Environmental heterogeneity constitutes an evolutionary challenge for organisms. While evolutionary dynamics under variable conditions has been explored for decades, we still know relatively little about the cellular and molecular mechanisms involved. It is of paramount importance to examine these molecular bases because they may play an important role in shaping the course of evolution. In this review, we examine the diversity of adaptive mechanisms in the face of environmental changes. We exploit the recent literature on microbial systems because those have benefited the most from the recent emergence of genetic engineering and experimental evolution followed by genome sequencing. We identify four emerging trends: (i) an adaptive molecular change in a pathway often results in fitness trade-off in alternative environments but the effects are dependent on a mutation's genetic background; (ii) adaptive changes often modify transcriptional and signalling pathways; (iii) several adaptive changes may occur within the same molecular pathway but be associated with pleiotropy of different signs across environments; (iv) because of their large associated costs, macromolecular changes such as gene amplification and aneuploidy may be a rapid mechanism of adaptation in the short-term only. The course of adaptation in a variable environment, therefore, depends on the complexity of the environment but also on the molecular relationships among the genes involved and between the genes and the phenotypes under selection.
How organisms adapt to their environment is a fundamental question in evolutionary biology. While our understanding of adaptation in a constant and simple environment has considerably progressed both theoretically and empirically over the past decades [1,2], the question of how adaptation proceeds in heterogeneous environments remains challenging [3,4]. One major challenge that remains is to determine what are the molecular and cellular consequences of the adaptive mutations in response to a changing environment and how these molecular changes translate into conditionally adaptive or deleterious phenotypes. The importance of this question extends further than our need for a fundamental understanding of evolution because predicting evolution would benefit all fields of life sciences, including medicine.
Microbial species offer exquisite models for the study of the molecular bases of adaptation to fluctuating environments. Their short generation time and generally small genome sizes allow evolution to be followed over relatively short periods of time in the wild and in the laboratory, for instance through the use of genetic engineering or experimental evolution. The combination of experimental evolution and whole-genome resequencing has greatly accelerated our understanding of how organisms respond to selection [5–8]. Most importantly, these experiments have allowed the identification of the molecular changes that may increase organismal fitness when the environment is changing. Here we explore adaptive strategies used by microorganisms and their molecular bases under various types of environmental changes in relation to their genetic architecture.
2. Adaptive strategies depend on the rate and predictability of changes
Because the environment is highly heterogeneous at microscales, microorganisms are challenged by fluctuating biotic and abiotic factors. The various parameters involved may include changes in pH , in inter- and intraspecific competition  and in nutrients and resources availability . A pertinent illustration is Escherichia coli in the digestive tract of mammals. Cells go through a succession of carbon sources such as lactose and maltose, and a succession of stresses such as increasing temperature and decreasing oxygen levels . The changes occur in time and space and vary in frequency, for instance within the time frame of a single generation .
Environmental heterogeneity encompasses two non-exclusive types: spatial heterogeneity, where conditions are changing across space and temporal heterogeneity, in which organisms are challenged in time within a generation. Because they perform well in their specific habitat, specialist genotypes tend to be favoured under constant or in spatially heterogeneous environments with limited dispersal. On the other hand, generalists tend to perform suboptimally but well in a wide-range of conditions, such that they are favoured under temporally heterogeneous environments [11,13–15]. The distinction between the two strategies is explained by trade-offs that characterize specialized genotypes: a beneficial or neutral allele in one environment may lead to deleterious effects in another [11,13,16].
A major environmental factor that determines what adaptive mechanisms would be favoured in a changing environment is its predictability. Adaptation in the case of predictably changing conditions may favour the evolution of molecular systems that can anticipate the changes that are coming, such as anticipatory gene regulation or the maintenance of past events in memory. For instance, seasonal variation can be predicted because there are periodic signals in the environment prior to the changes taking place [17,18]. In the case of random and non-predictable changes, selection may favour a bet-hedging strategy, where a genotype randomly visits different phenotypes and hence enhances the probability of adaptation to the coming condition . These various processes of anticipation, memory and bet-hedging are improving the capacity of microorganism to grow under fluctuating environments while diminishing trade-offs, and thus could be considered as mechanisms of a generalist or plastic strategy.
