Antonie van Leeuwenhoek73: 35–47, 1998. 35c 1998Kluwer Academic Publishers. Printed in the Netherlands.

Evolution of competitive fitness in experimental populations ofE. coli:What makes one genotype a better competitor than another?

Richard E. Lenski�, Judith A. Mongold, Paul D. Sniegowski1, Michael Travisano2,Farida Vasi, Philip J. Gerrish & Thomas M. SchmidtCenter for Microbial Ecology, Michigan State University, East Lansing, Michigan 48824, USA;1Present address:Department of Biology, University of Pennsylvania, Philadelphia, PA 19104, USA.2Present address: Departmentof Plant Sciences, University of Oxford, Oxford OX1 3RB, UK; (� author for correspondence)

Received 24 January 1997; accepted 1 October 1997

Key words:cell size, competition, evolution, fitness, nutrient specificity, selection

Abstract

An important problem in microbial ecology is to identify those phenotypic attributes that are responsible forcompetitive fitness in a particular environment. Thousands of papers have been published on the physiology,biochemistry, and molecular genetics ofEscherichia coliand other bacterial models. Nonetheless, little is knownabout what makes one genotype a better competitor than another even in such well studied systems. Here, wereview experiments to identify the phenotypic bases of improved competitive fitness in twelveE. coli populationsthat evolved for thousands of generations in a defined environment, in which glucose was the limiting substrate.After 10000 generations, the average fitness of the derived genotypes had increased by� 50% relative to theancestor, based on competition experiments using marked strains in the same environment. The growth kineticsof the ancestral and derived genotypes showed that the latter have a shorter lag phase upon transfer into freshmedium and a higher maximum growth rate. Competition experiments were also performed in environments whereother substrates were substituted for glucose. The derived genotypes are generally more fit in competition for thosesubstrates that use the same mechanism of transport as glucose, which suggests that enhanced transport was animportant target of natural selection in the evolutionary environment. All of the derived genotypes produce muchlarger cells than does the ancestor, even when both types are forced to grow at the same rate. Some, but not all, ofthe derived genotypes also have greatly elevated mutation rates. Efforts are now underway to identify the geneticchanges that underlie those phenotypic changes, especially substrate specificity and elevated mutation rate, forwhich there are good candidate loci. Identification and subsequent manipulation of these genes may provide newinsights into the reproducibility of adaptive evolution, the importance of co-adapted gene complexes, and the extentto which distinct phenotypes (e.g., substrate specificity and cell size) are affected by the same mutations.

Introduction

What phenotypic properties make one bacterial geno-type a superior competitor to another in a certain envi-ronment? Unfortunately, we cannot provide any gener-al answers to this question at the present time. Nonethe-less, we believe that this question is sufficiently impor-tant to the field of microbial ecology that it is worth-while to consider how one might go about getting ananswer, even if only in some special circumstances. Tothat end, we will summarize our progress to date in

understanding the phenotypic – and ultimately genet-ic – bases of enhanced competitive fitness in a mod-el experimental system. This review focuses on oneaspect of a long-term evolution experiment that is on-going in our laboratory. Other reviews on experimentalstudies of microbial evolution can be found in papersby Dykhuizen & Hartl (1983), Hall (1983), Levin &Lenski (1983), Mortlock (1984), Dykhuizen & Dean(1990), and Lenski (1992, 1995).

Before we describe our experimental system, let usconsider the general nature and scope of the problem.

36

Of course, it would be highly desirable to be able topredict the outcome of competition between bacterialgenotypes, especially in such managed ecosystems asbioreactors and agricultural fields. Yet who among uscan do so? Many thousands of person-years have beendevoted to investigations of the physiology, biochem-istry, and molecular genetics of one bacterial species,Escherichia coli(Neidhardt et al. 1996). But ask eventhe most knowledgeable student ofE. coli to predict,with any accuracy, which of two genotypes would pre-vail in competition in a particular environment. Asidefrom trivial cases – for example, that a Lac+ strainshould displace a Lac� mutant when lactose is the solesource of carbon and energy – most microbiologistswould readily admit that they could not do so.

Why not? Most microbiologists tend to ‘atom-ize’ an organism, studying one gene or pathway at atime, rather than viewing the organism as an integratedwhole. Microbiologists also usually pursue the tradi-tional genetic approach of determining structures andtheir functions by using defective mutants, whereasmany of the important differences between genotypesthat would influence the outcome of their competi-tion may involve subtle quantitative variation. Addi-tional complications arise because any two laboratory-derived genotypes may differ by mutations at severalloci, and two natural isolates of the same ‘species’may differ at the level of their nucleotide sequences byseveral percent (hence, tens of thousands of base-pairdifferences). All of these issues pertain to the complex-ity of the organisms, and many additional challengesarise from the complexity and unpredictability of theirenvironments. Perhaps, then, we should consider otherapproaches to studying the determinants of competi-tive fitness in bacteria – even if there are no guaranteesthat an alternative approach will succeed. In this paper,we examine one alternative approach, in which pop-ulations of bacteria are allowed to evolve in a simplelaboratory environment and the resulting phenotypicand genetic changes are then analyzed.

Overview of the experimental system

Twelve populations ofE. coli B were founded froma single ancestral strain, and these populations havenow been propagated for more than 10000 generationsin a relatively simple, defined environment (Lenski etal., 1991; Lenski & Travisano, 1994). The primarymotivation for this experiment was to address basicquestions about the dynamics of evolutionary adapta-

tion and divergence. Could one observe the process ofadaptation by natural selection, whereby the perfor-mance of the bacteria in the experimental environmentwould tend to improve over time? Would the rate oftheir improvement be essentially constant over time,or would the populations show episodes of faster andslower evolutionary change even in a constant environ-ment? Would all twelve replicate populations improvetheir performance to a similar extent, or would some ofthe populations find better solutions to the environmentthan would others?

