Next-generation sequencing as a tool to study microbial evolution


Michael A. Brockhurst, Fax: 0044 151 795 4408; E-mail:


Thanks to their short generation times and large population sizes, microbes evolve rapidly. Evolutionary biologists have exploited this to observe evolution in real time. The falling costs of whole-genome sequencing using next-generation technologies now mean that it is realistic to use this as a tool to study this rapid microbial evolution both in the laboratory and in the wild. Such experiments are being used to accurately estimate the rates of mutation, reveal the genetic targets and dynamics of natural selection, uncover the correlation (or lack thereof) between genetic and phenotypic change, and provide data to test long-standing evolutionary hypotheses. These advances have important implications for our understanding of the within- and between-host evolution of microbial pathogens.


Since the early 1970s, experimental evolution (EE) with microbes has been providing evolutionary biologists with a real-time window on evolutionary processes (Elena & Lenski 2003; Buckling et al. 2009). These studies pare evolution down to the bare essentials: starting with clonal populations in simple environments all observed evolutionary change results from the action of new random mutations and natural selection. This simple and powerful technique has been applied to a wide variety of evolutionary questions including the dynamics of adaptation and the evolution of diversity and sociality. The traditional motivation for such studies has been to model general evolutionary processes in convenient experimental systems. However, it is increasingly being realized that microbial EE may provide an even more powerful tool with which to examine the evolution of biomedically and ecologically relevant traits of microbes themselves, such as the emergence and maintenance of drug resistance and the evolution of virulence, with obvious practical implications.

To date, the majority of EE studies have been limited to characterizing evolution at the phenotypic level, but in some cases it has been possible to identify the mutation(s) underpinning key phenotypes. Indeed, in organisms with very small genomes, such as viruses, it has been both feasible and practical to examine whole-genome changes for sometime by traditional Sanger sequencing methods (e.g. Wichman et al. 1999; Cuevas et al. 2002; Novella & Ebendick-Corpus 2004; Betancourt 2009; for a recent review, see Elena & Sanjuan 2007). These studies have provided important insights into the process of viral evolution. The advent of next-generation sequencing (NGS) is providing an unparalleled opportunity to gain a sequence-level view of these evolutionary processes in both eukaryotic and prokaryotic organisms with larger, more complex genomes that encode for more complex life history and metabolism. Several studies have thus far utilized this new technology to obtain whole-genome sequences for clones or even entire populations from laboratory selection experiments. Here, we briefly review some of these studies highlighting the potential of EE and NGS to enhance our understanding of both evolutionary processes in general and microbial evolution in particular. We also suggest fruitful areas and questions for future research.

The raw materials of evolution

Mutation is the ultimate source of all genetic variation and provides the fuel for evolution. As a consequence, the rate of mutation and the distribution of mutational effects are key parameters in determining the properties of an evolving population. Reliable estimates of these parameters for microbes will provide important insight into how microbial populations may evolve. Early work suggested that despite larger variation at the nucleotide level, the rate of mutation at the genome level was surprisingly constant across DNA-based microbes (Drake 1991). Unfortunately, these parameters have been extremely difficult to determine empirically even for model laboratory organisms. The reasons are twofold: First they involve the detection of rare events of potentially small phenotypic effect, and second natural selection is continually acting as a filter, preventing mutations of modest effect from reaching high frequencies.

EE has been used effectively to circumvent these problems in the guise of the mutation accumulation (MA) experiment (see Halligan & Keightley (2009) for a recent review). As with other EE studies, MA experiments follow the evolution of replicate lines for many generations. However, MA experiments differ in one crucial respect; whilst in a typical EE study most evolutionary change is driven by natural selection, the aim of an MA experiment is to observe evolution under conditions where natural selection is reduced to a minimum. In practice, this is achieved by frequently bottle-necking the populations to very small size (one or a few individuals). By maintaining populations at such low effective population sizes, all but the most deleterious mutations are free to accumulate in lines in an essentially neutral fashion, meaning that such studies provide insight into the entire spectrum of mutational events. The rarity of individual mutation events is countered by maintaining lines over timescales of 100s of generations, thus ensuring that reasonable numbers of mutations occur during the experiment whilst minimizing natural selection, such that the mutations are a largely unbiased sample from the overall distribution. MA has been applied to a number of microbial systems including viruses (Chao 1990), Escherichia coli (Kibota & Lynch 1996) and yeast (Zeyl & DeVisser 2001; Joseph & Hall 2004), as well as several multicellular eukaryotes including Drosophila melanogaster (Mukai 1964), Arabidopsis thaliana (Shaw et al. 2000) and Ceanorhabdites elegans (Keightley & Caballero 1997).

