Proc. Natl. Acad. Sci. USA Vol. 96, pp. 6862–6867, June 1999 Genetics

The role of transient hypermutators in adaptive in

WILLIAM A. ROSCHE† AND PATRICIA L. FOSTER†‡, WITH APPENDIX BY JOHN CAIRNS§

†Department of Environmental Health, Boston University School of Public Health, Boston University School of Medicine, Boston, MA 02118; and §Clinical Trial Service Unit, Radcliffe Infirmary, Oxford OX2 6HE, United Kingdom

Communicated by Philip Hanawalt, Stanford University, Stanford, CA, April 23, 1999 (received for review January 10, 1999)

ABSTRACT Microbial populations under nonlethal se- that bear adaptive than among cells that do not (3). lection can give rise to mutations that relieve the selective In four cases in which this has been tested, the prediction has pressure, a phenomenon that has come to be called ‘‘adaptive been confirmed (3–6). However, it is not possible to determine mutation.’’ One explanation for adaptive mutation is that a from those results what proportion of the mutations that occur small proportion of the cells experience a period of transient during selection arise from hypermutating cells—that will hypermutation, and that these hypermutators account for the depend on the proportion of cells that are in the hypermutable mutations that appear. The experiments reported here inves- state and the degree to which their mutation rate is elevated tigated the contribution that hypermutators make to the (7, 8). Two independent measurements are needed to solve for .mutations occurring in a Lac؊ strain of Escherichia coli during these two unknowns selection for lactose utilization. A broad mutational screen, In the experiments reported here, we used a broad muta- loss of motility, was used to compare the frequency of non- tional screen, loss of motility, to compare the frequency of selected mutations in starved Lac؊ cells, in Lac؉ revertants, nonselected mutations in starved LacϪ cells, in selected Lacϩ and in Lac؉ revertants carrying yet another nonselected revertants, and in those few Lacϩ revertants that carried an mutation. These frequencies allowed us to calculate that the additional mutation. This procedure allowed us to solve the hypermutating subpopulation makes up Ϸ0.06% of the pop- equations given in the Appendix and thereby to estimate both ulation and that its mutation rate is elevated Ϸ200-fold. From the size of the hypermutating subpopulation, p, and the these numbers we conclude that the hypermutators are re- magnitude of its increase in mutation rate, M. Our results show sponsible for nearly all multiple mutations but produce only that, even though the hypermutating minority is responsible Ϸ10% of the adaptive Lac؉ mutations. for nearly all of the multiple mutations, it makes only a small contribution to the adaptive Lacϩ mutations occurring during Although spontaneous mutations occur at random while cells lactose selection in E. coli strain FC40. are proliferating, apparently static microbial populations un- der nonlethal selective pressure also accumulate mutations. In MATERIALS AND METHODS some cases, these mutations appear to be confined to genes that give an adaptive phenotype, and for this reason the Bacterial Strains. E. coli strain FC722 is a rifampicin- R Ϫ ⌬ phenomenon has come to be called directed or adaptive resistant (Rif ) derivative of P90C [F ara (lac proB)X111 thi Ј mutation (1). Most models for adaptive mutation invoke a (9)] that carries the F 128 episome with a mutant lac allele, ⌽ S genomewide process that produces genetic variants at random (lacI33-lacZ) (10), and a tetracycline-sensitive (Tet ) defec- but postulate that these variants are transitory unless the cell tive Tn10 element (4). FC722 is isogenic to strain FC40 (11) S achieves a mutation that relieves the selective pressure (2). The except for the episomal Tet element. FC1259 is FC722 but mutL::Kan [allele from strain GM4250 (12)] and was con- most extreme of these ‘‘trial-and-error’’ models is the ‘‘hyper- Ϫ mutable state’’ proposed by Hall (3). According to this model, structed by P1vir transduction. Strain FC691 is F but carries ⌽ most cells in a population under selection do not mutate, but the (lacI33-lacZ) allele on its chromosome. FC691 is a a minority transiently experience a high mutation rate and derivative of GM4270 (obtained from M. G. Marinus, Uni- soon die unless a useful mutation occurs. versity of Massachusetts Medical School) created by curing its Escherichia coli strain FC40 is unable to catabolize lactose episome with acridine orange (see ref. 13). FC29 is rifampicin- S Ϫ (LacϪ) but reverts to lactose utilization (Lacϩ) when lactose is sensitive (Rif ) and carries a nonrevertible Lac allele on its its sole carbon and energy source. Although this reversion has episome (11). been considered an example of adaptive mutation, during Media. M9 minimal medium (14) was supplemented with lactose selection nonselected mutations in a nearby gene 0.1% lactose (Lac), maltose (Mal) (both from Difco), glycerol accumulate in the LacϪ population at about the same rate as (Gly), or D-galacturonic acid (GA) (both from Sigma). When ϩ ␮ do Lac mutations (4). Thus, the mutational process in FC40 required, antibiotics were added at 100 g/ml rifampicin (Rif), ␮ ␮ does not meet the original definition of adaptive mutation, 10 g/ml tetracycline (Tet), 22.5 g/ml kanamycin (Kan), or 10 ␮ which specified that only adaptive mutations should appear. g/ml spectinomycin (Spc) (all from Sigma). Resistances to ϩ 5-fluorocytosine (5FC) and 5-fluorouracil (5FU) were However, we continue here to call the Lac mutations ap- ␮ pearing during lactose selection ‘‘adaptive’’ simply to distin- screened on M9-Gly medium supplemented with 20 g/ml 5FC ϩ ␮ guish them from the Lac mutations occurring during nonse- or 10 g/ml 5FU (both from Sigma). Mutators were screened lective growth and from nonselected mutations occurring on MacConkey Base Agar (Difco) supplemented with 1.0% during lactose selection. arabinose (Mac-Ara) or salicin (Mac-Sal). Motility was A specific prediction of the hypermutable-state model is that screened on LB medium (14) solidified with 0.35% Bacto Agar nonselected mutations occur at a higher frequency among cells (Difco).

