the origin of mutants under selection: interactions of...

Downloaded from www.asmscience.org by

IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

DOMAIN 6 EVOLUTION AND GENOMICS

The Origin of Mutants underSelection: Interactions ofMutation, Growth, and SelectionDAN I. ANDERSSON,1 DIARMAID HUGHES,2 ANDJOHN R. ROTH3

1Department of Medical Biochemistry and Microbiology, Uppsala University, S-75123Uppsala, Sweden2Department of Cell and Molecular Biology, Uppsala University, S-75124 Uppsala,Sweden3Department of Microbiology, CBS, University of California—Davis, Davis, CA 95616

ABSTRACT The classical experiments of Luria and Delbrück showed convincingly thatmutations exist before selection and do not contribute to the creation of mutations whenselection is lethal. In contrast, when nonlethal selections are used, measuring mutationrates and separating the effects of mutation and selection are difficult and requiremethods to fully exclude growth after selection has been applied. Although many claims ofstress-induced mutagenesis have been made, it is difficult to exclude the influence ofgrowth under nonlethal selection conditions in accounting for the observed increases inmutant frequency. Instead, for many of the studied experimental systems the increase inmutant frequency can be explained better by the ability of selection to detect smalldifferences in growth rate caused by common small effect mutations. A very commonmutant class, found in response to many different types of selective regimens in whichincreased gene dosage can resolve the problem, is gene amplification. In the well-studiedlac system of Cairns and Foster, the apparent increase in Lac+ revertants can be explainedby high-level amplification of the lac operon and the increased probability for a reversionmutation to occur in any one of the amplified copies. The associated increase in generalmutation rate observed in revertant cells in that system is an artifact caused by thecoincidental co-amplification of the nearby dinB gene (encoding the error-prone DNApolymerase IV) on the particular plasmid used for these experiments. Apart from the lacsystem, similar gene amplification processes have been described for adaptation to toxicdrugs, growth in host cells, and various nutrient limitations.

INTRODUCTIONMutation rates are kept low by multiple mechanisms evolved to protectgenetic information and these rates are not regulated in response to thephysiological needs of the cell. This broadly held consensus rests on classicalbacterial experiments interpreted as evidence that selective stress does notalter the mutation process (1, 2). This is not to say that all repair mechanismsare equally active at all times (3, 4). For example, mutation-repair mecha-nisms with different substrate specificities, such as MMR and VSR, are dif-ferentially active under different conditions of bacterial growth, although

Received: 12 January 2010Accepted: 16 March 2010Posted: 17 March 2011Supercedes previous posting at EcoSal.org.Editor: James M. Slauch, The School ofMolecular and Cellular Biology, University ofIllinois at Urbana-Champaign, Urbana, ILCitation: EcoSal Plus 2013; doi:10.1128/ecosalplus.5.6.6.Correspondence: Dan I. Andersson: [email protected]: © 2013 American Society forMicrobiology. All rights reserved.doi:10.1128/ecosalplus.5.6.6

ASMScience.org/EcoSalPlus 1

mailto:[email protected]

mailto:[email protected]

www.ASMScience.org/EcoSalPlus


IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

both use the MutL protein (5, 6). The interaction betweenmutation and selection is complex and has been centralto the history of biology. Before publication of Darwin’stheory of natural selection, induction of genetic changeby environmental stress was a popular idea (7). Trans-lated into microbiological terms, this idea posited that thestress imposed by growth limitation increases the rate atwhich new mutations arise (stress-induced mutagenesis).Darwin may have accepted the idea of stress-inducedmutation to explain the origin of the variability requiredfor natural selection (7). However, the lack of experi-mental evidence on this point contributed to the im-portance of the classic work of Luria and Delbrück (2)and of the Lederbergs (1) demonstrating that laboratoryselections can detect mutant organisms without causingtheir formation. These experiments validated the use ofpositive selection, which is fundamental to the wholefield of bacterial genetics. The experiments also provideda way to measure the formation rate of mutations duringnonselective growth. Later, these classic experimentswere more broadly interpreted as support for the currentconsensus that selective stress does not induce mutationsin natural populations.

Later, a flaw in the evidence for the broad interpretationwas pointed out by Cairns, Hall, and Shapiro (8, 9, 10).The classic experiments had used lethal selections todetect mutants whose resistance phenotype appearedonly several generations after the formation of the mu-tation. Thus, their selection could only detect mutantsthat arose well within the prior period of nonselectivegrowth. Any nonmutant cell exposed to selection wouldbe killed before a new mutation could arise and acquire aresistance phenotype. The experiments could not havedetected mutations generated in response to stress onthe selection plate. This reopened the question of stress-induced mutation in a slightly different form: “Do allmutations arise independently of selection (as did thosestudied in the classic experiments), or could some othermutations (missed by the classic experiments) arise inresponse to growth limitation?” The question raiseddoubts regarding the widely held underpinnings of evo-lutionary genetics and required experimental attention.

Direct evidence that growth limitation might increasemutation rate has been sought in many microbial systemsby using nonlethal selections. In most cases, the resultswere negative. However, a few systems showed behaviorthat has been interpreted as support for the idea of stress-induced mutagenesis, reviewed in references 11 and 12.

In these systems, populations of cells were plated ontoa selective medium that prevents growth but does notkill. Mutant colonies appeared over time and havebeen attributed to stress-induced mutagenesis (11, 12).The selection conditions superficially resemble standardstringent laboratory selection, and (as for laboratoryselections) increases in mutant number are attributed toincreases in mutation rate. To make such a conclusionpossible, the systems must adhere to the rules for labo-ratory selection validated by Luria and Delbrück andLederberg. That is, conditions must ensure that the in-crease in mutant frequency reflects new mutation eventsrather than the expansion of a preexisting mutant pop-ulation during growth under selection.

Belowwe describe how the relationship betweenmutationrate and mutant frequency can be obscured in naturalpopulations. First, we introduce the players—growth,selection, and the vastly different formation rates of var-ious mutant types. Then we outline the “Seven Pillars”—the essential rules that allow the field of laboratory mi-crobial genetics to infer mutation rates from mutant fre-quency. Finally, a series of experimental systems aredescribed and critically analyzed. Each of these systemshas been used as support for “stress-induced mutation.”Evidence is presented that each system violates one ormore of the “pillars” and thereby allows selection to in-crease the observed phenotype frequency (without mu-tagenesis). This is not to say that there are not conditionsunder which mutation rates will change (for example, UVirradiation or growth with an alkylating agent). Neitherdo we rule out that, in principle, a biological system withan environmentally regulated mutation rate might exist.However, we will argue that the experimental systems andphenomena that have been proposed to support stress-induced increases in mutagenesis in nongrowing cells canbetter be explained as the result of growth under selection.

MUTATION AND SELECTIONEach mutational event adds a new genetically distinct cellto the population. The phenotypic consequences ofmutations vary from none (most synonymous sub-stitutions) to lethality (loss of an essential function), butmany of these mutations are expected to generate aphenotype that can be detected by selection. One canexperimentally measure the frequency of mutant cells ina culture, but it is more difficult to assess the actual rate atwhich the responsible mutations arise, especially thosewith small phenotypic effects. It is especially difficult to

2 ASMScience.org/EcoSalPlus

Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

measure the formation rate of adaptive mutations in cellsgrowing under selection.

If all other factors are kept constant, an increase in mu-tation rate should cause an increase in the frequency ofcells with abnormal phenotypes. The reverse correlationis more difficult—and underlies the whole field of bac-terial genetics. Does an increase in mutation frequencyindicate mutagenesis? Genetics of bacteria relies on se-lection to detect rare mutants in huge populations—ituses the frequency of detected mutations to infer the rateof mutant formation. This procedure underlies mea-surement of mutation rate by fluctuation test, mutagenspecificity, and recombination rates. The reliability of thisprocedure depends on successful elimination of the manyfactors (especially selection) that obscure the relationshipbetween mutation rate and mutant frequency.

In contrast, the field of population genetics does not tryto establish a simple relationship between mutant fre-quency and rate but deals primarily with the interferingfactors. In natural populations, interference caused byselection and drift is so great that changes in phenotypefrequency are almost never attributed to mutagenesis.

Thus, these two areas of genetics use selection in verydifferent ways. In a natural population, cells grow and aresubject to selection before, during, and after formation ofa new mutation, while laboratory procedures sharplyseparate selection from growth and mutation. In a nat-ural population, a new mutant is immediately subjectto selective conditions that can make its growth ratefaster or slower than its parent. Selection can have hugeconsequences for phenotype frequency since selectioninitially causes exponential increases in phenotype fre-quency over time. In contrast, the process of mutation (inthe absence of selection) has a linear effect on mutantfrequency over time (see Fig. 1). It should be noted thatwe can never eliminate all selective effects from a growingpopulation, but we can find conditions that remove se-lection from particular mutant phenotypes.

Natural selection can detect minuscule differences ingrowth rate and thus can act on the most commonmutations, which have small effects on phenotype (seebelow). Thus, adaptation can occur rapidly in naturalpopulations because selection causes exponential in-creases in the frequency of mutation types that arise at ahigh rate. In contrast, laboratory genetics uses restrictiveconditions to ensure that selection detects only rare

preexisting large-effect mutations and has no effect ontheir frequency. Prejudices developed because of ourexperience with restrictive laboratory selections maymake it difficult to appreciate the speed at which geneticadaptation can occur once these restrictions are lifted.

Figure 2 depicts the behavior expected of natural popu-lations under various conditions. The vertical axis rep-resents a list of different cell lineages in a population, andthe width of each line represents the logarithm of thefrequency of each lineage. With neither mutation norselection, the frequency of each lineage remains constantover time (Fig. 2, top). With mutation but not selection,mutations can alter the genotype of any lineage but thefrequency of the (now altered) lineage does not changevis-à-vis others in the population (Fig. 2, middle). Byaffecting one lineage after another, mutation causes alinear increase in the frequency of mutant cells.

When mutation and selection occur together, somelineages acquire deleterious mutations and are removedfrom the population by selection (line becomes vanish-ingly thin). Lineages that acquire beneficial mutations arefavored by selection and expand exponentially, increas-ing the size of the lineage and thereby the frequency ofmutant types in the population (Fig. 2, bottom).

The Most Frequent Mutation Types Have theSmallest Phenotypic EffectThere is a wide distribution in the magnitude of mutantphenotypes and a wide variation in the rate of formationof different mutation types. The mutation types arising atthe highest rates are ones with the smallest phenotypic ef-fect. For example, one might consider changes in someenzyme activity and the fraction of the parent activity addedor lost by a mutation—mutations that cause small changesare more common. This may be in part a consequence ofhow selection has shaped the genetic code (13) and theseveral systems that prevent mutations and repair DNAdamage. Mutation types that most often escape cellularquality-control systems are those with the smallest phe-notypic consequences. Natural selection can detect van-ishingly small differences in growth rate and can thereforeact on a larger number of mutations. In contrast, experi-mental genetics focuses on large-effect mutations that arerelatively rare (among conventional point mutations). Theexperience of a laboratory geneticist with stringent selec-tions may generate an intuitive feel for the speed of geneticadaptation that is at odds with the situation in natural


The Origin of Mutants under Selection: Interactions of Mutation, Growth, and Selection



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

populations. A very high rate of genetic adaptation is pos-sible when natural selection acts on growing populationsand causes exponential increases in the frequency of com-mon small-effect mutations. We suggest that theseconsiderationsmay underlie some of the problems inherentin interpreting experiments on mutation under selection.

The basic idea is outlined in Fig. 3 (see below), whichclassifies 100 spontaneous mutations that might affect asingle gene. This table attempts to convey the nature ofgenetic variability at the moment it arises and beforefrequency has been acted upon by selection. The datapresented summarize results of a variety of studies andmust remain a crude estimate since every experimentalsituation used to generate such data uses a different se-quence and detects a different spectrum of mutant types.These frequency estimates are derived frommany sourcesplus our own experience in characterization of sponta-neous mutations (14, 15, 16, 17, 18, 19, 20). In most of

these studies, frequencies were estimated for cells growingin either rich medium or minimal medium with glucose.

It can be seen that, among base substitutions, transitions(A/T ← → G/C) are more common than transversions (allother substitution types). This seems reasonable sincetransitions involve a failure to discriminate between twopurines (or two pyrimidines), while transversions ex-change a purine for a pyrimidine. Because of the structureof the code, transitions are prone to cause synonymouscodon changes with no effect on the encoded proteins orconservative substitutions of structurally related aminoacids (13).

