General Stress Response Regulator RpoS in Adaptive Mutation and Amplification in Escherichia coli

Abstract

Microbial cells under growth-limiting stress can generate mutations by mechanisms distinct from those in rapidly growing cells. These mechanisms might be specific stress responses that increase mutation rates, potentially altering rates of evolution, or might reflect non-stress-specific processes in rare growing cells. In an Escherichia coli model system, both frameshift reversion mutations and gene amplifications occur as apparent starvation-induced mutations. Whereas frameshift reversion (“point mutation”) requires recombination proteins, the SOS response, and error-prone DNA polymerase IV (DinB), amplification requires neither SOS nor pol IV. We report that both point mutation and amplification require the stationary-phase and general stress response transcription factor RpoS (σ^S). Growth-dependent mutation does not. Alternative interpretations are excluded. The results imply, first, that point mutation and amplification are stress responses that occur in differentiated stationary-phase (not rare growing) cells and, second, that transient genetic instability, producing both point mutation and genome rearrangement, may be a previously unrecognized component of the RpoS-dependent general stress response.

MICROBIAL cells exposed to growth-limiting environments can, in some cases, acquire a beneficial or “adaptive” mutation that allows rapid growth. In bacteria and yeast, there are multiple mechanisms of mutation in growth-limiting environments, some demonstrably different from spontaneous mutation in rapidly growing cells (Foster 1999b; Rosenberg 2001). Also called “stationary-phase” mutation mechanisms, these generally occur in populations of cells that are in the stationary phase, defined as displaying no change in population size (Huisman  et al. 1996). The mechanisms and regulation of stationary-phase mutation in these systems are of considerable interest because they represent model systems for understanding interactions between the environment and the genome, particularly whether and how stressful conditions influence genetic change. Induction of genetic change in response to stress has broad relevance to mechanisms of evolution and, thus, to the treatment of bacterial infections; the evolution of bacterial pathogens (for example, the generation of antibiotic resistance); the host immune response; and the genetic instability underlying oncogenesis, tumor progression, and resistance to chemotherapeutic drugs (e.g., Loeb 1991; Strauss 1992; Martinez and Baquero 2000).

Since the inception of current interest in stationary-phase mutation mechanisms (Shapiro 1984; Cairns  et al. 1988), three broad classes of model have been considered to explain apparent increases in adaptive mutants under some selective conditions (reviewed by Foster 1999b; Rosenberg 2001). In “directed mutation” models (e.g., Cairns  et al. 1988), environmental stress was postulated to stimulate formation or retention of mutations specifically in genes whose altered functions were selected. In “cryptic growth” models, mutation rates are specified to be constant (unaffected by environmental conditions) and the emergence of mutants during stress to result from normal growth-dependent mutation mechanisms operating in rare growing cells (e.g., Partridge and Morgan 1988; Lenski  et al. 1989; Lenski and Mittler 1993; Galitski and Roth 1995; Andersson  et al. 1998; Hendrickson  et al. 2002; and see discussion). In “hypermutation” models, mutation rates are suggested to increase globally in response to stress, generating mostly deleterious, but some useful, mutations (e.g., Hall 1990; Ninio 1991; Torkelson  et al. 1997; Rosenberg 2001). The ability to increase mutation could itself be selected (for example, as an advantageous by-product of error-prone DNA repair) or not; either way, increased mutation is a stress response. These models can be viewed as representing different views of evolution (and so have been debated vigorously; see Chicurel 2001): the first, somewhat Lamarckian; the second, neoDarwinist in its adherence to constant mutation rates and gradual evolutionary change (e.g., Mayr 1982); and the third, compatible with Darwinism (which allows for changing rates of generation of heritable variations; Darwin 1859). In this article we examine stationary-phase mutation in a well-studied Escherichia coli model system in which directed mutation models were ruled out previously (and hypermutation models supported) by the frequent occurrence of neutral and deleterious mutations along with useful ones (Foster 1997; Torkelson  et al. 1997; Rosenberg  et al. 1998; Rosche and Foster 1999; Bull  et al. 2000, 2001; Godoy  et al. 2000). Whereas we have suggested that stationary-phase mutation in this system results from transient genetic instability induced in response to stress, a hypermutation model (e.g., Rosenberg 2001), others favor a cryptic growth model (e.g., Hendrickson  et al. 2002). Here, we examine the regulation of mutation in this system and address whether it is part of a stress response, i.e., a response to environmental conditions.