The publication costs of this article were defrayed in part by page charge Abbreviations: Lac, lactose; Rif, rifampicin; Tet, tetracycline; Mal, maltose; Gly, glycerol; 5FC, 5-fluorocytosine; 5FU, 5-fluorouracil; payment. This article must therefore be hereby marked ‘‘advertisement’’ in MMR, mismatch repair; Mac, MacConkey agar. accordance with 18 U.S.C. §1734 solely to indicate this fact. ‡To whom reprint requests should be addressed. e-mail: pfoster@ PNAS is available online at www.pnas.org. bu.edu.

6862 Downloaded by guest on October 1, 2021 Genetics: Rosche and Foster Proc. Natl. Acad. Sci. USA 96 (1999) 6863

Experimental Techniques. Lacϩ revertants were selected on the episome (strain FC722) or on the chromosome (strain M9-Lac plates as described (11). About 107 FC722 cells, 106 FC691) (Fig. 1). However, when the LacϪ allele is on the FC1259 cells (each with 109 FC29 scavengers), or 109 FC691 chromosome, adaptive mutation to Lacϩ has different genetic cells (without scavengers) were plated on each of 60, 48, and requirements and occurs at an Ͼ100-fold lower rate (ref. 13, 80 M9-Lac plates, respectively. Each plate received cells from and unpublished results; Fig. 1). In contrast, loss of methyl- an independent culture grown to saturation in M9-Gly me- directed mismatch repair (MMR) (16) increases the rate of ϩ dium. Each day for 5 days after plating, new Lac colonies adaptive mutation to Lacϩ 100-fold (17). Strain FC1259 is were counted, and their positions were marked. The plates isogenic to FC722 but defective in mutL, which encodes a were incubated at 37°C for an additional day to allow the component of the MMR system (18). We tested for nonse- colonies that appeared on day 5 to grow to the same size as lected mutations in Lacϩ revertants and starved LacϪ cells of earlier arising colonies. Plates were then stored at 4°C. Each each of these three strains. colony was assigned an identification number specifying its Mutations in as many as 55 genes, Ϸ1% of the E. coli position and day of appearance, and then some of its cells were , could affect motility (15, 19). Because motility is a transferred (patched) with a sterile wooden toothpick to a novel mutational screen, we tested our assay for reproducibility rectangular Nunc Omni tray containing 50 ml of solid M9-Lac (see Materials and Methods). Of the five mutant motility ϩ Rif medium for strains FC722 and FC1259 or M9-Lac phenotypes identified (here collectively called Mot*), MotϪ medium for strain FC691. The cells were patched in a pattern (no swimming) was the most reproducible, and so it is the only corresponding to the array of a standard 96-well microplate. phenotype given in Tables 1 and 3 and Fig. 2. Eighty-five After transfer, the Omni trays were incubated at 37°C for 48 percent of the Lacϩ clones grew vigorously on lactose; of these hours. ϩ Ϫ Ϫ strongly Lac clones, 2% had an apparent Mot phenotype in Starved Lac cells were obtained by taking plugs from the the initial screen, and 70% of these were confirmed to be lactose plates each day (11) and plating an appropriate dilution composed entirely of motility-defective cells on retesting (see on M9-Gly plates (plus Rif for strains FC722 and FC1259), below for the characteristics of the remaining clones). If the which were then incubated at 37°C for the duration of the cells were only weakly Lacϩ, this value dropped to 15%; poor experiment. These colonies were also assigned identification growth on the lactose medium used for the initial gridding numbers and then patched to Omni trays containing M9-Gly ϩ results in fewer cells being transferred to the motility-screening Rif medium for strains FC722 and FC1259 or M9-Gly medium, giving an apparent motility deficiency. medium for strain FC691. Of the strongly Lacϩ clones that had Mot* phenotypes Cells were transferred from the first Omni trays to screening (including MotϪ), about 16% proved to be composed of media by using a 96-pin replicator (Boekel Scientific, Feast- mixtures of wild-type and Mot* cells. Note that pure Lacϩ erville, PA). When transferring to motility agar, the cells were Mot* colonies would be produced only if the two mutations embedded in the agar by piercing the surface with the prongs occurred simultaneously or if the Mot* mutation occurred first. of the replicator. Motility (Mot) was scored after 16–19 hr of Lacϩ colonies with a large number of Mot* cells could indicate incubation at 37°C. Depending on