INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Several successive genetic alterations are required for a cancer cell to develop, proliferate, and be able to metastasize (70). In fact, as many as 6 to 10 mutations may be needed to turn a normal cell into a full-blown invasive tumor line (2), which is one reason why many cancers take decades to develop. Loeb (35) has postulated that an early step in the progression of some cancers may be the creation of a mutator cell, since the low spontaneous mutation rate might not account for the incidence of cancer. However, the elevated mutation rates in mutators would greatly increase the probability of cells accumulating all of the necessary mutations (27, 35), especially if repetitive rounds of clonal expansion and somatic selection occurred (48). In accord with this idea was the exciting finding that the inherited susceptibility to human nonpolyposis colon cancer (HNPCC) and ovarian cancer is due to a defect in one copy of one of the genes involved in the human counterpart to the bacterial mismatch repair system (3, 18, 30, 49). Presumably, when a somatic cell loses the other copy, the resulting cell is completely defective for mismatch repair and has a higher mutation rate. In fact, tumor lines from HNPCC patients are mutators with greatly increased repeat-tract or microsatellite instability (1, 26, 30, 36, 52, 53, 69), a propensity for frequent additions or deletions at repetitive nucleotide sequences (such as repetitive mono-, di-, tri-, and tetranucleotide repeats), in analogy with the repeat-tract instability seen in mismatch-repair-deficient strains of bacteria and yeast (11, 57, 66). These tumor lines also show elevated mutation rates in genes such as hprt (1, 3, 21). The realization that mutator cells are cancer prone (1, 30, 50, 51) has led to increased research in this area, including a search for new types of mutators and for a better understanding of how mutators proliferate.

Experiments with bacteria aimed at finding mutators (12, 23, 24, 34) or at examining the behavior of wild-type or mixed populations in chemostats (6, 9, 20, 46, 61, 65) showed that continuous selection increases the proportion of mutators in cell populations. In a previous paper, we demonstrated how easily a population of Escherichia coli cells could become principally or totally mutators in response to several successive selections (39). Here, we ask whether instead of changing medium in successive selections a single medium that selects directly for mutator colonies could be devised. By asking cells to overcome several barriers to growth on a medium with limiting amounts of each required nutrient, we developed a plate medium on which only mutator cells grow. We have employed this medium to examine the occurrence of mutators under several conditions, and we discuss how this mimics the situation in mammalian cells that need to overcome several growth restrictions before becoming proliferating cancer cells. We also show that one mutator can induce a second mutator that then stimulates useful genetic changes, as part of a mutator cascade.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Bacterial strains and plasmids. The strain CC107 carries an F'lacpro episome in the P90C (11, 43) strain background ara (gpt-lac)5. The lac region on the F factor carries a lacI mutation and also a frameshift in the lacZ gene that reverts by the addition of a GC base pair to a monotonous run of GC base pairs. This strain is described by Cupples et al. (11). AS18 is a Met derivative of CC107 carrying an ICR-191-induced frameshift mutation in the metE (or possibly metR) gene. AS18-29 is a Bgl derivative of AS18 carrying an additional ICR-191-induced frameshift mutation, this time in the blgA gene. AS210 is a Leu derivative of AS18-29, carrying a mutH-induced frameshift mutation at the leu locus and a Tn10kan insert near the wild-type mutH gene (zgh-3159 [62]). We constructed specific mutator derivatives of certain strains by using P1 transduction from strains in which a mini-Tn10 had integrated into either the mutT, mutH, mutS, mutL, or uvrD gene (45).

Genetic methods. Mapping experiments were carried out with P1 cotransduction with Tn10 transposons that had integrated near either the mutH, mutL, mutS, or uvrD gene, or near various nutritional markers (62). Auxotrophs were detected after ICR-191 mutagenesis by replicating Luria-Bertani (LB) plates spread with 100 to 300 colonies onto minimal medium plates and recognizing those colonies which failed to grow. Combinations of supplements were used to restore growth, and then individual supplements were employed. All other strains and bacterial genetic methods, such as determination of rifampin resistance (Rif^r), are described by Miller (43). We initially examined colonies for strong mutator activity by gridding them onto an LB plate and growing them overnight at 32°C before replicating them onto a second LB plate. The second plate was grown overnight, and then colonies were replicated onto an LB plate with 100 µg of rifampin per ml. After overnight growth, mutators defective in mismatch repair showed many Rif^r colonies growing out of the replicated patch, whereas nonmutator strains did not. More quantitative measurements were then carried out when relevant.

