Mechanisms and selection of evolvability: experimental evidence


Correspondence: Tim F. Cooper, Department of Biology and Biochemistry, University of Houston, Houston, TX 77204, USA. Tel.: +01 713 743 2552; fax: +01 713 743 2552; e-mail:


The vast number of species we see around us today, all stemming from a common ancestor, clearly demonstrates the capacity of organisms to adapt to new environments. Understanding the underlying basis of differences in the capacity of genotypes to adapt – that is, their evolvability – has become a major research field. Several mechanisms have been demonstrated to influence evolvability, including differences in mutation rate, mutational robustness, and some kinds of gene interactions. However, the benefits of increased evolvability are indirect, so that the conditions required for selection of evolvability traits are expected to be more limited than for traits that confer immediately beneficial phenotypes. Moreover, just because a trait can affect evolvability does not mean that it actually does so in a particular environment. Instead, some other function of the trait may better explain its success. Nevertheless, there is accumulating experimental evidence that some traits can increase the evolvability of a genotype and that these traits do influence evolutionary outcomes. We discuss recent theory and experiments that demonstrate the potential for traits that influence evolvability to arise and be selected.


The capacity of genotypes to adapt to an environment has been given the term ‘evolvability’. In studying this trait, different disciplines have tended to emphasize different definitions. Researchers focusing on macroorganisms have tended to emphasize the influence of existing genetic variation on relatively short-term responses to selection (Houle, 1992; Pigliucci, 2008). Populations with higher genetic variation are more evolvable because they have a greater capacity to respond to environmental changes (Fisher, 1930). Researchers who are interested in following differences in evolutionary potential over longer evolutionary time scales usually define evolvability in a different way: as the capacity of genotypes to generate de novo adaptive genetic variation (Wagner & Altenberg, 1996).

The term ‘evolvability’ is relatively new, but the idea that genotypes can differ in their capacity to adapt is old. Over 70 years ago, Waddington proposed that genes could interact with one another so that a gene at one locus could determine the effect of mutations at another locus on an organism's phenotype (Waddington, 1942). Waddington's interest was in how these interactions might allow organisms to be robust (his term was ‘canalized’) to genetic and environmental perturbations. Recent work has focused on the converse, how genotypes can differ in the probability that a newly occurring mutation will cause beneficial phenotypic change, thereby promoting adaptation to a particular environment. Here, we review some of the underlying mechanisms that can contribute to differences in the evolvability of distinct genotypes. We also discuss experimental evidence that evolvability does differ between genotypes and that it can play a role in their eventual evolutionary success. Throughout this review, we follow a definition of evolvability as the capacity to produce new adaptive variation. In general, this definition can be operationalized as evolvability corresponding to the mean fitness improvement of replicate populations beginning from a particular genotype and evolving for a certain time in a defined environment.

Measuring evolvability

Quantifying evolvability is experimentally challenging. A difference in the capacity of two genotypes to produce heritable beneficial variation can be defined in simulations and parameterized in models but, for two reasons, is difficult to measure directly. First, unlike traits that have immediate phenotypic consequences, evolvability can only be measured by considering effects accruing over evolutionary time scales. It is not possible to examine a single genotype and determine its evolvability. Instead, one must look to the fitness (or some other phenotype of interest) distribution of a genotype's offspring, usually following at least tens to hundreds of generations of evolution, measured relative to some control genotype. Second, differences in evolvability reflect a change in the rate or effect of beneficial mutations. Because the occurrence of all mutations is an inherently stochastic process, no two evolving populations will increase in fitness along the same trajectory, even if they begin with an identical genotype and remain equally evolvable (Johnson et al., 1995). To detect a signal of different evolvability among the stochastic noise caused by the chance occurrence and loss of new mutations requires analysis of replicated evolutionary outcomes.