During the course of this evolution experiment, webecame increasingly interested in the phenotypic andgenetic changes that made the derived bacteria bet-ter adapted to the experimental environment than weretheir ancestors. A mechanistic understanding of thesechanges would allow us to answer with greater preci-sion some of our basic questions about evolutionarydynamics. But we also saw a unique opportunity toaddress the question of what makes one genotype abetter competitor than another. On the one hand, thechanges that occur in the evolving populations are morenumerous and more subtle in their effects than the ‘one-at-a-time knockout’ mutations that are typically stud-ied by microbial geneticists. And on the other hand,these evolving populations and their environment arefar simpler than the natural systems that are typicallystudied by microbial ecologists.

The ancestral genotype in the evolution experimentwas prototrophic, but it also had several genetic mark-ers that allowed us to exclude the possibility of exter-nal contamination (Lenski et al., 1991). We also hadanother marker that was embedded in the experimen-tal design itself, which allowed us to exclude cross-contamination among the replicate populations. Thissame marker allowed us to put evolutionarily derivedgenotypes in competition with the ancestral strain ofthe opposite marker state and thereby directly measurethe changes in competitive fitness. Prior to experimentsto measure competitive fitness or any other property,both derived and ancestral genotypes were removedfrom storage at�80 �C and acclimated to the experi-mental environment. Thus, both ancestral and derivedgenotypes were in the same physiological state, andany significant differences between them must have agenetic basis.

The twelve evolving populations were seriallypropagated at 37�C in Davis minimal medium sup-plemented with 25�g of glucose per mL. Each day,0.1 mL of the previous day’s culture was transferredinto 9.9 mL of fresh medium. The 100-fold dilution

37

and regrowth allowed� 6.6 (= log2 100) generationsof binary fission each day, and the 10000generations ofexperimental evolution required 1500 daily transfers.Each day, the evolving cell populations experiencedlag, exponential growth, decelerating growth, and sta-tionary phases as the total size of each population fluc-tuated between� 5� 106 and� 5� 108 cells per 10mL culture.

The ancestral strain in our experiment did not carryany transmissible plasmids or viruses; thus, its evo-lution was strictly clonal (asexual). Mutations thatoccurred during the evolution experiment thereforeprovided all of the genetic variation that was availableto these populations. However, the mutational inputwas actually quite large. Given� 5� 108 cell replica-tions per day for 1500 days, and an estimated genomicrate of 0.0025 mutations per cell replication (Drake,1991), there must have been at least 109 mutationsin each of the evolving populations (Lenski & Trav-isano, 1994). By contrast, the genome size ofE. coliis only� 5 � 106 basepairs, and so there were morethan 100 mutations, on average, at each site in thegenome. These calculations may actually underesti-mate the number of mutations to the extent that certainloci are hypermutableand transposition events may notalways be counted. Of course, most of these mutationswould have been lost during each serial transfer andnot all sites are equally mutable. Even so, mutationclearly provided a tremendous amount of genetic vari-ation, which would allow the process of adaptation bynatural selection to proceed.

Experimental findings

Changes in competitive fitness

Figure 1 shows the trajectory for the grand mean fitness(averaged over the 12 replicate populations) during10000 generations of experimental evolution withE.coli. Fitness values were measured by placing derivedpopulations in competition with the ancestral strain(bearing the opposite genetic marker) in the same envi-ronment used for the experimental evolution. Fitnessis expressed relative to the common ancestor, and it iscalculated simply as the ratio of the growth rates real-ized by the two competitors during their head-to-headcompetition.

After 10000 generations, the derived genotypes are� 50% more fit than their ancestors, with about halfof this improvement occurring in the first 2000 gener-

Figure 1. Change in competitive fitness during 10000 generationsof experimental evolution withE. coli. Fitness is expressed relativeto the common ancestor. Each point is the grand mean averagedover twelve replicate populations. Error bars are 95% confidenceintervals. The dashed curve indicates the best fit of a hyperbolicmodel to the data. Data from Lenski & Travisano (1994).

ations. Based on more detailed analyses of the kineticsof mutation and selection, we estimate that only a hand-ful of mutations – about four, on average – can accountfor the improvementseen during these first 2000 gener-ations (Lenski et al., 1991; Lenski & Travisano, 1994;Elena et al., 1996; Gerrish & Lenski, 1998). Perhapsten or so mutations can account for all of the improve-ment in any population over the entire 10000 genera-tions. In addition to these beneficial mutations, othermutations may have drifted to high frequency in a pop-ulation by random sampling during the daily transfersor by hitchhiking along with a beneficial mutation thathappened to occur in the same cell. Using the theoreti-cal expectation that the steady-state substitution rate forselectively neutral alleles is equal to their mutation rate(Kimura 1983), then perhaps another twenty to thirtymutations may have been fixed in each population byrandom drift. This calculation is based on the totalgenomic mutation rate forE. coli (� 0.0025 mutationsper genome replication: Drake, 1991), and it furtherassumes that a substantial fraction of all mutations arein fact essentially neutral. In other words, despite thetremendous number of mutations that occurred in eachof these populations, only a relatively few are likely tohave increased to a high frequency.

How can we determine the mechanistic bases bywhich the derived bacteria have become better adapt-ed to their environment? We will briefly consider theadvantages and disadvantages of three approaches thatare widely used in molecular biology, genetics, and

38

evolutionary biology before describing the approachthat we have taken.