the area of cell-swimming that the hypermutational state lasts for several cell divisions from the center, the motility of each clone was scored as after a Lacϩ revertant begins to grow. Alternatively, the Mot* wild-type, increased, slightly reduced, reduced, very reduced, mutation could have persisted as uncorrected heteroduplex or minus. To confirm these phenotypes, roughly one-third of DNA, which would produce a 50:50 mixture of Lacϩ Mot* and the clones were retested. The entire original colony was Lacϩ Motϩ cells. Although potentially interesting, mixed resuspended, dilutions were plated on minimal medium, and at clones have been excluded from the results presented here. least five single colonies for each original colony were patched ϩ The reasons for nonreproducibility of the motility defects to Omni trays containing LB Rif or LB medium. These were among the remaining strongly Lacϩ clones are currently replicated to LB motility agar and scored as above. The various ϩ unknown but probably include broth-sensitive mutants, tem- Lac and motility phenotypes differed in their reproducibility, porary metabolic problems caused by nutritional shift, irreg- and the numbers reported below have been reduced by the appropriate correction factor (see Table and Fig. legends). A possible motility gene has been found near 5.3 minutes of the E. coli chromosome (15) and may be carried on FЈ128. To test whether any Mot phenotypes were caused by mutations in this gene conferring a dominant-negative phenotype, the epi- somes from 83 Lacϩ clones of FC722 that had motility defects were mated into a Motϩ ProϪ LacϪ StrepR recipient by selecting for Proϩ StrepR. All exconjugates were Lacϩ Motϩ. Cells were also replicated to M9-Mal, M9-GA, M9-Gly ϩ 5FC, M9-Gly ϩ 5FU, and LB ϩ Tet media. Because these selections picked up few potential mutants, we rechecked every one of them as described above. Mutants that were 5FCR or 5FUR were also tested for their ability to transfer their resistances by conjugation. Cells were screened for being heritable mutators on Mac-Ara, Mac-Sal, and LB ϩ Spc media; these screens were confirmed to identify mutators by testing them with derivatives of FC40 carrying the known mutator alleles mutL and dnaQ49. Starved cells were con- Ϫ ϩ firmed to be Lac by replicating them onto M9-Gly X-Gal ϩ ␤ FIG. 1. The accumulation during lactose selection of Lac rever- (5-bromo-4-chloro-3-indolyl -D-galactoside), on which me- ⌽ dium Lacϩ cells are blue. tants of the (lacI33-lacZ) allele on the episome and on the chromo- some. ■ (left axis), strain FC722 (episomal lac allele); E (right axis), strain FC691 (chromosomal lac allele). Values given are the means of RESULTS 60 FC722 and 40 FC691 cultures. (The Lacϩ counts from jackpots in ϩ two cultures of FC722 and five cultures of FC691 were eliminated from During lactose selection, Lac mutants accumulate at a nearly the results.) Error bars are SEMs, some of which are smaller than the constant rate whether the ⌽(lacI33-lacZ) allele is located on symbols. Downloaded by guest on October 1, 2021 6864 Genetics: Rosche and Foster Proc. Natl. Acad. Sci. USA 96 (1999)

ularities in the screening plates, and gridding and picking errors. The proportions of selected Lacϩ and starved LacϪ cells that had a MotϪ phenotype are given in Table 1. There was little difference among the three strains in the proportion of Lacϩ revertants that were MotϪ (␹2 ϭ 2.9, P ϭ 0.24). However, whether the lac allele was on the episome (strain FC722) or the chromosome (strain FC691), the frequency of MotϪ mutations among Lacϩ revertants was significantly elevated relative to starved LacϪ cells (␹2 ϭ 37 and 17, respectively, P ϽϽ 0.01). Thus, in these MMR-competent cells, selection for Lacϩ increased the frequency of nonselected MotϪ mutations at least 20-fold. In contrast, loss of MMR (strain FC1259 ϭ mutL, episomal lac allele) had little effect on the frequency of MotϪ mutations among the Lacϩ cells but increased the frequency of MotϪ mutations among the LacϪ cells 40-fold so that the frequency of MotϪ mutations was the same in both populations (␹2 ϭ 0.5, P ϭ 0.82). This result suggests that MMR (or some other MutL-dependent pathway) is wholly or partially deficient in the hypermutators (see Discussion). Fig. 2 shows the percentage of Lacϩ and LacϪ cells of FC722 (ϭ MutLϩ, episomal lac allele) that were MotϪ as a function of time after plating on lactose medium. Among the Lacϩ cells,

