The ability of a high frequency (10−2) of Escherichia coli to survive prolonged exposure to penicillin antibiotics, called high persistence, is associated with mutations in the hipA gene. The hip operon is located in the chromosomal terminus near dif and consists of two genes, hipA and hipB. The wild-type hipA gene encodes a toxin, whereas hipB encodes a DNA-binding protein that autoregulates expression of the hip operon and binds to HipA to nullify its toxic effects. We have characterized the hipA7 allele, which confers high persistence, and established that HipA7 is non-toxic, contains two mutations (G22S and D291A) and that both mutations are required for the full range of phenotypes associated with hip mutants. Furthermore, expression of hipA7 in the absence of hipB is sufficient to establish the high persistent phenotype, indicating that hipB is not required. There is a strong correlation between the frequency of persister cells generated by hipA7 strains and cell density, with hipA7 strains generating a 20-fold higher frequency of persisters as cultures approach stationary phase. It is also demonstrated that relA knock-outs diminish the high persistent phenotype in hipA7 mutants and that relA spoT knock-outs eliminate high persistence altogether, suggesting that hipA7 facilitates the establishment of the persister state by inducing (p)ppGpp synthesis. Consistent with this proposal, ectopic expression of relA′ from a plasmid was shown to increase the number of persistent cells produced by hipA7 relA double mutants by 100-fold or more. A model is presented that postulates that hipA7 increases the basal level of (p)ppGpp synthesis, allowing a significantly greater percentage of cells in a population to assume a persistent, antibiotic-insensitive state by potentiating a rapid transition to a dormant state upon application of stress.
In the 1940s, shortly after the introduction of penicillin, it was found that populations of Staphylococcus contain a very small fraction of cocci that survive prolonged exposure to penicillin after the vast majority of cells are killed (Hobby et al., 1942; Chain and Duthie, 1945). At the time, these surviving bacteria were recognized as an important contributor to therapeutic failure of penicillin, leading to recolonization of wound infections after antibiotic therapy was withdrawn. When recrudescence occurred, the bacteria isolated from the infected site were as fully susceptible to penicillin as the original cultures obtained from the wound. Bigger (1944) was the first to examine this phenomenon and demonstrated that approximately one in a million Staphylococcus cells was able to survive the lethal effects of penicillin. He also showed that surviving bacteria produced offspring that were as susceptible to penicillin as the parental culture, prompting him to differentiate the penicillin-sensitive survivors from ‘resistant’ bacteria by designating them ‘persisters’. Persisters, when exposed to penicillin, were suggested by Bigger to be ‘in a phase in which they are insusceptible to its action’. It was also proposed that persister cells represented dormant, non-dividing cells. In the past few years, it has become apparent that many bacterial species are able to generate persister cells as a survival mechanism when exposed to a variety of stress conditions. In particular, the extraordinary antibiotic resistance of biofilms has been attributed, in part, to the presence of persister cells (Lewis, 2000; Gilbert et al., 2002; Stewart, 2002).
In the 1980s, Harris Moyed began to investigate the molecular basis of persistence in Escherichia coli. Moyed screened for mutants that showed a higher frequency of persistence than wild-type strains and isolated the ‘high persistence’ mutants, which included the hipA7 mutant. The hipA7 mutant, when exposed to ampicillin, generated persisters at a frequency of 10−2 compared with the 10−6−10−5 frequency typically seen in wild-type cultures (Moyed and Bertrand, 1983). A very important discovery of Moyed's was the demonstration that cells exhibiting the high persistence phenotype were also capable of suppressing the lethal effects of several different stresses besides inhibition of peptidoglycan synthesis. In particular, hipA7 mutants showed greatly increased survival after inhibition of DNA synthesis by thymine starvation or after treatment with naladixic acid, a quinolone (Scherrer and Moyed, 1988). The hipA7 allele also increased survival of dnaAts, dnaBts or dnaCts mutants after incubation at the non-permissive temperature and rescued rpoH (htpR) mutants from prolonged heat shock (Scherrer and Moyed, 1988). In addition, hipA7 mutants demonstrated a cold-sensitive phenotype, with inhibition of cell division occurring at 20°C.
The high persistence phenotype was mapped to a single locus on the E. coli genome at approximately minute 34 (Moyed and Bertrand, 1983), a position directly opposite oriC in the chromosome near dif. The hip locus consists of an operon containing two genes, hipA and hipB (Moyed and Broderick, 1986; Black et al., 1991). hipB is transcribed first in the operon and encodes a small DNA-binding protein with a helix–turn–helix DNA-binding motif. The HipB protein acts as a transcriptional repressor and autoregulates hip operon expression by binding to four sites in the promoter region (Black et al., 1991). HipB also binds to the HipA protein, forming a tight complex in a one-to-one molar ratio (Black et al., 1994). Because it is not possible to obtain knock-outs of hipB alone and because HipB binds to HipA, it was proposed that HipA is toxic (Moyed and Broderick, 1986; Black et al., 1994) and that the HipA–HipB proteins possibly constitute a toxin–antidote module reminiscent of post-segregational plasmid stability systems (Falla and Chopra, 1999).
In an attempt to understand further how mutations in hipA might contribute to the establishment of the persistent state, we characterized the hipA7 allele and investigated the possibility that hipA mutants activate a generalized stress response in bacteria to confer persistence. Our studies revealed that the hipA7 allele contains two mutations, G22S and D291A, and unlike the wild-type allele, the hipA7 allele produces a non-toxic protein. Expression of hipA7 from a plasmid in a ΔhipBA background confers the same high level of persistent bacteria as found in Moyed's original mutants, suggesting that HipB plays no part in the establishment of the persistent state. When comparing persistence in different phases of growth, we discovered that the frequency of persisters in wild-type cells remained fairly constant throughout a wide range of cell densities, whereas hipA7 mutants generated a significantly higher frequency of persisters as the density of the culture increased. Finally, we demonstrate that loss of relA and spoT, which are essential for (p)ppGpp production, correlates with loss of the high persistent phenotype in hipA7 mutants and that expression of relA′ from a plasmid increases the frequency of persisters by 100-fold or more. We speculate that the presence of HipA7 primes bacteria for persistence by increasing the basal rate of (p)ppGpp synthesis, potentiating a rapid transition to quiescence under conditions of stress that ensures long-term survival of these non-dividing cells in the presence of certain antibiotics.