Initially, mutational parameters were inferred indirectly by applying elegant statistical machinery to the patterns of phenotypic variation observed across replicate lineages (Mukai 1964; Keightley 1994; Shaw et al. 2000). However, such indirect approaches are ultimately limited for a number of reasons. In particular, mutations of small phenotypic effects, a class of mutation that may have very important evolutionary consequences, will be extremely difficult to detect in this way. In addition, the procedures used to estimate mutational parameters rely on a number of unverified assumptions, such as the shape of the distribution of mutational effects. As a result, these indirect estimates will be both noisy and potentially biased. In principle, methods that allow mutations to be detected directly can provide better estimates of mutational parameters from MA lines. Initially, such methods relied on mutations that caused observable phenotypic change or allozyme variation. Increasingly advanced molecular methods of detecting mutations have been applied to both C. elegans (Denver et al. 2004) and D. melanogaster (Haag-Liautard et al. 2007) allowing higher-resolution direct estimates for these species. However, such studies are incredibly difficult to carry out and still only examine a relatively small portion of the genome.

The advent of high-throughput NGS makes examining the mutational input to MA lines directly, across large parts of their genomes, a practical reality. The most obvious advantage of this increased coverage is that it allows the detection of many more mutational events, increasing the precision of estimates of the spontaneous mutation rates that can be obtained from a set number of MA lines. And because these mutations are sampled from a far larger part of the genome than would be practical with other methods, these estimates will be less biased. Another advantage that is obvious only in the light of such studies is that this unbiased approach provides novel mechanistic insight into the causes of variable mutation rates, including the roles of adjacent sequences and genome position.

To illustrate the potential power of this approach, we will focus on a recent study in yeast. Lynch et al. (2008) applied NGS to the complete genome of four yeast MA lines that had accumulated mutations for approximately 4800 generations. They were able to examine mutational events directly across approximately half of the genome providing an estimate of the nuclear genome-wide mutation rate of 0.32 per cell division. This estimate is about 100 times higher than previous indirect estimates of deleterious mutation rates in yeast, suggesting that only about 1% of these mutations would be detectable in phenotypic assays.

However, the advantages of this approach are not limited to providing more reliable estimates of the overall mutation rate. Direct observation allows mutations to be classified into different classes. In yeast, the overall mutation rate could be broken down further into mutations of different type (0.0041 base substitutions, 0.0002 small insertions/deletions, 0.0019 micro-satellite mutations and 0.3094 homopolymer mutations). The data also showed clear differences between the mutational profiles of the nuclear and mitochondrial genomes, with the latter having a much higher rate of base substitutions and a much higher proportion of insertion/deletion mutations. Exploring the actual single base substitutions in more detail shows that the process of mutation itself is biased at this level, with G/C to A/T mutations much more likely than the reverse. Taken together this information provides us with a more detailed picture of the complete mutational process in yeast, something that would have been extremely difficult to obtain in any other way.

Outside of the microbial world, similar studies using NGS have been carried out on D. melanogaster (Keightley et al. 2009), C. elegans (Denver et al. 2009) and A. thalliana (Ossowski et al. 2010), allowing us to begin to examine how the mutational process in microbes compares to that of other species. Estimates of the rate of nuclear base substitution per base per cell division across yeast, D. melanogaster and C. elegans (Lynch et al. 2008; Denver et al. 2009; Keightley et al. 2009) are very similar, suggesting that this parameter may be evolutionarily conserved across a phylogenetically diverse range of taxa. However, initial evidence suggests that rates for other kinds of mutations (e.g. insertions and deletions) and in other parts of the genome (e.g. mitochondria) may be far more variable amongst species (Lynch et al. 2008). The increasing ease with which this kind of study can be carried out, coupled with the reducing costs, means that we will soon have data from a wider variety of species, allowing us to build a more complete picture of interspecific variation in mutation from which we may hope to understand how selection moulds mutation rates at this level.