Transversions are less common than transitions, butmore prone to alter protein structure. If one considersloss-of-function mutations, transitions are likely (30%) tobe synonymous and therefore silent. Relatively few basesubstitutions (3/64) result in the creation of nonsense

Figure 1 Effects of selection and mutation on mutant frequency. In the absence of selection, new mutants accumulate linearly with time. Whenselection acts, it causes an exponential increase in the frequency of a beneficial mutation. Thus, in a natural population, where both selection andmutation occur together, the frequency of a favorable mutant is more heavily impacted by selection than by recurring mutational events. It is oftendifficult to appreciate how powerful exponential increases can be.


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

codons with an essentially null phenotype. Frameshiftmutations (like nonsense) are likely to cause large loss-of-function phenotypes but are rarer than basesubstitutions. For gain-of-function mutations, the samegradient of effects are expected—large-effect mutationsare rare. Missense mutations that improve an existing

function or create a novel function are restricted toparticular changes at a small number of sites. Small im-provements are likely to be more common than thosewith a large positive effect. The general tendency is forcommon mutations to cause small phenotypic changes—whether losses or gains in function.

Figure 2 Effects of selection and mutation on mutant frequency. The vertical axis is a list of all the genotypes present when a population isinitiated. Mutation may change the nature of a particular lineage and the relative size of each lineage may change because of differences in theirrelative growth rate. Lineage size (log cell number) is depicted by the thickness of horizontal lines—some dwindle and some increase expo-nentially. Dashed lines indicate lineages altered by mutation but unaffected in growth rate.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

The same trend is seen for mutations that cause changesin copy number—deletions that eliminate a gene arerather rare, but duplications are among the most com-mon of mutations. Perhaps the most common mutationsare copy number changes in genomes with two or morecopies of a region. These events contribute to amplifi-cation or deamplification by unequal recombination be-tween identical copies in sister chromosomes.

Figure 4 diagrams these relationships. Note, in particular,that duplications (causing a twofold increase in genedosage) arise at a vastly higher rate than the large-effectmutations in common use in the laboratory. Even morefrequent are changes in gene copy number, which canadd or remove copies. This copy number changes almostalways involve repeats located adjacent to each other in

the same orientation (in tandem), but some copy numbervariants have inverse-order repeats (E. Kugelberg and J.R. Roth, unpublished data). Because common small-effect mutations arise (and recur) frequently they arepresent at highest frequency in a population before se-lection. Natural selection can detect these small-effectmutations making them major players in genetic adap-tation. Common small-effect mutations are not oftendetected in experimental genetics and therefore tend tobe ignored in thinking about the process of adaptation.(As seen below, laboratory genetics deliberately avoidsthese mutations.) Yet during growth under selectionthese mutations can initiate huge phenotypic changesthat occur by a cascade of successive small improvementswhose frequencies increase exponentially during growthunder selection (see below).

Figure 3 The frequency of mutation types as they arise.


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

The response of selection to copy number variants isparticularly rapid because the frequency of these mutantscomes rapidly to a very high steady-state frequency evenbefore selection is imposed. Typical loci are duplicated inabout 1 cell in a 1,000 of an unselected culture. Thissteady state is reached because formation rates are high(10−5/cell per division) and loss rates are even higher(10−2). In addition, duplications can have very significantfitness costs, which contribute to the rapid approach offrequency to a steady state. While duplications are com-mon before selection, higher copy number variants areexpected to also reach steady-state frequencies at asomewhat lower level (A. B. Reams and J. R. Roth, un-published data).

When Selection Acts on Common Mutations,Improvement Follows Pathways ofGenetic AdaptationIt is suggested above that the nature of the mutationprocess and the structure of the code dictate that geneticadaptation is likely to occur by a series of minor im-

provements. This seems likely because (i) the most fre-quent spontaneous mutations are those of smallphenotypic effect and (ii) natural selection can detectthese small differences and (iii) cause their frequency inthe population to rise exponentially.

Series of improvements are likely because increases in thefrequency of one mutation can produce a subpopulationlarge enough to permit a second common small-effectmutation to arise and provide an additional improve-ment. The most common secondary mutations (like theinitial improvement) are likely to provide small fitnessimprovements. Each improvement allows expansion of asubpopulation in which the next event can occur. As thefrequency of improved cells rises, the probability ofsubsequent improvement also rises, simply because thereare more cells in which they can occur. These selectivepathways allow a population to respond quickly. Thiscascading process is diagrammed in Fig. 5. It will be ar-gued below that such sequences of events underlie manyphenomena that have been interpreted as evidence forstress-induced mutation.

Figure 4 Rate of formation of different types of mutations and their phenotypic effects.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

A point to note in Fig. 5 is that several multisteppathways of improvement may operate in parallel andsome genotypes may arise by multiple routes. When apopulation adapts in this way, the final product may bevery hard to analyze genetically, since many mutationshave contributed to the final phenotype. Small-effectmutations early in the process may allow a populationlarge enough for a large-effect mutation to arise later. Insome cases the final phenotype may no longer require theearly events that allowed it to arise. The difficulty ofreconstructing these pathways is one that underlies theproblem of understanding the origins of cancer andunderstanding natural selection.

USES OF SELECTION IN LABORATORY GENETICSBacterial genetics relies on stringent positive selection todetect rare mutants in astronomically large populations(>109 cells). The frequency of detected mutants is used toestimate the rates at which the mutants formed. Thisprocedure is central to assessment of mutation rates,recombination rates, and mutagen activity. To achievethese ends, selection is used in a special way that detectsmutants without contributing to their frequency. This

essentially eliminates natural selection and is difficult toachieve because growth under selection is such a potentforce for changing mutant frequency (as outlined above).However, elimination of natural selection is essential tothe practice of experimental bacterial genetics.

To defeat natural selection, laboratory genetics separatesselection from growth and mutation. Mutations ariseduring an initial period of nonselective growth. This isfollowed by a sharply separated period of very strongselection on solid medium in which parent cells cannotgrow and no new mutations can arise (Fig. 6). The se-lection is made so strong that it also prevents growth ofthe many small-effect mutants that might otherwise growslowly and improve to produce a visible colony. Whenselection is too weak, the process of detecting preexistingmutants goes awry and the number of detected mutantsis inflated by common small-effect mutations whoseprobability of being detected is enhanced by growthunder selection. Use of very strong selection prevents thedetection process from getting enmeshed in the processof rapid adaptation diagrammed in Fig. 5. The two-stageprocess used in the laboratory (Fig. 6) allows mutantfrequencies to reflect only mutation rate (prior to selec-

Figure 5 Adaptive evolution proceeds by pathways of sequential frequent, small-effect mutations. During growth under selection, commonsmall-effect mutations increase in frequency exponentially and expand clones large enough to allow secondary improvements to occur.


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

tion) and prevents selection from influencing mutantfrequency (as it does in natural populations).

A caveat should be noted. Whenever cells grow andmutations can occur, natural selection is expected to act(e.g., any deleterious mutation will be disfavored). For thepurposes of this discussion, the assumption is made thatone can eliminate the effects of selection on the particularphenotype being studied. For example, rich medium isassumed to eliminate selection on auxotrophic mutants.

During the nonselective pregrowth period (Fig. 6, left),mutants with some particular selectable phenotype ac-cumulate linearly in the population; their frequency isdictated by mutation rate, by the timing of mutationalevents and by the target size for mutations with thatparticular phenotype (compare with Fig. 2, middle). Theimposed strong selection (Fig. 6, right) prevents growthof parent cells and small-effect mutants, but allows large-effect mutant cells to grow and form visible colonies. In

this way, selection detects the fully fit mutations thatarose during prior nonselective pregrowth, but does notcontribute to increasing the frequency of detectablemutants.

The next section (“Seven Pillars of Laboratory Selection”)describes the considerations described above as a set ofrules that support the whole edifice of bacterial genetics.These rules are seldom stated, but they are the under-pinning of bacterial genetics because only with these rulescan one selectively detect mutants in a culture withoutinfluencing the answer. The rules make it possible toanalyze genetic events in huge bacterial populations.They allow phenotype frequency to be a reliable indicatorof mutation rate. Extensive routine use of this proceduremay have led to trivialization of the underlying rules,thereby leading to misinterpretation of some experi-mental results. The rules are not legal obligations. Theymust be obeyed only if you need to reliably and accu-rately measure mutant frequency and rates.

Figure 6 The laboratory procedure that allows mutant frequency to indicate mutation rate. Selection is cleanly and completely separated fromgrowth and mutation. Mutations are allowed to accumulate (linearly) during the initial nonselective pregrowth period. Precautions eliminate theeffect of fluctuation or stochastic distribution of times at which mutations arise. Samples of nonselective growth culture are plated under strongselective conditions that block all growth of the parent type and all of the common small-effect mutations. Only rare large-effect mutations areallowed to grow under selection. These mutations are assumed to be fully fit and require no adaptive change to be detected on the selection plate.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

SEVEN PILLARS OF LABORATORY SELECTION

Assumptions That Underlie the Use of PhenotypeFrequency To Estimate Mutation RateThe following seven assumptions underlie the use ofphenotype frequency to estimate mutation rate:

1. No preexisting mutants Inoculate the preselectionculture with a population too small to include any ofthe mutant types of interest. This eliminates foundereffects and assures that all of the mutants that aredetected later by selection were formed during theprevious nonselective growth period

2. No growth defect or advantage during growth be-fore selection Assure that the mutant types to bedetected grow at the same rate as the parent cellsduring the preselection period—their frequency inthat population is dictated only by mutation.

3. No lag in expression Assume that the mutant phe-notype appears immediately when the mutation forms(no phenotypic lag). A delay in phenotype appearanceallows late-arising mutations to escape detection. Halfof the mutation events that occur in the pregrowthculture are expected to occur in the final generation ofthe pregrowth culture. A corollary assumption is thatmutant cells arising during pregrowth can (whenplated on selective medium) form a visible colony with100% efficiency. Thus, the phenotype frequency in theunselected population is the ratio of observed colonieson selection medium to the viable cells placed underselection.

4. There is no stochastic fluctuation in mutationevents Assume that the mutant frequency perfectlyreflects the mutation rate. This rule is difficulty toobey, but can be approximated. Although each mu-tation arises with the same probability, the exact mo-ment of its formation is subject to stochastic variation.Mutants that happen to arise early in the history of theculture will contribute more heavily to the final mu-tant frequency. As outlined below, the Luria-Delbrücktheory predicts that identical cultures will accumulatedifferent numbers of mutants simply because of vari-ation in the timing of mutations. One can correct forthis in a variety of ways (21). A rough, short-handcorrection is to simply use of the median phenotypefrequency of multiple cultures and thereby avoid themajor “jackpots” or “low-pots” cultures in whichmutation timing has caused an abnormally high or lowphenotype frequency.

5. Each mutant generates one colony Perform selec-tion on solid medium. By immobilizing cells of thepregrowth culture, it is assured that each mutant cell inthat population forms one distinct countable colony.

6. No parental growth Use selection conditions thatcompletely prevent growth of parent strain and anypreexisting small-effect mutants. This prohibits newmutations from arising during the period of selectionand prevents growth under selection from enhancingthe number of detected mutants. While this pillarsupports the whole history of bacterial genetics, it is asticking point for several systems used as evidence forstress-inducedmutagenesis. These systems assume that(contrary to this rule) nongrowing cells are subject tomutagenesis. The question of whether or notmutationsoccur in nongrowing cells will be discussed below.

7. No heterogeneity in mutant colonies Each mutantcolony is assumed to be a homogeneous population ofcells with the same genotype as the cell that initiated it.None of the counted colonies has been initiated by acommon small-effect mutant whose ability to form avisible colony depends on acquisition of additionalmutations during growth under selection.

How Growth Complicates Determination ofMutation Rate from Mutant Frequency(Even without Selection)Mutation rates are generally expressed as mutation/cell/division since it is usually assumed that most mutationsarise as errors in chromosome replication. Thereforewhen a laboratory selection is used to estimate mutationrate, it is essential to know howmany generations (acts ofDNA replication) occurred at which a mutation mighthave occurred. For example, a mutant would be definedas common if it arose during 100 acts of replication, butas rare if it arose only once in 109 acts of division. Thiscalculation is rather easy if one knows the inoculum sizeand the size of the final population in which mutants aredetected. To fairly estimate a mutation rate one mustknow how many divisions occurred.