the frequency of mutations increased linearly. The number of Ϫ ϩ MotϪ mutants among the nonselected LacϪ cells was too small FIG. 2. The accumulation of Mot mutations among Lac and LacϪ cells of FC722 during lactose selection. ■,Lacϩ revertants; ᮀ, to allow us to determine whether their frequency was changing Ϫ ϩ Lac cells. Because it takes 2 days for a Lac colony to become visible, with time. ϩ ϩ Ϫ the curve for the Lac cells has been displaced 2 days to the left. To To test whether any of the Lac or Lac clones were stable correct for efficiency of detection, the numbers of MotϪ clones were mutators, cells were replicated onto Mac-Ara and Mac-Sal to reduced by 30% in strongly Lacϩ clones, by 85% in the few weakly score for increased papillation and onto LB-Spc to score for an Lacϩ clones, and by 0% in LacϪ clones (see Materials and Methods). increase in the number of resistant colonies. Among the 3,168 Lacϩ revertants of FC722 (ϭ MutLϩ, episomal lac allele), 11 method used by Torkelson et al. overestimated the number of (0.35%) were stable mutators. Four of these also had motility 5FUR clones by allowing a density-dependent protection of defects (two being MotϪ), but none had any of the other some of the cells transferred onto the 5FU medium. Mutations screened phenotypes (see below). No stable mutators were in the episomal codBA operon give a 5FCR phenotype, whereas found among starved LacϪ FC722 cells, nor among Lacϩ or mutations in the chromosomal upp gene give both 5FUR and LacϪ cells of FC691 (ϭ MutLϩ, chromosomal lac allele). The 5FCR phenotypes (21) (although our 5FUR mutants were only 0.35% genotypic mutators among a selected population is in weakly 5FCR). Thus, our 8-fold difference in the frequencies good agreement with the 0.5% found in a previous study (20) of 5FCR and 5FUR mutants is consistent with the behavior of of the enrichment of mutators achieved simply by selecting for the ⌽(lacI33-lacZ) allele itself, which reverts at a lower rate new mutations. when on the chromosome (Fig. 1). ϩ ϩ In addition to motility, we screened Lacϩ and starved LacϪ Of 583 Lac revertants of FC691(ϭ MutL , chromosomal cells for five other phenotypes (Table 2).§ Three of these were lac allele), none had any of the third phenotypes given in Table also included in a previous study by Torkelson et al. (6). The 2; however, lacking the dTn10 element, this strain could not 0.06% MalϪ mutants and 0.25% 5FCR mutants among the have become TetR and was, for some unknown reason, weakly ϩ Lacϩ FC722 cells we obtained is in good agreement with their 5FCR.Of480Lac revertants of FC1259 (ϭ mutL, episomal Ϫ Ϫ results (0.07% and 0.27%, respectively). The major difference lac allele), one was Mal and three were 5FCR. None of Lac between the two studies is that we found a 10-fold lower starved cells screened in any background had any of these third frequency of 5FUR mutants among Lacϩ cells (0.03% versus phenotypes. 0.3%). A possible explanation is that the replica-plating Cairns (8) has modeled the contribution that a hypermu- tating minority would make to the frequency of selected §One Lacϩ colony of FC722 was composed entirely of TetR cells, which mutations among the population under selection. In this is a lower frequency of Lacϩ TetR double mutants than previously simplest of possible models, the population is assumed to reported for this strain (4). However, we found five apparent Lacϩ contain only two cell types: the majority plus a small propor- TetR double mutants that, when retested, segregated the two phe- tion (p) with a mutation rate (for every class of mutation) M notypes. Four of these were stable mutators. If these five are included, ϩ times higher than the majority. In this model, there are two the number of apparent Lac TetR doubles is not significantly different between the two studies (␹2 ϭ 2.13, P ϭ 0.14), suggesting unknowns (p and M), and they can be determined with the that many of the Lacϩ TetR double mutants obtained in the previous following two independent measurements (see Appendix): (i) Ϫ ϩ study were due to stable mutators. R1/0, the frequency of Mot mutations among the Lac

