Genes encoding toxin-antitoxin proteins are frequently found on plasmids where they serve to stabilize the plasmid within a bacterial population. The toxin-antitoxin proteins do not increase the likelihood of a progeny cell receiving a plasmid but rather function as post-segregational killing mechanisms which decrease the proportion of cells that survive after losing the plasmid. These toxin-antitoxin couples therefore act as plasmid addiction systems. Several new proteic toxin-antitoxin systems have been identified and these systems appear to be ubiquitous on the chromosomes of bacteria and archaea. When placed on plasmids, these chromosomal systems also have the ability to stabilize plasmids and in at least one case, chromosomal- and plasmid-based toxin-antitoxin systems have been shown to interact. Recent findings regarding toxin-antitoxin systems and questions that have arisen as a result of these findings are reviewed.
Bacterial plasmids are extrachromosomal elements which are not essential for the survival of their host cells and therefore may become lost at cell division. Many plasmids possess mechanisms which help to reduce their loss to well below the rate predicted from their copy number. Several mechanisms that reduce the plasmid loss have been discovered. For example, the par region of prophage P1 and the sop of plasmid F depend on the presence of a centromere-like region that appears to result in the pairing of plasmids that share this region [1,2]. During cell division, there is an active distribution of one member of the pair into each of the progeny cells. Site-specific recombination systems such as the parCBA system of RK2 (mrs/par locus) or the cer-xer of ColE1 ensure that plasmid multimers that arise during replication are resolved . This maximizes the number of independent units for segregation at cell division. Another type of mechanism are the plasmid addiction systems which result in the killing of progeny cells that have lost the plasmid. These post-segregational killing systems consist of two essential components, a stable toxin (or poison) and an unstable antitoxin (antidote). The toxins are usually proteins whereas the antidotes may be either proteins or antisense RNA [3–5]. Examples of systems where both toxin and antitoxin are proteins are the ccd system of plasmid F, the parDE system of RK2/RP4, the identical parD and pem systems of plasmids R1 and R100 and the phd/doc system of P1 (reviewed in ). The best studied example where the antitoxin is an antisense RNA molecule is the hok/sok system of plasmid R1 (reviewed in [5,6]). An unusual RNA-toxin RNA-antisense antidote system has been reported for the Gram-positive Enterococcus faecalis plasmid pAD1 .
Common features of the proteic toxin-antidote plasmid addiction systems
Proteic toxin-antidote plasmid stability systems usually consist of only two proteins, a long-lived toxin which is expressed at low levels and a short-lived, highly expressed antidote . On cell division, if a progeny cell fails to inherit a plasmid, it is nevertheless likely to inherit some of the cytoplasmic toxin-antitoxin complexes. Plasmid-free cells are unable to synthesize the unstable antitoxin and are therefore not able to counter the toxic effects of the stable toxin (Fig. 1). Cell killing is dependent on the differences in stability of the antitoxin relative to the toxin. A high turnover of antitoxins is due to their degradation by cellular proteases such as Lon or Clp [7,8]. Cellular targets of two of the toxins are known. The CcdB toxin (also called LetD) of plasmid F inactivates the Escherichia coli host gyrase by trapping the gyrase in an inactive DNA complex [9,10]. The crystal structure of CcdB has been determined and it has been proposed that the CcdB dimer binds to the central hole of the N-terminal portion of GyrA . CcdA (LetA) antidote is able to displace the CcdB toxin from the gyrase subunit A forming a CcdA-CcdB (LetA-LetD) complex and the gyrase is rejuvenated . The PemK toxin of plasmid R100 is believed to function as an inhibitor of DnaB preventing the initiation of DNA replication . Evidence has been obtained to suggest that the primary effect of the pem system is to inhibit cell division rather than killing of plasmid-free segregants . The most important features of the toxin and antidote proteins of the ccd system of plasmid F, the parDE system of RK2/RP4, the parD and pem systems of plasmids R1 and R100 and the phd/doc system of P1 have been reviewed by Jensen and Gerdes . Other plasmid toxin-antitoxin systems have been found on plasmid R485 from Morganella morganii and virulence plasmid pMYSH6000 of Shigella flexerni. In the absence of plasmid addiction systems, it is conceivable that cells which have become free of the metabolic burden imposed by the plasmid may outgrow plasmid-containing cells and the plasmid could become lost from the population.
