Evidence That Stationary-Phase Hypermutation in the Escherichia coli Chromosome Is Promoted by Recombination
Harold J. Bull, Gregory J. McKenzie, P. J. Hastings, Susan M. Rosenberg


Adaptive (or stationary-phase) mutation is a group of phenomena in which mutations appear to occur more often when selected than when not. They may represent cellular responses to the environment in which the genome is altered to allow survival. The best-characterized assay system and mechanism is reversion of a lac allele on an F′ sex plasmid in Escherichia coli, in which the stationary-phase mutability requires homologous recombination functions. A key issue has concerned whether the recombination-dependent mutation mechanism is F′ specific or is general. Hypermutation of chromosomal genes occurs in association with adaptive Lac+ mutation. Here we present evidence that the chromosomal hypermutation is promoted by recombination. Hyperrecombinagenic recD cells show elevated chromosomal hypermutation. Further, recG mutation, which promotes accumulation of recombination intermediates proposed to prime replication and mutation, also stimulates chromosomal hypermutation. The coincident mutations at lac (on the F′) and chromosomal genes behave as independent events, whereas coincident mutations at lac and other F-linked sites do not. This implies that transient covalent linkage of F′ and chromosomal DNA (Hfr formation) does not underlie chromosomal mutation. The data suggest that recombinational stationary-phase mutation occurs in the bacterial chromosome and thus can be a general strategy for programmed genetic change.

STATIONARY-PHASE (or adaptive) mutations occur in nondividing or slowly growing cells exposed to a nonlethal selection (reviewed by Drake 1991; Foster 1993; Hall 1993; Symonds 1993; Rosenberget al. 1994; Rosenberg 1997; Lombardoet al. 1999a; Lombardo and Rosenberg 1999). They differ from spontaneous growth-dependent mutations, which occur in dividing cells, before exposure to an environment selective for the mutation, and randomly in the genome (e.g., Luria and Delbrück 1943). In some assay systems for stationary-phase mutation, the mutations may occur preferentially in genes whose functions are selected (Wrightet al. 1999). In the system used here, genome-wide hypermutability appears to underlie adaptive mutations (i.e., those mutations selected) and produce non-adaptive mutations concurrently (Torkelsonet al. 1997; postulated by Hall 1990; Ninio 1991), although nonrandomness in the form of “hot” and “cold” sites for the mutation has been documented (Rosenberg 1997; Torkelsonet al. 1997). Stationary-phase mutations form via multiple different mechanisms, some of which clearly differ from spontaneous growth-dependent mutation (Maenhaut-Michel and Shapiro 1994; Hall 1995; Maenhaut-Michelet al. 1997; Rosenberg 1997; Taddeiet al. 1997; Wrightet al. 1999). The molecular mechanisms of mutation in nongrowing and slowly growing cells under stress provide important models for evolution of microbes in real-world, stressful environments, for mutations that confer resistance to antibiotics and chemotherapeutic drugs, and for mutations that initiate cancer in cells that are not growing actively. Elucidation of mechanisms of mutation in response to selection is modifying core concepts in biological evolution and development (e.g., Cairnset al. 1988; Culotta 1994; Thaler 1994; Shapiro 1997; Pennisi 1998; Caporale 1999). Understanding these mechanisms will illuminate their roles in evolution, development, cancer formation, and genome structure and function, all of which may be underpinned by such dynamic mutational processes.

The best-studied assay for stationary-phase mutation uses Escherichia coli cells carrying a revertible lac frameshift allele on an F′ sex plasmid and no lac genes in the chromosome (Cairns and Foster 1991). Growth-dependent Lac+ revertants, carrying mutations formed prior to plating on lactose minimal medium, appear after about 2 days of incubation on lactose plates. Additional Lac+ mutant colonies appear each day for several days and these carry mutations formed during starvation on the lactose medium (McKenzieet al. 1998; stationary-phase mutations). The stationary-phase mutations form via a unique molecular mechanism that differs from growth-dependent Lac+ mutations as follows:

  1. Homologous recombination functions recA, recB, ruvA, ruvB, and ruvC are required for stationary-phase, but not growth-dependent Lac+ mutation (Harris et al. 1994, 1996; Fosteret al. 1996).

  2. Because RecBCD loads onto DNA only at double-strand DNA ends (DSEs), DSEs are implicated as molecular intermediates in the mutagenic process (Harriset al. 1994).

  3. Formation of stationary-phase Lac+ mutations requires F-encoded transfer functions (Foster and Trimarchi 1995a; Galitski and Roth 1995), but not actual F plasmid transfer (Foster and Trimarchi 1995a,b; Radicellaet al. 1995; Rosenberget al. 1995). One possible explanation for this requirement is that the single-strand nick produced at the transfer origin by transfer (Tra) proteins develops into a double-strand break (DSB) and that this is the major DSB source on the F plasmid (Kuzminov 1995; Rosenberget al. 1995).

