Plasmid addiction modules and chromosome-encoded toxin–antitoxin systems
One of the most well-characterized examples of proteins with a function that is thought to elicit cell death is the family of antidote/toxin gene pairs known as ‘addiction modules’, which have been reviewed extensively elsewhere (Engelberg-Kulka and Glaser, 1999). These modules were initially discovered on extrachromosomal elements such as plasmids, and are responsible for a post-segregational killing effect in plasmid-free cells. In general, the toxin gene is transcribed at low levels, and encodes a stable, long-lived protein, whereas the antidote gene is transcribed at high levels and encodes a relatively unstable, short-lived protein. High-level transcription of the antidote gene in cells containing the plasmid harbouring the addiction module ensures that the toxin remains inactive. However, if the bacterial cell loses the plasmid, the cured cells are selectively killed by the stable toxin as a result of the relatively rapid degradation of the unstable antidote by a specific host-encoded protease (Engelberg-Kulka and Glaser, 1999). In this manner, the cell is ‘addicted’ to the presence of the plasmid or prophage harbouring the toxin–antidote gene pair. The toxins encoded by these modules appear to kill the cell by targeting specific cellular components, such as the GyrA subunit of DNA gyrase (Bernard and Couturier, 1992), or by inhibiting the initiation of DNA replication by interacting with the DnaB helicase subunits (Bernard and Couturier, 1992; Ruiz-Echevarria et al., 1995).
Several homologues of the plasmid-encoded addiction modules have also been discovered in the chromosomes of many bacteria (Masuda et al., 1993; Aizenman et al., 1996; Gotfredsen and Gerdes, 1998; Mittenhuber, 1999). One of the best characterized of these is the mazEF (also called chpAI and chpAK by Masuda et al., 1993) antitoxin/toxin system, which was originally discovered in Escherichia coli (Aizenman et al., 1996). These gene products are homologous with the PemI/PemK (also called Kis/Kid) antitoxin/toxin proteins found on the E. coli plasmid R1 (Ruiz-Echevarria et al., 1991), and display the characteristics of the plasmid-encoded addiction modules. Recent data have suggested that the MazF protein may inhibit both initiation of DNA replication as well as translation (Pedersen et al., 2002). The mazEF genes form an operon with the upstream relA gene, encoding an ATP:GTP 3′-pyrophosphotransferase [(p)ppGpp synthetase] (Aizenman et al., 1996). RelA is responsible for activating ppGpp synthesis in response to amino acid starvation, and the subsequent increase in intracellular ppGpp concentration acts as a signal for induction of the stringent response. Intriguingly, expression of mazEF is inhibited by high concentrations of ppGpp (Aizenman et al., 1996). This suggests that, under stressful conditions such as amino acid starvation, the labile antitoxin MazE would be rapidly degraded and, consequently, toxic MazF protein would be able to kill the cell. Further experiments demonstrated that mazEF-mediated cell death is induced by artificially overproducing ppGpp (Aizenman et al., 1996; Engelberg-Kulka et al., 1998), suggesting that mazEF may play a role in PCD in response to extreme starvation conditions. The idea that mazEF is involved in eliciting PCD in response to starvation has recently been strengthened by the discovery that mazEF appears to mediate cell death in response to thymine starvation (Sat et al., 2003). In this study, it was shown that a ΔmazEF strain was almost 100% viable when starved for thymine, whereas the wild-type strain underwent a significant loss of viability (over 90%) (Sat et al., 2003). Furthermore, the activity of the mazEF P2 promoter (primarily responsible for transcription and regulation of mazEF) was drastically reduced under conditions of thymine starvation (Sat et al., 2003).