Table 1. The frequencies of MotϪ clones Number of MotϪ clones͞total tested mutL Location of the Ratio of the frequency of Strain genotype lac allele Lacϩ revertants (%) LacϪ cells (%) MotϪ Lacϩ to MotϪ LacϪ FC722 ϩ Episome 48͞3,168 (1.5) 2͞2,878 (0.07) 22 FC691 ϩ Chromosome 12͞583 (2.1) 0͞956 (Ͻ0.1) Ͼ20 FC1259 Ϫ Episome 12͞480 (2.5) 12͞437 (2.8) 0.9 To correct for efficiency of detection, the numbers of MotϪ clones were reduced by 30% in strongly Lacϩ clones, by 85% in the few weakly Lacϩ clones, and by 0% in LacϪ clones (see Materials and Methods). Downloaded by guest on October 1, 2021 Genetics: Rosche and Foster Proc. Natl. Acad. Sci. USA 96 (1999) 6865

revertants (singles) divided by the frequency of MotϪ muta- Lacϩ singles than in the population as a whole, but it would not Ϫ ϩ tions among the Lac cells (zeros); and (ii)R2/0, the frequency be still higher in Lac doubles. Because we find that the of MotϪ mutations among Lacϩ revertants that bear another frequency of MotϪ mutations is far higher in doubles than in ϩ detected mutation (doubles) again divided by the frequency of singles, it follows that most Lac revertants are arising in the MotϪ mutations among the LacϪ cells (zeros). Our data in general population rather than in the hypermutating minority. Tables 1 and 2 for FC722 (ϭ MutLϩ, episomal lac allele) are The simplest model postulates that a certain proportion (p) sufficient to give values of R1/0 and R2/0 (see below). of all cells are hypermutators, and in them the mutation rate The least reliable numbers entering into these calculations is M-fold higher than in the rest of the population. Our are the number of MotϪ mutations in the LacϪ zeros, 2/2,878, experiments have given us two independent measurements: ϩ and in the Lac doubles, 2/13 (Table 2). Although each the factor R1/0, by which Mot mutations are more common in ϩ Ϫ frequency is significantly different from 48/3,168, the fre- Lac singles than in the general Lac population, and the Ϫ ϩ ϩ ␹2 ϭ factor R2/0, by which Mot mutations are more common in Lac quency of Mot mutations among Lac cells in general ( Ϫ 36.7, P ϽϽ 0.001, and ␹2 ϭ 8.4, P ϭ 0.004, respectively), 2 is a doubles than in the general Lac population. These two very small number. However, we can increase our confidence measurements allow us to calculate the values of the two in the estimates of p and M by including the cells that had other independent parameters, p and M (see Appendix). This and Mot phenotypes. These phenotypes were not always repro- other experiments then lead to certain conclusions about the ducible, but extrapolating from the clones that were retested hypermutating minority. (about 30% of the total), the frequency of all classes of Mot The Frequency and Mutation Rate of the Hypermutators. ϩ Ϸ mutations (Mot*) among the Lac cells (the singles) was Our data imply that 0.06% of all cells are hypermutators and Ϸ 210/3,168 (6.6%). Because their numbers were far fewer, all of that their mutation rate is 200-fold higher than that of the the apparent Mot* mutants found in the LacϪ zeros and the rest of the population. These estimates are not very precise Lacϩ doubles were retested, and the actual frequency of Mot* because they depend on some rather small numbers (see mutations was 10/2,878 (0.35%) and 8/13 (61.5%) among the Tables 1–3), but they are not significantly different from the zeros and doubles, respectively. Each of these three frequen- values suggested by Ninio (7) on theoretical grounds. cies is significantly different from the others (␹2 Ͼ 60, P ϽϽ We have assumed that there are only two populations of 0.001 in each case). cells, high and low mutators. However, there may be many The calculations of p and M based on the MotϪ data for populations of different sizes and with different mutation FC722 (ϭ MutLϩ, episomal lac allele) are given in Table 3. rates, which together give our values for p and M. To determine Because the frequency of multiple mutations increases linearly whether another population exists, a third independent vari- with time (Fig. 2) but different numbers of cells were screened able, e.g., R3/0, would have to be determined. each day, we corrected for this by calculating the frequencies The Proportion of Mutants That Arise in Hypermutators. If P ϭ 0.06% and M ϭ 200, Ϸ10% of all single mutations and per cell-day (the number of cells screened multiplied by the Ͼ number of days they were on the lactose plates before starting 95% of all double mutations will occur in hypermutators (see Eqs. 2 and 4 in the Appendix). Our preliminary results indicate to form a colony). From these calculations, the proportion of Ϫ that the sequence changes that revert the episomal lac allele hypermutators, p,is6ϫ 10 4, and their mutation rate, M,is (strain FC722) do not differ between singles and doubles 230-fold higher than that of the main population. If the more Ϫ4 (unpublished results). This suggests, but does not prove, that frequent Mot* mutations are used, p ϭ 8 ϫ 10 and M ϭ 190. ϩ the mechanism that generates Lac revertants of the episomal lac allele is the same in normal and hypermutating cells (see DISCUSSION below). The Source of the Hypermutating Minority. Both Ninio (7) Recent experiments (4, 6) have shown that nonselected mu- and Boe (22) suggested that hypermutators could be created tations in strain FC40 arise under adaptive-mutation condi- as the result of occasional errors in translation or transcription tions. Because nonselected mutations occur at a higher fre- ϩ leading to defects in the proteins used for DNA replication or quency among the selected Lac cells than among the non- Ϫ repair. Deficiencies in MMR might be particularly likely selected Lac cells, some or all of the mutations must be because the levels of certain MMR proteins have been found arising in a hypermutating subpopulation. To determine what to decline in stationary-phase cells (23, 24). Our data support proportion of cells are hypermutators and by how much their the hypothesis that hypermutators are MMR-deficient. We mutation rate is elevated, we measured the frequency of a Ϫ find that loss of MutL does not have a great effect on the common nonselected mutation, loss of motility (Mot ), in Ϫ ϩ Ϫ ϩ frequency of Mot mutations in Lac singles but raises the nonselected Lac cells (zeros), in selected Lac cells (singles), Ϫ Ϫ ϩ frequency of Mot mutations in Lac zeros to the level found and in selected Lac cells that also carried an additional in singles (Table 1). Thus, loss of MutL has a far smaller effect nonselected mutation (doubles). on the hypermutators than on the population at large, implying If all mutation were confined to a hypermutating minority, that hypermutators owe their high mutation rate to a complete every Lacϩ mutation would have arisen in a hypermutator. The Ϫ or partial deficiency in MMR (or some other repair pathway frequency of Mot mutations would then be much higher in that requires MutL). Although the majority of Lacϩ adaptive ϩ mutations arise in cells that are not hypermutators, the spectra Table 2. Additional phenotypes found among 3,168 Lac clones of FC722 of mutations that revert the episomal lac allele do not appear to differ between the two populations (as mentioned above). Mot Phenotype of the Therefore, the simplest hypothesis is that the mechanism that Double Mutants generates Lacϩ mutations on the episome is the same in all the Phenotype Number Found Motϩ MotϪ Mot* cells in the population but that a deficiency in MMR allows more of these to be retained in the hypermutators. Other, TetR 1001 Ϫ nonselected mutations on the episome and elsewhere would Mal 2012also necessarily be retained, although these may occur by R 5FC 8315different mechanisms. 5FUR 1100 Ϫ The Kinetics of Mutation in Hypermutators. Hall (3) was GA 1100the first to propose that mutation in populations under selec- Total 13 5 2 8 tive pressure depends on a hypermutating minority. He sug- Mot* are cells with any motility phenotype, including MotϪ. gested that the mutations appear to be specific to the selection Downloaded by guest on October 1, 2021 6866 Genetics: Rosche and Foster Proc. Natl. Acad. Sci. USA 96 (1999)