  4. Stationary-phase Lac+ mutations are nearly all −1 deletions in small mononucleotide repeats, whereas the growth-dependent Lac+ mutations are heterogeneous (Foster and Trimarchi 1994; Rosenberget al. 1994).

  5. The stationary-phase mutations are attributable to DNA polymerase errors made by the major replicative polymerase, PolIII (Fosteret al. 1995; Harriset al. 1997a).

  6. These errors persist under conditions of insufficient postreplicative mismatch repair (MMR) activity (Longerichet al. 1995), during which the MutL MMR protein becomes limiting (Harris et al. 1997b, 1999a).

The recombination-dependent stationary-phase mutations are proposed to result from DNA replication at sites of DSB repair via homologous recombination (Harriset al. 1994; reviewed by Rosenberg 1997; Lombardo and Rosenberg 1999) as follows: DSBs are suggested to occur during the stress of starvation on lactose medium (see Harriset al. 1994; Kuzminov 1995; Rosenberg et al. 1995, 1996; Bridges 1997; Seigneuret al. 1998, for suggestions on how DSBs could form). RecBCD loads onto DSEs and digests and unwinds the DNA, producing single-stranded DNA ends, which are used by RecA protein for strand invasion of a homologous DNA molecule (Figure 1). The D loops are proposed to prime DNA replication (Harriset al. 1994; Kogoma 1997; see Liuet al. 1999; Motamediet al. 1999) using DNA PolIII (Fosteret al. 1995; Harriset al. 1997a). Polymerase errors are suggested to persist due to transient MMR deficiency (Longerichet al. 1995; Harriset al. 1997b). These become Lac+ (and other) mutations.

The “adaptive” nature of these mutations can be accounted for by a modification of Hall's proposal in which adaptive mutations arise in a hypermutable subpopulation of cells exposed to selection (Hall 1990; see Ninio 1991). Both nonadaptive and adaptive (Lac+) mutations are proposed to form. However, the nonadaptive mutations might not be readily apparent in the main population either because of their low number or due to death of mutant cells that had not also acquired an adaptive mutation. This model was supported in the Lac system by the demonstrations of high frequencies of mutation at multiple sites, genome-wide, in Lac+ colony formers, but not in the main population of (Lac) cells exposed to selection (Torkelsonet al. 1997; Rosche and Foster 1999). These unselected mutations appear to form concurrently with the Lac+ adaptive mutations (not during growth of the Lac+ colony) as seen by their representation in all cells (not sectors) of the Lac+ mutant colonies.

Figure 1.

A model for formation of recombination-dependent stationary-phase Lac+ mutations. (A) A double-strand end (DSE) is proposed to occur [e.g., via processing of a Tra-dependent nick (Rosenberget al. 1995), disintegration (Kuzminov 1995) or stalling (Rosenberget al. 1996; Michelet al. 1997; Seigneuret al. 1998) of a replication fork, or other mechanism (e.g., Rosenberg 1994, 1997; Bridges 1997)]. (B) The DSE is processed by the RecBCD enzyme, creating single-stranded DNA ends (C) that become bound by RecA (small circles), which catalyzes invasion of a homologous duplex to produce a displacement loop (D). (E) The invading strand (in this example a 3′-ended single strand) serves as a primer and loading site for the replicative DNA polymerase PolIII (Liuet al. 1999). Errors produced by PolIII (X) may remain uncorrected due to a transient deficiency in methyl-directed mismatch repair (Longerichet al. 1995; Harriset al. 1997b). The error becomes genetically fixed giving a Lac+ mutation (X in F). An alternate outcome of intermediate D is that strand invasion (perhaps from 5′-ended single-strand invasions, which cannot serve as a primer) leads to (G) homologous recombination with no associated DNA replication (Harriset al. 1996; Rosenberg and Motamedi 1999; Motamediet al. 1999). (- - -) Newly synthesized DNA.

Although it is clear that (1) a fundamentally different mutation mechanism generates the Lac+ stationary-phase mutations, (2) the cells engaging in this mechanism are differentiated [transiently mismatch-repair deficient (Longerichet al. 1995; Harriset al. 1997b) and comprising a small hypermutable subpopulation (Torkelsonet al. 1997)], and also (3) chromosomal genes are mutated concurrently (Torkelsonet al. 1997; Rosche and Foster 1999), the possible relevance of the recombination-dependent stationary-phase mutation mechanism to mutation in the bacterial chromosome has been controversial (e.g., Foster and Trimarchi 1995a,b; Galitski and Roth 1995, 1996; Radicellaet al. 1995; Peterset al. 1996; Benson 1997, discussed below). The issue underlying this question is whether the recombinational stationary-phase mutation mechanism affects the bacterial genome in general.