Preparation of cultures. Unless otherwise stated, all cultures for the experiments reported here were prepared by inoculating a portion of a single colony into LB medium (43) or other medium and growing it overnight. A different single colony was used for each culture. Therefore, all mutants occurring in different cultures are of independent origin, since each single colony is derived from a single isolated cell. Typically, 5-ml cultures were used.

Mutagenesis. All methods of mutagenesis were exactly as described in the work of Miller (43). 2-Aminopurine (2AP; Sigma) was used at concentrations of 700 µg/ml. Cultures were prepared by subculturing 10⁴ to 10⁵ cells into 4 ml of LB with 2AP, and these were grown for 12 to 16 generations in LB with 2AP before plating. ICR-191 (Sigma) was used at concentrations of 10 µg/ml in minimal A medium (43) supplemented with 1 ml of LB broth, 2 ml of 20% glucose, 0.1 ml of 1 M MgSO₄, and 0.5 ml of B1 (thiamine hydrochloride) per 100 ml of medium.

Specialized media. Mutators were selected on lactose minimal A medium (43) supplemented with limiting amounts of required sugars or nutrients. We used 150 µg of glucose per ml as a limiting carbon source and 0.5 µg of required amino acids, purines, or pyrimidines per ml. For instance, in conjunction with strain AS18 or AS18-29 we used minimal A plates containing lactose, B1, and MgSO₄ (33), supplemented with 150 µg of lactose per ml and 0.5 µg of methionine per ml. When called for, these plates were also supplemented with 40 µg of X-Glu (5-bromo-4-chloro-3-indolyl--D-glucoside; Research Organics) per ml. This dye stains colonies with active -glucosidase deep blue. We find that colonies of Bgl strains on plates containing glucose as a carbon source and X-Glu will yield blue papillae without the addition of another substrate for -glucosidase. Thus, in the blue papillation method (47) as originally described for the lac system, phenyl--D-galactoside is added as a carbon source for papillae stained blue by X-Gal (5-bromo-4-chloro-3-indolyl--D-galactopyranoside). Here, this second carbon source is not necessary. Perhaps as a result of the lower concentration of the X-Glu in the role of carbon source, sugars are utilized in the following order by revertants of the Lac Bgl strains: glucose > lactose > X-Glu.

Determination of mutator frequency. The data from Tables 3 and 4 were used to calculate the spontaneous frequency of mismatch-repair-deficient mutants. Since the majority of cultures had no mutators, the fraction of cultures with no mutators was used to calculate the mean, m, for the Poisson distribution describing the distribution of mutants. In the case of cells at a density of 10⁵ cells per plate, the zero fraction is 327/355 = 0.921. Since for the Poisson distribution P_o = e^m, 0.921 is e^m, and m is 0.082, or, in this case, 0.082 × 10⁵. Accounting for the 10% plating efficiency gives a value of m as 0.82 × 10⁵. Similar calculations for cells at a density of 10⁴ cells per plate give 0.63 × 10⁵, for a plating efficiency of 100%. We find that the plating efficiency at 10⁴ cells varies somewhat more than that for 10⁵ cells, so we take 0.8 × 10⁵ as a more probable value. This value is for the fraction of mutators (or mutant frequency) in a culture grown for about 33 generations.