In the Lac system (Cairns and Foster 1991), E. coli cells unable to catabolize lactose (because of a +1 frame-shift mutation in lac genes carried on an F′ conjugative plasmid) are plated on lactose minimal medium, selecting for Lac⁺ mutants. Lac⁺ colonies appear after 2 days and continue appearing for >8 days. Most Lac⁺ colonies present before day 5 carry a compensatory lac frameshift mutation (“Lac⁺ point mutants”; Foster and Trimarchi 1994; Rosenberg  et al. 1994). The remainder (5-10%) are Lac⁺ because they carry amplified tandem arrays of the leaky lac allele (20-50 copies) that yield sufficient β-galactosidase for growth (“Lac⁺ amplified clones”; Hastings  et al. 2000). The occurrence of amplification in this Lac system (and others previously, see Horiuchi  et al. 1963; Tlsty  et al. 1984; Whoriskey  et al. 1987) was noted previously, but not examined closely (Foster 1994; Foster  et al. 1995). Recent work demonstrates that amplified clones make up 40-50% of the Lac⁺ colonies that appear after day 7 (Hastings  et al. 2000). Thus, both permanent genetic changes (point mutation) and reversible genome rearrangements (amplification) occur during starvation, making this system a useful model for both types of apparent environmentally induced genetic instability.

The mechanism of Lac⁺ stationary-phase point mutation differs from mechanisms of reversion of the same allele in rapidly growing cells in its requirements for homologous recombination and double-strand-break (DSB) repair proteins (Harris  et al. 1994, 1996; Foster  et al. 1996), the SOS response to DNA damage (Cairns and Foster 1991; McKenzie  et al. 2000), the SOS-regulated error-prone DNA polymerase IV (DinB; McKenzie  et al. 2001), F′ transfer functions (Foster and Trimarchi  1995a,b; Galitski and Roth 1995), and transiently limiting mismatch repair activity (Harris  et al. 1997, 1999; Foster 1999a). We have suggested that stationary-phase mutations are formed during error-prone DNA synthesis (involving pol IV) primed from recombination intermediates created as part of repair of double-strand (DS) ends (Harris  et al. 1994; Rosenberg 2001; and see Foster  et al. 1996; Rodriguez  et al. 2002). Adaptive amplification requires neither SOS nor pol IV (McKenzie  et al. 2001).

A key question concerns whether adaptive point mutation and/or amplification are stress responses, as predicted specifically by hypermutation models. In this work we examine the role of RpoS, a “general stress response” regulator, in mutation and amplification in the Lac system. RpoS is a sigma (transcription) factor of RNA polymerase that promotes the increased transcription of >50 genes in response to a variety of environmental stresses (Hengge-Aronis  1996, 2000; Loewen  et al. 1998). Inducers of the RpoS regulon include nutrient deprivation (leading to stationary phase), acidic pH, high temperature, oxidative stress, and osmotic stress (Hengge-Aronis 1996). Cross-protection is an important feature of the general stress response (Hengge-Aronis  1993, 2000). For example, when cells enter stationary phase due to carbon starvation, they become protected against oxidative stress and heat shock (Jenkins  et al. 1988). RpoS-deficient cells are less viable in long-term culture and are not protected against oxidative, heat, or osmotic stress in stationary phase (Mulvey  et al. 1990; Lange and Hengge-Aronis 1991; McCann  et al. 1991). Thus, induction of the RpoS regulon leads to the development of a differentiated, protected state.

We present evidence that RpoS is required for stationary-phase mutation and also amplification in the Lac system. This is the first documented genetic requirement for adaptive amplification and the first shared with adaptive mutation, suggesting a possible common early step(s) in both pathways. The results imply that stationary-phase mutation and amplification are stress responses and that both occur in differentiated stationary-phase cells, as hypermutation, but not cryptic-growth, models predict. Further, the data imply that transient genetic instability is a previously unrecognized component of the RpoS-dependent general stress response, at least in response to stationary phase and, perhaps, in response to other stressors as well.

MATERIALS AND METHODS

Bacterial strains and growth:  E. coli strains (Table 1) were constructed using standard techniques (Miller 1992). Bacterial growth was at 37°. M9 minimal medium (Miller 1992) had carbon sources added at 0.1% and thiamine (vitamin B1, or B1) at 10 μg/ml, and cultures were incubated for ∼40-48 hr to assure saturation. rpoS M9 glycerol cultures often saturated at cell densities 3- to 5-fold lower than those of rpoS⁺ and were sometimes concentrated 5-fold before plating. Rarely, all rpoS cultures of an experiment had normal turbidity but viable cell numbers 10- to 100-fold lower than expected; these experiments were unusable and the variable responsible is unknown. Antibiotics and other additives were used at the following concentrations (μg/ml): chloramphenicol, 25; kanamycin, 50; tetracycline, 10; rifampicin, 100; X-gal, 40; threonine, 50; and tryptophan, 50.