These adaptive strategies and the conditions that favour their evolution have been well studied and allow general conclusions to be drawn as to which one is expected to be favoured in one set of conditions versus another. However, the environment is only one part of the equation at play. The genotype of the organism could also play an important role by determining what mutations are available and how they may change cellular functions such that predictable changes can be detected and interpreted, or such that trade-offs are present or not in a new environment. A complete understanding of evolution in a changing environment, therefore, requires the study of the molecular bases of these strategies.
3. Molecular bases of specialists and generalists
(a) Negative pleiotropy and specialists
Experimental studies have demonstrated that generalists and specialists can evolve in the laboratory under controlled conditions, allowing for the identification of the molecular mechanisms underlying these two different strategies. To investigate what mutations may lead to fitness trade-offs, the budding yeast Saccharomyces cerevisiae was evolved under carbon limitation for more than 250 generations . The limitation of carbon sources led, among other mutations, to an increased copy number of the gene HXT6/7, which codes for hexose transporter proteins that perform transport of sugars across membranes . Yeast strains with this gene amplification show higher fitness compared with the ancestral state in limited glucose and other carbon source-limited media, consistent with an adaptive modification. The amplification, however, leads to lower fitness in rich glucose conditions , suggesting that increased influx of sugar has condition-specific effects on fitness. One potential mechanism for this trade-off is that a high concentration of glucose is toxic  and perturbs cell homeostasis. Another potential source of trade-off is the cost of expressing high levels of HXT6/7. The specialist strategy of overexpressing genes is thus accompanied by a loss of plastic response, which is deleterious in conditions where these genes would need to be downregulated. Because of the strong negative consequences of this gene amplification, it could be a transient solution to allow rapid adaptation to the population under limited glucose before the development of a more stable genetic adaptation in the long term. The transition from a costly molecular change like gene amplification, to a more sustainable mechanism is observed in bacteria during growth on limited carbon sources or drugs treatments . Gene amplification is, however, not always a long-term dead end. For instance, in the case of antibiotic-resistance mutations, the fitness cost in the absence of the drug can be sometimes compensated by gene amplifications .
Another study using a similar experimental approach showed that the loss of ability of cells to properly regulate signal transduction pathways is also implicated in fitness trade-offs. Kvitek et al.  evolved yeast strains in a nutrient-limited chemostat and followed the course of adaptive evolution . The loss of environmental sensing, for instance through disruptive mutations in major pathways such as glucose signalling and transport, and the cAMP/PKA pathway, was key to increased fitness  but was shown to be maladaptive during starvation, where nutrient availability is not constant. The authors proposed that in constant, nutrient-limited environments, constitutive commitment to cell division is beneficial and is achieved by mutations that uncouple the sensing and responsive parts of the pathways, which would be important in other conditions. The trade-off that evolved in the specialist is, therefore, inherently built into the cell regulatory machinery that needs to strictly balance two important cellular processes, leaving no options for alternative solutions.
These experiments are examples of how a specialist strategy evolves by the loss of the molecular mechanisms that would favour a generalist one. Again, a loss of plasticity may be key for the evolution of specialists but with associated trade-offs at the molecular levels. Most importantly, these pleiotropic effects are highly context-dependent and are, therefore, not universal. For instance, mutations in rpoB, which encodes a subunit of the RNA polymerase (RNAP), were shown to confer resistance to the antibiotic rifampicin with a fitness cost that is dependent on the species studied . More generally, the sign of pleiotropy varies depending on the mutated gene, on the particular mutation (e.g. different substitutions) and the epistatic interactions with the genetic background in which it occurs. Another example comes from antifungal resistance in yeast. Gerstein et al.  studied the molecular basis of antifungal resistance and identified the first mutational target of selection in 64 yeast lines. The replicated design used allowed the observation of a strong parallelism in the multiple mutations in four genes of the ergosterol biosynthesis pathways, suggesting that the mechanism of action is very specific and involves only few loci . The tolerance of resistant mutant lines was measured in different stresses, including ethanol, copper and salt, which in principle are not related to the antifungal used. This experiment revealed important gene-specific pleiotropic effects, with opposite signs: mutations in one gene could lead to decreased tolerance to copper while mutations in another gene could increase tolerance. While mutations in different genes may be equivalent with respect to adaptation to a first condition, they have distinct consequences in a changing environment. Whenever there are multiple mechanisms that can lead to the same adaptive phenotypes, even if these mechanisms are concentrated in the same molecular pathway, the nature of the genes and mutations involved determines whether these adaptations lead to fitness trade-offs in other conditions.