Several workers have pointed out that DNA restriction-modification systems also constitute a proteic toxin-antitoxin couple. When present on plasmids, these toxin-antitoxin couples have the effect of stabilizing the plasmid within a population because loss of the methylase ‘antitoxin’ results in a loss of DNA modification and lethal attack by the restriction enzyme [17,18]. Likewise, bacteriocins (such as colicins) and their immunity proteins are found on many plasmids and constitute a toxin-antitoxin couple which kills or inhibits related plasmid-free cells . A major difference between the colicins and other toxins is that the colicins are exported from cells and destroy plasmid-free cells from the outside. The effect of these systems is also to stabilize the plasmids on which they are present within a bacterial community .
The plasmid addiction system of plasmid pTF-FC2
Plasmid pTF-FC2 is a 12.2-kb broad host range, mobilizable plasmid that was isolated from a strain of the biomining bacterium Thiobacillus ferrooxidans. This strain formed part of the inoculum used for the pretreatment of a gold-bearing arsenopyrite concentrate from the Fairview mine (Barberton district, Mpumalanga, South Africa) . The plasmid has been sequenced and contains a replicon related to the IncQ plasmids, a mobilization region with a strong similarity to IncP plasmids and a transposon which is clearly Tn21-like . Located within the replicon between the genes for the RepB primase and the RepA helicase are three small genes, pasABC (see Fig. 2), which encode a plasmid addiction system .
The TF-FC2 pas is unique among toxin-antitoxin plasmid stability systems analyzed to date in that it consists of three rather than two proteins. PasA is an antitoxin, PasB is a toxin and PasC is a protein that appears to enhance the ability of PasA to neutralize the toxic effects of PasB. The size of the PasA antitoxin (74 amino acids (aa)) is of the same order as other antitoxin proteins (72–84 aa) and the PasB toxin (90 aa) is slightly smaller than other toxins (101–126 aa). In general, proteins of the toxin-antitoxin systems show a large amount of sequence variation and none of the toxin proteins has been reported to have a detectable sequence similarity. Antitoxin sequences are slightly more conserved with the PasA antitoxin providing the best example of this. The pTF-FC2 PasA antitoxin is poorly but clearly related (31% aa identity) to the ParD antidote of the parDE system of plasmid RK2 . The only other conserved plasmid-located antitoxins are the CcdA antitoxin of plasmid F and the Kid antitoxin of the ParD system of plasmid R1 which share 21% aa identity . PasC, the third protein of the pTF-FC2 plasmid addiction system is a 71-aa polypeptide with no detectable sequence similarity to any other protein. Inactivation of PasC by the introduction of a frame-shift mutation did not have a marked effect on the stability of pTF-FC2 in E. coli JM105 . An indication of the role of PasC came from a comparison of the toxicity of pasAB and pasABC constructs in which the pas genes were placed under the control of an IPTG-inducible tac promoter . The ability of the PasA antidote to neutralize the PasB toxin was greatly enhanced when PasC was present either in cis or in trans (on a co-resident plasmid). The conclusion was that pasAB antitoxin-toxin genes were on their own sufficient for plasmid stability but that E. coli cells containing only these genes would be slow-growing compared to plasmid-free cells. The function of PasC was to reduce the overall toxicity of the system and permit cells containing pas to compete with plasmid-free cells.