Estimating the number of divisions that serve as the basisfor the estimated mutation rate can be difficult especiallyin a population whose net size is not increasing. Severalsuch systems are used as evidence for stress inducedmutagenesis. Mutations are said to arise in a time-dependent manner with no acts of cell division. If this is aprominent phenomenon, many conclusions of bacterial


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

genetics are called into question. If this is not true, thenthe evidence for stress-induced (stationary phase) mu-tagenesis is called into question.

Cell growth and the occurrence of mutationsDNA replication provides an opportunity for mutation,and mutation rates are generally expressed per cell/pergeneration. Thus if a population is heterogeneous andincludes a subset with a faster growth rate, that sub-population is expected to acquire a disproportionatenumber of new mutations over a given time span simplybecause it undergoes more acts of replication andincludes an increasing proportion of the parent popula-tion. This is expected even if all cells have the sameprobability of making an error during one act of repli-cation. Thus interpreting increases in mutant frequency(assuming no selection) depends on knowing how manyacts of replication actually occurred to provide the cellswith the assayed mutant phenotype.

Growth can be a problem even in populationsshowing little net growthIn several of the experimental systems discussed below,growth of the parent population is blocked (as is true in astandard laboratory selection). However, in some casesthese populations are incubated for extended periods oftime before the number of mutant colonies is counted.Interpretation of these experiments has assumed thatall colonies appearing are due to large-effect mutationsthat either (i) arose during pregrowth of the culture, or(ii) were induced in nongrowing cells by a novel (stress-induced) mechanism after plating of the culture. That is,some mutations are assumed to arise in a nongrowingpopulation and to accumulate in a time-dependentmanner, without any scheduled DNA replication. When-ever a mutation is attributed to nongrowing cells in thismanner, the problem of growth becomes profound. It isdifficult to eliminate the possibility that the new mutantsactually arise in a growing subpopulation of cells. Estab-lishment of the idea of “stationary-phase” mutagenesisrequires elimination of the possibility of this crypticgrowth. The difficulty of detecting a growing subpopula-tion is illustrated in Fig. 7.

Suppose that selection limits growth of a population at108 cells, and a resistant mutant arises during the growthof that population. The new mutant can grow exponen-tially after growth of the parent cell type has ceased.

When the mutant subpopulation reaches 108 cells, thetotal population will have doubled which may be difficultto detect experimentally. Thus an apparently restingpopulation shows a substantial increase in mutationfrequency. It appears that a nongrowing population wasmutagenized by some novel (growth-independent, stress-induced) mechanism. Once one is aware of the existenceof a growing subpopulation, no mutagenesis is requiredto explain the frequency increase, which is due entirely tonatural selection. It should be noticed that at the end ofthe experiment in Fig. 7, half of the population will havegone through 27 extra generations of growth and have a27-fold higher probability of acquiring any mutation.

Another potential problem is seen in a population ofconstant size in which cells are dying and being replacedby slow growth of the population at large. This is dia-grammed in Fig. 8. Here a constant population may bereplicating continuously (with different subpopulationspossibly reproducing at different rates) and accumulatingmutations that appear to form in a stationary-phasepopulation.

Growth (and replication) contributes to theappearance of the phenotypeAfter a DNA sequence change, some timemay be requiredbefore the mutant phenotype appears (phenotypic lag).Escherichia coli cells growing under different growth con-ditions can have from 1 to 8 chromosomes per cell (22).This time lag may allow a recessive new mutant allele tosegregate from parental copies present in a sister chro-mosome or second bacterial nucleoid. It may allow newmutant enzymes to dilute out preexisting parental protein.If appearance of the mutant phenotype requires two celldivisions, then the phenotype frequency assessed at anymoment will fail to detect mutations that arose within thefinal two generations. If the population is growing expo-nentially, more that 75% of the mutational events will havearisen during that period andwill escape detection. (Half ofthe acts of chromosome replication occur at the final celldivision, and another quarter occur at the previous celldivision.) This phenomenon is particularly relevant in thecontext of the debate on “stress-inducedmutation” becausethe classic experiments demonstrating that mutations arisebefore selection all used phenotypes with a very long lag(streptomycin resistance and phage resistance). In bothsituations, the selected phenotype appears several gen-erations after the mutation arises. Any new mutationarising under selective conditions, would be immediately





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

counterselected and could never get through the lag periodand attain a resistant phenotype. Thus phenotypic lagbiased these experiments in favor of mutations that aroseseveral generations prior to exposure to selection. Allgrowth of the parent under selection was prevented by thelethal selection and any preexisting mutant that had notescaped lag (or any new mutant arising after plating) wassubject to the same fate.

Growth allows mutations to arise at variable times inthe history of the populationThe work of Luria and Delbrück clarified the kinetics ofphenotype accumulation in unselected populations bydemonstrating how the stochastic distribution of muta-tion occurrence times contributes to the frequency ofmutant phenotypes (2). The rate at which any mutationarises is essentially a probability statement, expressing thenumber of mutations expected to form per act of celldivision. In practice, any particular population growingfrom a small number of nonmutant cells will realize itsfirst mutation at a range of times, distributed around thatpredicted by the mutation rate. The phenotype fre-

quencies in a series of such cultures can vary widely(fluctuate) based on the time of first mutation. Theexpected statistical distribution was described by Luriaand Delbrück. The result of this fluctuation is that in-dependent cultures with the same mutation rate areexpected to contain widely different phenotype fre-quencies. There are a variety of ways to compensate forthis problem experimentally (21) and thereby to deter-mine the actual mutation rate.

How Can One Be Sure that the Pillars of WisdomHave Been Successfully Followed?The classical test to ensure that a selection is detectingpreexisting mutations is through a Luria-Delbrück fluc-tuation test. In doing this test, series of independentcultures are all initiated using a small inoculum. Eachculture is independently plated on selective medium. Ifcolonies appearing on the selective plate reflect coloniespresent in the plated population, then you expect to see afrequency fluctuation from one culture to the next. Theway to perform this test and treat the data has been de-scribed in detail (21).

Figure 7 Growth without apparent population size increase. The population diagrammed here reaches a plateau (108/ml) when some resource isexhausted. A mutant formed during the prior growth continues to divide (in this example for an additional 27 generations, resulting in a doublingof the total population).


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

If one sees fluctuation, the question is whether all mutantcolonies were initiated by preexistent cells or only somefraction (enough to cause the observed fluctuation). Away to test the observed fluctuation has been described(21). In this test, one plots the log of the number of mu-tants/tube (N) versus the log of the probability of a tubecontaining N or more mutants. If this plot gives a straightline of slope = −1, then the distribution ofmutants/tube inyour series of tubes reflects a Luria-Delbrück distributionand all observed colonies were initiated by mutants thatarose during the pregrowth culture. If some mutants ariseon the plate this plot shows a plateau followed by a sharpdrop—essentially a narrower variance.

There is a difficulty here in the case of copy number vari-ants. The above test was performed for the Cairns selectionand no fluctuation was noted. The conclusion was madethat all mutants arose on the plate (providing support forthe idea of stress-induced mutation). However, in the caseof duplications, fluctuation is not expected because thefrequency of duplications apLproaches a steady-state fre-quency before selection (18). That is, any differences in thetime of appearance of the first duplication are reduced be-cause frequencies all approach the same steady-state value.This prevented the fluctuation test from demonstrating thatthe revertants appearing (over 6 days of selection) are ac-tually initiated by mutant cells (with a lac amplification)that arose before selection. A test has recently been devisedwhich shows that the number of revertants in the Cairns

system is in fact determined by the number of cells in thepregrowth culture that have a lac amplification beforeselection (E. Sano and J. R. Roth, unpublished data).

Contribution of Genetic Drift and Founder Effects toDifferences in Phenotype FrequencyIf a population is initiated by an inoculum that alreadycontains mutant cells, these founders can provide ahigher phenotype frequency than one might expect to seeif all mutations arise only during the history of thepopulation. This can complicate interpretation of phe-notype frequencies and is systematically avoided by oneof the rules of laboratory selections described.

Founder effects are similar to consequences of genetic drift.In small populations, the frequency of mutant types canvary from generation to generation, simply by randomsuccess in reproduction or stochastic killing that is not in-fluenced by genetic differences. Drift will be ignored herebecause large microbial populations reduce its importance.

KEY SYSTEMS USED TO STUDY THE EFFECTS OFSELECTION ON MUTATIONOur understanding of the origins of mutations has beenstrongly influenced by a few key experiments. Luria andDelbrück’s experiment, and the Lederbergs’ follow-up ex-periments, showed that mutants could preexist selection in

Figure 8 A bacterial population of constant size in which cells are dying and being replaced by slow growth of the population at large.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

a bacterial culture. This was critically important in estab-lishing bacteria as a valid genetic system for analysis ofmutation rates. The Cairns-Foster experiments raised se-rious questions about the general validity of conclusionsbased on Luria and Delbrück’s experiment. Cairns-Fosterraised the issue of whether under nonlethal selection bac-teria were capable of “directingmutations” or of selectivelyincreasing mutation rates in response to environmentalpressure.Thequestionofwhetherbacteria could respond tostress by increasing mutation rates was taken further byRadman, Taddei, and Matic when they showed that bac-teria apparently increased mutation rates in response tonutrient limitation and aging. Below we examine each ofthese critical experimental systems to explore theirstrengths, limitations, and possible weaknesses.

1. The Luria-Delbrück and Lederberg Experiments—the Classic Evidence that (Some) Mutations Are NotInduced by StressIn some sense, the Luria-Delbrück experiment establishedthe “seven pillars” of laboratory selection (2). The exper-iment was designed to demonstrate that laboratory se-lections detect preexisting mutants and do not contributeto the creation of mutants. This experiment validates theuse of selection as a means of detecting mutants and istherefore fundamental to bacterial genetics. If exposure toselection actually caused mutations, the whole field ofbacterial genetics would be called into question.

The basic observation in the Luria-Delbrück experimentwas that the number of detected mutants in a series ofindependent preselection cultures varies in a manner thatcan be predicted mathematically if one assumes that thevariability reflects a stochastic distribution in the timingof mutational events during nonselective pregrowth ofthe plated cultures. This variability demonstrates that themutations arose before exposure to selection. The sameconclusion was reached by the Lederbergs’ using a moregraphic, nonmathematical method involving replicaprinting a master cell lawn to multiple plates of selectivemedium and seeing the same pattern of resistant cloneson each replica (1).

Use of lethal selections to detect mutants withphenotypic lagBoth Luria-Delbrück and Lederberg used lethal selections(phage resistance and streptomycin resistance, respec-tively) in which the detected mutant phenotype appearsonly several generations following occurrence of the

DNA sequence change. The selection conditions madethe results particularly clear because they enforced abroad separation of selection from growth and mutation.(They follow the “seven pillars” to the letter of the law.)No growth was allowed under selection (conditions werelethal). Large-effect mutations are detected. Mutationshad to arise several generations before plating and thuswere certain to arise within the nonselective growth pe-riod. Mutant frequency will be subject to the fluctuationsin frequency analyzed by Luria and Delbrück. The longphenotypic lag also ensures that, in the Lederberg ex-periment, cells resistant to streptomycin are likely to bepresent as sizable clones within the lawns of unselectedcells that were replica printed. Phenotypic lag made itextremely unlikely that any mutations arise followingplating on the lethal selection plates—cells would have toescape killing for several generations if a new resistantmutant were to be detected.

Avoiding the caveat inherent in theclassic experimentsWhile use of a lethal selection and mutations with phe-notypic lag allowed one to separate mutation from se-lection, it reduced the power of the experiments todemonstrate that all mutations arise independent ofstress. The design of the classic experiments ensured thatonly preexisting mutations could be detected. This caveatwas pointed out by several investigators (8, 10, 23).

To avoid the shortfall of the classic experiments one mustsomehow measure mutation rates in growth-limited(stressed) cells. This can be done if one limits growth inone way and follows accumulation of mutations that areneutral under those conditions. Such experiments havenot demonstrated a general increase in mutations ratesduring growth limitation (24). It is much more difficultto limit growth and measure the rate of formation ofadaptive mutations that are positively selected as soon asthey arise. (In principle, this could be done by fluctuationtests using the “zero-tube” method, but the growth andimprovement of small-effect mutations makes thistechnically difficult.) The Cairns-Foster experiment wasdevised to fill the gaps in the classic experiments.