Some alternative approaches to characterizing thechanges responsible for improved competitive fitness

A seemingly straight-forward molecular approachwould be to obtain and compare DNA sequencesfrom the ancestral and derived genotypes. Sequencingentire genomes, however, is still a formidable prob-lem. Unless the specific genes in which the changeshave occurred can somehow be identified, one mustuse methods, such as RFLP analysis, that can detectchanges in a sample of the basepairs around thegenome. Although these methods are now relativelysimple and inexpensive, the fraction of the genomethat can be sampled is still quite small. For example,a restriction enzyme that recognizes a particular 8-bpsequence can be expected to detect, on average, onlyabout one out of every 10,000 point mutations in theE. coli genome. (One would expect to detect a high-er fraction of insertion and deletion mutations by thisapproach.) But we estimated that there are probably nomore than about forty genetic differences between theancestral and derived genotypes, and so a large bat-tery of restriction enzymes would be needed to detectany changes that were due to point mutations. More-over, any genetic changes that were detected usingthis approach might have occurred by random drift orby hitchhiking with a beneficial mutation elsewherein the genome. Thus, having found a difference inDNA sequence between derived and ancestral geno-types, there is no guarantee that it is causally relatedto the difference in their relative fitness. Thus, while10000 generations is a very long time for an experi-ment, it is actually a very short time from the standpointof molecular evolution, and molecular changes do notnecessarily imply important phenotypic changes.

Another approach would be to use the classicalbacterial genetic methods of conjugation, transduction,and transformation to map the genes that enhance fit-ness. However, there are severe technical difficultieswith implementing this approach. Chief among themis the fact that these classical approaches work wellwhen there are qualitative differences in phenotype,such that a genotype either can or cannot grow on a par-ticular medium. In that case, one can cleanly counter-select against both parental strains and thereby find therecombinant ‘needle in a haystack.’ But the phenotypicdifferences that interest us are subtle changes in com-petitive fitness. One might circumvent this problem by

using parental strains that differed at a large number ofmarker loci that could be readily counter-selected, andthen perform a linkage analysis to find markers asso-ciated with fitness effects (in essence QTL mappingapplied to bacteria). However, fitness as a trait maybe influenced by so many genes that it seems likelythat the marker loci themselves would have individu-al or epistatic effects on fitness, thereby confoundingthe analysis and its interpretation. Thus, the classicalapproach to mapping bacterial genes is well suited tothe ‘plus or minus’ phenotypesused in studies that seekto identify structure and function by knocking out anunderlying gene; but this approach is not so useful formapping genetic changes that underlie the more subtlephenotypic refinements that are often important duringevolution.

Evolutionary biologists who study the effects ofnatural selection in contemporary populations of ani-mals and plants often employ a very different approach.In essence, this approach begins with the observationthat certain phenotypic traits appear to be especiallyimportant for the ecological performance of a popula-tion or species in the wild. For example, observationson one of Darwin’s finches,Geospiza fortis, suggestedthat beak size and shape might affect the birds’ successin foraging on several different kinds of seeds, whichbecame more or less abundant depending on patternsof rainfall (Price et al., 1984). To build a rigorous casethat natural selection is indeed changing these traits inan adaptive manner, one then faces a two-fold chal-lenge. First, one must show that variation in the traitsis at least partly heritable, so that offspring tend toresemble their parents owing to genetic similarity (asopposed to shared environmental influences). Second,one must show that differences in the traits are sig-nificantly associated with differences in the survivaland reproductive success of individual organisms. Inessence, it is necessary to fulfill these conditions inorder to infer heritable differences in fitness. Not sur-prisingly, such analyses are very difficult in most sys-tems given the lack of experimental control over bothgenetic factors and environmental variables.

By contrast, it has been relatively simple to demon-strate heritable differences in fitness between ancestraland derived genotypes in our experimental populationsof E. coli. In the following sections, we present theresults to date of a multifaceted ‘top-down’ analy-sis that seeks, first, to identify heritable phenotypicdifferences between the ancestral and derived geno-types, and then to explain the mechanistic basis forthe enhanced competitive fitness of the derived bacte-

39

Figure 2. (A) The derived genotypes have a significantly shorter lag phase, upon transfer to fresh medium, than does the ancestral strain. (B)The derived genotypes have a significantly faster maximum growth rate at high glucose concentration than does the ancestral strain. Error barsare 95% confidence intervals. Data from Vasi et al. (1994).

ria. As in all evolutionary studies, merely showing acorrelation between some phenotypic trait and fitnesswill not be sufficient to prove that changes in that traitwere responsible for the fitness gains in the derivedgenotypes. However, as putative phenotypic targets ofselection are identified, we hope eventually to incor-porate classical and molecular approaches to identifyand manipulate the genetic determinants of enhancedcompetitive fitness.

Changes in demographic parameters

Our first level of analysis sought to determine at whatstages in the population growth cycle the derived bac-teria gained an advantage relative to their progeni-tors (Vasi et al., 1994). Every day during the long-term evolution experiment, the bacteria were dilutedinto fresh medium, where they exhausted the avail-able glucose before the next transfer. This cycle of‘feast and famine’ was repeated during the experimentsused to quantify competitive fitness. In principle, thederived bacteria could have become better adapted tothis regime by shortening the duration of their lag phaseprior to growth, increasing their maximum growth rate,improving their growth rate at low glucose concentra-tions (i.e., reducing Ks), or by reducing their death rateafter the glucose was exhausted.