Cell death mediated by mazEF also appears to be triggered by antibiotics that inhibit transcription or translation, such as rifampin, chloramphenicol and spectino-mycin (Sat et al., 2001). In fact, mazEF-triggered death appears to be dependent on the ability of these antibiotics to inhibit either transcription or translation, as antibiotics that target other cellular processes kill wild-type and mazEF mutant cells at comparable levels (Sat et al., 2001). The presence of these antibiotics causes a reduction in the cellular level of the MazE antitoxin and presumably allows for the lethal action of MazF (Sat et al., 2001). Searches for other chromosomally encoded mazEF-like genes have revealed the presence of homologues in many Gram-positive and Gram-negative bacteria (Mittenhuber, 1999; Engelberg-Kulka et al., 2001). In general, it appears that many bacteria carry one or more genes encoding MazF-like toxins and that, in some cases, they lacked an identifiable upstream gene corresponding to MazE-like antitoxin (Mittenhuber, 1999). The mazF homologue in the Gram-positive organisms Staphylococcus aureus and Bacillus subtilis is the last gene of a putative seven-gene operon that also contains the alr gene, encoding alanine racemase (Kullik et al., 1998a; Mittenhuber, 1999). Interestingly, this operon is located immediately upstream of the sigB locus, which encodes the regulatory components of the sigma-B (σB)-mediated general stress response (Kullik et al., 1998a). Studies of expression of the sigB locus in S. aureus have revealed that the upstream mazF gene may be co-transcribed with the sigB operon under conditions of heat stress (Gertz et al., 1999). Collectively, the above observations suggest that mazEF may elicit PCD in response to stress conditions including nutrient deprivation, heat and exposure to certain antibiotics. Furthermore, many of these mazEF homologues appear to be associated with stress regulons in both Gram-negative and Gram-positive bacteria.
As an alternative to a role in bacterial PCD, recent data have suggested that MazF (ChpAK) induces a bacteriostatic condition in E. coli that is reversible by the MazE (ChpAI) antitoxin, rather than by actually killing the cells (Pedersen et al., 2002). This phenomenon was postulated to protect cells from long-term starvation and/or harsh environmental conditions (Pedersen et al., 2002). Indeed, an alternative explanation for the role of the chromosomally encoded toxin–antitoxin modules has been proposed, in which these systems are thought to be involved in regulating global rates of protein synthesis when the cell undergoes nutritional stress (Pedersen et al., 2002; Nystrom, 2003). However, it should be noted that, in the experiments described above, the expression of MazE and MazF did not simulate the way in which these proteins are normally expressed in the cell. Specifically, the MazE antitoxin was artificially overproduced, whereas the synthesis of the MazF toxin was inhibited. In its native context, MazE is constantly degraded and therefore exists in small amounts in the cell.
Recently, other types of putative antitoxin/toxin gene pairs have been identified with gene products that are not homologous to those of the addiction modules described above, but the action of which appears to be similar. For example, the entericidin locus of E. coli and Citrobacter freundii encodes a putative antidote/toxin pair that appears to be positively controlled by the stationary phase sigma factor RpoS and negatively controlled by the osmosensing EnvZ/OmpR two-component signal transduction system (Bishop et al., 1998). The entericidin locus contains two genes, ecnA and ecnB, which appear to encode two small cell envelope lipoproteins. This locus seems to be involved in promoting bacteriolysis in stationary phase under high osmolarity conditions, with the ecnA gene product acting as an antidote to EcnB (Bishop et al., 1998). Another gene pair organized in a similar fashion to the addiction modules described above is encoded by the spoIIS locus of B. subtilis (Adler et al., 2001). The spoIISA gene product is predicted to contain three transmembrane segments in the N-terminal half, with the last two-thirds of the protein located in the cytoplasm, whereas the spoIISB gene product appears to be a relatively small, positively charged, cytosolic protein (Adler et al., 2001). In the absence of SpoIISB, the putative antidote protein, cells undergoing sporulation acquire SpoIISA-induced lethal damage to their cell membrane shortly after asymmetric septation (Adler et al., 2001). Furthermore, forced expression of SpoIISA in exponentially growing cells or in the forespore leads to the same type of damage and cell death (Adler et al., 2001). The fact that both the entericidin and the spoIIS loci appear to be unique to the respective species in which they were characterized suggests that these cell death-mediating systems may have evolved to serve specific, and as yet unknown, killing functions in these organisms.