Recombination is a hallmark of this novel mutation mechanism. Here, we test the role of recombination in hypermutation of chromosomal genes that occurs concurrently with adaptive Lac+ reversion. We find that two recombination-altered alleles, both of which promote recombination-dependent stationary-phase mutation at lac (on the F′), also promote concurrent hypermutation of chromosomal genes. The data imply that recombination-dependent stationary-phase mutation is not strictly an F-plasmid-specific mechanism, but rather is a mechanism for genetic change at multiple sites throughout the genome. We observe that mutations at lac and chromosomal sites occur as independent events, supportive of the idea that these sites are not joined covalently (as an Hfr) at the time of mutation. In contrast, mutation of lac and another F′-borne site does not appear to be independent. The data support the idea that recombination-dependent stationary-phase mutation is a mechanism for genetic change at multiple sites throughout the genome and thus may be a general response to stress and a strategy for evolution.


E. coli strains: A strain unable to revert to Lac+ was used to scavenge carbon sources other than lactose (Cairns and Foster 1991). All other strains are derived from FC40 (Cairns and Foster 1991), which carries a large chromosomal deletion of the lac operon and neighboring genes, and an F′ sex plasmid carrying genes in the lac and proAB region. The lac allele on the F′ has a translational fusion of lacI with lacZ and a +1 frameshift mutation in lacI which is polar on lacZ. The recD derivative is SMR582 carrying recD1903::Tn10miniTet (Harriset al. 1994). The recG derivative is RSH316 carrying recG258::Tn10miniKan (Harriset al. 1996).

Mutation assays: Assays for Lac+ stationary-phase mutation were performed as described (Harriset al. 1996). Assays for unselected secondary mutations were performed by replica-plating Lac+ colonies, obtained in the Lac+ assay after 5 days of incubation, to various indicator and selective media as described by Torkelson et al. (1997). All presumptive secondary mutants were confirmed by streaking from the original Lac+ colony (master colony) for single colonies on the appropriate indicator plate. The purity of Lac+ colonies expressing fermentation mutations was determined by removing the master colonies with plugs of agar, suspending the cells in buffer, diluting, and spreading on minimal (M9 thiamine) lactose plates to obtain ~ 100 Lac+ colonies per plate. The resulting Lac+ colonies were replica-plated to the appropriate MacConkey indicator medium and the numbers of fermentation-defective mutants and fermentation-competent colonies were determined. Typically, >80% of the secondary mutant colonies assayed were pure in that all Lac+ colonies replica-plated were of the mutant phenotype. Mutations resulting in 5-fluorocytosine resistance (5FCr) map to codAB or upp whereas mutations resulting in 5-fluorouracil resistance (5FUr) map only to upp (Torkelsonet al. 1997). upp mutations were not useful in this study because we observed that both the recD and recG mutations are able to suppress the 5FUr and 5FCr phenotypes of a large (>80%) portion of the upp mutations (data not shown) and so only 5FCr 5FUs mutants were included. (Reconstruction experiments with known upp and codA mutations demonstrated that upp mutations that were suppressed for 5FUr by recD and recG were also suppressed for 5FCr. Thus all 5FCr 5FUs are at codAB.) 5FCr colonies were tested for purity as described above. Typically, >80% of 5FCr mutants identified in this manner were pure.

Unselected mutations in Lac starved cells were assayed as described (Torkelsonet al. 1997). Plugs of agar were removed from between visible Lac+ colonies each day and suspended in M9 buffer. Aliquots were spread on LBH and on MacConkey lactose plates and incubated. (This allowed detection of any Lac+ colonies that were not yet visible and had been picked accidentally.) The resulting Lac colonies (each derived from a Lac cell starved on lactose) were screened for unselected mutations by replica-plating.


Strategy for measuring stationary-phase mutation in the bacterial chromosome: Chromosomal mutations coincident with Lac+ stationary-phase mutation can be measured by replica-plating the Lac+ stationary-phase mutant colonies to media selective for particular loss-of-function mutants or to color indicator media for fermentation-defective mutants (Torkelsonet al. 1997). The hypermutation of chromosomal genes is observed in the Lac+ mutants only and not in the neighboring Lac cells, which were also starved on lactose and then rescued, grown into colonies, and replica-plated. Such “Lac stressed cell colonies” display low chromosomal mutation frequencies indistinguishable, in replica-plating assays, from Lac cells never exposed to selection (Torkelsonet al. 1997; Rosche and Foster 1999, and below). Therefore, to score stationary-phase hypermutation of chromosomal genes, we obtained Lac+ stationary-phase mutants to screen for the presence of additional mutations.

