that PAincubated with cells becomes nicked at this site (Leppla et al., 1988). Only PA63 remains bound to cells; PA20 can be detected in the supernate if this is
concentrated before analysis. PA cleavage on cells occurs at 4°C, a condition in which PAis retained on the cell surface. A PA mutant in which the 164RKKR167 sequence is deleted is not cleaved on the surface of cells and is non-toxic (Singh et al., 1989). The implication that PA63 is the active species needed for delivery of LF or EF was proven directly by showing that purified,
heptameric PA63 is toxic to macrophages when combined with LF. Finally, PAincubated with cells at 37°C and allowed to internalize forms oligomers that are stable to heating in SDS (Milne et al., 1994; Liu and Leppla, 2002). This occurs because acidification of endosomes causes the oligomer to insert into membranes to form a second, stable type of heptamer (to be discussed later).
Adetailed analysis of PAresidues 164–167 showed that cleavage by cellular proteases requires the minimum sequence RxxR (Klimpel et al., 1992). This result, combined with inhibitor studies, proved that the cellular protease that most rapidly activates PAis furin. This finding was consistent with evidence that a number of other bacterial toxins require proteolytic activation by furin (Gordon et al., 1995; Gordon et al., 1997). The strict requirement that PAbe proteolytically cleaved on the
cell surface has been exploited in the design of toxins dependent on other proteases, as will be discussed later.The other site that is uniquely sensitive to protease is the large loop in domain 2, aa 303–319. Cleavage in this region occurs in B. anthracis culture supernates, probably due to the action of a metalloprotease of the thermolysin type. Cleavage at the same site is obtained with 1 μg mL−1 of chymotrypsin or thermolysin. This cleavage occurs at the pair of Phe residues, aa 313–314 in the sequence SFFDI (Novak et al., 1992). Deletion of the pair of Phe completely inactivates PA(Singh et al., 1994). The deleted mutant PA binds to cells, becomes nicked, and internalizes LF to endosomes, but fails to tranlocate LF to the cytosol (Novak et al., 1992; Singh
et al., 1994). PA proteins mutated at this site have served as valuable controls when determining whether action of PA variants requires access to the cytosol. Receptor binding studies using radiolabeled PA that was nicked at residues 313–314 showed that only the C-terminal 47-kDa fragment was retained on cells. This result showed that the cell recognition domain is entirely contained in the 47-kDa fragment, residues 315–735. The PA63 heptamer binds tightly to LF, as can be demonstrated by several methods, including sedimentation equilibrium and gel electrophoresis. On nondenaturing 5% polyacrylamide gels run at pH 8.5, PA63 moves as a sharp band, much slower than PA. In mixtures of PA63 and LF, several very closely spaced bands migrating even more slowly than PA63 are seen. These contain oligomeric PA63 with increasing
numbers of bound LF molecules (Singh et al., 1999). Although the author’s initial studies were interpreted as showing that a PA63 heptamer can bind seven LF molecules, strong evidence against this came from mutagenesis studies showing both that the LF/EF binding site spans two adjacent PA63 monomers (Mogridge et al., 2002b) and that the large footprint of LF/EF on the heptamer precludes simultaneous use of two adjacent binding sites (Cunningham et al., 2002;
Mogridge et al., 2002a). It follows that a maximum of three LF/EF molecules can bind to the heptamer. However, in recent cryo-electron microscopy analyses, only a single LF was observed bound to the PA63 heptamer (Ren et al., 2004), indicating that the stoichiometry question deserves continuing study.
Functional sites of PAdefined by mutagenesis
Removal of residues 1–167 (domain 1a) by proteolytic activation and formation of the heptameric PA63 leaves a surface (domain 1b) to which LF and EF bind. Two groups have used mutagenesis to identify PA residues involved in LF binding. Nine residues on the surface that is thought to interact with LF were altered, and four of them (aa 202, 203, 205, and 207) were found
to be essential (Chauhan and Bhatnagar, 2002). Amore extensive mutagenesis study agreed that aa 205 and 207 are required for LF binding, and implicated five additional residues (Cunningham et al., 2002). More importantly, by introducing certain of the mutations into separate monomers that could only form dimers, the residues involved in the adjacent monomer’s subsites of the LF binding site were identified.