Autolysis and antibiotic-induced killing
Bacterial murein hydrolases are enzymes that cleave specific components of the bacterial cell wall, and serve many important physiological functions during cell growth and development such as cell division, sporulation and competence (Horne and Tomasz, 1985; Doyle and Koch, 1987; Holtje and Tuomanen, 1991; Romeis et al., 1991; Blackman et al., 1998; Smith et al., 2000). However, deregulated activity of these enzymes can also lead to self-lysis, or autolysis, of the bacteria under stressful circumstances such as those encountered during stationary phase and after exposure to antibiotics (Moreillon et al., 1990; Tobin et al., 1994; Heidrich et al., 2001). Although the cellular targets of most antibiotics have been well characterized, it is still unknown how the interaction between antibiotic and its target actually triggers cellular death. One recently proposed theory is that the deregulated autolytic activity observed in bacteria exposed to antibiotics such as penicillin and vancomycin is actually a manifestation of bacterial PCD (Lewis, 2000; Normark and Normark, 2002).
Much of the research directed at understanding the role of murein hydrolases in antibiotic-induced killing and lysis was pioneered by studying the killing effects of penicillin and vancomycin on the Gram-positive organism Streptococcus pneumoniae. Several studies of S. pneumoniae have reported that murein hydrolase activity is required for penicillin-induced lysis (Tomasz et al., 1970; 1988; Lopez et al., 1990). However, a study by Moreillon et al. (1990) demonstrated that penicillin-induced killing and penicillin-induced lysis are two separate and distinguishable events. For example, mutants of S. pneumoniae in which the amidase-encoding lytA gene was inactivated were still killed in the presence of penicillin, albeit at a slower rate than the wild-type LytA+ strains (Moreillon et al., 1990). Furthermore, a second, unidentified mutation unrelated to the amidase gene dramatically reduced the amount of penicillin-induced killing in both wild-type and lytA mutant strains, suggesting that penicillin kills pneumococci by both lytic and non-lytic mechanisms (Moreillon et al., 1990). As mentioned above, it is possible that one or both of these killing mechanisms represent steps in a PCD pathway that is induced in response to antibiotic exposure. If this is indeed the case, then what is responsible for initiating the non-lytic death pathway?
Signal transduction by a ‘death’ peptide in S. pneumoniae?
One hypothesis that has gained attention in recent years is that a two-component signal transduction system, VncR/S, triggers PCD in response to antibiotic exposure in S. pneumoniae (Novak et al., 1999; 2000). This two-component system was originally identified by screening an S. pneumoniae library of mutants for penicillin tolerance. Characterization of a mutant defective in the vncS gene, encoding the putative histidine kinase sensor, revealed that, in addition to penicillin, this mutant displayed tolerance to a wide range of antibiotics with mechanistically distinct targets (Novak et al., 1999). Interestingly, inactivation of vncR, encoding the response regulator, had no measurable phenotype (Novak et al., 1999). Despite this seemingly contradictory observation, these studies appeared to indicate that this two-component regulatory system was part of a signal transduction pathway leading to antibiotic-induced cell death.
Subsequent characterization of this system revealed that a small open reading frame (ORF) directly upstream of vncRS encodes a 27-amino-acid peptide (Pep27) that was detectable in the cytoplasm and supernatant of wild-type S. pneumoniae (Novak et al., 2000). Synthetic Pep27 appeared to be capable of inducing loss of viability when added to S. pneumoniae culture, and a pep27 mutant (generated by polar mutation due to insertion–duplication mutagenesis of the upstream vex3 gene) displayed tolerance to several different antibiotics (Novak et al., 2000). Furthermore, overexpression of VncS appeared to block autolysis, suggesting that VncS is a negative regulator (Novak et al., 2000). Based on these results, it was proposed that Pep27 is a ‘death signal peptide’ that somehow interacts with the membrane-bound histidine kinase VncS in response to antibiotic exposure (Novak et al., 2000).
However, a recent report has provided evidence that casts doubt on the role of VncRS and Pep27 in antibiotic-induced cell death (Robertson et al., 2002). In contrast to the previous findings mentioned above, individual mutations in vncS, vncR, vex3 or pep27 in three different genetic backgrounds of S. pneumoniae appeared to have no effect on the lytic response to vancomycin (Robertson et al., 2002). The vncS mutants failed to exhibit tolerance to vancomycin and exhibited wild-type sensitivity to other classes of autolysis-inducing antibiotics (Robertson et al., 2002). Furthermore, synthetic Pep27 did not induce autolysis or viability loss in these studies. Collectively, these findings challenge the earlier reports that Pep27 and the VncS/R system function as a general effector of self-induced cell death (Robertson et al., 2002). It is unclear why these two studies obtained such strikingly opposing data. However, one partial explanation for this discrepancy may lie in the finding that a reproducible tolerance to vancomycin was ob-served after exposure of the bacteria to sublethal concentrations of erythromycin (Robertson et al., 2002), levels that could easily have been reached by carry-over during subculturing.