We estimated all of these growth parameters forthe common ancestor and for derived genotypes sam-pled from each of the populations after 2000 genera-tions. As shown in Figure 2, the derived genotypes hadsignificantly shorter lag phases and significantly high-er maximum growth rates (Umax) than their ancestor.However, the derived genotypes showed no significantimprovement relative to their ancestor in either the

concentration of glucose required to support growthat half the maximum rate (Ks) or in their death ratesduring stationary phase. In fact, the derived genotypeshad significantly higher values for Ks than did theirancestor, although the ratio of Umax/Ks (which deter-mines the rate of growth as the glucose concentrationapproaches zero) had not changed significantly. Evi-dently, the derived bacteria had become better adaptedto exploiting the periods of ‘feast’ that occurred everyday, but they were not any better (or worse) than theirancestor in terms of tolerating the periods of ‘famine.’The failure of the derived bacteria to become betteradapted to these periods of famine probably reflectsthe fact that the ancestral genotype was already so welladapted to these conditions. That is, the ancestral strainexperienced little or no death during the 14–16 hoursafter glucose was exhausted (prior to the next transferinto fresh medium), and it could grow at greater thanhalf its maximum rate even after more than 99% of thedaily supply of glucose was depleted.

Although the derived genotypes increased in abun-dancerelativeto the ancestral genotype during compe-tition experiments, we also observed that the numericalyield (cfu per mL) of the derived genotypes was lowerthan that of the ancestor when they were grown in iso-lation. In principle, the reduced numerical yield of thederived bacteria could reflect several different changes:(1) the derived bacteria might have lower plating effi-ciency than the ancestor; (2) the derived bacteria mightgrow faster, but less efficiently, on glucose than theancestor; and (3) the derived bacteria might have cellsthat are larger in size but fewer in number than theancestor.

40

Figure 3. Change in average cell size during 10000 generations of experimental evolution withE. coli. Cell size is expressed volumetrically,where 1 fL = 10�15 L. Each point is the grand mean averaged over twelve replicate populations. Error bars are 95% confidence intervals. Thedashed curve indicates the best fit of a hyperbolic model to the data. Data from Lenski & Travisano (1994).

Changes in cell size

We used a Coulter particle counter to measure the vol-ume displaced by individual cells of the derived andancestral genotypes (Vasi et al., 1994; Lenski & Trav-isano, 1994). Derived genotypes from all twelve evolv-ing populations produced significantly larger cells thandid their ancestor. Figure 3 shows the evolutionary tra-jectory for the grand mean cell volume. After 10000generations, the derived genotypes produced cells thatwere, on average, approximately twice as large as thoseproduced by their ancestor. Thus, the evolved popula-tions have cells that are larger in size, but fewer innumber, than the ancestral population. However, theincrease in average cell size and the reduction in cellnumber do not exactly balance one another. Rather,there was a net increase in the biovolume yield of thederived populations (Vasi et al., 1994), indicating thatthe derived genotypes have become more efficient thantheir ancestors during growth on glucose.

One might have expected thatsmallercells wouldhave a selective advantage by virtue of their largerratio of surface area to cell volume, which would facil-itate nutrient transport. Of course, this ratio dependson the shape as well as the size of cells. Microscopicexamination shows that the derived genotypes signif-icantly increased both the length and the diameter oftheir cells relative to the ancestor (J.A.M. & R.E.L.,unpubl.). But it is noteworthy that the derived cellstypically increased their length to a somewhat greaterextent than they increased their width. Thus, while theratio of surface area to volume declined substantiallyrelative to the ancestor, it did so to a lesser degree than

if the cells had increased in size isometrically (i.e.,without changing shape).

It is unclear what advantage might accrue to largercells in the experimental environment. Perhaps larg-er cells have greater metabolic reserves, so that theycan respond more quickly to the sudden increase innutrient availability that occurs at each daily transfer.This explanation is consistent with the reduced lengthof the lag phase in the derived genotypes. Howev-er, there is another explanation that does not supposethat increased cell sizeper sewas the actual target ofnatural selection. Bacterial physiologists have shownthat there is a positive correlation between cell sizeand growth rate when bacteria are grown on differ-ent media (Schaechter et al., 1958; Bremer & Dennis,1987; Akerlund et al., 1995). This effect is strictlyphenotypic and does not involve any genetic change.By extension, it is possible that the larger cell size ofour derived genotypes reflects the simple fact that theygrow faster than does their ancestor. In other words, thelarger cell size of the derived genotypes might simplybe a correlated response to selection for faster growth,rather than reflecting any direct advantage to largercells (Figure 4A).

In fact, this hypothesized explanation for the evo-lution of larger cell size can be readily tested by forc-ing the derived and ancestral genotypes to grow at thesame rate by culturing them in separate chemostat ves-sels (Mongold & Lenski, 1996). When we did so, weobserved the previously reported relationship betweencell size and growth rate for both derived and ances-tral genotypes; that is, faster growing cells were alsophenotypically larger. However, the derived genotype

41

Figure 4. (A) Hypothetical non-adaptive mechanism for an evolutionary change in cell size due to a phenotypic correlation with growth rate.Filled circles: cell size of the ancestral genotype is phenotypically correlated with its growth rate. Open circle: a mutation that allowed a derivedgenotype to grow faster would cause a correlated increase in cell size, even though the mutation was in a gene that did not directly control cellsize. (B) This hypothetical non-adaptive mechanism is rejected by experiments in which ancestral and derived genotypes are forced to grow at thesame rate in independent chemostat vessels. Filled circles: relationship between mean cell size and growth rate in the ancestral genotype. Opencircles: relationship between mean cell size and growth rate in a derived genotype obtained after 2000 generations of experimental evolution.The derived genotype produces larger cells than the ancestor even when they are grown at the same rate. Panel B reprinted with permissionfrom Mongold & Lenski (1996).

was significantly larger than its ancestor when the twostrains were growing at the same rate, except at verylow growth rates (Figure 4B). Thus, the phenotypiccorrelation between cell size and growth rate cannotfully account for the evolutionary change in cell sizethat we observed, and some other explanation must besought.