However, the benefits of reduced costs will not be limited to exploring variation among species. The ability to carry out such studies at large scale will allow variation in mutational parameters within species and even populations to be examined too. To date, most empirical work on the evolution of mutation rates within species comes from experiments using bacterial mutator strains (De Visser 2002). These strains have mutation rates that are elevated by orders of magnitude compared to wild type, because of damaged DNA replication and repair systems. Whilst providing important insights into the forces acting on mutation rates, such experiments have provided a very coarse scale picture. Analysis of MA lines initiated with different strains of D. melanogaster suggests that more subtle quantitative variation in mutation rate may be widespread within natural populations (Haag-Liautard et al. 2007). The application of NGS to this problem provides the potential to explore this variation in detail and begin to understand how selective pressures will act when mutation rates vary more subtly within populations. Indeed, it is becoming feasible to combine such studies with more traditional EE approaches to allow direct empirical tests of theory and to examine how mutation rates and spectra evolve under specific selective pressures. Lind & Andersson (2008) have used this approach to highlight the role of particular DNA repair mechanisms in the evolution of genomic base composition bias in Salmonella typhimurium. We anticipate many more studies of this type in the future.

The targets of selection

In contrast to MA studies that characterize mutations that are fixed by genetic drift, the other side of EE has been driven by a focus to understand mutations fixed by natural selection (Elena & Lenski 2003; Buckling et al. 2009). Just as with MA, this aim will increasingly be well served by emerging sequencing technologies. Because of their ease of propagation and short generation times, it is possible to maintain microbial populations in the laboratory for hundreds or thousands of generations. This dramatic timescale has allowed, in a fundamentally new way, the consideration of the broad sweep of evolutionary change and enabled the direct tests of long-standing issues in evolutionary and ecological theory. How repeatable are patterns of adaptation (Lenski & Travisano 1994; Wichman et al. 1999; Cooper & Lenski 2000)? What is the shape of adaptive landscapes and how does this influence adaptation (Lenski et al. 1991; Burch & Chao 1999; Buckling et al. 2003; Colegrave & Buckling 2005; Rozen et al. 2008)? How are beneficial mutations distributed (Rozen et al. 2002; Sanjuan et al. 2004; Cowperthwaite et al. 2005; Barrett et al. 2006; Hegreness et al. 2006; Kassen & Bataillon 2006; Perfeito et al. 2007)? In large measure, these types of questions can and have been fruitfully analysed without consideration of their molecular details. Inclusion of molecular genetics, traditional Sanger and now NGS, however, has allowed these older issues to be reexamined and is allowing unprecedented insights into processes of molecular evolution.

What patterns, if any, are evident from this rapidly developing body of work? Most striking is the abundant repeatability of molecular change occurring in replicate cultures evolved in identical conditions (Treves et al. 1998; Cooper et al. 2003; Herring et al. 2006; Pelosi et al. 2006; Woods et al. 2006; Ostrowski et al. 2008; Conrad et al. 2009). For example, in the E. coli populations initiated by Richard Lenski (Lenski et al. 1991), which have been propagated in the laboratory in minimal glucose medium for over 50,000 generations, parallel changes are rife, with putative or experimentally confirmed beneficial mutations frequently found in 12/12 replicate populations (Cooper et al. 2003; Crozat et al. 2005; Woods et al. 2006; Barrick et al. 2009). The location of these mutations ranges from regulatory genes that modify the transcription of many downstream targets simultaneously, such as at the stringent response regulator spoT and which confers the benefits of nearly 10% (Cooper et al. 2003), to genes with more localized effects and smaller benefits, such as deletions in the ribose catabolism operon (Cooper et al. 2001). Similar parallelism has been found in studies with yeast (Gresham et al. 2008; Araya et al. 2010), indicating that this result is not because of a peculiarity of bacteria nor because of the environment in which these particular studies have been carried out.

This outcome differs markedly with the results of MA studies where predominantly deleterious or neutral mutations fixed by genetic drift in replicate populations are distributed across the genome. By contrast, large evolving populations are only minimally affected by genetic drift. Instead the expectation is that most mutations that have become fixed in these studies have done so because of their fitness benefits and natural selection. Repeated mutation in this context is taken as a hallmark of adaptive evolution, and in most cases where isogenic constructs containing single putatively beneficial mutations have been generated, this expectation is borne out.