(b) Positive pleiotropy and generalists
Trade-offs are not universal and mutations may also have positive pleiotropic effects. One mechanism is cross-protection, which corresponds to the enhanced survival to a second environmental stressor caused by a response to a first one [26–28]. Cross-protection refers to positive effects in two contexts. First, it refers to a response that does not involve genetic changes but rather to a physiological protection to secondary stresses. The first stressful condition appears to prepare cells to undergo other stresses that require responses that are physiologically similar. For instance, it was shown that heat-shocked E. coli cells are more tolerant to acid stress . This tolerance may be due to the synthesis of protecting enzymes induced by heat shock that are already present and protective in cells when facing the low pH condition . The second context is through the fixation of beneficial mutations that will later be beneficial in yet unseen conditions.
A simple explanation for these cross-protections is that they recruit the function of very similar sets of genes. From the cellular response point of view, acid exposure, osmotic stress or cold shock could be more or less similar, as they trigger at least partially identical molecular pathways. This suggests that there might be a general response to all stresses and thus exposure to any stress could cross-protect against other ones. Experimental investigation of this hypothesis showed that this is not always the case and could be more complex , suggesting that protecting prestresses may act in slightly different ways to provide cross-protection. Using a yeast experimental system where the role of nearly all the genes involved in cross-protection could be examined, Berry et al.  showed that each pretreatment may recruit the expression of largely different sets of genes, with only few genes overlapping between conditions. These few genes are involved in the regulation of Ras signalling and the TORC1 complex, as well as DNA damage repair, suggesting that major cell regulators may systematically be recruited, but without having the same importance relatively to the pretreatments .
The commonality of gene expression and physiological programmes often observed in response to stress suggests that cross-protection by adaptive mutations could readily evolve through mutations that affect these shared mechanisms (positive pleiotropy). An example of adaptive cross-protection is the acquisition of antibiotic resistance in bacteria in the absence of antibiotic selection, where resistance occurs by the acquisition of multiple stress resistances [32–34]. For instance, mutations in rpoB in E. coli that increase fitness under starvation or at high temperature confer antibiotic resistance to rifampicin [32,33,35]. In this case, by modifying transcriptional regulators, adaptive mutations lead to increased fitness in the selective environment and confer antibiotic resistance as a side effect. However, these changes were shown to be disadvantageous in other media and temperature conditions in a genetic background-dependent manner , showing that, as for negative pleiotropy, cross-protection is not universal.
4. Mechanisms of anticipation and memory
(a) Anticipation and associative learning
Environmental changes may occur at regular intervals or follow a predictable course. Adaptive strategies in this case could involve anticipation whereby cells use one signal to anticipate and gain protection against a stressor to come. The capacity to anticipate has been observed in several microorganisms including E. coli, S. cerevisiae, Candida albicans and Vibrio cholerae in predictable and repeatable successions of conditions [12,36–38]. For instance, the succession of heat shock and oxidative stress, correlated with the natural order of stressors during beer and wine fermentation, leads to unidirectional cross-protection in S. cerevisiae . During heat shock, a fraction of cells expresses genes essential for protection against oxidative stress, allowing these cells to be ‘pre’ acclimated to the new stressful condition to come . In the mammalian digestive tract, E. coli cells are exposed to a succession of lactose and then maltose. Mitchell et al.  showed that wild-type strains have the ability to anticipate the succession of carbon sources by the induction of maltose genes in response to lactose exposure (figure 1). In addition, after ingestion by mammals, E. coli cells immediately experienced increased temperature and the oxygen level drops to anaerobic conditions when they enter the intestine. Tagkopoulos et al.  used laboratory conditions to mimic the succession of conditions inside the mammalian digestive tract to determine the capacity of anticipation of E. coli cells. The transcriptional response to temperature upshift also repressed aerobic respiration to anticipate upcoming anaerobic conditions even if the oxygen level was still high . Adaptive strategies in a changing environment may thus involve the evolution of sensing systems that ‘learn’ to prepare cells to stresses to come. Such adaptive mechanisms are expected to evolve as a long-term strategy, but the rate of evolution of anticipatory gene regulation remains to be measured. In an attempt to do this, Dhar et al.  evolved yeast populations under fluctuating conditions of oxidative and salt stress during 300 generations . An asymmetric cross-protection or anticipation evolved where oxidative stress protects against salt stress but not the opposite. The authors could not clearly separate the two possible mechanisms but showed the possible emergence of anticipation within 300 generations, suggesting that this could evolve rapidly.