The role of bacterial holin-like proteins in controlling murein hydrolase activity (and PCD?)
In recent years, studies in our laboratory have focused on the regulation of murein hydrolase activity in Staphylococcus aureus. Initially, we identified a novel two-component regulatory system, lytSR, which was shown to affect the activity of several murein hydrolases and negatively regulate autolysis in the presence of penicillin and Triton X-100 (Brunskill and Bayles, 1996a,b). Subsequent studies revealed that this effect, at least in part, resulted from expression of the downstream lrgAB operon, transcription of which is activated by lytSR (Brunskill and Bayles, 1996a; Groicher et al., 2000). Specifically, a lrgAB mutant displayed increased extracellular murein hydrolase activity, indicating that this operon inhibits the expression or activity of these enzymes (Groicher et al., 2000). The lrgAB mutant also displayed greater susceptibility to penicillin in stationary phase, and expression of the lrgAB genes from a plasmid in both mutant and wild-type strains conferred increased penicillin tolerance in both exponential and stationary phase (Groicher et al., 2000).
The LrgA and LrgB proteins do not share significant sequence similarity with proteins of known function. However, the predicted secondary structure of LrgA contains characteristics shared with the bacteriophage holin family of proteins, including a relatively small size, two or more putative membrane-spanning domains, a polar N-terminal sequence and a charge-rich C-terminal domain (Young and Blasi, 1995; Groicher et al., 2000; Wang et al., 2000). Holins are small membrane proteins that control the timing and onset of bacteriophage-induced cell lysis, and have been reviewed extensively elsewhere (Young and Blasi, 1995; Wang et al., 2000; Young et al., 2000; Young, 2002). The most well-characterized member of the holin family is the lambda S holin, which oligomerizes within the cell membrane and allows access of the bacteriophage-encoded endolysin (murein hydrolase) to the cell wall substrate. Recent work on the S holin has suggested that these ‘holes’ cause a generalized membrane disruption rather than a regular oligomeric membrane pore (Wang et al., 2003). Intriguingly, holin-induced lethality can also occur independently of endolysin activity, and holin-induced viability loss occurs before cell lysis (Garrett and Young, 1982). This is postulated to result from collapse of the proton motive force (pmf) several seconds before host cell lysis (Grundling et al., 2001). Premature lysis mediated by the S holin is prevented by the presence of a second protein, termed the ‘antiholin’, which differs from the holin by only two or three amino acids at its N-terminus and functions as an inhibitor of cell lysis (Young et al., 2000). Translation of both the S holin and antiholin are controlled by a dual-start motif present in the holin gene, and the relative proportion of these two proteins is thought to regulate the timing of bacteriophage-mediated lysis (Young et al., 2000).
As discussed above, the S. aureus lrgAB operon negatively regulates murein hydrolase activity and promotes tolerance to penicillin. Given the structural similarity of LrgA to the holin family of proteins, we hypothesized that one or both of the lrgAB gene products function in a manner analogous to a bacteriophage antiholin, and proposed that a second operon, encoding the ‘effector holin’ component of this system, must also be present on the S. aureus chromosome (Brunskill and Bayles, 1996a; Bayles, 2000; Groicher et al., 2000). Indeed, we have recently identified and characterized what we believe to be the effector holin-like component of this system (Rice et al., 2003). This gene locus, designated cid, also contains two genes, cidA and cidB, with gene products that display 23% and 31% amino acid sequence identity with LrgA and LrgB respectively. Like the LrgA protein, CidA contains the characteristic features of the holin protein family. Furthermore, a cidA mutant displayed decreased extracellular murein hydrolase activity and increased tolerance to penicillin (Rice et al., 2003). This phenotype is opposite to that displayed by the previously characterized lrgAB mutant, supporting the idea that the cid locus encodes the holin-like component of this system (Rice et al., 2003).