Table 3. Calculation of the size of the hypermutating subpopulation and its increase in mutation rate Phenotype Cells, no. Cell-days MotϪ, no. MotϪ per cell, % MotϪ per cell-day, % LacϪ zeros 2,878 6,990 2 0.07 0.029 Lacϩ singles 3,168 7,006 48 1.5 0.68 Lacϩ doubles 13 34 2 15 5.9 p and M calculations are as follows. R1͞0 ϭ 0.68͞0.029 ϭ 24. R2͞0 ϭ 5.9͞0.029 ϭ 203. 2 4 Ϫ4 p ϭ R1͞0͞(R2͞0) ϭ 24͞4.2 ϫ 10 ϭ 5.7 ϫ 10 . M ϭ R2͞0[1 Ϫ (pR2͞0)] ϭ 203͞(1 Ϫ 0.12) ϭ 231. Proportion of single mutants that arise from the hypermutating subpopulation, pM͞[pM ϩ (1 Ϫ p)] ϭ 0.13͞[0.130 ϩ (1 Ϫ 5.7 ϫ 10Ϫ4)] ϭ 0.12. Proportion of double mutants that arise from the hypermutating subpopulation, pM2͞[pM2 ϩ (1 Ϫ p)] ϭ 30͞[30 ϩ (1 Ϫ 5.7 ϫ 10Ϫ4)] ϭ 0.97.