2. The Cairns-Foster Experiment—an Attempt ToAssess Mutation Rates under SelectionThe possibility that some fraction of mutations are stressinduced has been most aggressively addressed using asystem designed by Cairns and coworkers (8) and


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

modified by Cairns and Foster (25). In this experimentthe conditions used by Luria and Delbruck were varied inhopes of detecting mutations induced by stress. The se-lection is nonlethal, and the mutations scored show nophenotypic lag. Parent tester cells are mutants unable touse lactose. These cells are plated on selective medium(lactose), which prevents their growth but does not kill.Over 5 to 6 days, revertant (Lac+) colonies appear above alawn of nongrowing parent cells (Fig. 9).

In the Cairns experiment, as in the Luria-Delbrück ex-periment, parent cells are plated on selective mediummutant (revertant) colonies appear with time. Thesecolonies have been attributed to mutations induced bystress in the nongrowing parent population. The mainfeatures of the Cairns-Foster experiment are describedbelow and then discussed in terms of models offered toexplain them.

1. The plated Lac− tester cells (108) carry a revertable +1frameshift mutation.

2. The tester lac allele provides about 2% of the β-galactosidase level found in Lac+ revertant strains. Thisresidual function allows a single mutant cells to form acolony on standard lactose medium.

3. Growth of the tester strain is prevented by scavengercells having a lac deletion. A 10-fold excess of scav-enger cells is plated with testers and restricts growthof the tester cells, probably by consuming excretedmetabolites.

4. The testers are poised (by the titrated excess of scav-enger cells) on the brink of growth, such that any tinyincrease in β-galactosidase can allow them to grow.

5. During prolonged incubation (6 days on selectivemedium with scavengers), the lawn of tester cellsgrows very little.

6. Within 1 to 2 days under selection, a few coloniesappear which are attributed to mutant cells that aroseduring the growth before selection (as in a standard

Figure 9 The basic form of the Cairns experiment. As in the Luria-Delbrück experiment, parent cells are plated on selective medium mutant(revertant) colonies appear with time. These colonies have been attributed to mutations induced by stress in the nongrowing parent population.The main features of the Cairns-Foster experiment are described in the text and then discussed in terms of models offered to explain them.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

laboratory selection). These day 2 colonies show aLuria-Delbrück distribution and can be used to de-termined the unselected formation rate of full Lacrevertants as 10−8/cell/division.

7. Over the following 4 days, about 100 additional col-onies accumulate above the cell lawn. In some ex-periments, the accumulation is linear with time—inothers it shows some upward inflection.

8. The colonies arising after 4 days, require mutationalevents occurring after exposure to selective medium.(When independent cultures are tested, the number ofrevertant colonies does not show the fluctuationpredicted by Luria-Delbrück.)

9. The stressed parent population in the lawn is notmutagenized during the period of revertant accumu-lation. Samples taken from the lawn (between Lac+

revertant colonies) show no increase in genome-widemutant frequency.

The described accumulation of Lac+ revertants above anongrowing lawn has been interpreted as evidence thatstress induces a mechanism for genome-wide mutagen-esis. This proposed mechanism chooses a random subset(about 1/1,000) of the population and mutagenizes itheavily and nonspecifically. The mechanism is thought tohave evolved under selection because the mutations itprovides help cells adapt to stressful conditions. Supportfor these conclusions have been reviewed recently al-though it should be noted that there are different versionsof the model (11, 12). Most of this evidence assumes thevalidity of the basic experimental observations—thatmutations are in fact induced by stress. Mechanisms toexplain this mutagenesis have been proposed and tested,but the basic assumption—stress-induced mutation—hasnot been questioned or experimentally tested. Can thebasic results be explained by selection alone rather thanmutagenesis? To evaluate this question, we consider thebasic experiment in the light of the general con-siderations outlined above and then explore alternativeexplanations.

Comparing the Cairn-Foster experiment to astandard laboratory selectionThe Cairns-Foster experiment resembles a standardlaboratory selection in that (i) parent cells are plated onselective medium and (ii) the number of colonies arisingunder selection is used to estimate a mutation rate. In-ferring mutation rate from mutant frequency demands

that the rules (Pillars of Laboratory Selection) be obeyed.One must prevent selection from influencing the numberof detected mutants.

The Cairns-Foster experiment differs radically from astandard laboratory selection in that (i) mutations ariseunder selective conditions, (ii) mutations appear to arisein nongrowing cells, and therefore (iii) mutation rates areexpressed per unit time not per number of cell divisions.These features make it difficult to comparing mutationrates under selection with those measured during unre-stricted growth. The only barrier to selective enrichmentof mutant frequency is that the tester population appearsunable to grow. This leaves open the possibility that somesubpopulation grows and contributes to mutant yield.

The Cairns-Foster experiment violates the rules oflaboratory selectionThis experiment deviates from the rules for selection inthat the selective conditions are not strong enough toassure that only large-effect mutations (compensatingframeshifts) are detected (rules 6 and 7). Selection is weakbecause the parent tester cells have 2% of the revertantlevel of β-galactosidase before being exposed to selection.They are poised on the brink of growth by added thescavenger cells. Any small-effect mutant with a slightincrease in β-galactosidase activity can upset this poiseand initiate a slow-growing colony. Small-effect muta-tions are especially common since duplications andamplifications can serve to initiate a cascade of selectableevents (see below). Small-effect mutations can contributeto the number of counted visible revertant clones if theycan improve and form a visible colony. This is facilitatedbecause cells with residual β-galactosidase are held underselection for a very long time period (6 days). These extracolonies can inflate the revertant frequency on which themutation rate estimation is based.

Selection (without mutagenesis) can enhance the yield ofmutants in this experiment because it contributes to theimprovement of plated common small-effect mutants.During nonselective growth, these small-effect mutationswould escape detection. The frequency of the coloniesappearing under selection may be dictated by the fre-quency of these small-effect mutants and the probabilitythat they can improve so as to form a colony within 6days—a very long selection period. If the experimentcould ensure that every colony is initiated by a single, rare(lac → lac+) frameshift mutation formed on the plate, then


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

some induced mutagenesis would appear to be needed.However, if common small-effect mutants can initiateclones, then growth under selection can explain the en-hanced revertant yield.

As long as colonies can improve by common small-effectmutations occurring at standard mutation rates, thephenomenology can be explained by selection alone—rapid multistep genetic adaptation during growth underselection using variants that are available under all con-ditions. If this is true, the unusual set of assumptionsneeded to explain this system in terms of “stress-in-duced” mutagenesis are no longer needed.

1. No need to propose induced mutagenesis.

2. No need to propose mutagenesis of nongrowing cells.

3. No need to propose mutagenesis induced in a randomsubset of the population.

4. No need to propose an evolved mechanism to con-travene the elaborate known mechanisms that mini-mize mutation rates.

A model to explain the Cairns-Foster results byselection (without induced mutagenesis)The basic observations listed above and diagrammed inFig. 9 can be explained without recourse to stress-inducedincrease in the standard mutation rates. Duplications andfurther increases in copy number aremutations that occurat very high frequencies under all growth conditions (seeFig. 3). More than 0.1% of cells in an unselected popula-tion typically carry a duplication of any specified point inthe chromosome (26). This high duplication frequency isa steady state achieved because of the high formation rate,higher loss rate, and fitness cost of duplications (describedabove). Unselected population also carry higher copynumber variants at lower steady-state frequencies (20).The lac operon used in the Cairns experiment is dupli-cated in approximately 0.5% of the plated cells. Suchmutations double the small amount of β-galactosidaseproduced by the leaky mutant lac allele such. Thus, 105 to106 of the 108 plated cells have two or more lac copies.Further changes in copy number (increases or decreases)occur at a rate of about 0.1/cell/division by unequal re-combination between duplicated copies in sister strands.Thus, extremely common copy number increases mayserve as small-effect mutations allowing frequent colonyinitiation under selective conditions. These considera-tions underlie a model whose basic features are outlinedbelow and diagrammed in Fig. 10.

1. A duplication of the mutant lac operon is carried byabout 1 in 200 of the 108 plated tester cells. This lacallele is leaky, producing about 2% of the enzymeactive found in a lac± revertant. Thus, 106 of the platedcells have at least two lac copies; some may havemore, since the steady state duplication frequency isexpected to come to a steady state with higher am-plifications present at lower frequency (20). It is es-timated that roughly 100 of the plated cells may have5 lac copies. (A. B. Reams, M. Savageau, S. Maisnier-Patin, and J. R. Roth, unpublished results).

2. Cells with a sufficient number of lac copies initiatecolonies. Growth of cells within these colonies canlead to a visible colony if one lac allele reverts or at-tains a sufficiently high copy number by amplifica-tion under selection.

3. The duplication and amplification events are frequentevents under all conditions, even without selection.

4. The probability of a reversion (frameshift) event tolac+ is a function of the number of lac alleles on theselection plate. The allele number increases withinany developing colony because of increases in thenumber of lac alleles per cell and the number of cellsin the colony. For example, if 105 duplication-bearingcells initiate clones that grow to reach 1,000 cells with10 lac copies, a total of 108 acts of cell division and 109

replications of the lac allele occur. This is sufficient togenerate 10 reversion events on a plate with no in-crease in mutation rate. (The standard lac reversionrate is 10−8/cell/division.)

5. When a reversion event occurs, one fully functionallac+ allele is generated within an array of tandem mu-tant lac copies. Selection favors cells that retain thelac+ allele and lose the remaining mutant copies. Theextra copies impose a significant fitness cost and in-crease the likelihood of losing the lac+ allele. Ulti-mately a cell arises that is haploid for a lac+ allele andis fully Lac+ in phenotype and that suffers no loss dueto segregation or no fitness cost. These haploid re-vertant types overgrow the original colony. The stepslisted above are described on the left side of Fig. 9.

6. Colonies can form without reversion if they suffi-ciently amplify the mutant lac allele. This can occuronly if a deletion mutation reduces the size and thefitness cost of the lac amplification. A deletion of theIS3 duplication join point element places lac within asmaller repeated unit with a lower fitness cost (Fig. 9,





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

middle). Such cost reduction allows higher lac am-plification and faster growth on lactose.

7. Deletion events are common since each end pointcan lie anywhere in a 50-kb region of one copy of theduplicated region. Any higher amplification of thelarge IS3-duplication provides additional pairs ofcopies between which an effective deletion can occur.The number of paired copies that might form isexpected to increase as a function of copy number (n)by the function D = n(n − 1)/2.

8. Clones growing by virtue of amplification aloneachieve sufficient lac copy number that some rever-tant alleles are expected to arise in every clone. If aclone reaches 108 cells by having 100 copies of the lacregion, each population doubling is expected to add100 lac+ cells to the colony. Cells with such revertantalleles are likely to grow faster than cells with a highlyamplified mutant lac allele.

9. Two types of revertant colonies arise by the eventsabove. Colonies rich in stable Lac+ cells form whenreversion occurs early in one of the many small cloneswith large, low-copy amplification. In such colonies,lac+ haploid cells are able to form and overgrow to

produce a colony dominated by lac+ haploid cells.Colonies rich in unstable Lac+ cells arise when dele-tion mutations allow high amplification. These clonesare rich in unstable Lac+ cells that grew by virtue ofamplification alone. These bifurcating pathways aredescribed at the left side of Fig. 10 and correspond tothe pathways of adaptation diagrammed in Fig. 4.

10. A third pathway also appears to contribute to rever-sion. The events in this pathway are not yet fully elu-cidated, but amodel is described at the right side of Fig.10. It is duplication frequently arise as a symmetricaltandem inversion duplication (sTID), with extendedpalindromes at each of two junctions (27). Theseduplications are likely to be unstable and deleterious,but they provide multiple lac copies and are subject toamplification by exchanges between flanking directrepeats. While being maintained by selection for in-creases in lac copy number, these rearrangements canacquire deletions that render the junctions asymmet-ric and reduce the fitness cost of amplification.Amplifications of this type are estimated to underlieabout 30% of the unstable-rich Lac+ revertant colo-nies. The junctions of asymmetric derivatives (aTID)have been identified and sequenced (28).

Figure 10 Explaining the Cairns system with growth under selection (no mutagenesis).