It is intriguing to note that both penicillin G and cefotaxime (a third-generation cephalosporin) have been shown to depolarize the cell membrane in Streptomyces griseus (Penyige et al., 2002). Similar effects have also been shown to occur when S. aureus and Micrococcus luteus are treated with amoxicillin or streptomycin (Novo et al., 2000). Furthermore, Moreillon et al. (1990) speculated that the Cid mutation they generated in S. pneumoniae (conferring increased tolerance to penicillin) affected a gene with a product that would cause injury to the cell membrane when present at large concentrations, and that the cell death they observed in completely amidase-deficient cells was mediated by this proposed mechanism. Collectively, these results suggest that the loss of membrane potential could be a precursor to autolysis, similar to bacteriophage-mediated lysis (Penyige et al., 2002). In a similar vein, the regulation of murein hydrolase activity and autolysis in B. subtilis is proposed to be dependent on the energized state of the membrane and the resulting pH of the local cell wall environment (Jolliffe et al., 1981; Calamita et al., 2001; Calamita and Doyle, 2002). Specifically, loss of membrane potential would result in a leakage of protons from the cell, presumably altering the local cell wall pH and, in turn, regulating the activity of the murein hydrolases. Furthermore, the charged state of the bacterial membrane collapses in response to holin accumulation several seconds before host cell lysis (Grundling et al., 2001). These observations, in addition to our own study of the lrg and cid operons, have led to the proposal of a model in which the holin-like CidA/B and LrgA/B proteins may trigger PCD in response to penicillin and possibly other antibiotics and environmental stresses (Fig. 1). The validity of this model is currently under investigation in our laboratory.
Figure 1. Proposed holin model of CidA/B and LrgA/B function in bacterial PCD. In an unstressed cell, the lrgAB gene products are proposed to inhibit the function of CidA/B in a manner analogous to that of a bacteriophage antiholin. When a cell is stressed by the addition of an antibiotic such as penicillin, the CidA/B proteins are postulated to oligomerize within the bacterial membrane, creating lesions that disrupt the proton motive force (pmf), which in turn results in loss of cell viability. These lesions are also hypothesized to regulate the subsequent cell lysis by either allowing the murein hydrolases access to the cell wall substrate and/or regulating their activity by a possible change in the cell wall pH that results from loss of membrane potential.
Download figure to PowerPoint
Several other observations from our laboratory seem to support the idea that the cid and lrg operons also play a role in the PCD pathway of S. aureus, and possibly other bacteria. First, in addition to its effect on penicillin tolerance, preliminary studies have indicated that the S. aureus cidA mutation has a pleiotropic effect on antibiotic tolerance, as it appears to confer increased tolerance to other agents such as rifampin and mitomycin C (unpublished results). Secondly, a blast search of published microbial genomes has indicated the presence of at least one set of lrgAB homologues in many diverse bacterial species (Rice et al., 2003). The ubiquitous nature of these genes among a wide variety of bacteria suggests that they are probably important in bacterial physiology. Thirdly, recent work in our laboratory has shown that cidB and a downstream gene, designated cidC, comprise a transcriptional unit that is upregulated in S. aureus strains with a functional σB response (unpublished results). This suggests that one or more components of the cid operon may contribute to the bacteria's general stress response to both environmental signals and antibiotics. Finally, there is a striking functional similarity of the holin family of proteins to the eukaryotic Bcl-2 family of apoptosis regulatory proteins (Bayles, 2003). This ubiquitous family of membrane proteins includes both proapoptotic members such as Bax and Bid and antiapoptotic members such as Bcl-2 and Bcl-xL. In a manner analogous to bacteriophage holins (and possibly CidA), Bax functions to stimulate apoptosis by oligomerizing within the mitochondrial membrane and leading to membrane depolarization (Antonsson et al., 2000). This event in turn leads to the release of cytochrome c and subsequent triggering of the caspase cascade, the central effector of apoptosis. In addition, Bcl has been shown to inhibit the oligomerization of Bax and subsequent membrane depolarization, analogous to the action of an antiholin (and possibly LrgA) (Bayles, 2003).