The situation is not quite as clear as it may appear, however. To somewhat complicate matters, an exciting realization is that beneficial mutations occurring at the same locus do not typically occur at the same residue, nor do they necessarily confer equivalent benefits (Cooper et al. 2003; Herring et al. 2006; Ostrowski et al. 2008). This was shown clearly in Herring et al. (2006) who used NGS to sequence evolved clones from replicate populations of E. coli that had evolved in growth medium limited by the sugar glycerol. Of five independently evolved clones, all repeatedly fixed single mutations in glpK, a gene encoding glycerol kinase, but no two mutations across replicates were identical. Moreover, each variant had different consequences for enzyme kinetics and function, and each conferred fitness benefits of varying magnitude. It is interesting, with this new level of resolution to ask whether these changes really should be considered cases of parallelism? We do not aim to solve this issue here, but it is perhaps worth noting that while these mutations are all beneficial in the environments in which they arose (although to different degrees), they can confer sharply distinct phenotypes in alternative environments. In other words, their pleiotropic effects vary, indicating that these apparently parallel changes in fact represent very different sources of ‘latent heterogeneity’ should the environment change (Herring et al. 2006; Ostrowski et al. 2008). If mutational effects are integrated across environments, the examples of parallelism would appear to be less pronounced, both in terms of fitness and in terms of functional modules.

Many of the insights gained from sequencing genes from laboratory evolved microbes were derived from studies using a candidate gene approach, either because of a priori considerations of microbial physiology or because phenotypic assays had suggested loci of particular interest. Undoubtedly, many mutations are missed by this approach. Indeed, the other prominent pattern from locus-specific mutation detection in laboratory evolved microbes is how unpredictable the molecular responses seem to be. The functional explanation of fitness benefits of many mutations, even of genes that are otherwise well understood, is surprisingly poor. Given the broader coverage possible, this problem may actually be increased with NGS. Consider the recent studies of Fiegna et al. (2006) and Beaumont et al. (2009) who used NGS to identify mutations conferring novel social abilities and the origin of bet-hedging behaviour, respectively. In the first case, a single mutation upstream of a putative acetyltransferase restored social development in a cheater mutant of Myxococcus xanthus. While reconstructed strains carrying this mutation show expected benefits as well as large-scale reorganization of the developmental transcriptional programme (Fiegna et al. 2006; Kadam et al. 2008), the precise cause of these changes remains obscure. Similarly, while the molecular function of the bet-hedging mutation discovered in Pseudomonas fluorescens is known, it is still unclear how an enzyme that is part of the pyrimidine and arginine biosynthetic pathways allows the cells to epigenetically switch on and off the expression of its capsule (Beaumont et al. 2009).

There are, of course, cases where the function of mutations fixed in evolved bacterial or yeast populations is well understood (eg: Crozat et al. 2005; Notley-McRobb & Ferenci 1999; Pelosi et al. 2006). However, in many of these examples, just as in the two cases noted elsewhere, the loci involved would not have been considered a priori candidates for adaptive mutations. Finding such changes has represented one of the major difficulties in moving EE from phenotypic to genetic descriptions of evolution. NGS clearly overcomes this limitation by interrogating the entire genome in a single experiment with high resolution, thereby identifying every mutation, beneficial, deleterious or neutral, without consideration of any researcher knowledge or biases. This promises to save time and money and allows the job of marrying genotype and phenotype to progress more rapidly. It is notable that the advantages of this extend beyond evolutionary biology, because many of the identified beneficial mutations occur in genes with unknown function in addition to well-studied genes that have evolved altered functionality.

Another important benefit of NGS is that it allows entire populations, rather than just endpoints of adaptation, to be scanned for single nucleotide polymorphisms (SNPs), thus providing an unprecedented look into the hidden turmoil of adaptive dynamics. To pick one example, an area where EE has been very influential is in describing the statistical distribution and fixation dynamics of beneficial mutations (Rozen et al. 2002; Cowperthwaite et al. 2005; Barrett et al. 2006; Hegreness et al. 2006; Kassen & Bataillon 2006; Perfeito et al. 2007; Silander et al. 2007; Rokyta et al. 2008; Schoustra et al. 2009). Theoretical developments have, however, outstripped experimental power in this area (Gerrish & Lenski 1998; Desai et al. 2007; Jain & Krug 2007; Park & Krug 2007; Brunet et al. 2008; Kopp & Hermisson 2009; Sniegowski & Gerrish 2010), most clearly with regard to whether beneficial mutations fix singly or multiply and the degree to which fixation is impeded by competition among co-existing beneficial mutations, a process called clonal interference. By tracking the dynamics of markers linked to new mutations, it has been possible to observe fixation of new beneficial mutations as well as to determine that clonal interference is common. This has worked extremely well for common mutations near to fixation. But what of the rare, or even transiently common types, that ultimately fail to fix? By analysing whole populations of evolving cells for SNPs, Barrick & Lenski (2009) have shown how these rare types, even those destined to be outcompeted, can be observed. In addition to providing unprecedented insight into adaptive dynamics, this deeper resolution offers a huge new reservoir of fitness improving mutations for further study. If one of the broader aims of evolutionary biology is to develop a coherent understanding of the genotype/phenotype map, this massive collection of unstudied SNPs is a superb place to begin.