Experiments with microorganisms are well suited to address these difficulties (Colegrave & Collins, 2008) (Table 1). Evolutionary experiments lasting thousands of generations can be completed in months, allowing the capacity of genotypes to adapt to a defined environment to be evaluated directly. Moreover, genotypes can be used to begin multiple populations that are propagated in the same environment with sufficient replication to allow experimenters to distinguish between chance and deterministic evolutionary outcomes (Travisano et al., 1995; Blount et al., 2008; Woods et al., 2011) (Fig. 1). Approaches that exploit these attributes to test for differences in genotype evolvability are described below.

Table 1. Advantages of microbial experimental evolution for studying evolvability
Evolution in ‘real time’Allows evolvability to be examined as it affects the outcome of evolution
Known selective environmentsAddress alternative explanations for changes in apparent evolvability – for example due to differences in migration or environmental stress
Replicate evolving populations started with defined genotypesAllows robust statistical measure of (1) differences in evolvability and (2) effects of differences in ultimate evolutionary success
Genetic tools/convenient genome sequencingOpportunity to examine genetic and mechanistic basis of differences in evolvability
Figure 1.

Measuring evolvability. Two genotypes (red and gray) to be compared for differences in evolvability are grown and independently propagated as replicate populations in a defined environment. During this time, mutations occur and are selected, facilitating each population's adaptation to the evolution environment. Fitness assays allow the effect of these changes to be evaluated. A genotype that has a repeatedly higher relative increase in fitness is more evolvable.

We emphasize, however, that differences between laboratory and natural environments – most obviously, a large difference in the number of environments a population must face – seems likely to affect the magnitude of selection for different kinds of evolvability loci. For example, adaptation in a laboratory environment typically, though not always, involves losses of function in other environments, something that may be selected against in natural conditions (Cooper & Lenski, 2000; MacLean & Bell, 2002).

Mechanisms of evolvability

Changes in evolvability can occur through changes in the rate of mutational processes or changes in how genetic variation is mapped by regulatory and physiological processes to create a phenotype. We first consider several mechanisms that affect these processes and thus are candidates for influencing evolvability (Fig. 2). In subsequent sections, we consider evidence that these mechanisms can and have been selected to this end.

Figure 2.

Effect of different evolvability mechanisms in a fluctuating environment. A more-evolvable lineage (square) will tend to adapt more quickly than a less-evolvable lineage (circle), eventually causing the less-evolvable lineage to be driven extinct. Left, a higher mutation rate leads to greater genetic diversity (colored sections), increasing the likelihood of an adaptive variant being present following an environmental change (expansion of adapted section following environmental change). Center, if mutations cause a greater average beneficial effect, the same mutation (sector beginning in the middle of environment 1) causes a larger fitness increase (wider sector) in the more-evolvable lineage. Right, a difference in genetic architecture can increase the number of beneficial mutations that are available to a genotype, which will increase the relative number of adapted offspring produced and make it more likely that subsequent beneficial mutations will occur within the evolvable lineage.

Mutation rate

One of the best understood mechanisms that promotes evolvability in microorganisms is the ‘mutator’ phenotype, a global increase in mutation rate usually caused by a reduction in DNA proof-reading activity (Taddei et al., 1997; Tenaillon et al., 2000; Travis & Travis, 2002; Tanaka et al., 2003; Gerrish et al., 2007; Rajon & Masel, 2011). Mutators will tend to produce offspring with relatively higher genetic variation than will nonmutators. In turn, higher genetic variation leads to greater phenotypic diversity, making it more likely that beneficial phenotypes will be produced. Of course, a high genomic mutation rate will also cause more deleterious phenotypes to be produced, so that the net effect of a mutator phenotype is likely to depend on many factors, in particular, how well adapted a genotype is to its environment.