The presence of rpoS::Tn10 in SMR6541 was confirmed by PCR with primers 5′-TCAGCAACCGTAGCAATACGTACA-3′ and 5′-AGGCAATTATCGCGACCGCAG-3′ and also by PCR with those primers (separately) in combination with primer 5′-GACAAGATGTGTATCCACCTTAAC-3′, specific to a sequence present at both ends of Tn10. The presence of rpoS::Tn10 was confirmed similarly in a sampling of 10 day 2 and 10 day 5 rpoS Lac⁺ colonies, to exclude the possibility that they had lost the Tn10 and were rpoS⁺. Routinely, rpoS cultures were confirmed to be deficient in catalase activity (poor bubbling when hydrogen peroxide is applied to colonies; Bohannon  et al. 1991).

Sequencing: The rpoS gene of SMR4562 including its promoter region (Lange  et al. 1995) was sequenced using PCR-generated template DNA (primers flanking rpoS, above) and additional rpoS-specific primers. One strand was sequenced completely. The difference at codon 33 was confirmed by sequencing across that site in independent PCR products. The rpoS allele in SMR6577 was confirmed by sequencing across codon 33, also in PCR-generated DNA. Sequencing was performed by Lone Star Labs (Houston).

Reconstruction experiments: Reconstruction experiments were performed as described (Hastings  et al. 2000). Amplified Lac⁺ isolates were grown in M9 lactose to maintain selection for the amplified array of 20-50 copies of the lac region (Hastings  et al. 2000). The number of amplified cells was determined by plating nonselectively to LBH (1% tryptone, 0.5% NaCl, 0.5% yeast extract, 2 μg/ml thymine, pH 7, plate medium solidified with 1.5% agar) with X-gal to allow scoring of amplified cells (which form sectored blue-and-white colonies) or Lac^- cells (which form white colonies, having lost the amplified array; Hastings  et al. 2000 and references therein).

Mutation, conjugation, and transduction assays: Growth-dependent mutation rates to Lac⁺ were determined using fluctuation tests, as described (Harris  et al. 1997, 1999) and rates were estimated using a modified method of the median (Lea and Coulson 1949; von  Borstel 1978). A total of 29 or 30 independent cultures per strain were plated on M9B1 lactose and Lac⁺ colonies were scored. Rather than scoring only at 48 hr, we controlled for possible differences between genotypes in the time of colony formation and avoided including adaptive mutants (see Harris  et al. 1999) by plating four to eight independent Lac⁺ derivatives of each strain in parallel with the cultures being scored for growth-dependent mutants. Scoring of all plates at ∼1- to 2-hr intervals allows the determination of the median mutant frequency at a time when only 50% of the control Lac⁺ cells of that genotype have formed visible colonies (t₅₀). The t₁₀₀ is set at ∼48 hr when ≥95% of the Lac⁺ controls are visible for all genotypes. The mutation rates were determined from the mutant frequencies at the calculated t₅₀ and multiplied by two (to give the rates at t₁₀₀).

Stationary-phase mutation assays were performed as described (Harris  et al. 1996). All experiments presented had less than twofold net population change during days 1-3 after plating (see results for discussion of the importance of monitoring the Lac^- population).

For quantitative conjugation assays, donors were grown to saturation in M9B1 glycerol and then diluted 1:50 into M9 glucose and grown 3-4 hr. Donor and recipient cells (SMR3462, grown to saturation in LBH) were added to prewarmed M9B1 glucose proline and incubated for 30 min at 37°, during which time dilutions of each were plated on LBH to determine viable cells per milliliter. Mating mixes were vortexed and plated on M9B1 glycerol chloramphenicol to select for Pro⁺ Cam^R transconjugants. Donor cells were limiting by ≥10-fold to allow calculation of the frequency of conjugation as transconjugants per donor cell.

Transductions were performed as described (Miller 1992) with some modification. Recipient strains were grown to saturation in M9B1 glycerol with threonine or tryptophan added and cells were mixed with phage to give a multiplicity of infection (MOI) of ≤0.01. After 20 min of incubation, transduction mixes were washed once in M9 sodium citrate (20 mm) and plated on M9B1 glycerol 20 mm sodium citrate to select prototrophic transductants. Phage plating efficiencies were similar for both genotypes, indicating that rpoS cells are unimpaired for phage P1 infection and production. The phage titer on the rpoS⁺ recipient was used in determining the transductant frequency (transductants per phage) for both genotypes.