The dynamics of genome evolution

Characterizing the dynamics of phenotypic evolution has been the focus of a number of EE studies using a range of organisms (Elena & Lenski 2003; Buckling et al. 2009). The best studied of these are the long-term E. coli populations, for which, at regular intervals, Lenski and colleagues have measured a range of traits in each population (Lenski et al. 1991; Lenski & Travisano 1994). When plotted against time, these measures give insight into the dynamics of phenotypic evolution. For competitive fitness and cell size, a positive saturating relationship with time has been observed in replicate populations. These phenotypes evolved most rapidly in the first 2000 generations, after which time their rate of evolution slowed down dramatically (Lenski & Travisano 1994; Barrick et al. 2009).

What patterns of molecular evolution might underpin these dynamics of phenotypic evolutionary change? Two possibilities spring to mind: either the rate of beneficial mutations decreases as populations become more adapted or if the rate of beneficial mutation remains constant, the magnitude of their effects becomes smaller slowing their rate of fixation. Both possibilities suggest that the dynamics of molecular evolution should broadly correlate with the dynamics of phenotypic evolutionary change. In a recent study (Barrick et al. 2009), whole-genome sequences were obtained from one of these populations at 2K, 5K, 10K, 15K, 20K and 40K generations. Contrary to either scenario given earlier, these genome sequences revealed an almost constant rate of molecular evolution over the first 20K generations, with approximately two mutations fixed every 1000 generations (Barrick et al. 2009). Such clock-like regularity of substitutions is usually considered the signature of neutral evolution, but these substitutions were nonsynonymous and the population was clearly adapting to its environment. Furthermore, almost all of the mutations were found to have significant beneficial effects on fitness when introduced back into the ancestral bacterial genotype.

What process might explain this lack of correlation between the dynamics of molecular and phenotypic evolution? One possibility is that this is a consequence of clonal interference whereby coincident beneficial mutations in different clones must compete with one another until the eventual fixation of the most beneficial mutation. The relatively rare mutations of greatest beneficial effect will likely dominate the early stages of adaptation, whereas more common mutations of lesser beneficial effect will come into contention in later stages. This could have the effect of increasing the supply of contending mutations through time, leading to a relatively constant rate of substitution. This prediction is borne out by pooled sequences of the entire population (rather than single clones), which, by utilizing the extremely deep coverage that NGS is capable of, allow quantification of the level of polymorphism through time. This demonstrated that the number of contending mutations gradually increased through time peaking after approximately 10K–15K generations of evolution (Barrick & Lenski 2009).

As NGS techniques become more affordable, they simply become another empirical tool and experiments can be specifically designed with NGS in mind, opening up previously untestable hypotheses to rigorous analysis. In a recent study (Paterson et al. 2010), NGS was utilized to test the long-standing Red Queen Hypothesis, which posits that antagonistic species interactions should drive evolution through continual natural selection for adaptation and counter-adaptation (Van Valen 1973; Stenseth & Maynard Smith 1984). Populations of bacteriophage viruses were allowed to evolve against either a fixed bacterial host population (evolution treatment) or a co-evolving bacterial population (co-evolution treatment). Pooled population whole-genome sequences were obtained for six evolving and five co-evolving viral populations at the end of the experiment. Genetic distance travelled from the ancestral viral sequence was calculated for each population by scaling each mutation by its frequency in the population. This revealed that the genomes of co-evolving viruses were evolving twice as fast as those of their evolving counterparts confirming the central prediction of the Red Queen Hypothesis. Furthermore, co-evolving viral populations also contained higher levels of genetic polymorphism than evolving viral populations (Paterson et al. 2010).