Plating efficiency of mutators. Strain AS18-29 was plated at different densities on selective medium, and then dilutions of a mutH derivative were plated on the same plates and the percentages of mutator cells forming colonies were determined.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Finding mutants with useful selective markers. We looked for mutations which could be used in combination with other mutations to provide useful selections for mutator strains. We initially targeted mutators lacking the mismatch repair system. These strains have greatly enhanced rates of transitions and also of frameshifts at runs of single base pairs or repeating dinucleotides (11, 33, 57). Therefore, strains such as CC107 (11), which reverts from Lac to Lac⁺ via the addition of a G to a run of six G's, are useful for selecting for mutators, as we described previously (39). In order to find additional markers to use in concert with Lac strains such as CC107, we mutagenized CC107 and derivatives of CC107 and looked for auxotrophs and then characterized the mutants with regard to reversion rates in both wild-type and mismatch-repair-deficient (mutH or mutS) strain backgrounds. We also determined the nutritional requirement and mapped the mutation to one of the known loci on the chromosome in most cases (see Materials and Methods).

Detection of mutations in the bgl operon. As an additional indicator of mutator strains, we sought mutations in the bgl operon that revert in response to mismatch-repair-defective backgrounds. Since the bgl operon is cryptic in most E. coli K-12 strains (56, 58), we first scored spontaneous mutants of both CC107 and AS18 that could grow on salicin as a sole carbon source for intense blue color on glucose minimal plates with X-Glu (see Materials and Methods). We selected one Bgl⁺ mutant from each strain, mutagenized them with ICR-191, and screened for white colonies on X-Glu plates. These were then tested for enhanced reversion in a mutH strain in a variation of the blue papillation assay (47), adapted for X-Glu (see Materials and Methods). One Bgl mutant was selected for further use. The mutant derived from CC107 was termed CC107-17, and the mutant derived from AS18 was named AS18-29. In each case, the frequency of revertants to Bgl⁺ goes from approximately 10 per 10⁸ cells in a wild-type strain to approximately 5,000 per 10⁸ cells in a mutH strain. These strains form numerous blue papillae in a mismatch-repair-deficient background on X-Glu plates and thus serve as indicators for mutHLS and uvrD strains.

Sequential selection on a single medium detects mutators. We initially used strain AS18-29, which carries frameshift mutations in both lacZ and metE, to detect mutators. We experimented with different media to select colonies of strain AS18-29 that overcome both the Lac and the Met defects. We applied the principles depicted in Fig. 1 and 2. Lactose medium with limiting amounts of glucose and methionine should allow cells to grow to a small population size of approximately 10⁵ cells before exhausting the limiting carbon source (glucose). To grow further would require metabolizing the lactose, and this would require a mutation from Lac to Lac⁺. Then, the Lac⁺ cell could grow until it exhausted the methionine and again reach 10⁵ cells. Now, a second mutation to Met⁺ would be required to allow further colony growth. In cases where a mutator population of 10⁵ cells would have a Lac⁺ or a Met⁺ revertant, we would expect each mutator cell to give rise to a full colony under the right conditions, whereas only a fraction (about 10⁵ to 10⁴) of the nonmutator cells would form colonies (Fig. 1). We therefore examined the plating efficiency of a mutH derivative of AS18-29 on medium with different limiting amounts of glucose and methionine and found the best results with 150 µg of glucose per ml and 0.5 µg of methionine per ml (see Materials and Methods). On these plates, the mutH strain formed colonies with a 50 to 100% plating efficiency, while the wild-type derivative formed no colonies at all. We determined mutator colonies by either testing for enhanced frequency of Rif^r colonies or including X-Glu in the medium to indicate frequent mutation to Bgl⁺ by the appearance of many blue papillae. (In the absence of X-Glu, colonies are usually picked after 3 days, whereas in the presence of X-Glu an extra day is needed for optimal papillae formation.)

View larger version (13K): [in this window] [in a new window]   FIG. 1.   Selection for mutator colonies. Lac Met cells are plated on lactose medium with limiting amounts of glucose and methionine. Cells form a microcolony before exhausting the glucose. Revertants to Lac+ can grow further. Only 1% of the nonmutator microcolonies (left) will have a Lac+ cell, whereas 100% of the mutator microcolonies (right) will have a Lac+ cell. Further growth of the Lac+ cells yields microcolonies that exhaust the methionine. Again, 1% of the nonmutator Lac+ microcolonies (left) will have a Met+ cell, whereas 100% of the mutator Lac+ microcolonies (right) will. The very rare nonmutator Lac+ Met+ colonies have no Bgl+ papillae (left), but the Lac+ Met+ mutator colonies show microsatellite instability and give Bgl+ papillae (right; black dots). WT, wild type.