The high persistence phenotype in hipA7 mutants
The original strain encoding the hipA7 mutant (HM22) and its isogenic parent (HM21) are derived from AT984 (Moyed and Bertrand, 1983), which is a dapA mutant of KL16 (Bukari and Taylor, 1971). As far as can be ascertained, AT984 has undergone a minimum of 13 mutagenic treatments (Bachmann, 1996). To eliminate extragenic mutations in the chromosome that could potentially complicate the analysis of the hipA7 allele and to determine whether the high persistence phenotype could be reproduced in a more wild-type genetic background than in the strains originally used by Moyed, strains TH1268 hipA+, TH1269 hipA7 and TH1433 ΔhipBA were constructed from MG1655. The frequency of persister cells generated in these strains was determined by removing samples from an exponentially growing culture at a defined OD600 (in initial experiments, ≈ 0.3), diluting the cells and spreading aliquots on either LB plates (to determine cfu) or LB-ampicillin plates. After 24 h incubation at 37°C, the LB-ampicillin plates were sprayed with a fine mist of penicillinase and incubated overnight to allow persister cells to grow (Fig. 1A). Under these growth conditions, MG1655 derivatives carrying either hipA+ or ΔhipBA yielded persister cells at a frequency of ≈ 10−5 or less, whereas the isogenic strain carrying the mutant hipA7 allele produced persister cells at a frequency of ≈ 10−3, corroborating the original results obtained by Moyed. The observation that hipA7 demonstrates high persistence but hipA null mutants do not suggests that the high persistence phenotype is a gain-of-function associated with the hipA7 allele.
At the time of submission of this manuscript, the high persistence assay had been performed in this laboratory a total of 34 times using an OD600 of ≈ 0.3 as the harvest point. In the vast majority of experiments, the frequency of persister cells in hipA+ or ΔhipBA strains was observed to fall between the values of 1 × 10−6 and 9 × 10−5, with the hipA7 strain generating frequencies of 10- to 100-fold higher. The variability in persistence frequency observed with hipA+ or ΔhipBA strains appears to be consistent with Moyed's published values, as he typically reported a frequency for hipA+ strains as ‘10−6−10−5’ rather than reporting an exact frequency (Moyed and Bertrand, 1983; Black et al., 1991). However, in five experiments we conducted, the frequency of persisters in wild-type hipA or ΔhipBA strains was as much as 100-fold higher (10−4−10−3). The different factors that influence reproducibility of the persistence frequency are not yet known but, acting on the advice of Professor Kim Lewis (Northeastern University), we now know that the degree of aeration of the cultures improves, but does not entirely eliminate, the reproducibility problem. We are currently attempting to identify other factors that influence reproducibility.
One factor that we considered might influence reproducibility was cell density, especially in view of recent studies suggesting that quorum sensing may play an important role in the development of persistence in Pseudomonas (Spoering and Lewis, 2001). To investigate this possibility, the frequency of persistence in wild type and hipA7 mutants was determined at different cell densities in the growth cycle. As can be seen in Fig. 1B, the frequency of persisters in wild-type or ΔhipBA strains remained fairly constant or even dropped during the logarithmic stage of growth (from OD600 of 0.1 to 1.5), whereas in the hipA7 mutants TH1269 and TH1604, the frequency increased by more than 20-fold as the bacteria reached high cell density.
Wild-type HipA is toxic at low levels of expression
Moyed was the first to suspect that wild-type HipA may be toxic, as he was unable to clone the hipA gene when it was proximal to strong promoters and unable to introduce plasmids expressing hipA into strains lacking hipB (Moyed and Broderick, 1986; Black et al., 1991). Indeed, high-level expression of wild-type hipA can be toxic to cells, even when a functional hipB gene is present (Black et al., 1991; Falla and Chopra, 1998). On the other hand, intermediate levels of hipA induction in the presence of hipB only appeared to reduce cell growth (Falla and Chopra, 1998). To study the toxic effects of hipA using a well-controlled inducible gene expression system and to avoid the problems associated with very high levels of expression, we inserted the hipA alleles into the pBAD series of vectors (Guzman et al., 1995).
We encountered little trouble obtaining colonies when transforming strain TH1273 (hipA::kan) or TH1256 (ΔhipBA::res–npt–res) with either hipA+- or hipA7-containing pBAD vectors. However, a wide variation in growth characteristics was observed when transformants containing the wild-type hipA gene were restreaked and then transferred to LB broth for subsequent growth. In particular, a distinct difference was observed between TH1273 and TH1256 derivatives harbouring the wild-type hipA gene. The TH1273 derivatives grew normally in LB-glucose, but a definite lag in growth was observed when the LB was supplemented with arabinose to induce hipA expression (Fig. 2A). The bacteria eventually overcame this lag and resumed growth, reaching normal stationary phase cell densities. With the TH1256 derivatives harbouring pBAD24hipA, the growth patterns of independent colonies in LB were wildly different, even when bacteria were cultured in the presence of 0.4% glucose (Fig. 2B). Addition of arabinose to the culture medium of some colonies caused cessation of growth but, surprisingly, in others, the growth patterns in glucose and arabinose were not all that different. The variation in growth of hipA-expressing colonies could be improved if cells were grown in M9 minimal medium supplemented with glucose. In these conditions, colonies often grew to near normal cell densities with little apparent lysis. When the minimal media was supplemented with arabinose, cultures containing TH1256/pBAD24hipA underwent clearing, suggesting that cell lysis was taking place (Fig. 2C). Even though M9 minimal medium improved growth characteristics, it did not entirely eliminate variation in the growth pattern of colonies, and significant differences in growth were still observed from colony to colony. The basis for the differences in growth of individual colonies, in both LB and minimal medium, can be attributed to variations in expression from the ParaBAD promoter, which is known to differ significantly from cell to cell at subsaturating concentrations of inducer (Siegele and Hu, 1997). In summary, these results indicate that low-level expression of hipA from the PBAD promoter, even in the presence of glucose, is sufficiently toxic to affect cell growth or kill cells.