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

This model suggests that the Cairns-Foster experimentallows selection to affect the number of detected mutants,because the basic rules of laboratory selection have beenskirted. In particular, selection is too weak to preventgrowth of extremely common small-effect mutants withmultiple copies of lac. These cells initiate colonies thatsometimes succeed in generating a visible colony. Most ofthe observed colonies are initiated by common small-effect duplication mutations. The rare frameshift muta-tions, which are assumed to underlie these colonies, arenot required for colony initiation, but they may arise andeven predominate in the final colony because of selectivegrowth. The frequency of observable colonies is dictatedby common events that initiate the clone that becomesvisible in the course of 6 days on selective medium.

As expected for a natural population, genetic adaptationoccurs by a pathway of successive common events, eachproviding a minor growth improvement. Here the suc-cession of events that improve growth is (i) duplication ofthe mutant lac allele, (ii) stepwise further lac amplifica-tion, (iii) reversion of one copy to lac+, (iv) stepwise lossof nonrevertant lac alleles, and (v) appearance of a hap-loid cell with a revertant lac+ allele. As predicted forclones that improve in this way, each colony is a mixtureof cell types, some with an amplification of the lac region,and some that have reverted to lac+ and segregated tohaploid.

Very slight changes in experimental conditions of theCairns-Foster experiment can reestablish the laboratoryselection conditions (the Seven Pillars) and allow detec-tion of only rare, large-effect frameshift revertants arisingduring the nonselective pregrowth period. For example, ifthe parent frameshift mutation is rendered nonleaky by aribosomal mutation that reduces error frequency or byaddition of a competitive inhibitor of β-galactosidase(29), then the tester cells are no longer leaky. This makesselection strong enough to prevent growth of the com-mon small-effect duplications. Under these conditions,the Cairns experiment shows highly reduced accumula-tion of revertants with time, and detects primarily thepreexisting, large-effect revertants.

The Selection Model explains results cited as supportof “stress-induced” mutagenesisThe results most frequently cited in support of the severalmodels for stress-induced mutations (11, 12, 30) areconsistent with growth under selection associated with lac

amplification. Below we discuss and evaluate the evidenceprovided in support of stress-induced mutagenesis.

1. The stress-induced mutagenesis is attributed to theerror-prone DinB bypass polymerase. The sequencechanges that generate a lac+ allele under selection in-clude a high proportion of +1 frameshift mutations inbase runs; this is characteristic of mutations caused byDinB (31). Revertants arising under selection in theCairns system are more likely to carry associated,unselected mutations (in other genes) than revertantsarising during nonselective growth.

In the Selection Model these results reflect the fact thatthe dinB gene happens to be located close to lac onthe F′ plasmid and is included in all of the large IS-dependent lac duplications and in about 30% of thesmaller high-copy lac amplifications. This interpreta-tion is supported by the observation that associatedmutations require that dinB is located close to lac.Alleles of dinB located far from lac do not contribute torevertant yield or to associated mutations. While theeffect of amplifying dinB contributes about fourfold tothe yield of revertants under selection, selection stillenhances reversion to lac+ in strains devoid of dinBand that residual enhancement depends on recombi-nation. The observed mutagenesis is in essence anartifact due to the proximity of dinB to the lac alleleunder selection.

2. Stress-induced mutagenesis is attributed to error-prone recombinational replication. This conclusion isused to explain the fact that mutations arise in theapparent absence of chromosome replication and thatthe yield of mutants under selection requires recom-bination proficiency. In the Selection Model recom-bination is responsible for duplication, amplification,and segregation events that allow small-effect muta-tions to improve growth under selection.

3. Some revertant colonies include cells whose Lac+

phenotype is unstable. These cells include ampli-fications of the lac operon. Amplifications are attrib-uted to a separate stress-induced event that isindependent of frameshift mutation. In the Selectionmodel, the amplification cells are precursors to rever-sion that allow growth and enhance the probability of areversion even by providing more target sequences.

4. Short-junction amplifications and asymmetric TIDmutations are common amongunstable Lac+ revertantsin the Cairns system, but are rare in unselected popu-





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

lations. The stress-induced mutation model suggeststhat these events occur only when stress induces a novelmechanism that does not operate in unstressedcultures. The Selection Model proposes that thesemutations formby a series of genetic events that seldomgoes to completion without selection, because short-lived and deleterious intermediates are usually lostfrom the population. Under long-term selection forincreases in lac copy number, these precursors are heldunder selection long enough to acquire the deletionevents needed to complete the process.

5. The accumulation of mutants under selection in theCairns system requires that the lac mutation be leaky.The stress-induced mutagenesis model proposes thatthis residual function provides energy that is needed topermit a costly mechanism to create mutations innongrowing cells. The Selection Model proposes thatthis basal level allows copy number variants to in-crease lac enzymes and initiate clones under selection.

3. The Shapiro System—Appearance ofMu-Mediated ara-lac Deletions under SelectionPossibly the first system interpreted as evidence forstress-induced mutagenesis was one described byShapiro and coworkers (10, 32). In this system, a pro-moter-less silent copy of the lactose (lac) operon isinserted near the arabinose (ara) operon of E. coli with aMu prophage present between them. The strain wasplated on lactose medium, and Lac+ colonies appearedwith time. Each Lac+ revertant reflected a deletion mu-tation that allowed the lac operon to be expressed fromthe promoter of the ara region. The deletions wereformed by events induced by the Mu prophage. It wasinitially suggested that these revertants were induced bystress in the form of lactose limitation, since they seemedto be rare in populations that were not exposed to se-lection. Later work by Cairns and Foster (33) demon-strated by fluctuation tests that the number of selectedrevertant colonies shows a Luria-Delbrück fluctuationand thus must be determined during the nonselectivegrowth period prior to plating on selective medium andcould not be initiated by stress.

This would appear to eliminate stress-induced muta-genesis, but a question remains open. When Lac+ cloneswere isolated from the pregrowth culture by sib selections(no stress involved), the nature of the deletions was dif-ferent from that characterized in revertants isolated afterprolonged selection (34, 35).

The results have not been pursued but can be explained bythe selection model described here. That is, if mutantswith modestly improved ability to grow on lactose ariseprior to selection, these mutants may initiate colonies onselective mediumwithin which the population adapts andimproves. The improvements may lead to the deletiontypes ultimately characterized. If these deletions are gen-erated by multistep improvement, the process is expectedto operate with or without selection but be driven bymassaction when each successive change allows improvedgrowth on lactose. That is, without growth limitation, thesib selection should yield partially Lac+ cells that are un-likely to have completed the process. During growthlimitation, the process is driven to completion by higherrates of exponential growth of each intermediate.

4. The Radman-Taddei-Matic System, a NonlethalSelectionShowing theAccumulation of RifRMutantsin “Aging” ColoniesThe formation of mutants resistant to the antibiotic ri-fampicin (RifR) is widely used formeasuringmutation ratesin growing bacteria. The attractions of the method are thefollowing: (i) it can applied to a variety of bacteria withno need to construct special tester strains; (ii) it measuresbase substitutions (missense mutations) at a variety ofsites within an essential gene (rpoB, RNA polymerase β-subunit); (iii) since all RifR mutations affect a single gene,they can be easily characterized byDNA sequencing; (iv) atleast 69 different base substitution mutations, distributedamong 24 coding positions, result in a RifR phenotype (36);and (v) Rif resistance arises at a rate that is easy to measure(~10−8/cell/generation in E. coli and Salmonella).

Unselected mutation rates to RifR are measured by astandard laboratory selection (Fig. 6). Cells are grown inrich medium (lacking rifampicin), where it is presumedthat resistant and sensitive mutations grow equally well.Mutants form in these cultures and are detected by platedon selective medium containing rifampicin. Unlike theLuria-Delbrück experiment, this selection is not lethal,but it does reliably detect large-effect RifR without in-fluencing their frequency. Enumeration of these colonieswith allowance for frequency fluctuation allows one toestimate the rate of mutation formation during unse-lected growth.

Radman, Taddei, and Matic modified this system tostudy the effects of growth limitation on mutation rates(37, 38). Colonies were allowed to reach a terminal size


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

on rich medium, and the frequency of RifR cells wasassayed over time as the cells in the colony “aged”without growth. Increases in the frequency of RifR cells asa function of time were attributed to a stress-inducedincrease in mutation rate experienced by a nongrowingpopulation. Colonies were initiated by spotting 100 to1,000 cells on nitrocellulose filters placed on rich medi-um. The small starting population ensured that no RifR

mutants were present at the beginning of colony growth(no founder effects). Bacteria in any colony could beeasily retrieved for analysis. On the first day, viable cells(colony forming units, CFU) increased to ~109. By thesecond day, cell number per colony increased further to~1010. No further increase or decrease in colony popu-lation was seen over the following 5 days of incubation.Many artificial colonies were initiated in parallel so thatcell number and mutant frequency could be tested atvarious times over a period of about a week. It is sur-prising that the frequency of RifR cells increased duringthe aging period. Laboratory strains show increases upto 100-fold from day 1 to day 7 while the total CFUremained unchanged (37, 38). A set of 787 natural iso-lates of E. coli from diverse ecological niches (39) showedvariable responses with a median 7-fold increase in RifR

mutant frequency between days 1 and 7—a period inwhich the median CFU increased only 1.2-fold (39).

These results were interpreted as evidence of “stress-induced mutagenesis.” It was assumed that stressed cellsincreased their general mutation rate (37, 38, 39). Thephenomenon was referred to as ROSE (resting organismsin a structured environment) and MAC (mutagenesis inaging colonies). As in the Cairns system, there is noapparent growth to serve as the basis for comparingmutation rates in stressed and unstressed cells. That is,the stress of “aging” can not be imposed on growing cellsand there is no way to prevent growth without imposingstress. As in the Cairns system, the inferred stress-induced mutagenesis is based on using an increase inmutant frequency (over 7 days of aging) to infer an in-crease in mutation rate in a nongrowing population. Thephenomenon was assumed to reflect mutagenesis andexperiments were focused on how this mutagenesis wascaused and regulated, not on whether or not the fre-quency increase reflected mutagenesis.

Tests examined a series of mutant strains, each lackingsome candidate function, to determine which functionaffected accumulation of RifR mutants during aging.Accumulation depended on the ability to induce the SOS

system for DNA repair and on possession of an activeRpoS protein—a stationary-phase sigma factor known tocontribute to expression of a variety of genes in growth-limited cells. One particular natural isolate of E. coli,C4750, was tested for the predicted genome-wide mu-tagenesis that the model predicts (39). This isolateshowed a 77-fold increase in RifR cells between days 1 and7 but showed no increase in the frequency of mutantsresistant to streptomycin or nalidixic acid. The frequencyof mutants resistant to 5-fluorouracil, mecillinam, andfosfomycin increased only 3.4-, 1.9-, and 4.8-fold, re-spectively. While these results were interpreted as sup-port for general mutagenesis, they seem to suggest thatanother model is needed.

This system violates the rules of laboratory selectionwhich are essential here because an increase in mutantfrequency is used to infer an increase in mutation rate. Aconclusion requires knowing that the mutant frequencyincrease is due to new mutations forming over timeunder selection and not to faster growth of preexistingmutants. The only evidence for this is that the populationas a whole showed no significant growth between theinitial growth period (days 0 to 2) and day 7. Thesemeasurements could not detect mutations arising in agrowing subpopulation. More importantly, the tests didnot exclude growth of a RifR subpopulation in the vastlylarger population of nongrowing RifS parental cells.

Recently, Wrande et al. (40) readdressed the question ofhow RifR mutants accumulate in aging colonies. The basicphenomenon was reproduced in both E. coli and Sal-monella enterica, including clinical and natural isolates.As before, accumulation of RifR mutants depended on anactive rpoS gene. However, several striking and signifi-cant clues concerning the basis of this phenomenonemerged from new experiments.

1. Accumulation of RifR mutant cells continued for up toone month, reaching a frequency of >10−4.

2. The frequency increased exponentially (as expected forgrowth under selection) rather than linearly (as pre-dicted for mutagenesis of a nongrowing population).

3. The RifR mutant cells were localized in just one or afew small sectors of an aged colony, as expected if theyare clonal subpopulations growing from individualprecursor cells (Fig. 11). Mutagenesis of nongrowingcells predicts a normal spatial distribution of newmutants.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

4. All RifR cells in an individual sector carried the samerpoB sequence alteration, demonstrating their clonalrelatedness. In contrast, mutant cells from differentsectors within the same colony usually carried differ-ent sequence alterations.

5. In colonies with many RifR mutant cells, most of themutants were found in a smaller number of localizedsubclones, consistent with variation in mutant fre-quency reflecting differences in clone growth ratherthan differences in the number of clones.