Efficiency of addiction systems and strain-dependence
In general, the active partitioning systems are far more effective for plasmid stabilization (1000-fold for sop of F) than the toxin-antitoxin systems . A comparison of four systems using a mini-R1 conditional replication system was carried out in an E. coli CSH50 host strain. A >100-fold stabilization was conferred by parDE of RP4 (RK2), 100-fold by the hok/sok system of R1 and 10-fold by the ccd of F and parD of R1 . Using the same system and the same E. coli host strain, a 100-fold stabilization was achieved by the pas of pTF-FC2. However, if different E. coli host strains were used, stabilization by pas was also different. There was only a 2.5-fold increase in stability in E. coli JM105 with no increase detected in E. coli strains JM107 and JM109 . The reason for strain variation in the stability is unknown. However, the strains in which pas was most effective were also the strains in which PasB was most toxic.
Operon structure and autoregulation
The genetic organization of antitoxin-toxin systems is usually similar to the genes for antidote and toxin located in an operon with the gene for antidote preceding that for the toxin. An exception to this is the higB-higA system of plasmid Rts1 where the order of toxin-antitoxin is reversed . Autoregulation at the level of transcription is a general property of toxin-antidote sytems. For example, the ccd system of plasmid F is regulated by a CcdA-CcdB complex of 69 kDa and neither CcdA or CcdB alone is capable of autorepression . In contrast, the parDE system of RK2 is autoregulated solely by the ParD antitoxin . In other cases, although the antitoxin can serve as a repressor on its own, a combination of toxin and antitoxin is required for full repression. The parD locus of plasmid R1 is repressed only 30–40% by Kis on its own, but full repression occurs with the complete Kis-Kid complex . Regulation of the pTF-FC2 pas genes was found to be similar. The PasA antidote was able to repress the operon 25-fold but when PasA and PasB were both present, repression was enhanced to 100-fold. PacC had no detectable regulatory activity .
A question posed by Jensen and Gerdes  is whether autoregulation is essential for the function of plasmid addiction systems. It is conceivable that differences in half-lives of the antitoxin and toxin proteins together with differences in the level of toxin and antitoxin translation may on their own be sufficient to increase the plasmid stability. To test this, the promoter of the pas system of pTF-FC2 was removed and the pas genes were placed behind an IPTG-inducible tac promoter . The modified pas system was placed in a low copy number mini-R1 (pOU82) test plasmid system developed in the Gerdes laboratory . The stability of the tac-regulated pas-containing test plasmid with and without induction by IPTG was compared with a pOU82 control and pOU82 containing the unmodified autoregulated pas genes. Even without IPTG induction, the tac-regulated pas-containing plasmid was less stable than the test plasmid which lacked a plasmid addiction system. On IPTG induction, this instability increased still further. The conclusion was that an autoregulated control circuit was required for the proper functioning of the pTF-FC2 pas.
Relationship between plasmid toxin-antitoxin addiction systems
Gross similarities in the operon structure, function and regulation of plasmid addiction systems have led to the suggestion that they originated from a common ancestor . The discovery of a weak amino acid sequence similarity between the CcdA and Kid antitoxins of plasmids F and R1 as well as between the ParD and PasA antitoxins of plasmids RK2 and pTF-FC2 (see earlier) supports this view. More noticeable than the low levels of similarity is the great amount of toxin and antitoxin amino acid sequence divergence between addiction systems of different plasmids. Magnuson and Yarmolinsky  have suggested that this diversity could be explained by multiple origins of toxin-antitoxin genes, great age or a fast rate of change. The argument for a fast rate of change resulting in rapid sequence divergence is as follows. Once a toxin-antitoxin system has been acquired by a plasmid, it is likely to increase the fitness of that plasmid when competing for survival with other plasmids, especially related incompatible plasmids. Under such conditions, there is likely to be selective pressure for each plasmid to acquire its own copy of a toxin-antitoxin system. However, the value of such a system would decrease with an increasing frequency of occurrence. This would place strong selection for a rapid divergence in the components of plasmid addiction systems. A similar strong positive selection for divergence has been proposed for colicins .