Microscopic examination of samples from several different TH1258 cultures (ΔhipBA with pBAD24hipA) revealed many filaments and cell ‘ghosts’, even when cells were grown in LB-glucose. When arabinose was added to the culture, virtually all cells formed filaments and lysed, with some cells developing a mid-filament bulge before lysis (Fig. 3D). It should be noted that TH1258 colonies grown in LB supplemented with glucose were usually able to reach normal cell densities (see Fig. 2B), presumably because some cells in the population consistently expressed the hipA gene at sublethal concentrations. We draw this conclusion because subsequent examination of these overnight cultures demonstrated that these bacteria still retained a toxic copy of the hipA gene, as the bacteria were unable to grow or form colonies when streaked onto LB-arabinose plates.
In contrast, growth patterns of MG1655 or TH1256 (ΔhipBA) derivatives carrying a plasmid expressing hipA7 were unremarkable. Addition of arabinose to the cultures of TH1266 (ΔhipBA with pBAD33hipA7 ) had no effect on growth. Likewise, transformation of TH1433 (ΔhipBA) carrying the pBAD33hipA7 plasmid was equally successful on glucose or arabinose plates (Fig. 4), indicating that the mutant HipA7 protein was not toxic to cells even when HipB was absent. Finally, inspection of TH1266 samples by microscopy did not reveal filamentation, consistent with the loss of toxicity associated with this mutant.
The toxic effects of wild-type hipA expression were dominant when co-expressed in a ΔhipBA strain with the non-toxic mutant allele hipA7. This dominance was shown by transforming TH1256 with origin-compatible pBAD vectors expressing either wild-type hipA or hipA7 (i.e. pBAD24hipA and pBAD33hipA7) and plating transformants on arabinose plates (data not shown). The dominance of hipA toxicity over hipA7 is consistent with previous observations that the high persistence phenotype of hipA7 mutants was lost if a wild-type hipA gene was provided in trans (Moyed and Broderick, 1986).
Characterization of mutant hipA genes
The hipA gene is associated with three phenotypes that can be readily assayed: toxicity (wild-type hipA), high persistence (hipA7 ) and cold sensitivity (hipA7 ). It has been suggested that loss of toxicity might be a prerequisite for persistence, particularly if the HipA toxin mediates programmed cell death (Lewis, 2000). Also, Moyed was unable to separate the cold-sensitive and high persistent phenotypes of hipA7, suggesting that these two phenotypes were intricately linked (Scherrer and Moyed, 1988). To attempt to understand the inter-relationship of the different hipA phenotypes, we sequenced the hipA7 gene and assayed additional mutants of hipA to clarify the role of each phenotype.
The sequence of the hipA7 allele was determined from a polymerase chain reaction (PCR) product generated from the chromosome of the original hipA7 strain HM22 and from hipA7 genes cloned into pBAD vectors. Two mutations were identified in hipA7: G22S (G→A at the first position of the codon) and D291A (A→C at the second position of the codon). We subsequently used site-directed mutagenesis to introduce each mutation independently into hipA and characterized the resulting single-site mutants hipAG22S and hipAD291A with respect to toxicity, high persistence and cold sensitivity. We also characterized a mutant identified during the initial screening of PCR-amplified hipA genes, hipAD88N. In all cases, the characterized genes were expressed from a pBAD33 vector in a ΔhipBA background. A summary of the results is presented in Table 1.
Table 1. Characterization of wild-type and mutant hipA alleles for toxicity high persistence and cold sensitivity.
TH1601 WT hipA
. The frequency of persisters was determined by expressing the indicated hipA allele from a pBAD33 vector in TH1433 (ΔhipBA). For wild-type (WT) hipA and hipAD291A, which are toxic to the cell, these studies were conducted in glucose to minimize hipA expression and cell death. For the non-toxic alleles, the studies were conducted using both glucose and arabinose. Each value represents the average from at least three independent experiments.
We discovered that both G22S and D291A were required for full expression of all hipA7 phenotypes. G22S alone rendered the HipA protein non-toxic, but did not confer high persistence. Surprisingly, although D291A retained toxicity, at very low levels of expression (when cells were grown in glucose), it increased the frequency of persisters to the same level as the double mutant, indicating that this mutation alone was sufficient to confer persistence. The persister colonies that were generated by hipAD291A did not represent cells that had acquired a non-toxic protein, because subsequent restreaking of the hipAD291A persister colonies on arabinose plates prevented colony formation, indicating that the toxic form of the protein was still present. It was also determined that both G22S and D291A were required for the cold-sensitive phenotype associated with hipA7, as neither mutation conferred cold sensitivity by itself. Finally, a new hipA allele that was identified during our studies, hipAD88N, was also capable of conferring the high persistence phenotype, but was non-toxic and did not render cells cold sensitive.