6. The RifR mutants showed no significant increase in thenumber of secondary mutations (the frequency ofauxotrophs was tested), suggesting that they had notexperienced general mutagenesis.

7. Most of the isolated RifR mutants showed a growthadvantage over parent cells in colony reconstructionexperiments. This growth advantage explains why cellsaccumulate as colonies age.

Thus, the accumulation of RifR mutants in aging coloniesis real, but it is explained by selection, not mutagenesis.Apparently RifR mutants grow better in “aging” coloniesthan parent cells even when no rifampicin is present. Thedetected RifR mutants arise during early stages in theexperiment and increase in frequency by growing fasterthan the population at large (40). Mutations appear toarise in nongrowing populations because the total addedRifR cells are too few to be detected in the total population.

This gives the impression that new mutations accumulatein a nongrowing population when, in fact, growth is re-sponsible for the observed increase. Additional experi-ments will be required to explain exactly why RifRmutantsgrow in aging colonies. It seems likely that some mutantforms of RNA polymerase may allow expression of genesthat promote growth under aging conditions. Many RifR

mutations are known to affect transcription elongation ortermination (42, 43, 44, 45, 46, 47, 48), including two ofthe rpoB mutations shown to enhance growth in agingcolonies. Originally RpoS was also suggested to regulatethe proposedmechanism of “stress-induced”mutagenesis(37) and has also been suggested to controlmutagenesis inthe Cairns system (12). It seems more likely that RpoSsigma factor and mutant RNA polymerases can act to-gether to facilitate growth within stressed colonies.

SIX OTHER SYSTEMS CITED AS SHOWINGSTRESS-INDUCED MUTAGENESIS

The Hall System—Rapid Evolution of a Newβ-Galactosidase (EBG)In this extensively studied system, an E. coli lac deletionmutant acquires the ability to utilize lactose as a carbon andenergy source (9, 49, 50, 51, 52). This ability requires twomutations thatmodify a cryptic operon (ebg, for evolved β-galactosidase), whose normal function is unknown butdoes not include use of lactose. One mutation inactivates

Figure 11 The distribution of RifR cells in “aged” colonies is nonrandom.Most RifR mutants are found to be localized as jackpots in one or a fewsectors of the colony, and within each of these jackpots the mutants carry the same sequence alteration, consistent with selective growth of a RifR

mutant during the period of “aging.” Adapted from reference 41 with permission.


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

the EbgR repressor, thereby allowing constitutive operonexpression, and the second mutation alters the EbgACenzyme, increasing its ability to hydrolyze lactose. Eachmutation alone allows very slow growth on lactose (gen-eration times >3 days), but when both mutations arepresent cells grow rapidly (tgen = 20 min). The ebgR-inactivatingmutation is expected to be common (10−6/cell/generation) and the ebgAC point mutation rare (10−9/cell/generation). Based on these rates, cells with bothmutationsare expected to arise at a rate of 10−15/cell/generation and(as expected) are not detected in unselected cultures.

In contrast to this expectation, these double mutants arisecontinuously over a period of several weeks after 108 parentcells are plated on selective medium containing lactose.The selection plates also contain a small amount of glycerol(0.01%) that allows some growth, and the lactose-utilizingdouble mutants are detected as papillae appearing above asparse bacterial lawn. The lawn population (even afterperhaps a 10-fold increase at the expense of glycerol) is still6 to 7 orders of magnitude too small to ensure coincidentoccurrence of the two required mutations. The result hasbeen interpreted as evidence that stress induces mutations.Since the recovered lactose-using cells show no increase inthe frequency of other mutations, it is concluded that thestress-induced mutagenesis somehow directs mutations tosites that improve growth.

As was true for the Cairns system, this experiment re-sembles a standard laboratory selection altered so as toestimate the number of mutations occurring in a non-growing population. This system uses the frequency ofmutants (actually double mutants) to draw a conclusionregarding mutation rate. As described above, this pro-cedure requires that the selection conditions adhere tothe rules of selection and prevent all growth under se-lection. It seems likely that selection in this system is notstrong enough to prevent growth of small-effect muta-tions, which can initiate clones and improve their growthability during growth under selection. If this is true, thesystem violates the rules and the conclusions regardingmutation rates are unfounded. Below is one of manypossible scenarios by which selection might explain thebehavior of this system without recourse to changes inthe rate or site specificity of mutation.

Since the frequency of repressor mutants is high (10−6), itis predicted that about 100 null repressor mutants areincluded in the original population. Many more cellsmight carry more modest repressor defects. During

growth on glycerol, each of these cells could produce alarge population in which a duplication of the ebg operonarises. This would give multiple copies of a highly ex-pressed operon. Cells growing with an ebg amplificationcould produce many cells, each with multiple copies ofthe ebgAC genes. This increase in target gene number(more cells and more ebgAC genes per cell) could allow arare second mutation to arise without any change in therate or specificity of the mutation process.

In this experiment, no tests were performed to test for thesequential appearance of the two mutations within thepapillae. One might expect some cells within these clonesto show unstable amplifications (as seen in the Cairnssystem). Thus, the system does not eliminate growth ofsequential subpopulations and/or gene amplification.While the scenario above is speculative, it suggests thatthe behavior of this system may be explained by someform of the pathways of genetic adaptation used to ex-plain the Cairns system. This explanation may also ex-plain the trpAB revertants described below.

Growth and Selection Explains an Apparent Case ofStress-Induced Mutagenesis—Reversion of TwoSeparate Mutations (trpA and trpB) inEscherichia coliIn this system, missense mutations in the trpA and trpBgenes of E. coli appeared to revert to Trp+ more frequentlywhen starved for tryptophan than when tryptophan waspresent (53, 54). In a doublemutant carrying both of thesemutations, reversion to Trp+ was expected to be astro-nomically rare (3.3 × 10−19) based on the measured re-version rate of the individual mutations tested singly. Theobserved Trp+ reversion rate of the double mutant duringtryptophan starvation was 4.5 × 10−11 per cell per day.Thus, the trpA and trpB missense mutations reverted toTrp+ about 108-fold more frequently than would beexpected if the two mutations occurred independently ofeach other. The genome-wide mutation rate (as measuredby valine resistance and lac reversion) did not increaseduring Trp starvation and there was no effect on Trpreversion when cells were starved for another amino acid,cysteine. Thus it appeared that stress was inducing mu-tagenesis and directing mutations preferentially to sitesthat improve growth.

Hall observed early on that the stress-induced hypermu-table state model (which he proposed) was inconsistentwith the behavior of the trp mutations, because the





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

magnitude of the genome-wide mutagenesis required toaccount for the reversion would certainly be lethal for anycell. Each base would have a probability of around 0.04 ofmutating during the selection period and a vast number oflethal mutations would be expected (as they are expectedin the Cairns system). Hall therefore examined the pos-sibility that the two mutations occur sequentially withintervening growth. This analysis was based on the sug-gestion that the trpA, trpB double mutant is expected toaccumulate the intermediate indole-glycerol-P (due to thetrpA mutation). This compound breaks down spontane-ously to indole, which the TrpB enzyme can convert totryptophan. Thus, if the trpB mutation reverted first, theresulting trpA, trpB+ cell could use the spontaneouslygenerated indole and form a growing clone in which thesecond mutation (to trpA+) could occur. Hall measuredthe growth rates of single-mutant cells within double-mutant colonies concluded that the most likely explana-tion of the enhanced reversion of the double mutant was asequence of two steps, each giving progressively improvedgrowth. That is, trpA, trpB → indol accumulation →, trpAtrpB+

→ growth on indole → trpA+, trpB+ (55). This illus-trates one of the myriad pathways by which selectiveconditions that incompletely limit growth can give theappearance of stress-induced mutagenesis. It also makesclear the point that considerable detective work may berequired to see how the devious hand of selection canupset even the most careful attempts to foil it.

Stress-Induced Accumulation of DeleteriousMutations in Stationary-Phase Cultures of E. coliBehavior of this system suggests that stressed cells in anongrowing liquid culture show a stress-induced increase inthe formation rate of deleterious mutations (56). Replicatebacterial cultures are grown to full density in liquid mediumand then further incubated for up to 100 days with very littleadditional population increase. Samples were removed fromthe resting cell suspension at various times and used to in-oculate new cultures in fresh medium. These secondarycultures were used to estimate fitness (growth rate). Themedian decrease in fitness and the increase in varianceamong the sample lineages were used to infer mutation ratesin the resting cell suspension. The genomic mutation rateinferred for cells in prolonged stationary phase was about0.03/genome/day, about 10-fold higher than values extra-polated from exponential cultures.

While this experiment is not a standard laboratory se-lection (not done on solid medium), it does share many

features with systems used as evidence for stress-inducedmutation. That is, a population is held under conditionsthat allow no net growth and increases in mutant fre-quency (assessed indirectly) are used to infer increases inmutation rate in the nongrowing population.

To fairly interpret these results requires knowing theamount of growth that might be occurring within theresting population and whether selection for growth orsurvival in the resting period could affect mutant fre-quency. As described in Fig. 7 and Fig. 8 above, largecultures that show no net change in population size maysupport substantial growth of subpopulations in whichmutations can arise. The growth may be populations tosmall to detect or balance growth and death such that nonet change in viable cells occurs. One potential explana-tion for this result that does not involve starvation-induced mutagenesis is undetected growth.

Critics of this experiment have suggested that it involvesunappreciated growth where specific subpopulations growextensively (57) without any change in the population as awhole—the GASP phenomenon (58, 59). This crypticgrowth provides added opportunities for random muta-tions to form at normal rates leading to fitness decreasesthat are not surprising in view of the large fraction of totalmutations expected to be deleterious. A more extremepossibility is that mutations are positively selected duringstationary-phase incubation. These mutations may en-hance survival in stationary phase but be deleterious underthe growth conditions used in the fitness tests.

As in several other systems that involve nongrowingpopulations, mutation rates cannot be compared with anunstressed control population other than the one growingwithout restriction. The authors express their apparentmutation rates based on a clock-like view of mutation(mutations/day), whereas their stress-free control ratesare based onmutations/cell/generation. It is not obvious ifand how clock-like and generation-dependent mutationrates can be directly compared.

Experiments Interpreted as Evidence thatTranscription DirectsMutations to Specific Genes inNongrowing CellsYasbin and colleagues made Cairns-type experimentsasking whether stress-induced mutagenesis also occurs inBacillus subtilis, a gram-positive species. They measuredthe accumulation of late-appearing prototrophs from


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

three different auxotrophic mutants (his nonsense, metnonsense, and leu missense) and over 10 days observedincreases of ~20-fold for His+, <5-fold for Met+, and 2-fold for Leu+ (60, 61, 62). As in most experiments of thisgenre the authors made the assumption that the increasein the frequency of prototrophs with time could beinterpreted as due to an increase in the rate of mutationin nongrowing cells. No direct tests for increased muta-genesis (for example, by testing for an increase in thefrequency of unselected mutations at other locations)were made. Instead, attention was focused on identifyinga mechanism to explain the presumed increase in sta-tionary-phase mutation rate. They were intrigued whenthey found that the accumulation of late-appearing (aswell as growth-associated) prototrophs was partly de-pendent on the gene mfd (63). On this basis they pro-posed that the late accumulation of prototrophs wasassociated with transcription of the mutant genes. Theidea that transcription might “direct” mutations to aparticular target, like lac, was proposed shortly afterpublication of the original Cairns experiment (64). Therelevance ofmfd for this class of model is that it codes forTCRF (transcription-coupled repair factor) (65). WhenRNA polymerase gets stalled in the process of tran-scription by bulky DNA lesions, the stalled RNAP can bedislodged by TCRF protein, which then recruits Uvr(A)BC to the lesion site, thus promoting nucleotide excisionrepair. TCRF activity reduces the rate of mutation on thetranscribed strand of DNA relative to that on thenontranscribed strand (65). The observation by Yasbinand colleagues that the absence of the TCRF genedecreases the frequency of prototrophs is not readilyexplained by the known activities of TCRF, and no modelhas been proposed to explain the phenomenon.