RESULTS

An rpoS mutation decreases stationary-phase mutation and amplification: An rpoS null allele (Loewen and Triggs 1984; Ivanova  et al. 1992) decreases Lac⁺ colony yield substantially (Figure 1A, “total” values). Multiple experiments give similar results (Figure 1D), with some variability. The rpoS mutation decreases both point mutation and amplification (Figure 1, A and B; a second experiment gives similar results, data not shown). Lac⁺ point mutant and amplified colonies can be distinguished by streaking to nonselective indicator medium (see materials and methods). The rpoS phenotype in amplification is first detectable on day 5 as a threefold reduction, with greater reductions on later days (five-, six-, and sixfold on days 6, 7, and 8, respectively). RpoS appears to be required for both point mutation and amplification; however, this conclusion depends on control experiments (below) that address other possible explanations for the rpoS phenotype.

rpoS cells are viable during the stationary-phase mutation assay: Mutations in rpoS can decrease survival in stationary-phase cultures, at least in liquid culture (Mulvey  et al. 1990; Lange and Hengge-Aronis 1991; McCann  et al. 1991), suggesting that death of the rpoS cells might explain the decreases in observed Lac⁺ point mutants and amplified clones. We monitor viability by counting viable cells in plugs of agar removed from the selective plates each day, beginning the day after plating (day 1). Because Lac⁺ point mutant cells (the majority of Lac⁺ before day 7) take 2 days to form a colony (Table 2; McKenzie  et al. 1998), any detected growth or death of the population is expected to be reflected in increased or decreased numbers of Lac⁺ colonies 2 days later. The viability data for the experiment in Figure 1, A and B, are presented in Figure 1C. The data are presented offset by 2 days because the Lac⁺ point mutant cells that give newly visible colonies on any given day were present in the population 2 days earlier (for example, the day 1 viability is plotted on day 3 because Lac⁺ cells in the day 1 population grew into the Lac⁺ colonies observed on day 3). The data demonstrate that the rpoS mutant strain is not poorly viable under these conditions. It is as viable as rpoS⁺ until 5 days after plating (plotted as day 7 in Figure 1C). This is consistent with previous reports of viability of rpoS strains on solid medium (Bridges and Timms 1998; Bridges  et al. 2001). We conclude that net population death does not occur and cannot account for the lack of Lac⁺ point mutant colonies in the rpoS strain.

The effect of rpoS on formation of Lac⁺ amplified colonies is also not attributable to altered viability relative to rpoS⁺. It appears in Figure 1 that lack of growth of the rpoS strain when the rpoS⁺ begins to grow (late in the experiment) may correlate with the lack of amplified Lac⁺ colonies in the rpoS strain. However, amplified Lac⁺ cells take 3-5 days to form a colony (Table 2; Hastings  et al. 2000). Thus, changes in the population size 5 days after plating (day 7 in Figure 1, at which point rpoS⁺ shows growth and the rpoS mutant does not) do not affect amplified colony yield until at least day 8, whereas the difference in amplified colony yield in rpoS is clear even by day 5. If the rpoS Lac⁺ amplified values are adjusted to reflect the viability 3, 4, or 5 days earlier, the effect of rpoS on amplification is even greater (data not shown). Also, we observe no loss in viability of the rpoS cells, relative to rpoS⁺, between plating and day 1 (data not shown). Therefore, poor viability of the population does not cause the rpoS phenotype in mutation or amplification.

Previous data imply that a hypermutable cell subpopulation gives rise to point mutants (Torkelson  et al. 1997; Rosche and Foster 1999; Godoy  et al. 2000), but not amplified Lac⁺ clones (Hastings  et al. 2000). We cannot assess the viability of any subpopulation directly. However, it is unlikely that rpoS reduces point mutation by causing inviability of the hypermutable subpopulation because rpoS also decreases amplified clones, which are not derived from it (Hastings  et al. 2000).

rpoS phenotype cannot be attributed to poor colony formation or loss of amplification: Perhaps Lac⁺ mutations and amplified arrays form normally in rpoS cells, but cells carrying them form colonies slowly, giving the appearance of decreased mutation and amplification. We performed reconstruction experiments (as previously, McKenzie  et al. 1998; Hastings  et al. 2000) with multiple day 2 (presumed growth-dependent) and day 5 (stationary-phase) Lac⁺ isolates to determine how efficiently and rapidly rpoS Lac⁺ cells form colonies under the exact conditions of the stationary-phase mutation assay. Both day 2 and day 5 rpoS Lac⁺ point mutant cells form Lac⁺ colonies with unit efficiency (∼100% of viable cells plated yield Lac⁺ colonies) and as rapidly as the rpoS⁺ controls (Table 2A). To control for the possibility that rpoS Lac⁺ cells had acquired mutations that facilitate colony formation under selective conditions, we performed similar experiments with rpoS⁺ day 2 Lac⁺ point mutants that were transduced with rpoS:: Tn10. These gave similar results (data not shown).