Deletion of relA and spoT eliminates high persistence in hipA7 mutants
Any model for the role of hipA in high persistence must account for the ability of hipA7 mutants to demonstrate increased survival after exposure to several stress conditions besides ampicillin treatment, including fluoroquinilone treatment, thymine starvation or heat shock in hipA7 rpoH double mutants (Scherrer and Moyed, 1988). One key global response of E. coli that is known to induce ‘antibiotic tolerance’ and affect the expression of numerous genes involved in heat shock, DNA replication and cell division is the stringent response (Cashel et al., 1996). The stringent response is mediated by the alarmone (p)ppGpp, which is synthesized by relA and spoT gene products. To test the possibility that the high persistence phenotype requires (p)ppGpp production, both relA and relA spoT strains containing the hipA7 allele were assayed. As can be seen in Fig. 5, loss of relA alone correlated with a 100-fold decrease in the high persistence phenotype at high cell densities, but the frequency of persister cells in the relA strain was similar to that in the relA+ strain at low cell densities. In contrast, the ppGpp° strain TH1685, which lacks both relA and spoT, did not generate increased numbers of persisters at any cell density, indicating a collapse of the high persistent phenotype in the absence of (p)ppGpp production.
If production of (p)ppGpp is required for the establishment of persistence and constitutes the primary pathway by which cells enter into the persistent state, it follows that bacteria lacking relA and spoT would demonstrate an even lower frequency of persisters than wild-type cells. As shown in Table 2, this is exactly what we observed. Strains that were hipA+ but lacked either relA or relA spoT generated persister cells at a significantly lower frequency than the relA+spoT + strain.
Table 2. Effect of relA spoT on high persistence in hipA+ strains.
No. of cells screened
No. of persisters
Frequency of persisters
.P = 0.001.
.P = <0.001.
Tests for significance between the results obtained for MG1655 and the other two strains were performed by a chi-square analysis with Yate's correction for continuity.
Expression of the relA′ gene restores high persistence in hipA7 relA double mutants
As a final test of the role of (p)ppGpp synthesis in the generation of persisters, hipA7 mutants lacking relA were transformed with plasmid pHM675, which encodes the relA′ gene under the control of an IPTG-inducible Ptac promoter. The hipA7 relA double mutant (TH1686) was chosen for these studies because expression of the relA′ gene in the hipA7 relA spoT triple mutant is lethal (Sarubbi et al., 1988). The relA′ gene encodes a truncated form of the RelA protein that synthesizes (p)ppGpp independently of ribosome association and has a half-life of only 7.5 min compared with 2–3 h for full-length RelA (Schreiber et al., 1991). As a control for these studies, a plasmid in which most of the relA′ gene had been deleted (pHM675Δ) was also introduced into a hipA7 relA strain (TH1708) and into a hipA+relA strain (TH1709). The frequency of persisters generated by the three strains was then determined at different cell densities and with different concentrations of IPTG. As seen in Fig. 6B, the empty vector pHM675Δ had no influence on the frequency of persisters generated by the hipA7 relA or the hipA+relA strains at any concentration of IPTG or at any cell density. However, the presence of pHM675 increased the frequency of persisters in the hipA7 relA mutant by 10- to 100-fold even in the absence of IPTG, presumably because of leaky expression of the Ptac promoter. The addition of IPTG to 12.5 µM, which had little effect on the growth rate of the bacteria (Fig. 6A), increased the frequency of persisters in the TH1686 hipA7 relA strain to 2.8 × 10−1 at an OD600 of 1.5 compared with 10−4 for cells of the same genetic background carrying the empty vector. Addition of IPTG to 25 µM did not measurably increase the frequency of persisters above that seen with 12.5 µM, but did have a more pronounced effect on cell growth. Thus, increased synthesis of (p)ppGpp to a level that does not significantly alter the growth rate restores the ability of hipA7 relA mutants to generate very high frequencies of persisters.
HipA is well conserved in other Gram-negative bacteria
The existing literature suggests that hipA has homologues only in the Rhizobium symbiosis plasmid pNGR234 (genes Y4mE and Y4dM) (Falla and Chopra, 1999). However, a recent search of the microbial genome databases at NCBI identified 83 hits in more than 45 different bacterial species. hipA-like genes were identified in the chromosomes of the Gram-negative bacteria Erwinia chrysanthemi (52% identity over 423 residues), Vibrio vulnificus (47% identity over 430 residues), Burkholderium fungorum (two genes with 40% and 41% identity to E. coli hipA over 433 residues), Yersinia enterocolitica (41% identity over 435 residues), Ralstonia solanacearum (41% identity over 444 residues) and Rhodospririllum rubrum (41% identity over 425 residues). A search of the genome sequences of these organisms in the regions adjacent to the hipA genes revealed hipB-like genes just upstream in every case except B. fungorum. Other species showing hipA-like genes with reasonably high identity were Burkholderia mallei, Bordetella pertussis, Nitrosomonas europaea and Klebsiella pneumoniae. A clustalw alignment (Thompson et al., 1994) of hipA genes from several different species that also contain adjacent hipB-like genes is shown in Fig. 7.
It is clear from the alignment that certain regions of HipA are highly conserved across these divergent species. Presumably these regions correspond to domains required for protein function, such as binding to HipB or interaction with the cellular target of HipA, but what role HipA might play in the physiology of these organisms has yet to be determined. We also noted that residue D291 (position 301 in Fig. 7) is conserved in all species represented in the figure, and G22 (position 28) is conserved in all but Burkholderia fungorum.