View larger version (44K): [in this window] [in a new window]   FIG. 2.   A representation of mutator colonies growing to full size after the selection described in the legend to Fig. 1. Many small microcolonies (open dots) are seen in the background, but only two full-grown colonies appear, and these have many Bgl+ papillae (dark dots).

View larger version (63K): [in this window] [in a new window]   FIG. 3.   Selection for mutators. (A) Strain AS18-29 was grown overnight in LB broth with 2AP and plated on X-Glu plates. Here, one mutator colony (see insert) is evident, as indicated by the numerous blue Bgl+ papillae. (B) A photograph of a plate like that diagrammed in Fig. 2. Two full-size mutator colonies grow on selective plates onto which several thousand cells have been spread. Minute microcolonies in the background are nonmutator colonies. The mutator colonies appear blue because of thousands of blue Bgl+ papillae (see magnification in panel C). (C) A magnification of part of the plate shown in panel B. Here, the nonmutator microcolonies are evident, as are the blue Bgl+ papillae, representing the microsatellite instability of the full-size mutator colony.

Detection of spontaneous mutators. We employed strain AS18-29 to detect spontaneous mutators in cultures grown for three generations. We examined 355 cultures, by plating 10⁵ cells from each culture, and an additional 344 cultures, by plating 10⁴ cells from each culture. Colonies were detected on selection medium without X-Glu and purified once before being tested for the frequency of Rif^r colonies. Because the frequency of spontaneous mutators is on the order of 10⁵ in the population, some Lac⁺ Met⁺ colonies are due to spontaneous mutations occurring in nonmutator cells. These colonies also appear on the order of 10⁵ in the population. As Table 4 shows, about 50% of the colonies detected were mutators. About 92% of the plates with 10⁵ cells yielded no mutator colonies. Applying the 10% plating efficiency determined from reconstruction experiments (Table 3), we can calculate the mutator frequency from the fraction of cultures with no mutators. This gives a frequency of mutators of 0.8 × 10⁵ (see Materials and Methods). For the cultures plated with only 10⁵ cells per plate, which gives a plating efficiency of 50 to 100%, approximately 94% of the cultures yielded no mutator colonies, which translates to 0.6 × 10⁵ to 1.2 × 10⁵ for the mutator frequency. These values are close to estimates based on other measurements we have made, as described in the Discussion.

Identification of mutators detected spontaneously. We mapped the mutation causing the mutator phenotype in 45 of the spontaneous mutators detected in the above experiments, by P1 transduction (see Materials and Methods). All of the mutations fell into one of the four mismatch repair genes. However, as Table 5 shows, the mutations were not distributed equally among mutL, mutH, mutS, and uvrD. Instead, 19 of the mutations were in mutH, 18 were in mutL, 7 were in mutS, and 1 was in uvrD.

Different combinations of markers are effective in selection. The strain AS18-29 carries a mutation in the lac region on an F' plasmid and a met mutation in the chromosome. The selective medium is not limited to those markers, nor does it depend on having a mutation on an F' plasmid (data not shown). Several different combinations of markers were made for use in the successive selection medium. Some of the markers are derived from the experiment whose results are shown in Table 1. A second set of markers was derived by first making a mutH derivative of AS18-29 and obtaining auxotrophs (in the presence of methionine) induced by the mutator effect of the mutH allele. Then, the mutH allele was crossed out, and the reversion rates were compared. The new strains from this selection carry mutations that are not limited to additions or deletions at monotonous runs of G's or C's. They may contain additions or deletions at runs of A's or T's, or at repeating dinucleotides, as well as certain base substitutions. Table 6 shows the useful strains obtained from this selection. One additional mutation used in some experiments (data not shown) is the argE amber mutation derived from strain XAC (43). This mutation reverts via base substitutions that either restore the UAG codon to a sense codon or else create an amber or ocher suppressor.