Several more focused mutation rate increasing mechanisms have also been described. Contingency loci are short DNA sequences – often di- or trinucleotide repeats – that lead to locally high mutation rates (Moxon et al., 1994, 2006). These loci are present, for example, in some pathogens where they facilitate evolutionarily rapid switching of surface antigen expression, which might provide a benefit in evading the host immune response (Richardson & Stojiljkovic, 1999; Ancel Meyers et al., 2003). Recently, it has been demonstrated that mutation rates can increase transiently through induction of an error-prone DNA polymerase, DinB, by a combination of stress responses. This polymerase acts to increase the frequency of mutations occurring in the vicinity of double-strand DNA breaks that it repairs (Gonzalez et al., 2008; Shee et al., 2011). By concentrating the potential cost of a high mutation rate to times of stress, when the benefits might also be the highest, stress-induced mutagenesis has the potential to broaden the parameter range over which the underlying mutation rate increasing allele can invade a population (Ram & Hadany, 2012).

Mutational robustness

Mutational robustness describes the capacity of a genotype to produce the same phenotype in the face of mutational perturbation (de Visser et al., 2003). Not surprisingly, differences in a genotype's mutational robustness are thought to influence its evolvability – however, this relationship is complex (Masel & Trotter, 2010). Perhaps counterintuitively, high mutational robustness can increase long-term evolvability by facilitating accumulation of genetic variation that would otherwise be removed by selection. Accumulation of genetic variation can lead to the eventual generation of genotypes that express novel phenotypes (de Visser et al., 2003; McBride et al., 2008; Draghi et al., 2010). Lower robustness can increase short-term evolvability by increasing the fraction of mutations that have some phenotypic effect, increasing a population's response to selection (Draghi & Wagner, 2008; Cuevas et al., 2009).

Many underlying mechanisms can affect mutational robustness, including differences in effective pleiotropy – that is, the typical number of phenotypic traits affected by a mutation (Kirschner & Gerhart, 1998; Wagner & Zhang, 2011), structural stability of mutagenized macromolecules (Soskine & Tawfik, 2010), and the activity of molecular chaperones (Fares et al., 2002; Sangster et al., 2004; Tokuriki & Tawfik, 2009). Some combination of these mechanisms can mean that genotypes in some regions of genetic space are less likely to suffer phenotypic consequences following mutation than are individuals in other regions (Sanjuán et al., 2007). A related mechanism involves evolutionary ‘capacitance’, in which genetic variation is hidden until it is released by a mutational or environmental trigger (True & Lindquist, 2000; Bergman & Siegal, 2003; True et al., 2004; Masel, 2005). A well-characterized example of capacitance is the expression of accumulated genetic variation in Saccharomyces cerevisiae cells that express the [PSI+] prion (True & Lindquist, 2000). When the [PSI+] prion is active, ribosomes become more likely to read through stop codons, resulting in the expression of previously untranslated sequence in the 5′ region of many genes (Firoozan et al., 1991).

Gene interactions

Whereas mutational robustness will usually reflect gene interactions that act generally to reduce the phenotypic effect of mutations, more specific types of gene interactions can also influence the generation and expression of genetic variation. Indeed, gene interactions that cause a mutation's effect to depend on the presence of other mutations – a phenomenon known as epistasis – are probably widespread (reviewed in de Visser et al., 2011). It is easy to imagine such interactions causing the effect of a newly arising mutation to change from being deleterious to beneficial, dependent on the presence of a previous mutation. Because it would have more available beneficial mutations, a genotype that had the earlier mutation would be more evolvable than one that did not. Examples of this kind of gene interaction come from work on the TEM1 β-lactamase gene in which particular stabilizing mutations are required before subsequent mutations can act to change enzyme specificity (Weinreich et al., 2006; Salverda et al., 2011) and from examining the basis of a new citrate utilization phenotype in an experimentally evolved population of Escherichia coli (Blount et al., 2008). More indirect interactions are also possible. For example, several recent experiments suggest that less-fit genotypes might be generally more evolvable than fitter genotypes (Barrick et al., 2010; Chou et al., 2011; Khan et al., 2011). In two experiments, this was demonstrated to be a consequence of less-fit genotypes receiving a disproportionately greater benefit on adding a new beneficial mutation (Chou et al., 2011; Khan et al., 2011).