because every hypermutator rapidly dies unless it produces a constant with time for our E. coli strains. Here we show this is mutation that allows growth to resume. This hypothesis pre- true both for the episomal and the chromosomal lac allele (Fig. dicts that, although nonselected mutations may occasionally be 1) and is also true for the accumulation of MotϪ mutations in picked up as fortuitous hitchhikers, their frequency in the the Lacϩ population (Fig. 2). Second, no fewer than 5% of the population of selected mutants should not increase with time. cells must be mutating. This is because the proportion of cells Ϫ In fact, we find that the frequency of Mot mutations among that are mutating cannot be smaller than 1/R1/0 and R1/0 is selected Lacϩ revertants increases linearly with time (Fig. 2). about 20 (Table 3). The simplest hypothesis consistent with The simplest interpretation is that the hypermutators are not these results is that during lactose selection all cells are capable continually turning over to a significant degree, at least during of producing mutations, and that, for the majority of cells, their the course of our experiments. Nevertheless, few cells would be probability of doing so is constant with time. expected to survive for long with a mutation rate so high that 2 of every 13 of them suffered a null mutation in a group of We thank J. Cairns for invaluable aid in analyzing our data, genes comprising only 1% of the genome (Table 2). Thus, it preparing this paper, and for the mathematical analysis given in the will be interesting to determine what allows the hypermutating Appendix. We also thank M. Marinus and J. Miller for strains and minority to survive its ever-increasing number of mutations. helpful discussions. This work was supported by National Science The Mutational Targets for the Hypermutating Minority. Foundation Grant MCB 97838315. The hypermutating minority can be identified whether the allele undergoing selection, ⌽(lacI33-lacZ), is on the episome APPENDIX or on the chromosome (Table 1), although the frequency and ϩ If a small proportion of all cells are high mutators and have a genetic requirements for adaptive Lac mutation are different mutation rate that is significantly greater than that of the rest in the two cases. Thus, the pathways for the two classes of of the population, any given class of mutation will be more mutation can each be divided into two components, one that frequent in cells that are mutant for some other trait (singles) partially overlaps (so that hypermutators are hypermutators than in the population as a whole (zeros). R , the frequency for both) and one that is specific to the position of the lac allele 1/0 of mutation B in A-mutants divided by the frequency of (because different gene products are involved). This is con- mutation B in all cells, will be simply a function of the sistent with the hypothesis that all mutations, no matter how proportion (p) of cells that are high mutators and the factor, generated, are more likely to be retained in hypermutators. M, by which their mutation rate is raised. Our data also indicate that a conjugal plasmid is not required The relation between R , p, and M has been shown to induce the hypermutable state. 1/0 elsewhere (8) to be The Frequency of Transient and of Genetically Stable ϭ 2 ϩ Ϫ ϩ Ϫ 2 Hypermutators. Continuous selection for phenotypes that R1/0 [pM (1 p)]/[pM (1 p)] , [1] increase competitive fitness (25, 26) or repeated selection for a succession of novel phenotypes (20) selects for cells that are and the proportion of any class of mutant cells that arise in high stable mutators. But in the real world, cells may more often be mutators is confronted with some particular barrier to growth rather than ϩ Ϫ a succession of hurdles. To meet these complex challenges, pM/[pM (1 p)]. [2] which may require several mutations, transient mutators may be more important than genetically stable mutators. (Indeed, R2/0, the frequency of mutation C in double-mutant AB cells of the population enriched for hypermutators, the Lacϩ MotϪ (doubles) divided by the frequency of mutation C in all cells doubles, we found only 2% to be genetically stable.) In (zeros), is similarly simply a function of p and M and can be addition, cells that succeed because of a transient increase in shown to be mutation rate would not continue to be burdened with a [pM3 ϩ (1 Ϫ p)] ϭ potentially deleterious mutation rate. R2/0 2 ϩ Ϫ ϫ ϩ Ϫ , [3] Implications for the Mechanism of Adaptive Mutation. The [pM (1 p)] [pM (1 p)] lac mutations that revert the episomal allele during lactose and the proportion of any class of doubly mutant cells that arise selection are caused by DNA polymerase errors (27). Cur- in high mutators is rently, there are two theories for the origin of this de novo DNA Ϫ synthesis—amplification of the Lac allele (17, 28) and DNA pM2/[pM2 ϩ (1 Ϫ p)]. [4] replication initiated during recombinational repair of double- strand breaks (18, 29). The results presented here do not R1/0 and R2/0 can be shown to reach their maximum values distinguish between these two hypotheses, but do limit them in when p is 1/(M ϩ 1) and 1/(M3/2 ϩ 1), respectively. two ways. First, as previously reported (11, 30), the rate at p and M are the two unknowns, and it is possible to calculate ϩ ϽϽ Ϸ which Lac mutations appear during lactose selection is their values by measuring R1/0 and R2/0.Ifp 1, then R1/0 Downloaded by guest on October 1, 2021 Genetics: Rosche and Foster Proc. Natl. Acad. Sci. USA 96 (1999) 6867