The crystal structure shows that residues 259–487 form the central core of the heptameric PA channel. Recognition that the large loop including aa 303–319 had alternating hydrophilic and hydrophobic residues led to the proposal that its insertion into the membrane would produce a β barrel (Figure 18.3) like that seen in the structure of staphylococcal alpha-hemolysin (Petosa et al., 1997). Extensive mutagenesis of residues 302–325 confirmed that the hydrophobic side chains
interact with lipid while the hydrophilic residues are accessible to solvent, as expected if the structure followed the example of the alpha-hemolysin structure (Benson et al., 1998). subsequent studies extended this analysis and suggested that the β barrel involved the
entire region from aa 275 to 352 (Nassi et al., 2002), implying that a major rearrangement of the heptamer takes place upon membrane insertion, and that the β barrel extends beyond the bilayer.
The requirement for oligomerization implies that dominant negative mutant proteins can exist, ones in which mutations within a single monomer could inactivate an entire heptamer. A systematic mutagenesis study of the entire PA63 region was designed to identify candidate dominant negative mutants (Mourez et al., 2003). At least 33 residues in the entire PA63 sequence were found to be essential for toxicity. Amore targeted search for dominant negative mutants identified several having that behavior (Sellman et al., 2001; Singh et al., 2001). By definition, such mutants are able to form the heptameric prepore, but this fails to convert
to a functional protein-conducting channel. In theory, one could also expect that mutation of key residues lining the lumen of the channel may also yield dominant negative proteins, but none have yet been reported. Systematic mutagenesis of domain 2 also identified several residues within the extended β barrel region that were required for toxicity (Mourez et al., 2003). Of particular interest are residues 337, 342, and 346, which were later shown to participate in binding of the prepore heptamer to the cellular receptor (Santelli et al., 2004; Lacy et al., 2004a), as will be discussed later.
Systematic mutagenesis studies (Mogridge et al., 2001; Mourez et al., 2003) identified very few residues in domain 3 (aa 488–595) that are essential for activity. An E515C mutant was inactive but an E515A mutant retained activity. The loops containing aa 483–486 and 510–518 interact with the adjacent monomer in the heptamer prepore structure (Lacy et al., 2004a), and substitutions in these loops had modest effects on function. However, evidence to date suggests that the principal role of domain 3 is to stabilize the heptamer. It can be noted that the only residues (aa 536 and 571) that vary among the natural B. anthracis PA isolates (see above) are located in domain 3. If these residues are a target of neutralizing antibodies, as suggested above,
then those antibodies might neutralize by blocking heptamer formation.
The C-terminal domain 4 of PAwas initially implicated in receptor binding by analysis of mutants truncated to varying extents (Singh et al., 1991). Later studies suggested that the effect of these truncations was indirect, perhaps by causing changes to the folding of domain 4. Codon-based mutagenesis was used to identify important residues in the two solvent-exposed loops of domain 4 that were predicted to interact with the cellular receptor (Varughese et al., 1999; Brossier et al., 1999). Substitution of N657 or N682 had the largest effects in decreasing PAbinding to receptor. Further mutagenesis of the small loop containing residues 679–693
showed that this region contains four residues for which single alanine substitutions decreased toxicity more than 10-fold (Rosovitz et al., 2003). The D683 residue was shown to be critical, which was later explained by its interaction with the divalent metal ion in the receptor, as discussed later. The systematic mutagenesis study referenced above (Mourez et al., 2003)
identified residues 656, 657, 665, 682, 683, and 687 as important for PAactivity.