In certain conditions, recombination can promote adaptation by generating optimal combinations of mutations. For example, combining beneficial mutations from separate genotypes into a single genotype can promote the spread of both mutations, increasing the rate of population fitness increase (Kim & Orr, 2005; Cooper, 2007). Conversely, combining deleterious mutations into a single genotype can increase the efficacy of selection in removing them from the population (Kondrashov, 1993). The benefits of these mutation combinations are indirect consequences of recombination itself and so can be thought of as reflecting a difference in evolvability. Several studies have focused on the influence of transformation – the mechanism of recombination driven by bacteria themselves, rather than by accessory elements, which may have divergent selective pressures – on evolvability. While this work has generally supported the potential for transformation to increase evolvability, it is important to note that this is not equivalent to claiming that transformation has been selected as an evolvability mechanism. While in some circumstances transformation can act indirectly to increase evolvability (Baltrus et al., 2007; Levin & Cornejo, 2009), it can also provide a direct benefit through allowing the uptake and subsequent catabolism of DNA (Redfield et al., 1997; Palchevskiy & Finkel, 2006). Indeed, the advantages of recombination as an evolvability trait may be insufficient to allow it to invade a population when it is rare (Levin & Cornejo, 2009).

Recombination may also be able to select for genetic architectures – the overall map between genotype and phenotype – that promote evolvability. Two examples come from work on digital organisms. In one experiment, recombination selected for an architecture that caused deleterious mutations to tend to interact negatively, facilitating their removal by selection and increasing population fitness (Azevedo et al., 2006). In other experiments, recombination selected for a modular genome organization (Earl & Deem, 2004; Misevic et al., 2006). Such genomes can be more evolvable in at least some selective environments (Watson et al., 2010).

Selection of evolvability

It is one thing for a trait to have the capacity to increase evolvability; it is another for this effect to be the mechanism that actually drives its evolutionary dynamics – determining a trait's success or failure in a population. Indeed, several traits that can increase evolvability are likely to also confer immediate benefits that might better explain their success. In this case, any indirect benefit conferred by a trait increasing evolvability can be thought of as a ‘byproduct’ of direct selection on some other function (Sniegowski & Murphy, 2006). For example, efficient chaperones might provide an immediate advantage under times of environmental stress by promoting proper protein folding, as well as confer an indirect benefit by decreasing the mean phenotypic effect of mutations, so that long-term population genetic variation increases (Tokuriki & Tawfik, 2009). Similarly, transformation genes might provide a direct benefit of allowing DNA to be used as an energy source as well as an indirect benefit through promoting recombination of beneficial allele combinations. In short, the fact that a focal trait is capable of increasing evolvability does not necessarily mean that it was selected because it increased evolvability. With this in mind, we discuss how a trait that confers an increase in the capacity of a genotype to generate adaptive variation, but that has no direct benefit, might be selected.

Common sense suggests that the eventual fate of a mutation will depend on its effect on the fitness of the individual(s) in which it is present. If the mutation confers some benefit (cost), we expect that it will tend to increase (decrease) in frequency. (If the mutation is neutral, we predict an equal probability of increasing and decreasing in future generations, but if it is initially rare, it is much more likely to go extinct than to fix.) Indeed, a central tenant of Darwinian evolution is that evolution is not a forward-looking process. A trait is selected based on its immediate effect, not on whether or not it might contribute to future fitness. How, then, can a trait that increases evolvability, but that confers no immediate advantage, be selected?