For three reasons, we infer that these additional chromosomal mutations occurred during transient, stationary-phase hypermutability and not during subsequent growth of the Lac+ mutant cell into a colony: first, the Lac+ colonies with additional mutations are mostly pure, not mixed (sectored), for the additional mutation, implying that the initial colony-forming cell carried the mutation (Torkelsonet al. 1997; and shown again here, see materials and methods). Second, the Lac+ mutants are not heritable mutator mutants (Longerichet al. 1995; Torkelsonet al. 1997) and, third, they are not heritable stationary-phase mutator mutants (Rosenberget al. 1998); thus they must have descended from a transiently mutable subpopulation. Lac stressed cells, which show low frequencies of additional mutation (Torkelsonet al. 1997; Rosche and Foster 1999, and below), make up the main population.

In recombination-defective strains, no Lac+ stationary-phase mutants arise (Harris et al. 1994, 1996; Fosteret al. 1996). Therefore we tested the role of recombination in chromosomal hypermutation using recombination-proficient cells with elevated stationary-phase Lac+ mutation, recD and recG null mutants.

Rationale for use of recD and recG mutants: We tested whether two recombination gene defects that promote recombination-dependent stationary-phase mutation of lac on the F′ affect mutability of chromosomal genes in stationary phase. recD null alleles confer hyperrecombination (Chaudhury and Smith 1984; Amundsenet al. 1986; Biek and Cohen 1986; Thaleret al. 1989) and enhance stationary-phase mutation (Harriset al. 1994; Foster and Rosche 1999). Strains carrying recG null mutations are hypermutable in Lac+ stationary-phase mutation (Fosteret al. 1996; Harriset al. 1996), and several lines of evidence imply that RecG protein, which is a Holliday junction branch migration helicase (Whitbyet al. 1994), interferes with those recombination intermediates that promote replication (Whitbyet al. 1993; Al-Deibet al. 1996; Harriset al. 1996; McGlynnet al. 1997). Thus both recD and recG mutations increase numbers of the strand-exchange recombination intermediates thought to promote replication and both promote recombination-dependent stationary-phase mutation (see Figure 1).

recD and recG increase mutability of chromosomal genes in Lac+ stationary-phase mutants: Otherwise isogenic rec+, recD, and recG strains were starved in parallel on lactose minimal medium. Following the fifth day of lactose selection, the Lac+ colonies were replica-plated to appropriate indicator and selective media to reveal chromosomal loss-of-function mutants. Chromosomal mutations assayed were among those detected previously by Torkelson et al. (1997; and see materials and methods). The results of three separate experiments are presented in Table 1 and Figure 2. The recD null mutant showed approximately twice as many xylose (Xyl) and maltose (Mal) fermentation-defective mutations per Lac+ colony as did rec+ cells. The increased frequency of chromosomal mutation coincident with Lac+ is similar to the increase in Lac+ stationary-phase mutant frequency in the recD background (Figure 2). Because recD strains are hyperrecombinagenic (Chaudhury and Smith 1984; Amundsenet al. 1986; Biek and Cohen 1986; Thaleret al. 1989), these data suggest that the increased mutability of chromosomal loci and F-borne loci is due to recombination.

In the recG null strain, the frequency of Xyl and Mal mutations per Lac+ mutant was, respectively, 4.6-fold and 6.0-fold higher than in the rec+ control (Table 1 and Figure 2B). Also, fructose fermentation-defective (Fru) mutations, which previously (Torkelsonet al. 1997) and here were so infrequent as to be undetectable in rec+ cells, were detected in the recG strain (Table 1 and Figure 2B). Thus, loss of RecG increases the frequency of chromosomal mutations concurrent with Lac+ stationary-phase mutation. The total increase (Xyl plus Mal plus Fru) is at least 5.7-fold (Table 1 and Figure 2B). Because RecG is a helicase that can unwind and abort recombination intermediates (Whitbyet al. 1993; Al-Deibet al. 1996; Harriset al. 1996; McGlynnet al. 1997) and that inhibits recombination-dependent stationary-phase mutation (Fosteret al. 1996; Harriset al. 1996), these data suggest that allowing recombination intermediates to enter a replication-promoting pathway in the RecG-deficient strain promotes chromosomal mutation, in agreement with the recD data (above).

recD and recG strains increase mutation at unselected chromosomal loci to an extent similar to their effect on Lac+ colony formation (Figure 2). However, their effect on unselected mutations is smaller than that seen on Lac+ colony formation. It may be that chromosomal genes cannot be mutated with the same efficiency as F′-borne genes, perhaps because of some sequence specificity of the mutation mechanism. The apparent difference in recD and recG could reflect varying susceptibility to mutagenesis for the chromosomal loci. We have observed hot and cold sites for unselected chromosomal mutation (Rosenberg 1997; Torkelsonet al. 1997).