Changes in cell composition

We have also begun experiments to examine whetherthe macromolecular composition of the evolving cellshas changed and, if so, how this scales with cell size(J.A.M., unpubl.). The derived genotype has more pro-tein, RNA, and DNA per cell than does its ancestor,which is not surprising given that the derived cells aregenerally larger. When the amounts of protein, RNA,and DNA are expressed per unit of cell volume, the con-centrations of protein and RNA are indistinguishablebetween the derived and ancestral genotypes, whereasthe concentration of DNA is actually somewhat lowerin the derived genotype. Thus, the ratio of protein toRNA has remained constant, while the ratio of RNAto DNA appears to have increased in the derived geno-type. Of course, most of the RNA in a cell is ribosomal.

Comparative data indicate the copy number of ribo-somal (rrn) operons is evolutionarily labile; whileE.coli typically has sevenrrn operons, other bacterialspecies have between 1 and 14rrn operons (Schmidt,1996). Given the increased ratio of RNA to DNA inthe derived genotype, we wanted to see whether the

rrn copy number might have increased in the evolvingpopulations (T.M.S., unpubl.). To that end, we isolatedgenomic DNA from the reciprocally marked ancestralgenotypes and from derived genotypes sampled fromall twelve populations after 10000 generations. ThisDNA was cut with restriction enzymePvuII and thenprobed with a DIG-labeled transcript from positions874 to 1382 of the 16S RNA-encoding gene of theE.coli rrnB operon. This enzyme cuts a unique site ineachrrn operon, and that site is located such that evena tandem duplication of anrrn operon should yield anadditional fragment that could be detected using thatprobe. As shown in Figure 5, all twelve independentlyderived genotypes retain the seven ancestralrrn oper-ons, without any additional copies being present. Thus,we can reject an increase in the number of ribosomaloperons as the explanation for the increased ratio ofRNA to DNA in the derived genotype.

It will also be interesting to obtain information onother aspects of cellular composition, including cellenvelope proteins as well as total carbohydrate andlipid content. An increase in the stationary-phase levelsof storage molecules would corroborate the hypothesisthat larger cells have greater metabolic reserves, suchthat they can respond more quickly to the daily influxof nutrients. Also, the stoichiometry of the envelopemight be such that the amounts of key proteins involvedin glucose transport (see below) cannot be raised with-out simultaneously increasing the surface area of thecell. If so, this could provide an alternative explanationfor the evolutionary trend to larger cell size.

42

Figure 5. Ancestral and evolutionarily derived genotypes ofE. colihave the same number ofrrn operons.PvuII-restricted genomicDNA was hybridized with a labeled fragment ofE. coli 16S RNA.The resulting Southern blot indicates the constancy ofrrn copynumber as well as the position ofPvuII restriction sites adjacentto the rrn operons. Shown from left to right are: genotypes fromsix of the evolving populations after 10000 generations; the tworeciprocally marked ancestral genotypes; and genotypes from theother six evolving populations after 10000 generations. Unpublisheddata of T.M.S.

In summary, the evolving populations underwentdramatic changes in average cell size and more sub-tle changes in their macromolecular composition.We have devised and performed further experimentsthat allow us to reject certain explanations for thesechanges. However, we have not provided independentevidence either to confirm or refute other plausibleexplanations for these changes.

Changes in competitive fitness on substrates otherthan glucose

The estimates of competitive fitness shown in Figure 1were all obtained in the same glucose-supplemented

minimal medium in which the experimental popula-tions had evolved. To gain further insight into thephenotypic bases of the derived genotypes enhancedcompetitiveness, we also examined the fitness of thederived genotypes relative to their ancestor in mediain which various other substrates replaced glucose asthe sole source of carbon and energy (Travisano et al.,1995; Travisano & Lenski, 1996).

Maltose is a disaccharide of glucose, and the degra-dation of maltose for energy requires a set of enzymesthat are almost identical to those required for the degra-dation of glucose. ButE. coli B uses completely dif-ferent sets of proteins for the uptake of glucose andmaltose, as shown schematically in Figure 6. Glucosediffuses through the outer membrane via OmpF porins,and it is then actively transported across the innermembrane via the phosphotransferase system (PTS).By contrast, maltose does not readily diffuse throughOmpF porins, but it induces expression of LamB porinsthat allow maltose (and other maltodextrins) to diffuseinto the cell. Maltose also uses an inducible permease,which is not part of the PTS, for its transport across theinner membrane. If the derived bacteria have adaptedto the experimental environmentprimarily by improve-ments in glucose transport, then one would not expecttheir higher fitness to be manifest when maltose is sub-stituted for glucose. By contrast, if the derived bacteriahave adapted by improving almost any other aspect oftheir physiology (catabolism, biosynthesis, etc.), thenwe would expect to find a similar competitive advan-tage in maltose as in glucose.

Figure 7 summarizes the results of competitionexperiments between derived and ancestral genotypesin glucose and maltose. All twelve derived genotypesare more fit in glucose than in maltose. The derivedgenotypes are also much more variable in their compet-itive fitness in maltose than in glucose. These outcomessupport the hypothesis that improvements in glucosetransport contribute substantially to their greater fit-ness.