Triple selection. We utilized strain AS210 (Table 6), which carries frameshift mutations in lacZ, metE, and leu. Even though three successive selections occur, colonies were obtained on plates with limiting amounts of glucose, methionine, and leucine. After mutagenesis with 2AP, 97% of the colonies (26 of 27) were found to be mutators, and 90% of the colonies detected spontaneously were mutators. All of the mutator colonies tested had defects in one of the mismatch repair genes. The plating efficiency of a mutator on this medium, however, does not exceed 10%.

Mutator cascade. Some mutators may not exhibit an increased rate of mutations at certain sequences, such as frameshifts at runs of identical bases of repeated sequences. However, they might show an increased rate of mutation at a second mutator locus, the resulting secondary mutators now being able to stimulate the specific sequence change. This type of mutator cascade might be significant in creating certain phenotypes. We can examine this phenomenon by looking at our tester strain, AS18-29, into which we have crossed a mutation at the mutT locus. The resulting mutators show an increased incidence of only one specific transversion, A:TC:G (10), because of the failure to hydrolyze the oxidatively damaged DNA synthesis precursor 8-oxo-dGTP (38). That mutT cells do not greatly increase mutations at the frameshifts we used to monitor mutations in AS18-29 is evidenced in Table 7. Comparing the wild-type and mutT derivatives of AS18-29 with respect to reversion of either the lac, met, or bgl frameshift mutation shows either only slight or no differences compared with the mutH derivative of AS18-29, which shows an enormous increase (Table 7). Yet, when we look at the incidence of mutations in the mutHLS pathway that create mutators that can now stimulate the frameshifts in lac, met, and bgl, we see that mutT increases the spontaneous level of these mutators from 8 × 10⁶ to 3 × 10⁴.

View larger version (107K): [in this window] [in a new window]   FIG. 4.   Appearance of mutator subpopulations as colony sectors. A mutT derivative of strain AS18-29 was plated on minimal glucose plates with methionine and X-Glu. Mismatch-repair-deficient (MMR) sectors (e.g., mutH) are revealed by frequent blue Bgl+ papillae (microsatellite instability). (A) An MMR mutant has arisen at the first cell division, resulting in half of the colony having microsatellite instability. (B) An MMR mutant has arisen at approximately the fifth cell division, resulting in a thinner sector of the colony displaying microsatellite instability.

View larger version (15K): [in this window] [in a new window]   FIG. 5.   The effects of a mutator cascade are depicted in a schematic diagram of a mutT derivative of strain AS18-29 (Lac Met Bgl) growing on lactose minimal medium with trace amounts of glucose and methionine and the indicator X-Glu. The starting cell (A) forms a microcolony, in which a mutH cell arises (B). As the microcolony grows, exhausting the glucose in the medium, the subpopulation of mutH cells also slowly proliferates (C), until a Lac+ cell arises within the mutH subpopulation (D). The Lac+ cells can grow further on the lactose in the medium until the trace methionine is exhausted. If a Met+ cell appears (E), it can now grow without the restriction of limiting methionine or glucose, expanding rapidly and showing microsatellite instability by the appearance of many blue Bgl+ papillae (F).

View larger version (70K): [in this window] [in a new window]   FIG. 6.   Mutator cascade. The figure shows selection for the growth of a mutator arising from a microcolony. The starting microcolony is a mutT derivative of strain AS18-29. The mutT mutator cannot revert the frameshift mutations in lacZ and metE, but can generate mutations in the mutH, -L, or -S genes, resulting in mismatch-repair-deficient (MMR) mutators that can revert the frameshifts and break free of growth restrictions. (A and B) MMR mutator colonies growing out from a microcolony. The MMR mutators have many blue papillae, demonstrating microsatellite instability (see legend to Fig. 4). (C) An MMR mutator generated on a more crowded plate will grow in an unrestricted fashion, proliferating over the plate and overrunning the nonmutator microcolonies.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