Sequence analysis of the B. subtilis prototrophs showedthat only a minority of the late-appearing mutants wereactually caused by reversion of the auxotrophic alleles.The majority were due to external suppressors (possiblymutant tRNAs), and some of these were late-appearingbecause they grew slowly (63) and may therefore havebeen preexisting growth-associated mutations. The leak-iness of the three auxotrophic alleles was not tested, butnonsense alleles in other systems are frequently highlyleaky and could plausibly support slow growth, especiallyif the mutant genes were amplified in a subpopulation ofcells, as observed in the lac system. Indeed, the two allelesfor which the frequency of late-appearing prototrophswas highest are both nonsense mutations. No tests weremade for gene duplication or amplification, or for the

presence of genetic heterogeneity in prototrophic colo-nies. TCRF activity was only associated with the infre-quent reversion prototrophs and not with the much morefrequent external suppressor prototrophs. We do notknow why TCRF should associate with a small increase inthe frequency of reversions: one possibility we suggest isthat the new DNA synthesis following nucleotide exci-sion repair is error prone, fortuitously causing occasionalreversion mutations. This has not been tested.

In summary, as in other cases documented here, theseexperiments are based on the assumption that stress-in-duced increases in mutation rate were an established fact;the alternative possibility, of selection acting on slow-growing subpopulation, was not tested. The experimentalsystem breaks the rules of laboratory selection by notclearly separating selection from growth. The accumu-lation of late-appearing prototrophs in this system canplausibly be explained by selection and growth of sub-populations of cells carrying leaky alleles, possibly asso-ciated with gene amplifications.

Adaptive Reversion of a LYS2 Frameshift Mutation inSaccharomyces cerevisiaeA Cairns-type experiment was made to test whether thestress of stationary phase caused increased mutation ratesalso in eukaryotes (66). This experiment measured theaccumulation of late-appearing prototrophs from aconstructed lys frameshift mutation in S. cerevisiae andwas, in design, methodology, and result, almost identicalto the classic Cairns-lac reversion experiment (8).

Lys+ colonies appeared on the selective agar plates after 3days of incubation, with additional colonies appearing oneach of the following 5 days. The numbers of coloniesappearing from independent cultures showed a distri-bution of mutants that fitted well to expectations forrandom (preexisting) mutations on day 3, calculatedaccording to Lea and Coulson (67). At later times (days 5to 8) the distribution of mutants fitted well to a Poissondistribution, as expected for mutations occurring afterplating and in response to selection. Genetic and physicalanalysis of the Lys+ mutants showed that 90% contained acompensating frameshift mutation within the LYS2coding sequence, while the remaining 10% were ap-parently revertants, a conclusion supported by directsequence analysis in a later study (68). After testingand ruling out cross-feeding and slow-growing mutantsas causes of the late-arising Lys+ mutants, Steele and





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

Jinks-Robertson (66) concluded that at least some of theevents leading to the Lys+ phenotype must have occurredafter selective plating.

The experiment showed that there was no significantincrease in total cell number as a function of the numberof days a plate was incubated. However, these experi-ments did not test the possibility that subpopulations ofcells were growing (those that eventually gave rise toprototrophic colonies) or that within those subpopula-tions there might have been amplification of the LYS2target gene. Either or both of these events would haveincreased the genetic target for reversion or suppressormutations without necessarily causing any appreciableincrease in the size of the general population of platedcells. By analogy with the Cairns-Foster lac system, wesuggest that the LYS2 frameshift mutation might beleaky, and that it could be subject to duplication andamplification under selection for lysine biosynthesis. Toour knowledge no one has checked this experimentalsystem for evidence that the LYS2 mutant allele is leaky,or that it is amplified during the postplating selectionperiod before the visible growth of late-appearing Lys+

revertants.

Interestingly, the authors define the late-appearing Lys+

revertants as “adaptive” because they occur specifically inresponse to lysine starvation. Thus, if cells were starvedfor other amino acids it did not select for Lys+ revertants.This showed that starvation did not induce genome-widemutagenesis. This is in agreement with data that star-vation in Salmonella does not cause a general increase inmutation rates (24).

Steele and Jinks-Robertson also selected Lys+ revertantsfrom populations of cells that were UV-irradiated beforeplating. The numbers of early- and late-appearing Lys+

revertants were both increased by the irradiation, 5.5-foldand ~10-fold, respectively. Interestingly, the 10-fold in-crease applied to revertants appearing on days 4, 5, and 6,whereas on day 7 it fell back to a 6-fold increase. But ifirradiated cells were allowed to grow before plating, orwere held with another starvation before plating, then thenumber of late-appearing Lys+ revertants was signifi-cantly reduced relative to the numbers found whenplating directly after irradiation. These data can easily beexplained by the gene amplification model because UVirradiation would increase the number of recom-binogenic DNA ends, increasing the probability of du-plication and amplification of the LYS2 allele. This would

increase the number of cells in the population that couldgrow slowly in the absence of lysine, and be substrates forgene amplification and eventual selection of a reversionmutation in the LYS2 gene. If, however, cells were notplated on the selective medium directly after UV irradi-ation, then DNA repair systems would have time to re-pair lesions, unstable and nonselected duplications oramplifications would be lost, and the number of cells inthe population with a genome configuration suitable forgrowth selection to make lysine would be reduced.

Our conclusion is that the late-appearing Lys+ mutants,selected in the absence of lysine, can best be explainedwithin the parameters of the model for growth advantageassociated with gene amplification, essentially as de-scribed for the lac system. Experiments to test this modelhave not been made but we would predict that the LYS2frameshift allele is leaky and that some late-appearingLys+ mutant colonies should be heterogenous and con-tain cells with duplications and amplifications of themutant gene.

Mutation underSelection for RecessiveMutations ina Chromosomal GeneThis system resembles that of Cairns and Foster in thatone selects for Lac+ revertants of a Lac− parent strain (69,70). The system differs in that it does not involve an F′plasmid and appears to demand recessive (loss of func-tion) chromosomal mutations, while the Cairns-Fostersystem demands dominant (gain of function) mutations.This system uses a lac operon inserted within a chro-mosomal purine gene (purD) and transcribed from itspurine-regulated promoter. Since the purD promoter isnormally repressed in the presence of excess purines, thestrain is phenotypically Lac− (and requires purine be-cause of the purD disruption). When this strain is platedon minimal lactose medium with added purine (ade-nine), about 100 revertant colonies (Lac+) arise over thecourse of 6 days above the lawn of plated cells. The mostfrequent revertants are expected to be common recessivemutations in the gene for the purine repressor protein(purR− 1,000 bp) or rare mutations in the tiny PurR-binding operator sequence (16 bp) near the purD pro-moter. Either type of mutation should activate tran-scription of purD and allow expression of the inserted lacgenes, resulting in a Lac+ phenotype.

In the initial description of this system (70), the testerlawn showed very little growth and about 100 Lac+ re-


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

vertant colonies accumulated over the course of 6 days ofincubation. Results were interpreted as evidence forstress-induced mutagenesis of nongrowing cells. Thisconclusion was supported by the finding that the ratio ofOc to R− mutants was much higher than one would ex-pect based on their target sizes. Among day 2 revertants,Oc mutants were 34% of total. Among day 6 revertants Oc

mutants were 8% of revertants; the operator target is only1.6% of the purR target. The high relative frequency of Oc

mutants is interpreted as evidence that stress directsmutations to sites near those that limit growth (the un-expressed lac operon); the higher frequency on day 2 isnot explained. In this experiment, the absolute frequencyof Oc mutants is also surprising—approximately 10 per108 plated cells—this is approximately 100- to 1,000-foldhigher than one would expect for a target of 10 bp.

The conclusion that this system reflects stress-inducedmutation rests on the pillars of laboratory selection. Thatis, growth of the parent strain is must be prevented byselection, and revertant colonies rmust reflect new singlelarge-effect mutations that confer full fitness and are notsubject to further improvement during maturation of thecolony. Recent analysis of this system (S. Quinones-Sotoand J. R. Roth, unpublished) suggests that substantiallawn growth occurs (about 3 generations) and the pro-cess by which Lac+ colonies arise requires multiple mu-tations that arise as individual steps in a pathway ofgenetic adaptation during growth under selection.Growth of the parent on lactose is restricted not only byrepression, but also by polar effects of a stem-loopstructure present upstream of lac in the inserted se-quence. Neither a purR nor an Oc mutation alone issufficient to allow a revertant colony to appear underselection. All revertants have lost the stem-loop structure.Thus, the recovered revertants have genotypes that areexpected to arise at frequencies of less than 10−15, yetabout 100 such revertants arise from 108 slow-growingplated cells. Direct tests of these revertants reveal noevidence that they have experienced any global muta-genesis during selection. Revertant colonies include cellswith an unstable Lac+ phenotype. These cells have anamplification of the lac operon due to recombinationbetween the rrnH and rrnE loci –5 kb direct sequencerepeats that flank the purD::lac region. The yield ofrevertants in this system is reduced about fivefold by arecA mutation. While many details of the reversionprocess remain to be elucidated, it is clear that the excessof Oc mutations and frequency of stem-loop deletionscan be explained by amplification of the purD::lac region

under selection during growth of the developing rever-tant colonies. This amplification allows growth andprovides additional copies of the growth-rate-limiting lacregion, thereby enhancing the likelihood of mutationsthat improve growth under selection. This system seemsto exemplify the amplification selection process de-scribed for the Cairns-Foster system and involves noincrease in mutation rate.

CONCLUDING REMARKSPublication of the Cairns system in 1988 (8) sparkedrenewed interest in the process by which mutations arise.Work following up on this initial report has revealedmethodological pitfalls associated with measuring mu-tation rates and the difficulty of unambiguously sepa-rating the effects of mutation and selection. Thus, asoutlined in this chapter, the inference that stress inducesmutagenesis can generally be explained by an inability inthe different experimental setups used to completelyprevent the effects of growth when applying nonlethalselections; i.e., the claims of stress-induced mutagenesishave not strictly excluded the influence of growth andselection in increasing mutant frequencies. Instead manyof these observations can be explained by the ability ofselection to detect and amplify small differences ingrowth rate caused by common small-effect mutations.One very common class of mutations that will be im-portant during any adaptive response in which increasedactivity of some gene can enhance growth is gene am-plification. In the widely studied lac system of Cairns,amplification of lac and occasional coamplification of thenearby dinB (coding for the error-prone DNA poly-merase, PolIV), can account for the apparent increase innumber of Lac+ revertants without any need to invoke anincreased mutation rate.

From a more general perspective, gene amplifications areexpected to be a major immediate response whenevergrowth is limited by an external factor (e.g., low nutrientlevels or the presence of a toxic factor) or an internalfactor (e.g., a deleterious mutation) and when increasedlevels of a gene product might mitigate the problem.Furthermore, gene amplification could facilitate theprocess of acquiring stable adaptive mutations by in-creasing the number of selected target genes in thepopulation. These stable changes could occur outside ofthe amplified region and increase in frequency by pop-ulation expansion alone, or they could affect the ampli-fied genes, in which case target number is increased by





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

both cell growth and added copies per cell. Since theamplification may be dispensable after a stable muta-tional improvement occurs, it is conceivable that theintermediate amplifications leave no trace in the genome.Thus, we suggest that many adaptive phenomena in factoccur with amplifications being transient intermediatesthat, because of their intrinsic instability, are rapidly lostand therefore easily underestimated during studies ofadaptive evolution. Apart from the lac system, recentlydescribed examples of such amplification processes ofmedical significance are provided by the sequential evo-lution of cephalosporin resistance by TEM β-lactamases(41), resistance to the tRNA synthetase inhibitor mu-pirocin (71), and resistance of tumor cells to cytostaticdrugs (72). Similar processes of sequential selection ofcommon mutations of small effect (including gene am-plification) are likely to apply also for malignant trans-formations, where cells successively escape the variousspatial and temporal control mechanisms that keepgrowth of individual cells in check (73).

ACKNOWLEDGMENTSNo potential conflicts of interest relevant to this review were reported.