To further test this hypothesis, we performed com-petition experiments using a total of twelve differentgrowth substrates. Several of these substrates sharewith glucose the use of OmpF and the PTS for transportacross the outer and inner membranes, respectively,whereas the other substrates differ at one or both steps.To compare the responses to all twelve substrates, wecalculated an index of parallel adaptive evolution foreach one. The numerator of the index is the propor-tion of the derived genotypes which are more fit thantheir ancestor in a given substrate, and the denomina-

43

Figure 6. Schematic illustration of the gene products involved in the uptake of glucose and maltose byE. coli. Maltose is a glucose disaccharide;glucose and maltose are transported into the cell by completely different mechanisms, but their subsequent catabolism is virtually identical.Reprinted with permission from Travisano et al. (1994).

tor is the genetic variance in fitness among the derivedlines. We expect a higher value of this index for thosesubstrates that are most similar to glucose in terms ofthe mechanisms that underlie adaptation of the derivedgenotypes to the experimental environment. As shownin Figure 8, the derived genotypes show the highestdegree of parallel adaptive evolution for glucose, fol-lowed by the five other substrates that use both OmpFand the PTS for their uptake by the cell; those sub-strates that differ at one or both steps in transport givethe lowest values. These results provide additional sup-port for the hypothesis that improvements in glucosetransport are largely responsible for the gains in com-petitive fitness that occurred during the experimentalevolution.

Changes in mutation rates

Spontaneous mutations (including point mutations,transpositions, etc.) provide the genetic variation thatis required for evolutionary change in our experimen-tal populations. Therefore, we sought to determinewhether the rates of mutation themselves might haveundergone evolutionary change (Sniegowski et al.,1997). To that end, we performed a series of fluctua-tion tests (Luria & Delbr̈uck, 1943) to estimate the ratesat which the ancestral and derived genotypes mutate ateach of three different loci (at which mutations producedistinct phenotypes that can be readily scored). Theseexperiments indicate a mutation rate of about 10�10

per basepair per generation for the ancestral strain,which is similar to values that have been reported for‘wild-type’ E. coli (Drake, 1991). We obtained simi-lar estimates for nine of the derived genotypes isolatedafter 10000 generations of evolution. For three of the

44

Figure 7. Competitive fitness of derived genotypes from twelvereplicate populations ofE. coli, measured in glucose (filled symbols)and maltose (open symbols), after 2000 generations of experimentalevolution in a medium containing only glucose. Fitness is expressedrelative to the common ancestor. All twelve lines are more fit inglucose than in maltose. Note also that the derived lines are muchmore variable in their competitive abilities in maltose than in glucose.Data from Travisano et al. (1995).

derived genotypes, however, mutation rates were esti-mated to be about two orders of magnitude higher atall of the loci examined. We used samples that hadbeen periodically taken and stored from each popula-tion to determine when the mutator phenotype aroseand whether it was stably maintained. All three pop-ulations with elevated mutation rates had acquired themutator phenotype by 8000 generations, and each oneretained that phenotype for thousands of generations.Evidently, there were dramatic and stable increases inthe genomic mutation rate in some, but not all, of theexperimental populations.

We have begun to characterize the mutations thatare responsible for producing these mutator pheno-types. A number of loci are known that can increasethe genomic mutation rate when they are mutated (Cox,1976; Horiuchi et al., 1989). Genetic complementa-tion tests indicate that increased mutation rates in thethree derived genotypes have resulted from defects inthe methyl-directed mismatch repair system. But evenknowing the genetic basis of the increased mutationrate, one does not know the population genetic processby which the mutator genotypes became establishedin the evolving populations. On the one hand, a high-er mutation rate would be expected to accelerate theappearance of rare beneficial mutations,perhaps givingsome long-term evolutionary advantage to genotypeswith higher mutation rates. On the other hand, manymore mutations are harmful than are beneficial, and so

Figure 8. Twelve nutrients are ranked on the basis of an index ofparallel adaptive evolution. The numerator of the index is the pro-portion of the derived genotypes that are more fit than the ancestralstrain during competition for that nutrient. The denominator of theindex is a measure of the variability in competitive fitness amongthe derived genotypes. Vertical hatching denotes a nutrient that usesOmpF as its primary mode of transport across the outer membrane,and horizontal hatching denotes a nutrient that uses the phospho-transferase system for transport across the inner membrane. Thederived genotypes showed the most parallel adaptive evolution inglucose, followed by the five other nutrients that share with glucosetheir mechanisms of transport across both membranes. Reprintedwith permission from Travisano et al. (1996).

the short-term effect of having a higher mutation rateshould be to reduce a genotype’s mean fitness. Thus,it would seem to be difficult for mutator genotypes tobecome established, even if they might provide somelong-term advantage (Chao & Cox, 1983).

We can suggest two possible explanations for howmutator genotypes became established in some of theexperimental populations, despite their presumablyincreased load of deleterious mutations. One possi-bility is that the mutator phenotype is a side-effectof a mutation whose primary effect confers a selec-tive advantage. For example, selection for more rapidgrowth might have favored the loss of DNA repairfunctions, because their expression may slow the rateof DNA replication. The other possibility is that, byrecurrent mutations, the mutator genotypes reacheda sufficiently high frequency (despite their harmfulshort-term effect) that eventually a secondary bene-ficial mutation occurred in a mutator background. Themutator gene could then have hitchhiked to high fre-quency in association with the beneficial secondarymutation (Taddei et al., 1997). Importantly, neither ofthese hypotheses requires selection for faster evolutionper se, although both could have that consequence.

45

It should be possible to distinguish between thesetwo hypotheses by moving the mutator and correspond-ing ‘wild-type’ alleles to isogenic backgrounds, andthen measuring the short-term effect of the mutatorallele on competitive fitness. If the mutator allele pre-vails during short-term competition experiments, thenthat would support the hypothesis that the mutator phe-notype is a side-effect of a mutation whose primaryeffect is beneficial in the experimental environment.Alternatively, if the mutator allele is at a competitivedisadvantage when placed in isogenic backgroundsin the experimental environment, then that outcomewould support the hitchhiking hypothesis.