To determine whether rpoS slows colony formation of Lac⁺ amplified cells, we introduced rpoS::Tn10 into 9 independent amplified Lac⁺ strains with different amplified arrays that confer varying growth rates on lactose medium (Hastings  et al. 2000). The rpoS derivatives retain the sectoring phenotype (Figure 1 legend) that is the hallmark of lac amplification, and the time required for colony formation is similar (Table 2B). Thus, rpoS does not decrease numbers of mutant and amplified colonies by slowing colony formation. Rather, rpoS cells appear to be impaired in forming mutations and amplified arrays.

The retention of the amplified (sectored colony) phenotype when rpoS::Tn10 is introduced by transduction also indicates that rpoS does not cause loss of amplified arrays, which presumably occurs by homologous recombination and segregation. This contrasts with recA mutations, which stabilize preformed lac amplification in E. coli (Tlsty  et al. 1984) and Salmonella (Andersson  et al. 1998). Apparently rpoS mutation does not promote loss of amplified arrays, but instead inhibits their formation.

rpoS and growth-dependent Lac⁺  mutation: Several genes required for Lac⁺ stationary-phase mutation (e.g., recA) are not required for reversion of the same allele in rapidly growing cells (growth-dependent mutation), providing clear distinctions between the stationary-phase and growth-dependent mutation mechanisms (Cairns and Foster 1991; Harris  et al. 1994, 1996; Foster  et al. 1996; McKenzie  et al. 2001). We find that growth-dependent Lac⁺ mutation is higher in rpoS cells than in rpoS⁺ (Table 3; paired Student’s t-test on the three sets of data, t =-4.33, d.f. = 2, P = 0.050), which is the opposite of its effect on stationary-phase mutation (Figure 1). This is the first direct evidence we are aware of that RpoS can affect mutation in growing cells. We conclude that RpoS is required specifically for Lac⁺ mutation and amplification in stationary phase and that it may suppress mutation in growing cells.

The rpoS genotype of the scavenger cells does not affect Lac⁺  colony yield: In adaptive mutation experiments, the lac frameshift-bearing cells are plated with an excess of nonrevertible (Δlac) “scavenger” cells to remove nonlactose carbon sources (Cairns and Foster 1991). rpoS cells might appear deficient in mutation because the rpoS⁺ scavenger cells might, for example, prevent growth of rpoS Lac⁺ mutant colonies. If so, rpoS::Tn10 scavengers might have given a different result. However, we find that the yield of Lac⁺ colonies in rpoS⁺ and rpoS::Tn10 strains is unaffected by the rpoS allele of the scavengers (data not shown). This excludes a key role for the scavengers in the rpoS mutation/amplification phenotype.

DNA recombination and F transfer are not defective in rpoS cells: The results presented above indicate that RpoS is required specifically for formation of stationary-phase mutations and amplified arrays and exclude several trivial reasons for the rpoS phenotype. The experiments in this and the following section address several possible mechanisms by which RpoS (and the genes it controls) might promote stationary-phase mutation and amplification.

Stationary-phase mutation of the F′-borne lac allele requires F transfer functions (Foster and Trimarchi  1995a,b; Galitski and Roth 1995) and RecBC-dependent homologous recombination proteins (Harris  et al. 1994, 1996; Foster  et al. 1996). We find that neither F transfer nor RecBC-dependent recombination is deficient in rpoS cells. rpoS⁺ and rpoS stationary-phase cells are similarly active as donors in F transfer assays (consistent with previous results; Frost and Manchak 1998). The average numbers of transconjugants per donor cell are (experiment 1 and experiment 2; values are the means of three cultures): rpoS⁺ (strain SMR4562), 7.7 × 10^-3, 4.0 × 10^-3; rpoS (SMR6541), 14 × 10^-3, 3.8 × 10^-3. Also, the frequency of transduction (a RecBC-dependent process; Lloyd and Low 1996) is not different for rpoS⁺ and rpoS minimal B1 glycerol stationary-phase cultures (Table 4; also true for minimal B1 glucose and rich medium cultures, data not shown). These are the only experiments we are aware of that address a role for RpoS in homologous recombination. Because rpoS mutation affects neither F transfer nor homologous recombination, the rpoS requirements in stationary-phase mutation and amplification are unlikely to be to due to defects in either of these processes.