A full answer to this question requires mathematical modeling and depends on many variables, not all of which are well understood. Nevertheless, an intuitive explanation is quite simple. Imagine an evolvability trait that confers no immediate fitness benefit in any environment. In an asexual population, individuals that acquire this trait (e.g. through random mutation or lateral gene transfer) have an increased probability of producing offspring with increased fitness. If the evolvability trait persists in the population long enough that offspring with higher fitness are in fact produced, the evolvability trait can increase in frequency by virtue of being linked to the new beneficial mutation – a process known as genetic hitchhiking (Fig. 3) (Maynard Smith & Haigh, 1974). The idea that the dynamics of a focal mutation depends on the effects of other linked mutations in the same genetic background has been studied extensively, especially in the context of deleterious mutations influencing the fate of linked beneficial mutations (reviewed in Charlesworth, 2012).

Figure 3.

Selection of an evolvability trait. A mutation that causes a genotype to have a higher frequency of beneficial mutations (red circle), increasing evolvability, arises in a population. If a descendent of this genotype acquires a beneficial mutation (blue triangle), the increase in frequency of the beneficial mutation will increase the frequency of the evolvability mutation through genetic hitchhiking. Note that selection for the evolvability mutation is indirect in that by itself it does not confer any fitness advantage.

A classic test of this idea was performed by Chao & Cox (1983). They mixed otherwise isogenic clones of E. coli that differed only by the presence of a mutator allele and a neutral marker. The mix of clones was propagated until just after the marker began to change in frequency. In the large populations used in the experiment, this change indicated that at least one new beneficial mutation had occurred in an individual that had the marker type that was increasing. Clearly, in this kind of experiment, as long as there are beneficial mutations available to the experimental population, one marker type will eventually win. In fact, if the competing genotypes have equal evolvability, the probability that a particular genotype will win is simply proportional to its initial frequency. By contrast, if one genotype is more evolvable, it will win at a probability greater than expected based on its initial frequency alone.

Chao and Cox found that mutators were more evolvable. As long as they were initially present above a frequency of approximately 1 × 10−4, they outcompeted their nonmutator competitor. Such a threshold is exactly what is predicted if the mutators did not confer a direct benefit but were indirectly selected by hitchhiking in the same genetic background as beneficial mutations they made more likely to occur (Tenaillon et al., 1999). (If the mutator allele conferred a direct benefit, it would be expected to be successful when introduced at all but the very lowest initial frequencies, where chance effects tend to dominate selection.) Interestingly, the exact frequency of the invasion threshold is difficult to predict, depending on the strength of the mutator allele, the relative frequency of deleterious and beneficial mutations, and the outcome of complex dynamics caused by multiple competing beneficial mutations arising in both the mutator and nonmutator subpopulations (de Visser & Rozen, 2006; Desai & Fisher, 2011).

An important caveat to the above explanation for the indirect selection of mutator alleles as they act to increase evolvability is that there is no recombination. In the presence of recombination, the evolvability trait can be separated from its consequences (the beneficial mutations) and, if it confers any immediate cost, will tend to be lost from the population (Tenaillon et al., 2000).

What kind of selective environments might favor the spread of traits that increase evolvability? In general, the greater the number of new beneficial mutations available to be substituted in a population, the greater the opportunity for an evolvability trait to arise and increase in frequency through linkage with those mutations. A situation exemplifying this condition is evolution in fluctuating environments. In simulations, evolvability mutator alleles often fix when the environment fluctuates between two extremes that require different sets of adaptive mutations, so that the population is periodically selected to move to a new fitness peak (Travis & Travis, 2002; Tanaka et al., 2003). Consistent with this model, the frequency of mutators increased over 500-fold in an experiment that tracked mutator dynamics following sequential selection to two antibiotics (Lane et al., 1997). As worryingly, the application and withdrawal of a strong antibiotic selection also represents an environmental fluctuation that can select for mutator strains (Perron et al., 2010), although antibiotics can themselves be mutagenic (Kohanski et al., 2010), complicating the application of standard models.

Evidence for the evolution of evolvability

A common limitation of studies of evolvability is the difficulty of testing the influence of evolvability on the eventual success of a particular trait. Indeed, while evolutionary biologists must always be vigilant against ‘adaptationism’ – ascribing a ‘common-sense’ explanation for a trait's success without robust supporting evidence – this caution is even more important when considering a trait proposed to be selected only indirectly, via linkage with the beneficial mutations it causes.