Conclusions and future directions

Microbial ecologists have long been interested inunderstanding what attributes make one bacterial geno-type or species more competitive than another in aparticular environment. Yet it remains unclear howeven to address this problem in any general way,despite decades of study devoted to the physiology,bio-chemistry, and molecular genetics of certain bacterialspecies, such asE. coli. There are many reasons for thedifficulty in making substantial progress on this front.Whereas other fields of microbiology have successful-ly employed a reductionist approach to characterizeone gene or pathway at a time, the competitive fitnessof an organism may depend on the integration of allits component genes and pathways. Also, natural envi-ronments are notoriously variable and unpredictable,making it difficult even to quantify reproducibly therelative fitness of two bacterial competitors, which isessential before one can properly begin to ask why onetype prevails over another.

To circumvent these and other problems, we soughtto identify what makes one bacterial genotype a bettercompetitor than another by employing a highly sim-plified experimental system. Twelve populations ofE.coli were founded from a common ancestor and thenpropagated for 10000 generations in a defined labo-ratory environment consisting of serial transfer in aglucose-supplemented minimal medium. During thistime, the twelve populations adapted genetically, sothat at the end of this period the derived genotypeswere demonstrably much more fit in this experimentalenvironment than was the ancestral strain. The derivedgenotypes grow� 50% faster than does the ancestorwhen they are placed in direct competition.

We have subsequently characterized the changesthat occurred during this experimental evolution in var-

ious traits that might affect their competitive fitness.We showed, for example, that the derived bacteria hadachieved a higher maximum growth rate in the experi-mental environment, and they also experienceda short-er lag phase prior to exponential growth upon transferinto fresh medium. Unexpectedly, we also found thatthe derived bacteria produce cells that are much larg-er, on average, than those produced by the ancestralstrain. The difference in cell size persists even whenthe derived bacterial genotypes are forced to grow atthe same rate as the ancestral strain, so that it cannotbe explained by the well-known phenotypiccorrelationbetween growth rate and cell size. The derived bacteriamay also have a somewhat higher ratio of total cellularRNA to DNA than does the ancestor; we rejected onepossible explanation for this change, which was thatthe derived bacteria had acquired an additional copy ofanrrn operon.

We also asked whether the advantage enjoyed bythe derived bacteria persisted when the competitionexperiments were performed in eleven other environ-ments in which glucose was replaced by different sub-strates. In general, the derived genotypes had high-er and more uniform competitive abilities for sub-strates that share with glucose the mechanisms of theirtransport into the cell; whereas the derived genotypeswere less fit and more heterogeneous when competingfor substrates that use other mechanisms of transport.These findings suggest strongly that changes in glu-cose transport involving the porins, the phosphotrans-ferase system, or both, contributed to the improvedcompetitive fitness of the derived bacterial genotypes.[See also Dykhuizen & Dean (1990) for a reviewof some other experiments which provide compellingevidence that transport functions often limit bacterialgrowth.] Another unexpected change in three of thetwelve derived populations is that the mutation ratehas increased by some two orders of magnitude. Theincreased mutation rate is a genome-wide effect, andin each of the three cases it is caused by defects in themethyl-directedmismatch repair system. We do not yetknow why these mutator genes increased in frequency.One possible explanation is that the bacteria were ableto grow faster by disabling this repair process; anotherpossibility is that the mutator gene ‘hitchhiked’ alongwith some beneficial mutation at another locus. In anycase, these changes in mutation rate may affect (forbetter or for worse) not only the competitiveness of thederived bacteria but also the further evolution of theircompetitiveness.

46

Evidently, we have had some success in identifyinga number of phenotypic traits that have changed as thebacteria have adapted to, and become better competi-tors in, the environment in which they were propagatedfor 10000 generations. Some of the changes seem tocontribute quite directly to the higher competitive fit-ness of the derived bacteria, for example, enhancedglucose transport and a shorter lag phase upon transferinto fresh medium. But in other cases, it is unclear howthe phenotypic changes contribute to the competitivesuccess of the derived bacteria. Do large cells have anintrinsic competitive advantage over smaller cells and,if so, why? Or is a larger cell size simply a correlatedconsequence of some physiological change that, onceidentified, will be readily understood as contributingto improved fitness? For example, improvements inglucose transport might cause an increase in averagecell size merely as a consequence of changing the rel-ative rates of resource uptake and genome replication.Similarly, have mutation rates increased in some ofthe evolving populations because disabling a systemof DNA repair allows faster growth, or for some otherreason?

Clearly, our findings have raised at least as manyquestions as they have answered. We are optimisticthat we can, in fact, address these questions. Wenow have two sets of candidate loci that can be com-pared between the derived and ancestral genotypes,those involved in glucose transport and those encod-ing the methyl-directed mismatch repair system. Oncewe have identified precisely the genetic changes thatoccurred during the experimental evolution, we canthen perform genetic manipulations to construct iso-genic strains that will enable us answer the questionsposed above. For example, we might find that thesame mutations that enhance glucose transport alsoincrease cell size as a side-effect, thus obviating theneed for some separate adaptive explanation for theevolution of larger cells. Similarly, we might find thatthe mutations causing defects in methyl-directed mis-match repair allow more rapid growth, thereby pro-viding them with a direct selective advantage in theexperimental environment.

Thus, the ‘top-down’ approach that we have usedto characterize changes in phenotype that took placeduring the experimental evolution has generated newhypotheses that can be tested using a ‘bottom-up’approach in which specific genes are manipulated.Ultimately, a combination of these two approachesmay reveal whether the competitive fitness of bacte-ria can be understood by examining one trait (or one

gene) at a time, or whether it is essential to study bacte-ria in a more integrated fashion in order to understandtheir ecological capabilities.