Persistence is becoming increasingly recognized as an important survival strategy that bacteria use to escape the lethal effects of antibiotics. Although this phenomenon was first reported in the 1940s, the extremely small fraction of persister cells that arises in bacterial populations after antibiotic exposure made the problem intractable. More recently, the proposal that the extraordinary resistance of biofilms to antibiotic treatment may arise in part from the production of persisters (Brooun et al., 2000; Spoering and Lewis, 2001; Gilbert et al., 2002) has sparked a renewed interest in the persistence phenomenon and the mechanisms that govern the establishment of the persistent state. To date, the only well-characterized mutations known to increase the frequency of persistent cells are the hip mutants of E. coli, which were isolated by Harris Moyed and coworkers (Moyed and Bertrand, 1983). Moyed succeeded in identifying and characterizing the gene responsible for the ‘high persistence’ phenotype, hipA, but retired from science before uncovering the mechanism by which the high persistence phenotype is manifested.
We have sought to extend Moyed's observations on the role of the hip genes in persistence by further characterizing the wild-type and mutant hipA gene products and examining possible pathways by which hipA7 expression might induce the establishment of the persistent state. We have confirmed that the wild-type hipA gene encodes a toxin and report here for the first time that hipA7 encodes a non-toxic protein containing two mutations, G22S and D291A. Both mutations are required for the expression of all three hip phenotypes. The G22S mutation renders the normally toxic HipA protein non-toxic, but is unable to confer high persistence or cold sensitivity. The D291A mutation retains toxicity and confers high persistence, but does not confer cold sensitivity. We also demonstrate that strains carrying the hipA7 allele produce persister cells at an increasing frequency as the cell density in the bacterial population increases, possibly because of entry into stationary phase or through the action of a quorum-sensing system. Finally, the ability of hipA7 strains to generate a high frequency of persistent cells relies on functional relA and spoT genes, suggesting that hipA7 facilitates the development of the persistent state through the production of (p)ppGpp.
The wild-type HipA protein appears to be a potent toxin. Low-level expression of wild-type hipA from pBAD vectors in the presence of glucose was sufficient to induce filamentation and cell death in significant numbers of ΔhipBA cells, suggesting that very few molecules of HipA are required to exert toxicity. This result correlates well with Moyed's original observation that insertion of hipA into a plasmid proximal to a strong promoter was not possible, even when the plasmid was introduced into a host strain expressing a copy of hipB (Moyed and Broderick, 1986; Black et al., 1994). On the other hand, pBAD vectors containing hipA could easily be introduced into hipB+hipA+ strains, most probably because low-level hipA expression from pBAD vectors in the presence of glucose could be neutralized by increased expression of the chromosomal hipB gene. We also found that full induction of pBAD-controlled hipA expression by arabinose could be accomplished in hipB+ strains, but growth was slowed, causing cultures to take two to three times as long to reach stationary phase (Fig. 2A). These results are consistent with the results of Falla and Chopra (1998), who cloned hipA into a pET30b vector and saw a dose-dependent effect of IPTG induction on cell division. The mechanism by which this reduction in cell division occurs is not known, but could result from either the release of non-lethal quantities of free HipA, as suggested by Falla and Chopra (1998) or an effect mediated by high intracellular concentrations of HipA–HipB complex.
The toxicity of wild-type hipA and the high persistent phenotype conferred by the hipA7 allele represent an unusual combination of properties for hip alleles: having the ability to alternately kill or rescue cells. The toxin–antitoxin relationship of hipA and hipB places these genes in the same category as other chromosomal toxin–antitoxin addiction modules, such as mazEF, relBE and chpB, which have been suggested to mediate programmed cell death (Bugge Jensen and Gerdes, 1995; Lewis, 2000; Sat et al., 2001). Recent evidence suggests that the ‘toxins’ encoded by relE and mazF are not bacteriocidal but, rather, exert a bacteriostatic effect on cells when synthesized in the absence of the antitoxin (Pedersen et al., 2002). In the case of relE, the static state is apparently imposed by RelE-mediated cleavage of cellular mRNA in the A-site of ribosomes (Christensen and Gerdes, 2003; Pedersen et al., 2003). Once the static state has been induced by unbalanced expression of either relE or mazF, cells can be revived by induction of the cognate antitoxin (Pedersen et al., 2002). It is striking that mazEF is located in the same operon as relA and that relBE was originally identified as a delayed growth mutant after amino acid starvation, as this links both these toxin–antitoxin modules to the stringent response. Although the significance of this correlation remains to be elucidated, it will be of some interest to determine whether high persistence by hipA7 is also mediated by bacterial stasis similar to that imposed by mazF and relE, as this would lead to an antibiotic-resistant state.
Lewis (2000) made the intriguing suggestion that exposure of cells to antibiotics or other stresses might induce programmed cell death (PCD), thereby allowing a ‘majority of cells to “choose” to die’. If so, it is predicted that loss of toxicity in hipA mutants, if it constitutes a major PCD pathway, would lead to increased survival upon antibiotic challenge. The results reported here support this proposal in part. However, the establishment of persistence appears to be more complex than can be accounted for by loss of toxicity by the hipA7 mutant or by inhibition of a programmed cell death pathway. The hipAG22S allele encodes a non-toxic protein, yet is unable to confer the high persistence phenotype. Likewise, the single mutant hipAD291A retains toxicity, yet is still able to induce persistence at a level comparable to hipA7 strains. Although it remains a possibility that a non-toxic HipA protein improves the chances of cell survival under conditions of stress, the overall picture that emerges is that loss of toxicity is neither sufficient nor necessary for establishment of the high persistence phenotype.
Evidence is also presented here that HipB is not required for high persistence. The hipA7, hipAD291A and hipAD88N alleles were all capable of conferring the high persistence phenotype in strains deleted for hipBA. This result was surprising because Moyed and coworkers isolated a high persistent mutant hipB2 that was complemented by plasmids expressing hipB only (Black et al., 1991). This mutant could not be transformed with plasmids expressing hipA, suggesting that hipB2 encoded a defective protein that no longer completely neutralized the toxic effects of HipA. The ability of this mutant to establish high persistence could be related to the proposal made by Falla and Chopra (1998) that low levels of free wild-type HipA could actually increase survival of cells exposed to ampicillin or fluoroquinilones. However, our results demonstrated that low-level expression of wild-type hipA in ΔhipBA strains had no effect on the frequency of persisters when cells were exposed to ampicillin (Table 1). Unfortunately, the mechanism by which the hipB2 mutant confers high persistence will probably never be known, as it appears that this mutant has been lost.