rpoS and the SOS response: Additional evidence supporting the involvement of RpoS in the SOS-dependent point mutation pathway (Cairns and Foster 1991; McKenzie  et al. 2000) comes from epistasis analysis of lexA3 and rpoS mutations. lexA3 encodes a noncleavable LexA repressor; a lexA3 strain is therefore unable to induce the Lex-dependent genes of the SOS response (Walker  et al. 2000). lexA3 prevents most stationary-phase mutation (Cairns and Foster 1991; McKenzie  et al. 2000), but has no effect on amplification (McKenzie  et al. 2001). If RpoS and the SOS response act in separate pathways leading to point mutation, then the rpoS lexA3 double mutant should have a more severe decrease in Lac⁺ yield than either single mutant. We find, in contrast, that the double mutant is similar to the rpoS strain (Figure 2), indicating that the RpoS and the SOS regulons act in the same pathway. That amplification also requires RpoS (Figure 1; Table 2), but not SOS (McKenzie  et al. 2001), suggests that RpoS might act upstream of SOS before a common pathway branches into SOS-dependent point mutation and SOS-independent amplification routes. However, it is not yet clear whether RpoS plays the same role in amplification as it does in point mutation.

Mutation and amplification do not require a special rpoS allele: Many E. coli laboratory strains carry rpoS mutations that attenuate RpoS function (Ivanova  et al. 1992; Visick and Clarke 1997; Atlung  et al. 2002), and the acquisition of an rpoS^att allele is a key feature of the growth advantage in stationary phase (GASP) phenotype (Zambrano  et al. 1993; Zambrano and Kolter 1996). In GASP, cells in long-term culture acquire advantageous mutations, the first being rpoS^att, and then outcompete their parental genotypes in successive waves, each requiring additional mutations (Zambrano and Kolter 1996). Might Lac⁺ stationary-phase mutation require a special allele of rpoS?

We find that the rpoS allele of the lac frameshift strain SMR4562 [an independent construction of FC40 (Cairns and Foster 1991) from the same parental strains] has a mutation relative to the sequence in the E. coli K12 “wild-type” strain MG1655 (Blattner  et al. 1997). rpoS codon 33 in SMR4562 is TTG, encoding leucine, instead of the CAG, encoding glutamine, in MG1655. This codon varies in laboratory strains (glutamine or an amber codon are common; Ivanova  et al. 1992; Visick and Clarke 1997; Atlung  et al. 2002). However, this variant allele is not important for stationary-phase mutation and amplification because neither is altered in an SMR4562 derivative carrying a nonattenuated rpoS⁺ allele (Atlung  et al. 2002) from wild-type MG1655 (Figure 1E and additional data not shown). Therefore, the FC40 variant rpoS allele does not specifically promote mutation and amplification, and further, attenuated rpoS function conferred by mutation, which underlies GASP, does not underlie these stationary-phase mutations.

DISCUSSION

Summary: We found that RpoS, which controls a large regulatory network induced in response to various environmental stresses, including the onset of stationary phase (Loewen and Hengge-Aronis 1994), is required for most stationary-phase point mutation and amplification in the Lac system (Figure 1; Table 2; results reported in the text). RpoS is required specifically for mutation in stationary phase and not in rapidly growing cells (Table 3). RpoS and the SOS regulon act in the same pathway leading to stationary-phase point mutation (Figure 2), with RpoS possibly acting upstream of SOS at a point before the amplification route branches from the point mutation route (e.g., Figure 3). These results support the idea that transient genetic instability in this system is a stress response, specifically one induced in stationary phase.

These conclusions are possible only because other explanations for the rpoS phenotype (poor viability, poor colony formation, and loss of amplified arrays) were excluded experimentally. A previous study reported data indicating that an rpoS mutation reduced late Lac⁺ colonies, but the authors dismissed the phenotype because it correlated with altered growth relative to rpoS⁺ before day 4 (Bridges  et al. 2001). Their experimental conditions differed sufficiently as to allow growth during the early days of the experiment, which confounds interpretation of the data. For example, if the population grows, some unknown number of Lac⁺ colonies must be attributed to non-stationary-phase (rec- and din-independent) mutation mechanisms. Here, we controlled carefully for effects on growth and viability and other alternative explanations, including RpoS effects on mutation in growing cells, and conclude that RpoS is required specifically for most stationary-phase mutation and amplification.