Next-generation sequencing and pathogen evolution

By stripping evolutionary and ecological scenarios to their bare essentials EE in the laboratory has highlighted fundamental elements of these fields. However, microbes in nature live in complex environments with a broad range of genetic and ecological inputs that modify and constrain evolutionary patterns. Accordingly, it has been challenging to extrapolate from the laboratory to explain the evolution and diversity of microbes in nature and especially of microbial pathogens.

There are two complementary approaches to understanding microbial pathogen evolution in ‘real-world’ environments. First, there is an increased recognition that it is possible to carry out long-term controlled studies using microbial pathogens in natural or novel hosts to examine the within- and between-host evolutionary changes. This has great potential to illuminate the important aspect of pathogen evolution (Ellis & Cooper 2010; Racey et al. 2010) and when allied with NGS could provide a frame of reference for studies of pathogen genome evolution in the wild. For example, (Hall et al. 2010) examined how compensatory adaptation might reduce the costs of rifampicin resistance in P. aeruginosa in antibiotic-free environments. By sequencing rpoB, the gene mutated to confer drug resistance, they were able to identify several of the compensatory mutations, but this approach is clearly limited to finding the subset of compensatory mutations that occurred in the resistance gene. NGS offers the potential to extend such studies to identify compensatory sites elsewhere in the genome, thereby improving our mechanistic understanding of antibiotic resistance evolution with important implications for the persistence of antibiotic resistance in pathogens once the selective pressure is removed. Second, NGS can provide a window on real-time microevolution of pathogen strains causing infections or epidemics in host populations. Smith et al. (2006) used traditional Sanger sequencing to compare the genomes of two P. aeruginosa isolates from 6 months and 96 months postinfection of the lungs of a cystic fibrosis patient. Evidence of positive selection was reported across the genome, including selection against a range of putative ‘virulence factors’. NGS methods would feasibly allow the comparison of a far greater number of isolates to decipher the tempo and mode of molecular evolution of pathogen strains across broader spatial and temporal scales. Along these lines, (Harris et al. 2010) have recently applied NGS to a large global collection of 63 clinical isolates of the ST239 strain of methicillin-resistant Staphylococcus aureus (MRSA). Rather than re-sequence each whole genome, they use NGS to map the SNPs and indels in each genome against a reference genome sequence, strain ST239. Using this technique, 6714 SNPs were identified at 4310 sites. There was a predominant pattern of divergence, with little convergent evolution between isolates and almost 30% of such homoplasies being linked to antibiotic resistance, underlining the importance of clinical selection. Furthermore, a very strong geographical signal was detected such that strains isolated in the same region grouped together on the phylogeny, suggesting clonal expansion of single variants within regions. Exploiting the fact that isolates were collected over a period of ∼25 years, Harris and colleagues were also able to estimate the rate of substitution as 3.3 × 10−6 per site per year, which is somewhat higher than that observed in other bacterial species, perhaps suggesting that the population bottlenecks inherent to the life of a pathogen may lead to MA in MRSA.

The Harris et al. (2010) study highlights the extra information that can be obtained by NGS from a traditional strain collection. However, finer scale information on pathogen microevolution could be obtained by employing different sampling regimes. NGS of temporal samples from single infections, particularly chronic infections, would reveal the dynamics and mechanisms by which pathogenic microbes adapt to their hosts. Conversely, NGS of samples from multiple patients infected with the same strain would reveal the diversity of evolutionary trajectories available. Finally, NGS of pooled population samples would reveal the diversity present within infecting populations of pathogens, which is so often lost by picking a single colony as is often carried out in clinical diagnostic laboratories.


The declining costs and broader availability of NGS have meant that this powerful approach can now be used by laboratories with modest budgets studying nonmodel organisms. We have briefly touched on areas where NGS is having a marked and growing impact on understanding the dynamics and mechanisms of adaptation in the laboratory populations of microbes. Increasingly, we anticipate that NGS will be used to generate and test novel ideas, both to further our understanding of fundamental evolutionary processes and as a means to integrate laboratory studies of microbes with patterns of evolution in microbial pathogens in complex environments.


We thank Steve Paterson and Stuart Piertney for inviting us to write this review and two anonymous reviewers for their comments and suggestions.

The authors use experimental evolution with microbes to test evolutionary theory.