Inactivation of relA diminishes the ability of hipA7 to confer the high persistence phenotype, whereas deletion of both relA and spoT genes eliminates both the capacity to synthesize 3′,5′-bispyrophosphate [(p)ppGpp] and the ability of the hipA7 allele to confer high persistence. The 100-fold higher frequency of persisters in the relA hipA7 strain versus the relA spoT hipA7 strain at all cell densities suggests that residual synthesis of (p)ppGpp by spoT can partially supplement the loss of relA in establishing the persistent state. These results conflict with Moyed's report that a relA mutation had little effect on the frequency of persister cells generated in hipA7 mutants (Scherrer and Moyed, 1988). However, we have confirmed by PCR and phenotypic studies that the appropriate mutant alleles, hipA7, relA251::kan or spoT207::cam, are present in strains TH1681 and TH1685. We suspect that the conflicting results obtained in our studies and Moyed's published studies result from two differences: (i) the fact that Moyed used strain CP79 (Fiil and Friesen, 1968), which was only relA and not ppGpp°; and (ii) the fact that Moyed might have harvested cells at a low OD600 where the differences between the relA+ and relA strains are not so pronounced.
The ability of relA spoT mutations to eliminate the high persistence phenotype in a hipA7 mutant and significantly reduce the frequency of persisters in hipA+ strains, along with the corresponding increase in persister cells when relA expression is increased in a hipA7 relA mutant, suggests that the stringent response may form the basis of high persistence. The connection between persistence and the stringent response correlates well with the known ability of amino acid-starved cells to enter into a non-growing phase where peptidoglycan synthesis is inhibited, thus rendering cells impervious to penicillin treatment (discussed below). Also, it has been reported recently that Mycobacterium tuberculosis persistence to TB drugs and long-term survival under nutrient-limiting conditions are correlated with increased expression of the relA gene and involve the stringent response (Primm et al., 2000; Betts et al., 2002). Other manifestations of the high persistent phenotype in E. coli, including increased survival after fluoroquinilone treatment, thymine starvation or incubation of dnaBts mutants at the non-permissive temperature (Scherrer and Moyed, 1988), could also be accounted for by induction of (p)ppGpp synthesis, as it is known that initiation of DNA replication is inhibited in the presence of high intracellular levels of (p)ppGpp (Schreiber et al., 1995).
The results presented here may connect two phenomena that were originally described in the 1940s. It is possible that persistence (or high persistence) is simply another manifestation of the well-known phenomenon called penicillin tolerance (Davis, 1948; Lederberg and Zinder, 1948), which is associated with induction of the stringent response and (p)ppGpp production (Goodell and Tomasz, 1980; Kusser and Ishiguro, 1985). Nutrient-starved E. coli or E. coli in which (p)ppGpp is expressed from a plasmid enter into a non-growing state that allows them to tolerate lethal concentrations of ampicillin for extended periods of time, because of an inhibition of peptidogylcan synthesis (Rodionov and Ishiguro, 1995). Antibiotic tolerance bears a strong resemblance to high persistence, and the possibility exists that the two physiological states share a common mechanism of action. However, one crucial difference between our experiments and experiments examining antibiotic tolerance is that antibiotic tolerance is established only after a period of (p)ppGpp synthesis is allowed to occur. For example, in the studies using amino acid starvation to induce the stringent response or a plasmid-encoded relA gene product, relA expression was induced 15 min before the addition of antibiotic to allow tolerance to develop (Goodell and Tomasz, 1980; Rodionov and Ishiguro, 1995). Without this period of adaptation, lysis of cells was rapid, and viability dropped by 30-fold within 30 min of antibiotic treatment (Goodell and Tomasz, 1980). In comparison, our assays for high persistence use bacteria that are taken from exponentially growing cultures and plated directly on LB-ampicillin agar following dilution. Under these circumstances, cells would have to transit rapidly from maximal growth and elongation to a non-growing state to escape the effects of the antibiotic.
We offer a new model (Fig. 8) for the development of the high persistence phenotype that incorporates the observations presented in this paper. In this figure, HipA* represents a high persistent form of HipA, such as HipA7. Because HipB is not required for the high persistence phenotype, we speculate that significant amounts of HipA* are free in cells (1), possibly because of decreased binding interactions with HipB. Because the high persistence phenotype is a gain-of-function mutation, we propose that the mutation causes an altered interaction with the normal cellular target of the wild-type HipA toxin (2) or introduces a new activity for HipA* (3) that is independent of interactions with the normal target. In either case, the gain-of-function activity associated with HipA* increases the basal level of (p)ppGpp synthesis (4), which in turn leads to altered gene expression (5), priming cells for entry into the persistent state. It should be noted that the postulated low-level synthesis of (p)ppGpp is sufficient to alter gene expression without affecting the rate of cell growth, as hipA7 mutants grow at the same rate as wild-type hipA cells (Moyed and Bertrand, 1983). Upon application of stress (6), production of (p)ppGpp increases rapidly, leading to high levels of (p)ppGpp accumulation (7) and promoting entry into the persistent state (8). Alternatively, the increased basal synthesis of (p)ppGpp and altered gene expression lead directly to a subpopulation of dormant, persister cells (5→8), which are refractory to the effects of penicillin antibiotics and other stresses that affect growing cells.