Acknowledgments

We thank A. Bennett, D. Dykhuizen, L. Forney, and M.Rose for stimulating discussions of the issues raised inthis paper; M. Blot, M. Stanek, and M. Thomashow fornew collaborations to identify the molecular changesthat underlie the phenotypic evolution reviewed here;and a reviewer for helpful comments on our paper. Ourresearch is supported by the NSF Science and Tech-nology Center for Microbial Ecology (BIR-9120006)and an NSF grant (DEB-9421237) to R.E.L.

References

Akerlund T, Nordstr̈om K & Bernander R (1995) Analysis of cellsize and DNA content in exponentially growing and stationary-phase batch cultures ofEscherichia coli. Journal of Bacteriology177: 6791–6797

Bremer H & Dennis PP (1987) Modulation of chemical compositionand other parameters of the cell by growth rate. In: Neidhardt FC,Ingraham JL, Low KB, Magasanik B, Schaechter M & UmbargerHE (Eds)Escherichia coliandSalmonella typhimurium: Cellularand Molecular Biology, Vol 2 (pp 1527–1542). American Societyfor Microbiology, Washington, DC

Chao L & Cox EC (1983) Competition between high and low mutat-ing strains ofEscherichia coli. Evolution 37: 125–134

Cox EC (1976) Bacterial mutator genes and the control of sponta-neous mutation. Annual Review of Genetics 10: 135–156

Drake JW (1991) A constant rate of spontaneous mutation in DNA-based microbes. Proceedings of the National Academy of Sci-ences USA 88: 7160–7164

Dykhuizen DE & Dean AM (1990) Enzyme activity and fitness:evolution in solution. Trends in Ecology & Evolution 5: 257–262

Dykhuizen DE & Hartl DL (1983) Selection in chemostats. Micro-biological Reviews 47: 150–168

Elena SF, Cooper VS & Lenski RE (1996) Punctuated evolutioncaused by selection of rare beneficial mutations. Science 272:1802–1804

Gerrish PJ & Lenski RE (1998) The fate of competing beneficialmutations in an asexual population. Genetica: in press

Hall BG (1983) Evolution of new metabolic functions in laboratoryorganisms. In: Nei M & Koehn RK (Eds) Evolution of Genes andProteins (pp 234–257). Sinauer Associates, Sunderland, Massa-chusetts, USA

Horiuchi T, Maki H & Sekiguchi M (1989) Mutators and fidelity ofDNA replication. Bulletin Institut Pasteur 87: 309–336

Kimura M (1983) The Neutral Theory of Molecular Evolution. Cam-bridge University Press, Cambridge

Lenski RE (1992) Experimental evolution. In Lederberg J (Ed) Ency-clopedia of Microbiology, Vol 2 (pp 125–140). Academic Press,San Diego

Lenski RE (1995) Molecules are more than markers: new directionsin molecular microbial ecology. Molecular Ecology 4: 643–651

47

Lenski RE & Travisano M (1994) Dynamics of adaptation and diver-sification: A 10,000-generation experiment with bacterial popu-lations. Proceedings of the National Academy of Sciences, USA91: 6808–6814

Lenski RE, Rose MR, Simpson SC & Tadler SC (1991) Long-termexperimental evolution inEscherichia coli. I. Adaptation anddivergence during 2,000 generations. The American Naturalist138: 1315–1341

Levin BR & Lenski RE (1983) Coevolution in bacteria and theirviruses and plasmids. In: Futuyma DJ & Slatkin M (Eds) Coevo-lution (pp 99–127). Sinauer Associates, Sunderland, Massa-chusetts, USA

Luria SE & Delbr̈uck M (1943) Mutations of bacteria from virussensitivity to virus resistance. Genetics 28: 491–511

Mongold JA & Lenski RE (1996) Experimental rejection of a non-adaptive explanation for increased cell size inEscherichia coli.Journal of Bacteriology 178: 5333–5334

Mortlock RP, Ed (1984) Microorganisms as Model Systems forStudying Evolution. Plenum Press, New York

Neidhardt FC, Curtiss R, Ingraham JL, Lin ECC, Low KB, Mag-asanik B, Reznikoff WS, Riley M, Schaechter M & UmbargerHE, Eds (1996)Escherichia coliandSalmonella typhimurium:Cellular and Molecular Biology, 2nd Edition. American Societyfor Microbiology, Washington, DC

Price TD, Grant PR, Gibbs HL & Boag PT (1984) Recurrent patternsof natural selection in a population of Darwin’s finches. Nature309: 787–789

Schaechter M, Maaløe O & Kjelgaard NO (1958) Dependence onmedium and temperature of cell size and chemical compositionduring balanced growth ofSalmonella typhimurium. Journal ofGeneral Microbiology 19: 592–606

Schmidt TM (1996) Multiplicity of ribosomal operons in prokaryoticgenomes. In: de Bruijn FJ, Lupski JR & Weinstock G (Eds)Bacterial Genomes: Physical Structure and Analysis. Chapmanand Hall, New York

Sniegowski PD, Gerrish PJ & Lenski RE (1997) Evolution of highmutation rates in experimental populations ofEscherichia coli.Nature 387: 703–705

Taddei F, Radman M, Maynard Smith J, Toupance B, Gouyon PH &Godelle B (1997) Role of mutator alleles in adaptive evolution.Nature 387: 700–702

Travisano M & Lenski RE (1996) Long-term experimental evolutionin Escherichia coli. IV. Targets of selection and the specificity ofadaptation. Genetics 143: 15–26

Travisano M, Vasi F & Lenski RE (1995) Long-term experimentalevolution inEscherichia coli. III. Variation among replicate pop-ulations in correlated responses to novel environments. Evolution49: 189–200

Vasi F, Travisano M & Lenski RE (1994) Long-term experimentalevolution in Escherichia coli. II. Changes in life-history traitsduring adaptation to a seasonal environment. The American Nat-uralist 144: 432–456