Increased mutation is limited to the hypermutable subpopulation: The increase in chromosomal mutations per Lac+ stationary-phase mutant in recD and recG strains indicates that recD and recG loss increases mutability in cells that become Lac+. We wished to know whether the elevated mutability is specific to the hypermutable subpopulation cells or whether loss of recD or recG increases the mutability of all cells exposed to starvation on lactose medium. Because of the large numbers of replica-plated colonies required to detect chromosomal mutations among Lac starved cells [one to two orders of magnitude less frequent than among Lac+ colonies, at 10−4 to 10−5 of the whole population (Torkelsonet al. 1997)], we tested only the recG strain. Stationary-phase mutation is elevated so dramatically by recG that a recG-promoted increase in chromosomal mutability should be readily detectable even in the Lac cells.

View this table:

Chromosomal mutations per Lac+ adaptive mutant are increased in recG and recD cells

To assay chromosomal mutations among Lac stressed cells, those cells were recovered from between visible Lac+ colonies after prolonged starvation and replated nonselectively to form colonies that were then replica-plated to screen for chromosomal mutants (see materials and methods and Torkelsonet al. 1997). The data in Table 2 indicate that the low frequency of Mal and Xyl mutations per Lac stressed cell colony [one to two orders of magnitude lower than per Lac1 mutant (Table 2; also reported by Torkelsonet al. 1997)] is not increased detectably by the recG mutation. By contrast, Mal and Xyl mutations are increased per Lac+ colony (Tables 1 and 2; Figure 2). These data imply that promotion of mutation by the absence of RecG is limited to the hypermutable subpopulation cells.

Figure 2.

(A) Lac+ stationary-phase mutants accumulated over 5 days of selection in rec+, recD, and recG strains. (B) The frequency of unselected chromosomal mutations (Mal, Xyl, and Fru) per Lac+ stationary-phase mutant in rec+, recD, and recG strains. The values are the mean of three separate experiments, with rec+, recD, and recG tested in parallel each time [Table 1, total mean ± 1 SE (error bars)]. Values obtained for B are from the Lac+ colonies reported in A.

Some condition present in the subpopulation, but not the main population, appears to be necessary for high levels of chromosomal stationary-phase mutation, as is the case for Lac+ adaptive mutation. The condition that makes the subpopulation cells mutable could be the occurrence of DNA DSBs or DSEs at which recombination would occur (Harriset al. 1994), limiting MMR (Harriset al. 1997b), or other (Torkelsonet al. 1997; G. J. McKenzie, R. S. Harris, P. L. Lee and S. M. Rosenberg, unpublished results).

Independent events underlie mutation of lac and chromosomal but not F-linked genes: Previously, F′-linked as well as chromosomal genes were hypermutated in Lac+ stationary-phase mutants (Torkelsonet al. 1997). These F′-linked mutations associated with Lac+ could result from recombination-dependent mutation, but might not show increased mutability in recG or recD, if the recombination event that leads to their formation is the same event responsible for mutation at lac (Figure 3). For example, mutations at lac and the nearby locus codAB might occur via polymerase errors made during the same act of DNA synthesis, from the same recombinational DNA repair event (Figure 3A). If so, their coincident frequency would not be increased by conditions that increase only the number of recombination (and synthesis) events, without increasing the error rate per base synthesized (Figure 3Bii). Both recD and recG null alleles are expected to increase replicative strand-exchange intermediates (discussed above) and not error rate per base synthesized, as neither has a general mutator phenotype (Harris et al. 1994, 1996; Fosteret al. 1996).

View this table:

recG increases mutation specifically in the Lac+ population

Figure 3.

Models for mutation at a locus linked to lac on the F′. (A) Proposal that mutation at lac and the linked codAB locus result from a common DNA replication event. lac and codA codB are located near each other on the F′. Among all selected mutational events that give rise to Lac+ colonies, some frequency of the time (1/3, as a simple model) an unselected mutation in codA or codB also occurs, giving rise to a Lac+ FCr colony. (B) Two models to account for the increase in Lac+ mutations in a recG strain. (i) If recG increased the error rate of DNA synthesis (for example by disabling DNA polymerase proofreading or postsynthesis mismatch repair), then the number of mutations per base synthesized would increase. As a consequence, the frequency of unselected mutations at codA, codB would increase among Lac+ colonies. This is not observed (Table 3). (ii) If recG increases the number of synthesis events without affecting the error rate, then the ratio of Lac+ 5FCr (3/9) will not change compared with the rec+ situation (1/3). This latter model is supported by the data in Table 3. (- - -) Newly synthesized DNA.