Although pre-existing, quiescent cells might account for the formation of persisters in a wild-type population of bacteria, we think that the majority of persisters in hipA7 mutants arise through a rapid transition to the dormant state after the application of stress. This postulation is based on the difference in the frequency of persisters that are generated in the hipA7 relA/pHM675 strain (2.8 × 10−1) versus the hipA7 relA/pHM675Δ strain (2.2 × 10−4) under similar growth conditions (OD600 of 1.5, 12.5 µM IPTG; Fig. 6B, centre). It is unlikely that a 1000-fold greater proportion of cells exists in a dormant state in the TH1686 (hipA7 relA/pHM675) cell population versus the TH1708 (hipA7 relA/pHM675Δ) population when the growth rates of the two strains are such that the time required to double to an OD600 of 3.0 was 60 and 54 min respectively.
The extremely low frequency of survivors after ampicillin treatment in wild-type bacterial populations (10−6−10−5) suggests that the establishment of persistence is a stochastic process that probably arises from the diversity of gene expression that was demonstrated recently in genetically homogeneous bacterial populations (Elowitz et al., 2002). Based on the fact that null mutants of hipA have the same low frequency of persisters as wild-type cells, we postulate that the hipA7 allele facilitates the development of this unique physiological state rather than directly causing establishment of the persistent state. This postulation implies that high persistence conferred by hipA7 constitutes the same physiological condition that generates low-frequency persisters in wild-type cells of E. coli. Koch (1996) suggested that phenotypic differences in bacterial populations can give rise to rare dormant cells. As described above for antibiotic tolerance, these dormant cells would resist the effects of antibiotics and could possibly arise at random from elevated levels of (p)ppGpp synthesis. If so, one cell per million in a bacterial population would have sufficiently elevated levels of (p)ppGpp to resist lysis when antibiotics are applied. However, until more is known about the cellular target of HipA7 and how the interaction between HipA7 and its target affects the physiological state of bacteria, the exact nature of the conditions that gives rise to persisters in wild-type or hipA7 mutant bacteria will remain an open question.
Falla and Chopra (1999) proposed that hipA and hipB are not widely distributed in the Enterobacteriaceae and that the gene pair most probably represents a post-segregation plasmid stabilization system. This proposition was based on the observation that the only hipA homologues identified in the databases at that time were located in the Rhizobium symbiosis plasmid pNGR234. In addition, these authors examined 40 clinical isolates of E. coli by PCR amplification, and ≈ 20% of the isolates lacked either a hipA or a hipB gene by this criteria (Falla and Chopra, 1998). However, since these papers were published, the explosion in bacterial genome sequencing has radically altered this picture. A recent search of the databases revealed 82 homologues to hipA in a wide variety of Gram-negative bacteria and included chromosomal hipA homologues in 26 different bacterial species that scored higher than the Y4dM gene of plasmid pNGR234. Furthermore, in this search, the Rhizobium genes were the only plasmid-associated hipA homologues identified, arguing against the proposal that the hipA–hipB genes represent a simple plasmid addiction module. Instead, the identification of chromosomal-encoded homologues of hipA with regions of high amino acid identity argues that this gene plays a well-conserved role in the physiology of other Gram-negative organisms. Further examination of the mechanism by which hipA mutants confer high persistence and identification of the target of wild-type hipA in E. coli should help to elucidate the function that hipA performs in these other bacteria.
Media and chemicals
Bacteria were grown on LB agar plates or in LB broth. In strains carrying pBAD vectors, 0.4% glucose was used to suppress expression and 0.4% arabinose to induce expression. Antibiotics were used at the following concentrations: ampicillin 100 µg ml−1, chloramphenicol 50 µg ml−1, kanamycin 25 µg ml−1 and tetracycline 12.5 µg ml−1. Penicillinase (Sigma) was resuspended at 2500 U ml−1 in sterile 50 mM Tris-Cl (pH 7.5 at 25°C), 50 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol (DTT) and 20% glycerol and frozen at −80°C in 100 µl aliquots. Before use, the penicillinase was diluted to 50 U ml−1 in LB broth and filter sterilized.
Strains and plasmids
Strains and plasmids used in this study are listed in Table 3. The hipA7 allele was transferred from HM22 (Moyed and Bertrand, 1983) into the wild-type E. coli K-12 strain MG1655 by co-transduction with the zde264::Tn10 marker to produce TH1269. The isogenic wild-type strain TH1268 was constructed in a similar manner using a P1vir lysate from HM21. The presence of the hipA7 mutation in TH1269 was confirmed by testing for high persistence and by screening for cold sensitivity. To produce TH1604, the zde264::Tn10 transposon in TH1269 was eliminated by fusaric acid selection (Maloy and Nunn, 1981). The relA spoT double mutant TH1685 was routinely checked for inability to grow on minimal glucose plates, a characteristic of ppGpp° strains, to ensure that suppressor mutations had not developed.
pHM675 with amino acids 35–438 of the relA′ gene deleted
pMAK705 (repAts) with regions adjacent to hipBA inserted on either side of a res-npt-res cassette (Kristensen et al., 1995)
To study hipA toxicity, a hipBA knock-out was constructed that replaces the hipBA genes with a res–npt–res cassette, which contains a kanamycin resistance gene flanked by the res sites from the plasmid RP4 (Kristensen et al., 1995). To generate this deletion, 800–900 bp fragments from either side of the hipBA operon were amplified by PCR. Primers used to amplify the downstream region were 5′-TCTCTCTC TCAAGCTTACTGTCGGATAAGTTATAAAA and 5′-TCTCTCT CTCGCGGCCGCTAATGAGCATGACAATCATGACC. Primers for the upstream region were 5′-TCTCTCTCTCGCGGC CGCACGTCCACAGCAAGTTTATCCG and 5′-TCTCTCT CTCCTGCAGATCTTGGCGTGATGGCAGG. The downstream PCR product was digested with HindIII and NotI, the upstream PCR product with NotI and PstI. The two PCR products were simultaneously cloned into the HindIII–PstI sites of pMAK705 (Hamilton et al., 1989), generating a single NotI site at the junction of the upstream and downstream fragments. The res–npt–res cassette was purified from NotI-digested pCK155 (Kristensen et al., 1995) and inserted into the NotI site between the upstream and downstream hipBA fragments to generate pMAKΔhipBA::kan. This plasmid was digested with ScaI, and the linear fragment was used to transform TH996 (MG1655 recD::Tn10). Kanamycin-resistant, chloramphenicol-sensitive colonies were analysed by PCR and Southern blotting to confirm the presence of the deletion. The ΔhipBA::res–npt–res was then moved into MG1655 by P1 transduction to produce TH1256. Finally, the npt gene of ΔhipBA::res–npt–res in TH1256 was removed by conjugational mating with strain S17-1 (λpir)/pJMSB8 (Kristensen et al., 1995) to generate strain TH1433.