Location of addiction systems on plasmids
It is possible that plasmid addiction systems may be more effective when located at a particular position on a plasmid relative to important functions like the origin of replication or the origin of conjugal transfer (oriT). For example, an advantage for an addiction system to be located in the leading region following the origin of transfer is that during conjugation, the addiction genes would enter the recipient first. This would enhance the possibility of an addiction system being expressed early in the transfer process and minimize the ability of the host to survive should it be successful in eliminating the plasmid on entry (for example though the use of a restriction endonuclease). Anti-restriction proteins have been found to occur in the leading region of certain plasmids so that during conjugal transfer, their early expression assists in plasmid establishment . Interestingly, the hok-sok protein-antisense RNA addiction system of plasmid R1 (R100, NR1) and an equivalent system of plasmid F occupy a similar position to the anti-restriction proteins . Examination of the location of proteic plasmid addiction systems has not revealed a pattern of placement. For example, the parDE genes of the IncPα plasmids RP4/RK2 are positioned close to the multimer resolution active partition system. This is 15 kb from the oriT and 17 kb from the origin of replication of the 60-kb plasmid . In at least three plasmids, the toxin-antidote system is located close to the origin of replication. The ccd system of plasmid F is positioned between ori-1 and ori-2 and about 20 kb from the oriT site , while the genes for the Kis/Kid (PemI/PemK) proteins are within 2 kb of the origin of replication of plasmid R1 (also called NR1) [34,37].
In the case of pTF-FC2, the pas is also closely associated with the IncQ-like replicon and occurs between the repB (primase) and the repA (helicase) genes (Fig. 2). It may be questioned whether the pas is situated in this position by chance or whether there is an advantage to this placement. Circumstantial evidence suggests that the location of pas on pTF-FC2 may be by chance. When the pasA gene (antitoxin) was inactivated through the introduction of a frame-shift mutation, E. coli cells containing the construct produced small, slow-growing colonies presumably because of the effects of the unneutralized toxin . However, occasionally, large strong-growing colonies were produced from these mutants and the region containing the pas system of plasmids isolated from these cells was sequenced. Two types of deletion were detected, one in which pasAB and most of pasC was missing and the other in which the promoter region of pas was deleted. In both cases, the plasmid copy number was unchanged. Therefore, in spite of their presence between repB and repA, the pas genes play no role in plasmid replication. This is in contrast to the IncQ plasmids (RSF1010, R1162) where one of the two small genes (cac) situated between repB and repA has been shown to be a regulator of plasmid replication . This pair of genes does not also function as a toxin-antitoxin plasmid addiction system . It has been suggested that plasmids are composed of modules that have been acquired independently of each other. The location of the pTF-FC2 pas supports this view as the toxin-antitoxin module appears to have become inserted within the replicon in a way that has not interfered with plasmid replication.
Chromosomally located proteic toxin-antidote genes
Several toxin-antitoxin genes of which both products are proteins have been identified on the chromosome of E. coli. These include the chpAI–chpAK genes located downstream of the relA locus and the chpBI–chpBK genes found in the ppa region . When cloned into plasmids, these genes result in plasmid stabilization by post-segregational killing of plasmid-free cells in a manner similar to plasmid-based systems. Not only are the chromosomal chpA and chpB systems related to the parD/pem systems of plasmids R1/R100 but it has also been shown that these systems can functionally interact . The chpAI–chpAK gene products appear to be homologues of the Kis-Kid (PemI–PemK) proteins of plasmid R1 (R100). Cells containing a temperature sensitive Kis antitoxin mutant were unable to grow due to the inability to neutralize the Kid toxin at 42°C. In the presence of the ChpAI antitoxin, growth at 42°C was partially restored. This partial functional complementation occurred even though the sequences of the antitoxin proteins have only 22% amino acid identity. In vitro mutagenesis was used to isolate mutants with an improved ability to neutralize the Kid toxin. These mutations were all located in the N-terminal region of the antitoxin and did not alter the amino acid sequence identity between the ChpAI and Kis antitoxins . This observation indicates that although proteins of the plasmid and chromosomal toxin-antitoxin systems might appear to be dissimilar, they may nevertheless be able to functionally react with each other. This may be taken as evidence of a common ancestry for chromosomal and plasmid toxin-antitoxin systems which is not apparent from direct sequence comparisons.