Unselected mutations at codAB, on the F′, were assayed in Lac+ stationary-phase mutants (materials and methods; Torkelsonet al. 1997). In the case of recG cells (Table 3), there is a 2-fold increase in the frequency of codAB mutation in Lac+ mutants, whereas Lac+ mutation itself was increased 26-fold. These physically linked sites do not show elevated coincident mutation, as the unlinked chromosomal and lac sites do (Tables 1 and 2; Figure 2). This implies that the recombination events that generate Lac+ and codAB mutations are not independent events.

We infer from these data that Lac+ and chromosomal mutations occur during independent events and that both events are stimulated in recD and recG cells. One implication of these results is that lac and the sites mutated in the chromosome do not need to be joined physically (as they would be in an Hfr cell) during chromosomal hypermutation (discussed below).

In recD, mutation at codAB was increased just over twofold relative to rec+, whereas Lac+ mutation increased fourfold (Table 3). Recall that a twofold increase in secondary mutation frequency in recD was also seen for chromosomal loci (Table 1 and Figure 2). This suggests that the absence of RecD affects codAB and lac independently at least for some of the mutation events. Possible bases for these results are discussed below.


The results reported here can be summarized as follows:

  1. Absence of either RecD or RecG increases concurrent mutation of chromosomal sites in Lac+ stationary-phase mutants (Table 1 and Figure 2).

  2. This increase is similar to the increase in Lac+ mutation in recD and recG strains (Figure 2).

  3. The increase in chromosomal mutation frequency is specific to cells that experienced a Lac+ mutation and is not seen in Lac starved cells (at least in the case of recG; Table 2). Because both recD and recG are predicted to promote strand-exchange recombination intermediates leading to replication (Figure 1, and reviewed above), these data support models in which (some) sites on the bacterial chromosome are accessible to recombination-promoted mutation in stationary phase.

  4. The increase in mutability in recG is observed at chromosomal sites but not at codAB on the F′ (Table 3). This implies that mutations at sites linked to lac do not usually occur independently of the Lac+ mutation event. This also implies that lac and chromosomal sites are not linked during mutation of chromosomal sites. We suggest that the same recombination events that lead to Lac+ mutation also lead to mutation of nearby genes, perhaps in the same DNA recombination-replication event. Evidence that recombination events promote DNA replication directly is reported elsewhere (Motamediet al. 1999 and references reviewed therein; see also Courcelleet al. 1997; Kogoma 1997; Liuet al. 1999 for further discussion).

    View this table:
    TABLE 3

    Frequencies of coincident mutation at F′-linked loci codAB and lac in recD and recG strains

Recombination-promoted mutation in the bacterial chromosome: It was suggested that recombination-dependent stationary-phase mutation might be confined to sex plasmids because mutations at lac require F′ transfer (Tra) proteins (Foster and Trimarchi 1995a; Galitski and Roth 1995), though not actual transfer (Foster and Trimarchi 1995b; Radicellaet al. 1995; Rosenberget al. 1995), and because the lac operon on the chromosome is cold for recombination-dependent mutation (Foster and Trimarchi 1995a; Radicellaet al. 1995; Rosche and Foster 1999; M.-J. Lombardo and S. M. Rosenberg, unpublished results). Previous evidence arguing against F′ specificity included, first, hypermutation of chromosomal genes during Lac+ adaptive mutation (Torkelsonet al. 1997; Rosche and Foster 1999; this study) and, second, the demonstration of chromosomal hot and cold spots for mutation (Rosenberg 1997; Torkelsonet al. 1997), which can explain why not all chromosomal sites mutate recombinationally.

The demonstrations that recG and recD promote coincident chromosomal mutation (Figure 2; Tables 1 and 2) suggest that chromosomal sites are susceptible to recombination-dependent mutation. Note that we cannot test recombination dependence directly because blocking recombination via, e.g., loss of RecA, RecB, or RuvA, B, or C functions abolishes stationary-phase Lac+ mutation (Harris et al. 1994, 1996; Fosteret al. 1996), and we have scored chromosomal mutations only in cells that are also Lac+. Thus, although further, direct evidence is required to demonstrate conclusively that recombination-dependent mutation occurs in the E. coli chromosome, the current information is most easily explained by such a model.