Mutator cells, defined as cells with higher rates of mutation than those of wild type, were first detected in Drosophila melanogaster in the early 1940's (54) and in bacteria in the early 1950's (68). Now, mutations in more than 20 different genes in bacteria such as E. coli have been shown to result in the mutator phenotype (see reviews in references 8, 13, 19, 44, and 64). Most of these cause defects in repair or damage avoidance systems, although some affect less well defined pathways (7, 42, 63). Higher cells also display mutator phenotypes when certain genes are inactivated, and this can lead to increased incidence of disease. For instance, inheriting one copy of a defective gene involved in the human counterpart to the bacterial mismatch repair system leads to increased susceptibility to colon cancer, HNPCC-related endometrial cancer, and ovarian cancer (4, 18, 30, 49; see review in reference 37). The resulting tumor lines are mutators that show increased repeat-tract instability (see reviews in references 27 and 35) and higher mutation rates in genes such as hprt (1, 3, 21). Loeb had already postulated that creation of a mutator cell would be an early step in some cases of carcinogenesis, since a mutator can generate the multistep mutations faster than can a normal cell with lower mutation rates (27, 36). How then do mutators arise in a population of cells, and how do they proliferate?

Clearly, spontaneous mutants constantly arise in a population, and some of these are mutators. The balance of selective advantage and disadvantage will affect the proportion of mutators in any given environment, but the continued generation of new spontaneous mutations will provide a constant source. Using the selection system described in this work and discussed further below allows us to measure the proportion of cells in a growing population of E. coli cells that have defects in one of the four mismatch repair genes. This number of just under 10⁵ (0.8 × 10⁵) mutants is in a population growing in broth for about 33 generations. This number is in agreement with estimates from our previous study (39), in which Lac⁺ frameshift mutants detected at 2 × 10⁷ were scored to reveal that 0.5% (5 × 10³) were mutators. These values predict that mutants that are both Lac⁺ and defective in one of the four mismatch repair loci are present at 10⁹ in the population (the product of these latter two frequencies). However, since the Lac⁺ revertant rate in each mutator subpopulation is close to 10⁴ (9, 30), then the mutators are estimated to be in the population at 10⁹ divided by 10⁴, or close to 10⁵. (Interestingly, a subsequent analysis of mismatch repair mutators in Salmonella typhimurium shows that they occur spontaneously an order of magnitude less frequently than found here for E. coli [32]).

Looking at the distribution of mutations among the four mismatch repair loci (Table 5) that leads to a combined rate of 0.8 × 10⁵ mutators in the population makes us realize how these measurements can vary, depending on the specific gene being monitored. The mutant frequencies for each of the four loci are as follows: mutH, 3.4 × 10⁶; mutL, 3.2 × 10⁶; mutS, 1.2 × 10⁶; uvrD, 1.8 × 10⁷. It is not clear why only 1 of 45 mutations is in the uvrD gene. Reconstruction experiments indicate that the efficiency of plating of strains with uvrD mutations in the selection employed is the same as for strains with mutations in mutH (data not shown), suggesting that the effect is due to rates of mutations themselves, but it is possible that other hidden experimental biases conspire to reduce the appearance of UvrD mutants.

We should note at this point that mutagenesis from any of a number of sources serves to increase the proportion of mutators to near or above 10³ (see, for instance, reference 39). However, in addition to mutagenesis, selection for mutant phenotypes can serve to increase the frequency of mutators in a population. Chemostat experiments show that extensive continued growth in a full-nutrient environment can still lead to the selection of fitter strains and increase the proportion of mutators, since they can give rise to fitter variants more rapidly than can the wild type (6, 9, 20, 46, 61, 65). Computer simulations also argue that mutators are selected for in continued growth in chemostats (67). In fact, some mutators were originally found by procedures designed to screen mutagenized cells after single or successive selections (12, 23, 24, 34) or by observation of cells selected for a specific phenotype (60, 61). Several investigators have found that natural isolates of bacteria have several percent mutators (22, 28, 31), suggesting that populations in the wild might be undergoing constant selection. Cebula and coworkers have argued that the several percent mutators found among E. coli and Salmonella strains isolated from patients might be related to pathogenicity (31), although others have disputed this correlation (41).