As noted above, several laboratory experiments have examined the fate of alleles that are predicted to influence evolvability when they are introduced in competition with alternative ‘less-evolvable’ alleles (e.g. mutator vs. nonmutator mismatch repair alleles). There are, however, at least two shortcomings of these experiments when it comes to a full understanding of the potential for selection of evolvability. First, these competition experiments typically begin with the evolvability allele at a relatively high frequency, so that they address the influence of an established allele on the fate of a population (de Visser et al., 1999) or competing subpopulation (Chao & Cox, 1983; de Visser & Rozen, 2006), but rarely the potential for it to initially invade from when it first arises. Theoretical work underlines the importance of making a distinction between these attributes: alleles that can provide benefits when common may be unable to reach this threshold frequency and thus be unlikely to provide any benefit in practice (Levin & Cornejo, 2009; Desai & Fisher, 2011). Second, any experiment that begins with populations having some deliberate mix of traits is likely to choose those traits on the basis of some expectation of their function. Thus, such experiments are unlikely to identify traits that contribute to evolvability in an unexpected way, for example, through an unanticipated effect on a genotype's genetic architecture.

Addressing these issues, several laboratory evolution experiments have observed the de novo evolution of evolvability traits. Most commonly, these involve increases to the overall genomic mutation rate. For example, in a long-term evolution experiment propagating E. coli in a minimal glucose environment, three of 12 replicate populations became mutators – having a 10- to 100-fold increased mutation rate relative to their ancestor – within 10 000 generations (Sniegowski et al., 1997). Mutators can also reach high frequencies in populations of E. coli evolved in chemostats (Notley-McRobb et al., 2002) and co-evolving bacteria-bacteriophage populations (Pál et al., 2007). Importantly, in these cases, the mutator trait arose within the evolving population and was not deliberately introduced. The implication is that the mutator alleles must confer some benefit or they would be unlikely to repeatedly reach high frequencies in the large experimental populations.

In principle, the benefit of mutator alleles could derive from two sources. It could be indirect, resulting from linkage with beneficial mutations that it makes more likely to occur. In this case, the mutator allele would be acting as an evolvability allele, increasing the capacity of an organism to adapt, while not conferring any direct fitness benefit. Alternatively, it could be direct, perhaps due to an increase in growth rate caused by reducing the energetic overhead required to repair mostly neutral mutations (Sniegowski et al., 2000). In this case, despite being able to act as a evolvability allele, mutator alleles need not have been selected for this reason.

To distinguish between these possibilities, Shaver et al. (2002) began by testing for a direct fitness benefit of a ∆mutS mutator allele (Shaver et al., 2002). They found that this allele conferred no direct benefit when added to the ancestor of a population in which it subsequently fixed. Moreover, although the dynamics of mutator allele fixation in the original evolving populations was quite complex, their spread was typically associated with a fitness increase, consistent with the mutator alleles hitchhiking with beneficial mutations that they produced. This combination of results is consistent with mutator alleles rising to high frequency through indirect selection as a result of increasing genotype evolvability.

Other experiments have found evidence for changes in evolvability that have to do with differences in genetic architecture. Unlike mutator alleles, which are relatively easy to model as a general increase in mutation rate, changes in genetic architecture can be idiosyncratic, altering the rate of occurrence and/or effect of a subset of all possible mutations.

A classic experiment that implicates a change in genetic architecture as a cause for a difference in evolvability was performed by Burch & Chao (2000). These authors started two replicate populations from a single genotype of bacteriophage Φ6 and selected them for improved growth on Pseudomonas syringae host cells. Each population was then divided into several new populations, which were propagated for an additional 100 generations. There was a significant and repeatable difference in the ability of the two starting populations to continue to adapt, indicating a difference in their evolvability. The basis of this change was not determined, but seems likely to result from a change in genetic architecture. Nevertheless, it was not possible to determine whether the difference in evolvability was an influential driver of dynamics within the original populations.