Independence of Lac+ and coincident chromosomal mutations: The finding that the coincident mutation frequency of lac and chromosomal sites increases in recD and recG cells (Tables 1 and 2; Figure 2) implies that the mutation frequency at each site is increased by these alleles. These results bear on the possibility that although it occurs in the chromosome, hypermutation during Lac reversion might actually require integration of the F′ into the chromosome. This occurs when Hfr chromosomes form (e.g., Lloyd and Low 1996). We found previously that Lac+ mutants carrying chromosomal mutations are not enriched for Hfr's (Lombardoet al. 1999b). However, we could not rule out the possibility that chromosomal mutations form in short-lived Hfr cells, which subsequently re-form the F′ (Lombardoet al. 1999b). The mostly independent stimulation of mutation in chromosomal and lac genes by recD and recG (Tables 1 and 2; Figure 2) does not support such models.

Site-specificity and the role of the F: We have suggested that the key feature that allows some sites, and not others, to mutate recombinationally is occurrence of DNA DSBs at which RecBCD loads (Harriset al. 1994; Rosenberget al. 1995; Rosenberg 1997; Torkelsonet al. 1997). In this view, Tra proteins activate the F′ by nicking the origin of transfer (Rosenberget al. 1995), and hot and cold sites on the chromosome correspond with sites that are more or less susceptible to DSBs (reviewed by Rosenberg 1997).

Although the results presented here suggest that the F is not needed in cis with the DNA that mutates (discussed above), it remains possible that trans-acting functions encoded by the F are required for mutation of chromosomal genes. The F encodes several proteins that interact with DNA, including its own single-strand DNA binding protein, a topoisomerase-like double-strand endonuclease, components that modify the bacterial SOS response, and many of the transfer proteins (reviewed by Bagdasarianet al. 1992; Frostet al. 1994; Yarmolinsky 1995; Firthet al. 1996). Whether recombination-dependent stationary-phase mutation and hypermutation of unselected genes can occur in the absence of sex plasmids is not yet known (see Note added in proof).

recD and coincident mutation in the Fand chromosome: recD null mutants are hyperrecombinagenic (Chaudhury and Smith 1984; Amundsenet al. 1986; Biek and Cohen 1986; Thaleret al. 1989), hypermutable in recombination-dependent stationary-phase Lac mutation (Harriset al. 1994; Rosenberget al. 1994), and recently have been seen to increase F′ copy number relative to the chromosome (Foster and Rosche 1999). The stationary-phase hypermutation at lac in recD cells might have resulted from hyperrecombination in recD cells (Harriset al. 1994) or from more lac copies available for mutation in those cells (Foster and Rosche 1999) or from both. The finding that chromosomal gene mutability increases about as much as lac does in recD cells supports the recombinational idea and does not support the idea of an effect based purely on increased F′ copy number relative to the chromosome.

A perplexing result is that, unlike recG, the recD effect on chromosomal and F′ sites was similar (Tables 1 and 2). This could indicate a global (stationary-phase specific; Harriset al. 1994) twofold mutator activity in recD strains. However, another interpretation is possible. Loss of the RecD subunit changes RecBCD enzyme (Amundsenet al. 1986; Palas and Kushner 1990) and prevents Chi recognition by the enzyme (Chaudhury and Smith 1984; Thaleret al. 1989). Whereas most recombination models include RecBCD-mediated digestion of DNA from a double-strand end up to a Chi site followed by recombination at Chi (e.g., Rosenberg and Hastings 1991; Myers and Stahl 1994; Anderson and Kowalczykowski 1997), in recD (exonuclease-defective) cells, the RecBC(D) enzyme promotes recombination immediately at the DNA end at which it loads (Thaleret al. 1989). This would change the position of strand-invasion events and, in models in which recombination primes replication, would alter the positions of synthesis tracts (Figure 4). Two loci might be synthesized on the same tract in rec+ cells and on different tracts in recD cells (Figure 4), leading to uncoupling of lac and codAB mutation in recD cells.

Implications for the hypermutable subpopulation:

  1. Recombination and the hypermutable subpopulation: Previously, lac and an F′-borne gene were observed to show no increase in coincident mutation in recG cells, leading to the suggestion that recG somehow increases the size of the hypermutable subpopulation, rather than the mutability per subpopulation cell (Foster 1997). Our results for the F′ (Table 3) agree with those reported previously (Foster 1997). However, the data we have obtained on chromosomal site mutability (Table 1 and Figure 2) do not support the idea that recG increases subpopulation size, but rather imply that the mutability of subpopulation cells is increased by promoting strand-exchange intermediates. We suggest that at linked sites, the secondary mutation event and the primary Lac+ mutation event are not independent, such that their coincident mutation frequency does not reflect the mutability per cell.

    Figure 4.

    DSEs and therefore DNA synthesis tracts primed in double-strand break repair would fall in different places in rec+ and recD cells. This is so because the RecBCD enzyme is a double-strand DNA exonuclease whereas the RecBC(D) enzyme lacks exonuclease activity, but is recombination-proficient (Chaudhury and Smith 1984; Amundsenet al. 1986; Biek and Cohen 1986). Dashed lines represent newly synthesized DNA. See text for discussion.