In a previous study (39), we showed how quickly the proportion of mutators can rise in a population, and how several successive selections can result in the entire surviving population being mutator. This underscores the consequences of certain chemotherapies, in which the resistant cells may be enriched for mutators. In the previous work (39), the successive selections involved transferring cells to different media. We have extended that study in the work reported here by designing a single medium that selects for several phenotypes in succession. Cells with multiple nutrient requirements are challenged to form colonies by eventually overcoming each of the growth barriers. In the most studied case, strain AS18-29, carrying frameshifts in the lacZ gene on the F plasmid and in the chromosomal metE gene, must revert both mutations in order to form full-size Lac⁺ Met⁺ colonies. A second strain adds a third growth requirement via a frameshift mutation in the chromosomal leu operon. The medium contains very small amounts of glucose as a carbon source and small amounts of methionine (and also of leucine when relevant). This permits a microcolony of nearly 10⁵ cells to form, which if derived from a mutator cell will contain enough cells to have a mutant that can now use, for example, the lactose in the medium as a carbon source to initiate a new microcolony of 10⁵ cells before it exhausts the methionine. A mutator cell will now have a second mutation in this new microcolony that reverses the metE mutation, allowing it to form a full-size colony (in the case of the derivative with the mutation in leu, a third round of successive microcolony formation would be required). Full-size colonies can be picked and analyzed or directly visualized by using reversion of a frameshift mutation in the bgl operon to decorate mutators with scores of blue papillae. The results, shown in Fig. 3, reveal that cells having to overcome several growth requirements in succession on a single defined medium give rise to mutator colonies. After mild mutagenesis, 90 to 100% of the colonies on this medium are mutators with defects in the mismatch repair system, depending on whether two or three growth requirements are employed. Even without a mutagen, spontaneous mutators constitute 50% of the colonies of cells that break through the growth restrictions.

It is interesting to compare the emergence of mutator colonies on plates selecting for overcoming several successive growth restrictions with the emergence of a cancer cell that undergoes successive mutations to break free of growth restrictions, particularly in the case of colon cancer. In both cases, frameshift mutations at runs of mono- or dinucleotides are involved in creating mutants. In the case of the bacterial strain shown here, the frameshift mutations restore the normal gene, whereas in the carcinogenesis model, frameshifts inactivate certain genes. Many colon cancer cell lines have mutations in the APC gene (25), the rII gene encoding the negative growth suppressor transforming growth factor  II (40, 50), and the apoptosis-associated BAX gene (55). The APC gene contains runs of A residues and an AG dinucleotide repeat that are frameshifted in sequenced mismatch-repair-deficient tumor lines (25). In these lines, most of the mutations inactivating transforming growth factor  are at a run of 10 A's or at a threefold repeat of a GT sequence in the rII gene (40), and frameshifts at a run of eight G's are found in the BAX gene (55). It is easy to see the parallels between successive selections that enrich for mismatch-repair-deficient strains in bacteria and those in human tumor lines that have to overcome several growth restrictions.

Figures 4 to 6 portray the events that occur when a mutHLS mutator cell arises within a colony of cells growing very slowly under restrictive conditions. As all the cells grow, the patch of cells derived from the mutHLS mutator cell (Fig. 4) reaches a sufficient size to allow the appearance of a mutant that can overcome subsequent growth restrictions (Fig. 5), proliferate during unrestricted growth, and exhibit repeat-tract instability (Fig. 6). In this series of experiments, the appearance of the mutHLS mutator is accelerated by the presence of a different mutator, in this case mutT, that cannot revert the frameshifts (Table 7) to overcome the growth restrictions but that can generate mutations that inactivate the mutH, mutL, or mutS gene. This produces a mutator cascade, where one mutator induces a second one. We have also found similar results with the mutA mutators, which result from miscoding tRNAs (data not shown). It will be interesting to see whether any examples of such a mutator cascade are found among tumor lines, or whether any cancer susceptibilities are found to result from the inheritance of a defective copy of a different repair gene that by itself leads to a mutator effect without microsatellite instability but which can generate mismatch-repair-deficient mutators. There is precedent for inactivation of the mismatch repair genes as a second step in the development of colon cancer. A significant fraction of sporadic colon cancer tumor lines that show microsatellite instability suffer inactivation of mismatch repair genes as a consequence of the epigenetic gene silencing produced by hypermethylation of both copies of the relevant promoters (29).