REFERENCES1. Lederberg J, Lederberg EM. 1952. Replica plating and indirectselection of bacterial mutants. J Bacteriol 63:399–406.2. Luria SE, Delbrück M. 1943. Mutations of bacteria from virussensitivity to virus resistance. Genetics 28:491–511.3. Feng G, Tsui HC, Winkler ME. 1996. Depletion of the cellularamounts of the MutS and MutH methyl-directed mismatch repairproteins in stationary-phase Escherichia coli K-12 cells. J Bacteriol178:2388–2396.4. Harris RS, Feng G, Ross KJ, Sidhu R, Thulin C, Longerich S,Szigety SK, Winkler ME, Rosenberg SM. 1997. Mismatch repairprotein MutL becomes limiting during stationary-phase mutation.Genes Dev 11:2426–2437.5. Bhagwat AS, Lieb M. 2002. Cooperation and competition inmismatch repair: very short-patch repair and methyl-directed mis-match repair in Escherichia coli. Mol Microbiol 44:1421–1428.6. Drotschmann K, Aronshtam A, Fritz HJ, Marinus MG. 1998. TheEscherichia coli MutL protein stimulates binding of Vsr and MutS toheteroduplex DNA. Nucleic Acids Res 26:948–953.7. Mayr E. 1982. The Growth of Biological Thought: Diversity, Evo-lution, and Inheritance, p 738–744. Harvard University Press,Cambridge, MA.8. Cairns J, Overbaugh J, Miller S. 1988. The origin of mutants.Nature 335:142–145.9. Hall BG. 1988. Adaptive evolution that requires multiple sponta-neous mutations. I. Mutations involving an insertion sequence. Ge-netics 120:887–897.10. Shapiro JA. 1984. Observations on the formation of clones con-taining araB-lacZ cistron fusions. Mol Gen Genet 194:79–90.11. Foster PL. 2007. Stress-induced mutagenesis in bacteria. Crit RevBiochem Mol Biol 42:373–397.

12. Galhardo RS, Hastings PJ, Rosenberg SM. 2007. Mutation as astress response and the regulation of evolvability. Crit Rev BiochemMol Biol 42:399–435.13. Freeland SJ, Hurst LD. 1998. The genetic code is one in a million.J Mol Evol 47:238–248.14. Freeland SJ, Hurst LD. 2004. Evolution encoded. Sci Am 290:84–91.15. Leong PM, Hsia HC, Miller JH. 1986. Analysis of spontaneousbase substitutions generated in mismatch-repair-deficient strains ofEscherichia coli. J Bacteriol 168:412–416.16. Li W-H, Graur D. 2000. Fundamentals of Molecular Evolution,2nd ed. Sinauer Associates, Sunderland, MA.17. Markiewicz P, Kleina LG, Cruz C, Ehret S, Miller JH. 1994.Genetic studies of the lac repressor. XIV Analysis of 4000 alteredEscherichia coli lac repressors reveals essential and non-essentialresidues, as well as “spacers” which do not require a specific sequence.J Mol Biol 240:421–433.18. Miller JH. 1980. Genetic analysis of the lac repressor. Curr TopMicrobiol Immunol 90:1–18.19. Miller JH. 1983. Mutational specificity in bacteria. Annu RevGenet 17:215–238.20. Reams AB, Kofoid E, Savageau M, Roth JR. 2010. Duplicationfrequency in a population of Salmonella enterica rapidly approachessteady state with or without recombination. Genetics 184:1077–1094.21. Rosche WA, Foster PL. 2000. Determining mutation rates inbacterial populations. Methods20:4–17.22. Akerlund T, Nordstrom K, Bernander R. 1995. Analysis of cellsize and DNA content in exponentially growing and stationary-phasebatch cultures of Escherichia coli. J Bacteriol 177:6791–6797.23. Hall BG. 1990. Spontaneous point mutations that occur moreoften when advantageous than when neutral. Genetics 126:5–16.24. Hughes D, Andersson DI. 1997. Carbon starvation of Salmonellatyphimurium does not cause a general increase of mutation rates.J Bacteriol 179:6688–6691.25. Cairns J, Foster PL. 1991. Adaptive reversion of a frameshiftmutation in Escherichia coli. Genetics 128:695–701.26. Anderson P, Roth J. 1981. Spontaneous tandem genetic dupli-cations in Salmonella typhimurium arise by unequal recombinationbetween rRNA (rrn) cistrons. Proc Natl Acad Sci USA 78:3113–3117.27. Kugelberg E, Kofoid E, Andersson DI, Lu Y, Mellor J, Roth FP,Roth JR. 2010. The tandem inversion duplication in Salmonellaenterica: selection drives unstable precursors to final mutation types.Genetics 185:65–85.28. Kugelberg E, Kofoid E, Reams AB, Andersson DI, Roth JR.2006. Multiple pathways of selected gene amplification during adap-tive mutation. Proc Natl Acad Sci USA 103:17319–17324.29. Andersson DI, Slechta ES, Roth JR. 1998. Evidence that geneamplification underlies adaptive mutability of the bacterial lac operon.Science 282:1133–1135.30. Hastings PJ. 2007. Adaptive amplification. Crit Rev Biochem MolBiol 42:271–283.31. Kim SR, Matsui K, Yamada M, Gruz P, Nohmi T. 2001. Roles ofchromosomal and episomal dinB genes encoding DNA pol IV intargeted and untargeted mutagenesis in Escherichia coli. Mol GenetGenom 266:207–215.32. Shapiro JA, Brinkley PM. 1984. Programming of DNA rear-rangements involving Mu prophages. Cold Spring Harbor SympQuant Biol 49:313–320.33. Foster PL, Cairns J. 1994. The occurrence of heritable Muexcisions in starving cells of Escherichia coli. EMBO J 13:5240–5244.


Andersson et al.



IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

34. Maenhaut-Michel G, Blake CE, Leach DR, Shapiro JA. 1997.Different structures of selected and unselected araB-lacZ fusions. MolMicrobiol 23:1133–1145.35. Maenhaut-Michel G, Shapiro JA. 1994. The roles of starvationand selective substrates in emergence of araB-lacZ fusion clones.EMBO J 13:5229–5239.36. Garibyan L, Huang T, Kim M, Wolff E, Nguyen A, Nguyen T,Diep A, Hu K, Iverson A, Yang H, Miller JH. 2003. Use of the rpoBgene to determine the specificity of base substitution mutations on theEscherichia coli chromosome. DNA Repair (Amst.) 2:593–608.37. Taddei F, Matic I, Radman M. 1995. cAMP-dependent SOSinduction and mutagenesis in resting bacterial populations. Proc NatlAcad Sci USA 92:11736–11740.38. Taddei F, Halliday JA, Matic I, Radman M. 1997. Geneticanalysis of mutagenesis in aging Escherichia coli colonies. Mol GenGenet 256:277–281.39. Bjedov I, Tenaillon O, Gerard B, Souza V, Denamur E, RadmanM, Taddei F, Matic I. 2003. Stress-induced mutagenesis in bacteria.Science 300:1404–1409.40. Wrande M, Roth JR, Hughes D. 2008. Accumulation of mutantsin “aging” bacterial colonies is due to growth under selection, notstress-induced mutagenesis. Proc Natl Acad Sci USA 105:11863–11868.41. Sun S, Berg OG, Roth JR, Andersson DI. 2009. Contribution ofgene amplification to evolution of increased antibiotic resistance inSalmonella typhimurium. Genetics182:1183–1195.42. Fisher RF, Yanofsky C. 1983. Mutations of the beta subunit ofRNA polymerase alter both transcription pausing and transcriptiontermination in the trp operon leader region in vitro. J Biol Chem258:8146–8150.43. Jin DJ, Gross CA. 1988. Mapping and sequencing of mutations inthe Escherichia coli rpoB gene that lead to rifampicin resistance. J MolBiol 202:45–58.44. Jin DJ, Gross CA. 1991. RpoB8, a rifampicin-resistant termina-tion-proficient RNA polymerase, has an increased Km for purinenucleotides during transcription elongation. J Biol Chem 266:14478–14485.45. Landick R, Stewart J, Lee DN. 1990. Amino acid changes inconserved regions of the beta-subunit of Escherichia coli RNA poly-merase alter transcription pausing and termination. Genes Dev4:1623–1636.46. Sagitov V, Nikiforov V, Goldfarb A. 1993. Dominant lethalmutations near the 5′ substrate binding site affect RNA polymerasepropagation. J Biol Chem 268:2195–2202.47. Singer M, Jin DJ, Walter WA, Gross CA. 1993. Genetic evidencefor the interaction between cluster I and cluster III rifampicin resistantmutations. J Mol Biol 231:1–5.48. Yanofsky C, Horn V. 1981. Rifampin resistance mutations thatalter the efficiency of transcription termination at the tryptophanoperon attenuator. J Bacteriol 145:1334–1341.49. Hall BG. 1993. The role of single-mutant intermediates in thegeneration of trpAB double revertants during prolonged selection.J Bacteriol 175:6411–6414.50. Hall BG. 1995. Adaptive mutations in Escherichia coli as a modelfor the multiple mutational origins of tumors. Proc Natl Acad Sci USA92:5669–5673.51. Hall BG. 1997. On the specificity of adaptive mutations. Genetics145:39–44.52. Hall BG. 1998. Adaptive mutagenesis at ebgR is regulated byPhoPQ. J Bacteriol 180:2862–2865.

53. Hall BG. 1991. Adaptive evolution that requires multiple spon-taneous mutations: mutations involving base substitutions. Proc NatlAcad Sci USA 88:5882–5886.

54. Hall BG. 1999. Spectra of spontaneous growth-dependent andadaptive mutations at ebgR. J Bacteriol 181:1149–1155.

55. Hall BG. 1998. Activation of the bgl operon by adaptive mutation.Mol Biol Evol 15:1–5.

56. Loewe L, Textor V, Scherer S. 2003. High deleterious genomicmutation rate in stationary phase of Escherichia coli. Science 302:1558–1560.

57. de Visser JA, Rozen DE. 2004. Comment on “High deleteriousgenomic mutation rate in stationary phase of Escherichia coli.” Science304:518; author reply 518.

58. Finkel SE, Kolter R. 1999. Evolution of microbial diversity duringprolonged starvation. Proc Natl Acad Sci USA 96:4023–4027.

59. Zambrano MM, Kolter R. 1996. GASPing for life in stationaryphase. Cell 86:181–184.

60. Pedraza-Reyes M, Yasbin RE. 2004. Contribution of the mis-match DNA repair system to the generation of stationary-phase-induced mutants of Bacillus subtilis. J Bacteriol 186:6485–6491.

61. Sung HM, Yasbin RE. 2002. Adaptive, or stationary-phase, mu-tagenesis, a component of bacterial differentiation in Bacillus subtilis.J Bacteriol 184:5641–5653.

62. Sung HM, Yeamans G, Ross CA, Yasbin RE. 2003. Roles of YqjHand YqjW, homologs of the Escherichia coli UmuC/DinB or Y su-perfamily of DNA polymerases, in stationary-phase mutagenesis andUV-induced mutagenesis of Bacillus subtilis. J Bacteriol 185:2153–2160.

63. Ross C, Pybus C, Pedraza-Reyes M, Sung HM, Yasbin RE,Robleto E. 2006. Novel role of mfd: effects on stationary-phase mu-tagenesis in Bacillus subtilis. J Bacteriol 188:7512–7520.

64. Davis BD. 1989. Transcriptional bias: a non-Lamarckian mech-anism for substrate-induced mutations. Proc Natl Acad Sci USA 86:5005–5009.

65. Selby CP, Sancar A. 1993. Molecular mechanism of transcription-repair coupling. Science 260:53–58.66. Steele DF, Jinks-Robertson S. 1992. An examination of adaptivereversion in Saccharomyces cerevisiae. Genetics 132:9–21.67. Lea DE, Coulson CA. 1949. The distribution of numbers ofmutants in bacterial populations. J Genet 49:264–285.68. Heidenreich E, Wintersberger U. 2001. Adaptive reversions ofa frameshift mutation in arrested Saccharomyces cerevisiae cells bysimple deletions in mononucleotide repeats. Mutat Res 473:101–107.69. Yang Z, Lu Z, Wang A. 2001. Study of adaptive mutations inSalmonella typhimurium by using a super-repressing mutant of atrans regulatory gene purR. Mutat Res 484:95–102.70. Yang Z, Lu Z, Wang A. 2006. Adaptive mutations in Salmonellatyphimurium phenotypic of purR super-repression. Mutat Res 595:107–116.71. Paulander W, Maisnier-Patin S, Andersson DI. 2007. Multiplemechanisms to ameliorate the fitness burden of mupirocin resistancein Salmonella typhimurium. Mol Microbiol 64:1038–1048.72. Fojo T. 2007. Multiple paths to a drug resistance phenotype:mutations, translocations, deletions and amplification of coding genesor promoter regions, epigenetic changes and microRNAs. Drug ResistUpdat 10:59–67.73. Albertson DG. 2006. Gene amplification in cancer. Trends Genet22:447–455.





IP: 169.237.98.98

On: Mon, 30 May 2016 19:52:48

the origin of mutants under selection: interactions of...

Documents