  2. How many mutable subpopulations? Torkelson et al. (1997) reported that single (Lac+), double (Lac+ plus an additional mutation), and triple mutants (Lac+ plus two additional mutations) fit a Poisson distribution if a mutation rate of 5 × 10−3 mutations per cell per day occurred in 10−4 to 10−5 of the cells of the whole population. The ability to fit the data to a Poisson distribution was taken to imply that mutation to Lac+ and the formation of associated mutations occur at about the same frequency. This is compatible with the hypothesis that Lac+ and associated mutations occur by the same mechanism and arise from the same subpopulation, but does not exclude the possibility that there are two or more mechanisms affecting different but overlapping subpopulations. A different data set and method of calculation led to the conclusion that only 10% of the Lac+ mutations result from the hypermutating subpopulation that gives rise to the secondary mutations (Rosche and Foster 1999). However, the data on associated mutations were few, such that with 95% confidence limits applied to them, as many as 98% of the Lac+ mutants could have arisen from the hypermutable subpopulation.

    Nevertheless, the general concept of different but overlapping subpopulations may be applied to the results presented here. The recD and recG mutations might increase the size of the subpopulation undergoing mutation to Lac+ such that the subpopulation now includes a higher proportion of those cells of the subpopulation that gives rise to associated mutations. This would have the effect of increasing the frequency of associated mutations among the Lac+ mutants without increasing the mutation rate in the hypermutating subpopulation. Invoking two populations and two mechanisms is a more complicated and thus less attractive model.

  3. Subpopulation size and mutation rate: The proposed mutation rate of 5 × 10−3 mutations per cell per day of Torkelson et al. (1997) may seem lethally high, and yet no net cell death is observed (Cairns and Foster 1991, and many subsequent references). It should be noted, however, that first, even massive death of a subpopulation of 10−5 of the cells would be unnoticeable when measuring cell viability and, second, because only some (hot) sites are mutable (see discussion of hot and cold sites above in Rosenberg 1997), many essential genes may be spared, so death might not occur (supported by data of Foster 1997).

Significance: The findings reported here suggest that recombination-dependent stationary-phase mutation is a mechanism of genetic change under stress that can alter at least some of the cell's primary genetic reserve, the chromosomal genes. This inference will hold whether or not components on the F′ are found to be required for the chromosomal hypermutation. Sex plasmids are natural genetic elements and if they provide such conditional mutability to their hosts, this could be an advantageous, selected feature for their host cells.

Several aspects of recombination-dependent stationary-phase mutation may also be general to other organisms and circumstances. Mutation promoted by DSB-repair recombination in yeast has been demonstrated (Strathernet al. 1995; Holbeck and Strathern 1997), as has recombinational involvement in mutation in vertebrates including mammals (reviewed by Maizels 1995; Harriset al. 1999b). Findings suggestive of this association abound in many organisms (Demerec 1962, 1963; Magni and von Borstel 1962; Paszewski and Surzycki 1964; Esposito and Bruschi 1993). Additionally, the MMR system, which becomes limiting during stationary-phase mutation (Harris et al. 1997b, 1999a), is conserved in eubacteria and eukaryotes. Its loss of function is also a powerful force of genetic change in other organisms (reviewed by Radmanet al. 1995; Kolodner 1996; Modrich and Lahue 1996), and its transient diminution would be potentially more important in multicellular organisms that suffer more drastic consequences from mutagenesis of component cells. The mechanism of action, control, and scope of this stationary-phase mutation mechanism in E. coli will illuminate a path toward understanding conditional mutagenesis, programmed or accidental, in all of these systems.


We thank R. S. Harris, M.-J. Lombardo, and J. Petrosino for comments on the manuscript. This work was supported by National Institutes of Health grants R01 GM53158 and R01 AI43917, and in part by an Alberta Cancer Board Postdoctoral Fellowship (H.J.B.), by a D.O.D. Breast Cancer Research Program Scholarship (G.J.M.), and by Alberta Heritage Foundation for Medical Research Senior Scholarship and Medical Research Council of Canada Scientist awards to S.M.R.


  • Note addeed in proof: Recent work of Godoy et al. (2000) both confirms previous findings of chromosomal hypermutation during Lac+ adaptive mutation (Torkelsonet al. 1997; Rosche and Foster 1999) and indicates that there is indeed an F′-supplied function that promotes stationary-phase mutation. As discussed above, the results presented here imply that any F′-related function would act in trans in mutation, not via Hfr formation (above, and Lombardoet al. 1999b).

  • Communicating editor: P. G. Young

  • Received September 5, 1999.
  • Accepted December 10, 1999.


View Abstract