Using PCR, the wild-type hipA gene was amplified from MG1655 and cloned into the EcoRI–PstI sites of expression vectors pBAD24 and pBAD33, which are ColE1- and p15A-based vectors respectively (Guzman et al., 1995). Primers for hipA were 5′-TCTCTCTCTCGAATTCACCATGCCTAAACT TGTCACTTGGATG and 5′-TCTCTCTCTCCTGCAGTCACT TACTACCGTATTCTCGGC. Using the same primers, the hipA7 gene from HM22 was amplified and cloned into pBAD24 and pBAD33. All cloned genes were sequenced to ensure that no additional mutations had been introduced during amplification.
Site-directed mutagenesis was performed using the QuikChange kit (Stratagene) or by methods described by Henderson et al. (2001).
The plasmid pHM675 was a generous gift from Michael Cashel at NIH. This plasmid is based on the vector pGB2, which is derived from pSC101 (Churchward et al., 1984) and contains a PCR-generated fragment from plasmid pALS13 (Svitil et al., 1993) containing the lacI–Ptac–relA′ gene region inserted into the EcoRI and BamHI sites of the pGB2 multiple cloning site. The ‘empty’ vector pHM675Δ was constructed by PCR amplification of the entire plasmid except for the region of relA′ encoding amino acids 35–438. Primers used for PCR amplification were 5′-TCTCTCTCTCGCGGCCGCG CATCGGGGCAAAAATTGGCGGG and 5′-TCTCTCTCTC GCGGCCGCAAGCACTCACACGACTTCTGGC. The PCR product was digested with NotI to cleave at the NotI sites incorporated into the primers, and the fragment was recircularized to generate pHM675Δ.
Assays for HipA toxicity, cold sensitivity and high persistence
The toxicity of mutant hipA alleles was determined by introducing pBAD24 or pBAD33 derivatives expressing the desired hipA allele into either TH1256 or TH1433. Overnight cultures of strains containing the plasmids were diluted, aliquots spread on LB-glucose or LB-arabinose plates supplemented with the appropriate antibiotic, and the plates were incubated overnight at 37°C. Colony-forming units were counted for both types of plates, and the frequency of survival was determined. Alternatively, a simple survival assay was performed by streaking individual colonies onto LB-glucose and LB-arabinose plates and incubating the plates overnight at 37°C to determine whether growth was suppressed on the arabinose plates.
Cold sensitivity was assayed by streaking strains onto LB agar with appropriate antibiotics and incubating at 20°C for 3–4 days. The absence of growth on any part of the plate except in the primary streak was scored as cold sensitive.
The assay for high persistence was performed according to the method of Moyed and Broderick (1986) with the following modifications. For measuring high persistence of chromosomal-encoded hipA mutants, 25 ml of LB broth in 250 ml baffled flasks was inoculated with 50 µl from an overnight culture and incubated at 37°C with shaking at 200 r.p.m. Except where indicated otherwise, bacteria were harvested at an OD600 between 0.3 and 0.35. Dilutions were prepared, and 20–100 µl of a 1:102 or 1:101 dilution was spread on LB-ampicillin plates in triplicate, and 20–100 µl of a 1:104 or 1:105 dilution was spread on an LB plate. After overnight incubation at 37°C, colonies on the LB plates were counted to determine cfu, and the LB-ampicillin plates were sprayed with ≈ 0.3 ml of a penicillinase solution (50 U ml−1) using a MartiniMister™ to produce a fine mist. The LB-ampicillin plates were returned to the 37°C incubator, and persistent colonies were recorded on the following day. For ΔhipBA strains carrying pBAD33 with arabinose-inducible hipA genes, high persistence was assayed in a similar manner using LB-ampicillin plates supplemented with glucose or arabinose.
Because some strains generated low frequencies of persisters, data points in figures and tables were calculated by a ‘pooled data’ method, unless stated otherwise. The pooled data method used the (sum total of persisters ÷ sum total of cells screened) from all samples taken in two to four independent experiments to calculate the persister frequency. This method was most useful for strains generating frequencies of persisters of < 1 × 10−6, as two or more independent experiments were usually required to screen enough cells to obtain a sufficient number of persisters to calculate an accurate frequency. In the case of strains with frequencies > 1 × 10−6, the ‘pooled data’ method gave values that were virtually identical to the method of averaging values obtained from several independent experiments.
This manuscript is dedicated to the memory of Harris Moyed. We gratefully acknowledge Mike Cashel (NIH) for sending us strain CF1693 and plasmid pHM675 and for helpful advice. We also thank Kim Lewis (Northeastern University) for the suggestion that increased aeration would improve the reproducibility of the high persistence assay, and Pat Carr (UNDSMHS) for assistance with statistical analyses. This work was supported by grants from the Invitrogen Corporation and the Office of Research and Program Development at UND.