Mutation and amplification are stress responses—implications for hypermutation and cryptic growth models: Three broad classes of model, two of which persist, were proposed as possible explanations for adaptive mutation in this and other assay systems (reviewed by Foster 1999b; Rosenberg 2001). One class, postulating mutations directed to genes under selection (directed mutation models, e.g., Cairns  et al. 1988), was rejected for the Lac system (see Introduction; Rosenberg 2001). The two extant classes of model evoke either a global increase in mutation rates in a stress response to the environment (hypermutation models) or “normal” mutation rates and mechanisms acting on multiple DNA copies in rare growing cells (cryptic growth models, an example is discussed below). In the Lac system, hypermutation models are supported not solely by detection of mutation at non-lac sites (see Introduction; reviewed by Rosenberg 2001), but also by requirements for components well known to contribute to increased mutation in other contexts and/or assay systems (Friedberg  et al. 1995; Kim  et al. 1997; Wagner and Nohmi 2000; Walker  et al. 2000): the mutagenic SOS response (Cairns and Foster 1991; McKenzie  et al. 2000), the low-fidelity DNA pol IV (McKenzie  et al. 2001), and limiting mismatch repair (Harris  et al. 1997, 1999).

A current cryptic growth model (Andersson  et al. 1998; Hendrickson  et al. 2002) proposes preexisting spontaneous amplification of the lac region as a mechanism to achieve cryptic growth and mutation in some rare cells: multiple lac copies (more β-galactosidase) allow growth; mutation rates are normal, and the increased Lac⁺ mutants in stationary phase result mainly from growth and extra lac copies. Recent versions of this model account for the extensively documented genome-wide hypermutation in Lac⁺ mutants (Torkelson  et al. 1997; Rosche and Foster 1999; Bull  et al. 2000, 2001; Godoy  et al. 2000) and for the requirement for SOS-inducible DNA pol IV (McKenzie  et al. 2001) by invoking induction of the SOS response by amplification (Hendrickson  et al. 2002) or by amplification of the gene encoding pol IV (Kofoid  et al. 2003). This cryptic growth model thus makes hypermutation a by-product of, rather than central to, formation of Lac⁺ adaptive point mutants. Lac⁺ mutations are proposed to result mainly from growth and increased replication of multiple gene copies, occurring more robustly with increasing growth rate, not as part of a stress response or in differentiated stationary-phase cells. The requirement for an RpoS-dependent stress response (Figure 1; Table 2) and the implication that the mutating cells are in stationary phase are incompatible with such cryptic growth models that allow no special circumstances (other than growth) and instead support hypermutation models, which feature increased mutation rates as part of a stress response in growth-limited cells (e.g., Figure 3).

RpoS and models for transient genetic instability: In Figure 3, we suggest that RpoS acts upstream in a common stress response pathway that branches to yield either point mutation or amplification, rather than downstream in promoting survival of cells carrying mutation or amplification. This order of action is supported by the well-characterized role of RpoS in establishing the differentiated state of stationary phase (Hengge-Aronis 2000) and by the epistasis of rpoS to a lexA(Ind^-) mutation (Figure 2). Although it is economical to hypothesize a single role for RpoS in the two genetic instability routes in the common part of a branched pathway, it is also possible that RpoS plays different roles in point mutation and amplification.

The two branches (to point mutation or to amplification; Figure 3) are clearly delineated: the SOS response and DNA pol IV are required only for mutation, not for amplification (McKenzie  et al. 2001), and amplified isolates do not display hypermutation as point mutants do (Hastings  et al. 2000). The point mutation pathway is better defined, requiring DSB-repair proteins (Harris  et al. 1994, 1996; Foster  et al. 1996), F transfer proteins (Foster and Trimarchi  1995a,b; Galitski and Roth 1995), an SOS response (Cairns and Foster 1991; McKenzie  et al. 2000), SOS-induced DNA pol IV (DinB; McKenzie  et al. 2001), and limiting mismatch repair (Harris  et al. 1997, 1999), and is associated with transient genome-wide hypermutation (Torkelson  et al. 1997; Rosche and Foster 1999; Bull  et al. 2000, 2001; Godoy  et al. 2000). In contrast, amplification requires neither SOS nor pol IV (McKenzie  et al. 2001) and is not associated with genome-wide hypermutation (Hastings  et al. 2000). We suggest that point mutation occurs in a transiently hypermutable cell subpopulation in which the SOS response is induced, leading to increased production of error-prone pol IV. The mutations could result from error-prone synthesis (using pol IV) primed during homologous recombinational DSB repair (Harris  et al. 1994). Amplification might result when DS ends (possible sources reviewed by Rosenberg 2001) are repaired beginning with a nonhomologous end-joining event, creating either a duplicated segment or a rolling circle intermediate, using high-fidelity DNA synthesis in non-SOS induced cells (Hastings and Rosenberg 2002).