2 ϩ ϩ 2 2 ϾϾ Ϸ ϩ (pM 1)/(pM 1) , and if pM 1, then R2/0 M/(pM 13. Foster, P. L. & Trimarchi, J. M. (1995) Proc. Natl. Acad. Sci. USA 1). 92, 5487–5490. It follows that 14. Miller, J. H. (1972) Experiments in Molecular Genetics (Cold Spring Harbor Lab. Press, Plainview, NY). Ϸ 2 p R1/0/(R2/0) [5] 15. Blattner, F. R., Plunkett, G., III, Bloch, C. A., Perna, N. T., Burland, V., Riley, M., Collado-Vides, J., Glasner, J. D., Rode, and C. K., Mayhew, G. F., et al. (1997) Science 277, 1453–1479. 16. Modrich, P. & Lahue, R. (1996) Annu. Rev. Biochem. 65, Ϸ Ϫ M R2/0/[1 (R1/0/R2/0)]. [6] 101–133. 17. Foster, P. L. & Cairns, J. (1992) Genetics 131, 783–789. In other words, by measuring the frequency of some class of 18. Foster, P. L., Trimarchi, J. M. & Maurer, R. A. (1996) Genetics mutation in the population as a whole, in cells mutant for one 142, 25–37. other trait, and in cells mutant for two other traits (i.e., 19. Berlyn, M. K. B., Low, K. B. & Rudd, K. E. (1996) in Escherichia measure R1/0 and R2/0), it should be possible to determine p coli and Salmonella: Cellular and , eds. Nei- and M. Once p and M are known, it becomes possible to dhardt, F. C., Curtiss, R., III, Ingraham, J. L., Lin, E. C. C., Low, determine the magnitude of the contribution of high mutators K. B., Magasanik, B., Reznikoff, W. S., Riley, M., Schaechter, M. to single-mutant and double-mutant cells. & Umbarger, H. E. (Am. Soc. Microbiol., Washington, DC), 2nd The above analysis is based on the assumption that there Ed., pp. 1715–1902. are only two classes of cell, the high-mutators and the 20. Mao, E. F., Lane, L., Lee, J. & Miller, J. H. (1997) J. Bacteriol. low-mutators and therefore only two independent variables 179, 417–422. (p and M). To establish that there are indeed only two 21. Neuhard, J. & Kellin, R. A. (1996) in Escherichia coli and classes, it would be necessary to measure the frequency of Salmonella: Cellular and Molecular Biology, eds. Neidhardt, F. C., fourth mutations in cells bearing three mutations (i.e., to Curtiss, R., III, Ingraham, J. L., Lin, E. C. C., Low, K. B., determine R ). Until that has been done, the possibility Magasanik, B., Reznikoff, W. S., Riley, M., Schaechter, M. & 3/0 Umbarger, H. E. (Am. Soc. Microbiol., Washington, DC), 2nd remains that the population contains many classes of cell and Ed., pp. 580–599. many values for both p and M. 22. Boe, L. (1992) Mol. Gen. Genet. 231, 469–471. 23. Feng, G., Tsui, H.-C. T. & Winkler, M. E. (1996) J. Bacteriol. 178, 1. Cairns, J., Overbaugh, J. & Miller, S. (1988) Nature (London) 335, 2388–2396. 142–145. 24. Harris, R. S., Feng, G., Ross, K. J., Sidhu, R., Thulin, C., Annu. Rev. Microbiol. 47, 2. Foster, P. L. (1993) 467–504. Longerich, S., Szigety, S. K., Winkler, M. E. & Rosenberg, S. M. 3. Hall, B. G. (1990) Genetics 126, 5–16. (1997) Genes Dev. 11, 2426–2437. 4. Foster, P. L. (1997) J Bacteriol. 179, 1550–1554. 25. Cox, E. C. & Gibson, T. C. (1974) Genetics 77, 169–184. 5. Boe, L. (1990) Mol. Microbiol. 4, 597–601. 6. Torkelson, J., Harris, R. S., Lombardo, M.-J., Nagendran, J., 26. Sniegowski, P. D., Gerrish, P. J. & Lenski, R. E. (1997) Nature Thulin, C. & Rosenberg, S. M. (1997) EMBO J. 16, 3303–3311. (London) 387, 703–705. 7. Ninio, J. (1991) Genetics 129, 957–962. 27. Foster, P. L., Gudmundsson, G., Trimarchi, J. M., Cai, H. & 8. Cairns, J. (1998) Genetics 148, 1433–1440. Goodman, M. F. (1995) Proc. Natl. Acad. Sci. USA 92, 7951–7955. 9. Coulondre, C. & Miller, J. H. (1977) J. Mol. Biol. 117, 525–575. 28. Andersson, D. I., Slechta, E. S. & Roth, J. R. (1998) Science 282, 10. Calos, M. P. & Miller, J. H. (1981) J. Mol. Biol. 153, 39–66. 1133–1135. 11. Cairns, J. & Foster, P. L. (1991) Genetics 128, 695–701. 29. Harris, R. S., Ross, K. J. & Rosenberg, S. M. (1996) Genetics 142, 12. Aronshtam, A. & Marinus, M. G. (1996) Nucleic Acids Res. 24, 681–691. 2498–2504. 30. Foster, P. L. (1994) Genetics 138, 253–261. Downloaded by guest on October 1, 2021