Barrick et al. (2010) investigated the possibility of some general relationship between the fitness of a genotype and its evolvability. They isolated eight rifampicin-resistant strains of E. coli from a single ancestral strain. These resistance mutations had the correlated effect of reducing fitness by 3–30% in the absence of the antibiotic. Each of these strains was used to found 11 new populations, which were evolved in the antibiotic-free environment for 640 generations. Over this time, new beneficial mutations had the opportunity to arise and fix in each population. By tracking the dynamics of a neutral marker embedded into the experimental design, Barrick et al. were able to estimate the effective size and rate of the beneficial mutation driving the initial marker dynamics in each population – that is, the evolutionary parameters that directly underlie the differences in evolvability. Interestingly, estimates of the beneficial mutation effect size scaled positively with the initial fitness cost conferred by the rifampicin resistance mutations; genotypes that were initially less fit tended to have larger fitness increases at their first adaptive step. By contrast, no relationship was found between a strain's initial fitness and the rate at which new beneficial mutations occurred. Together with recent studies that find a trend for beneficial mutations to interact negatively, this work points to the possibility of an underlying tendency for fitter genotypes to be less evolvable (Kryazhimskiy et al., 2009; Chou et al., 2011; Khan et al., 2011).

A case study in demonstrating the evolution of evolvability

Just what is required to demonstrate the evolution of a new genetic architecture that affects evolvability and determined the evolutionary success of a lineage arising within a population? First, it should be established that the lineage is more evolvable than its contemporaries. Second, the more-evolvable genotype should not have an immediate advantage over contemporaries. If it does, then its selection is more easily explained as a direct result of this fitness advantage, rather than from the indirect effect of producing offspring with a higher capacity to adapt. Third, the evolutionary competition between the lineages must be repeated to distinguish between chance and deterministic explanations for the original outcome. These points are clearly shown in a recent study by Woods et al. (2011) who document the indirect selection of a more-evolvable genotype within a long-term E. coli evolution experiment.

Woods et al. followed the dynamics of a number of mutations during the early adaptation of a single evolving population. These dynamics revealed the presence of two lineages after 500 generations of evolution. One lineage – the eventual winner (EW) – contained two mutations, in the gene topA and in the rbs operon, that went on to fix in the population. The other lineage – the eventual loser (EL) – was characterized by having alternative mutations, also in topA and the rbs operon, that ultimately went extinct. Surprisingly, despite its eventual success, clones isolated at 500 generations from the EW lineage were less fit than clones from the EL lineage.

So does the observation that the initially less-fit EW lineage was ultimately successful indicate that it was more evolvable? Not necessarily. It could also be that the EW lineage happened to get lucky in the single evolutionary iteration captured in the evolving population. By chance, the EW lineage might have been first to acquire one or more new beneficial mutations that increased its fitness, displacing the EL lineage before similar mutations could occur there. To test whether the success of the EW lineage was due to luck (the stochastic occurrence of new beneficial mutations) or greater evolvability (a quasi-deterministic increase in producing beneficial genetic variation), Woods et al. performed a ‘replay’ experiment, evolving 10 new replicate populations from each of two EW and two EL 500 generation clones. This replay experiment went on for long enough that each replicate population had the opportunity to acquire new beneficial mutations. Comparing the fitness increase in the replay EW and EL populations revealed that the EW populations not only increased in fitness by a larger amount than did the EL populations, but also attained a higher absolute fitness, overcoming their initial handicap. This finding supports that the EW lineage was more evolvable and that this greater capacity for generating beneficial variation contributed to their success in outcompeting the EL lineage.