In this model (Figure 3), acquisition of either a Lac⁺ point mutation or amplified lac DNA allows growth, curtails the RpoS-dependent stress response, and restores genetic stability. That amplification is an endpoint, rather than an intermediate en route to point mutation (as in single-pathway models, e.g., Hendrickson  et al. 2002), is supported by previous findings that amplified clones are neither hypermutated nor particularly mutable and do not lead readily to formation of adaptive point mutants under selection (Hastings  et al. 2000).

RpoS in other stationary-phase mutation mechanisms: RpoS is required for stress-induced mutation in another assay system in E. coli (Bjedov  et al. 2003). In that system mutation in aged colonies also requires RecA, but in contrast to mutation in the Lac system does not have strong dependence on either RecB or the LexA regulon. Thus this appears to be a different mechanism, but one that also requires the RpoS regulon, supporting a general role for RpoS in multiple stationary-phase mutation mechanisms (although not in all stationary-phase mutation systems; see Bridges and Timms 1998). RpoS is required for mutation-dependent adaptation to long-term culture conditions that can give some cells a growth advantage in stationary phase (see results; Zambrano and Kolter 1996), as are pol IV and two other SOS-inducible polymerases (Yeiser  et al. 2002). Also, apparent stationary-phase mutations in Pseudomonas seem to require RpoS (Saumaa  et al. 2002) as do stationary-phase genome rearrangements associated with transposable elements in E. coli (Gomez-Gomez  et al. 1997; Lamrani  et al. 1999) and other bacteria (Ilves  et al. 2001; Saumaa  et al. 2002).

Candidate genes in the RpoS regulon: RpoS-regulated gene products (Loewen  et al. 1998) that might be involved at the DNA level in mutation and amplification include proteins involved in base excision repair (Exo-nuclease III), methylation damage repair (AidB), chromosome segregation/cell division (FtsQ, FtsZ), and prevention of oxidative damage (catalase HPII, Dps). In addition, RpoS downregulates methyl-directed mismatch repair in stationary phase, apparently by decreasing the levels of MutS and MutH (Feng  et al. 1996; Tsui  et al. 1997). An apparent insufficiency of functional MutL facilitates stationary-phase mutation in this system (Harris  et al.  1997, 1999; and perhaps amplification, because the work was done prior to the discovery of amplification as a major component on later days).

Transient genetic instability and the general stress response: The data presented here imply that transient genetic instability in the Lac system occurs in cells that are differentiated by RpoS-dependent gene expression. The well-characterized role of the RpoS regulon (Hengge-Aronis 2000) in protecting cells (particularly at the DNA level) from environmental stresses may seem at odds with a role in supporting genetic instability. However, the evidence that only a subpopulation of cells experiences genetic instability in this system (reviewed in Rosenberg 2001) means that the population as a whole is not threatened. Only a subpopulation is subject to the potentially damaging effects of increased point mutation rates. Recent work demonstrates that the majority of wild E. coli experience stress-inducible mutation, supporting the relevance of increased mutation rates to bacterial survival (Bjedov  et al. 2003; Rosenberg and Hastings 2003). The role of RpoS as the regulator of the general stress response (Hengge-Aronis 2000) suggests that stationary phase might not be the only condition that stimulates mutation and genetic rearrangements such as amplification. Other inducers of the general stress response might also cause transient genetic instability and so promote bacterial adaptation to stressful environments.

Note: A prepublication report released during the review of this manuscript shows that RpoS positively regulates the DinB error-prone DNA polymerase (pol IV; Layton and Foster 2003). Because DinB is required for adaptive point mutation (McKenzie  et al. 2001), this could account for some or all of the role of RpoS in adaptive point mutation. Because DinB is not required for adaptive amplification (McKenzie  et al. 2001), RpoS must play some other role in that process.

Acknowledgement

For providing strains, we thank Mary Berlyn and the E. coli Genetic Stock Center, Roberto Kolter and Steve Finkel, Joe Peters, and P. J. Hastings. For comments on the manuscript, we thank Bryn Bridges, Harold Bull, Susan Gottesman, Steve Finkel, P. J. Hastings, Megan Hersh, Greg McKenzie, Rebecca Ponder, Miroslav Radman, Peg Riley, Andrew Slack, and Shirley Yang. This work was supported by National Institutes of Health grants F32-GM19909 (M.-J.L.) and R01-GM53158.

Footnotes

LITERATURE CITED

Author notes