More recently, it has been demonstrated that the relBE genes present on the chromosome of E. coli K-12 belong to a new family of toxin-antitoxin genes . These authors demonstrated that relE encodes a toxin and that relB encodes an antitoxin. relBE stabilized a mini-R1 test plasmid when present in an E. coli relBE chromosomal mutant and relBE was autoregulated with the RelB antitoxin serving as repressor and the RelE toxin as co-repressor. Database searching revealed that a second copy of related genes (dinJ and yafQ) was present on the chromosome of E. coli K-12 and that relBE gene equivalents were present on the chromosomes of Haemophilus influenzae, Vibrio cholerae and an enterotoxin encoding plasmid p307 . As with other toxin-antitoxin systems, the degree of amino acid identity of the relBE family was rather low with the degree of RelB amino acid sequence identity ranging from 20 to 48% and that of RelE from 14 to 55%. Besides the Gram-negative bacteria, RelBE-like systems have been found to be widespread on the chromosomes of Gram-positive bacteria and archaea .
The role of toxin-antitoxin systems
As described earlier, poison-antidote plasmid stability mechanisms are rather ineffective compared to active plasmid stability mechanisms . Furthermore, the killing of plasmid-free segregants would represent a reproductive loss for the plasmid and therefore be less favorable than an active partitioning system. This raised the question whether plasmid stability is their only function. Discovery of the role of toxin-antitoxin systems on the chromosomes of bacteria may indicate alternate roles for toxin-antitoxin couples and possible origins of the plasmid toxin-antitoxin systems.
There are clear indications that at least some chromosomal proteic poison-antitoxin systems may play a role in the stringent-relaxed response which occurs when bacteria such as E. coli face amino acid starvation. In their discussion on the role of RelBE, Gotfredsen and Gerdes  describe previous studies which demonstrated that E. coli relB mutants exhibit the so-called delayed relaxed response. This is based on the observation that during amino acid starvation, stable RNA synthesis stops, but then resumes after a delay of about 10 min. It was speculated that the RelE toxin may be a protein synthesis inhibitor and that inactivation of RelB antitoxin may allow active RelE to decrease protein synthesis. This would in turn lead to recharging of tRNA and result in the delayed relaxed response observed.
Expression of the E. coli relA-associated chromosomal chpA locus (also called mazEF) has also been linked to the E. coli stringent response . The promoter of mazEF is inhibited by a (p)ppGpp-dependent mechanism. Amino acid starvation might lead to the induction of mazEF and cell killing . Killed cells would release nutrients which would permit other members of the culture to survive until conditions improved. It is interesting that the product of the bacteriophage λrexB gene, one of the few to be expressed by λ when it is in the lysogenic state, inhibits degradation of the MazE antitoxin . This allows phage λ to counter the effects of the mazEF system and to survive under conditions of nutrient stress.
Evolution of plasmid addiction systems
Given current information, one may speculate on how plasmid toxin-antitoxin addiction may have evolved (Fig. 3). Whatever the function of chromosomal toxin-antitoxin systems is found to be, such systems could have provided the source of genes for plasmid addiction systems. It is well-established that as a result of transposon or insertion sequence activity, genes are able to move fairly readily between chromosomes and plasmids. It may be speculated that genes which have a physiological function on the chromosome have been captured by plasmids where they provide a selective advantage for the plasmid. Once captured on a plasmid, the addiction systems would be under a strong positive selective pressure to diversify to ensure that the advantages to be gained from the system were available to a particular plasmid and not to competing plasmids. The promoter regions may then have been modified so as to become independent of regulation by cellular functions such as the stringent response, thereby becoming more suited to their role in enhancing plasmid survival. It would be interesting to test whether any of the plasmid-based systems retains remnants of control by (p)ppGpp and stringent response regulators. Evidence that the plasmid Kis/Kid and chromosomal ChpAI-KI systems can interact supports this evolutionary model. It is possible that the acquisition of toxin-antitoxin genes occurred in the reverse direction from plasmids to the chromosome. However, since plasmid addiction systems are not very efficient in plasmid stabilization, this possibility is less likely.