The replay populations also provided the opportunity to address the genetic basis of the difference in evolvability. Whole-genome sequencing of clones isolated from eight of the replay populations started with the EW and EL clones identified a number of newly arising mutations and also allowed the complete genotype of the starting clones to be inferred. Two results are notable. First, the EW and EL clones differed by a number of mutations in addition to those in topA and rbs that had already been identified. These did not, however, appear to affect the overall mutation rate of any clone, judged by the accumulation of similar numbers of mutations that accumulated in the EW and EL replay populations. Second, a mutation in spoT, the gene in which the next mutation occurred in the course of the original experimental population, occurred in two of the four fully sequenced EW replay populations and was subsequently found in four others. By contrast, it was not found in any of the EL replay populations. This distribution is unlikely to occur by chance, suggesting that the spoT mutations were more likely to be selected in the EW genotypes, presumably because they conferred a larger benefit. Subsequent strain constructions confirmed this expectation, demonstrating that the same spoT mutation conferred a ~ 10% benefit in combination with the topA allele found in the EW clones, but only a ~ 3% benefit in combination with the topA allele found in the EL clones. This observation identifies the different topA alleles as causally responsible for a difference in the genetic architecture of the EW and EL clones that influenced their ability to adapt.

Difference in evolvability of natural microbial populations

It is hard to test for the role of evolvability in natural microbial populations. A closely related reference strain will not usually be available, and it is difficult to distinguish between observed evolutionary outcomes reflecting a genotype being more evolvable or just being the lucky player in the great beneficial mutation lottery. Nevertheless, some observations do suggest that differences in evolvability may play a role in adaptation of bacteria to natural environments.

A good example comes from the long-term longitudinal study of Pseudomonas aeruginosa populations that chronically infect lungs of patients with cystic fibrosis. These infections can be quite stable, with the same lineage of P. aeruginosa often being associated with a patient over many years, providing an opportunity for long-term adaptation and study (Smith et al., 2006; Yang et al., 2011). One striking observation of P. aeruginosa isolated from patients with cystic fibrosis is the presence of a much higher fraction of mutator clones than are found among environmental isolates (Kenna et al., 2007; Mena et al., 2008). These mutators are significantly more likely to be resistant to antibiotics than are nonmutator competitors, consistent with the mutator phenotype increasing the likelihood of resistance mutations that, in turn, provide the mutator with an opportunity to hitchhike to high frequency. This is the same general mechanism proposed to explain the spread of mutator alleles in laboratory populations. Consistent with this interpretation, in mouse models, mutators tend to confer a selective advantage under antibiotic pressure as is expected if they contribute to the generation of better adapted genotypes (Alcalá-Franco et al., 2012).

Summary and future prospects

Laboratory evolution experiments that are explicitly designed to examine the basis and evolution of evolvability have and will continue to make significant contributions to the field of evolutionary biology. Early work identified the importance of evolvability from the fixation of alleles – particularly mutators – for which there was a mechanistic justification for thinking they would provide a future, but not current, benefit. With whole-genome sequencing techniques enabling researchers to identify and track mutations occurring within evolving populations, it is becoming possible to identify co-existing lineages and follow the dynamics of their competition. Combined with genetic reconstruction techniques and the ability to restart populations with sufficient replication to determine whether evolutionary outcomes are dictated by chance or qualities intrinsic to distinct genotypes, it seems certain that new examples of evolvability will continue to be found. Perhaps most excitingly, these experiments also provide the opportunity for follow-up experiments that examine the mechanistic basis of differences in evolvability.

An understanding of the genetic and physiological basis of evolvability will have practical implications relevant to fields such as vaccine and antibiotic design, and biotechnology, where bacterial evolvability will often be something to be countered or exploited, respectively. For example, strategies to reduce evolvability – for example caused by high mutation rates – during antimicrobial treatments may be useful as a means to combat the evolution of antibiotic resistance (reviewed in Smith & Romesberg, 2007).


We thank Ricardo Azevedo and Daniel Stoebel for helpful discussion and the US National Science Foundation for funding (DEB-0844355 to TFC).