Microorganisms often switch between carbon sources by catabolite repression: the preferred nutrient is first consumed, which enables high growth rate until depletion, followed by transcriptional reprogramming that allows the use of other, less preferred nutrients [39–42]. In this context, the memory of past activation may accelerate reactivation in face of a new carbon source. In S. cerevisiae, an anticipatory induction of metabolic pathways allows certain cells to adapt more rapidly and lead to fitness gain in environments with novel carbon sources [40,41]. Evolved and natural strains have been used to demonstrate consistent results: as long as glucose is available, expressing a second sugar metabolic pathway is costly in a constant environment but allows more rapid growth and more efficient acclimation when the environment changes from glucose to the other carbon source [39–41]. Experimental evolution of strains under variable carbon sources showed that mutations in HXK2, a gene that encodes a glucose sensor, and STD1, a gene interacting with glucose sensors to regulate carbon source utilization, provided better recovery after the shift from glucose to maltose . Many of the mutants displayed a reduced catabolite repression of maltose genes, which most likely explains the reduced recovery time after the abrupt shift to maltose . In natural isolates of S. cerevisiae, a better recovery after a shift from glucose to galactose is influenced by the early induction of galactose utilization (GAL) genes, which prepares the transition to galactose metabolism in some strains . In the reverse situation, when glucose is added to the system containing galactose, post-transcriptional regulation is involved through the degradation of galactose network transcripts .
These results show that the ability to display memory and anticipation varies in natural populations and that it could respond to selection in experimental evolution. To test this hypothesis, Segrè et al.  evolved strains of S. cerevisiae under repeated glucose–galactose transitions and identified mutations that increased fitness under this selective regime. The authors found that adaptive mutations shared one of the same target genes, GAL80 . GAL80 encodes a transcriptional regulator that represses the genes involved in galactose metabolism when galactose is absent and glucose is present. This adaptive mutation reduces the repression of the galactose metabolic pathway and thus enhances the capacity of cells to rapidly cope with changes in carbon sources. These results show that repression could be tuned to better anticipate the change in conditions when variation is frequent.
Overall, these observations suggest that microbes have complex cellular machineries that enable them to cope with environmental changes by ‘learning’ the transition between events, for instance through transmission of information from mother to daughter cells, which could be achieved by chromatin remodellers and the coupling of feedback signalling pathways [45–48]. In S. cerevisiae, the cytosolic catalase Ctt1p appears to be involved in H2O2 tolerance after mild salt stress . The mechanism in this case seems to be the transmission of Ctt1p from mother to daughter cells, which never faced salt pretreatment . Such transmission of long-lived protein is also reported in the memory of galactose exposure in yeast, with the inheritance of the signalling factor Gal1p [45,47].
5. Stochasticity of events and cell fate decisions
Environmental stochasticity may come from biotic as well as abiotic factors and their combination . A strategy that has emerged under unpredictable changes in microorganisms is stochastic phenotypic switching, also called bet-hedging. Theoretical and empirical studies have shown that this phenotypic noise allows a single genotype to produce randomly different phenotypes, which facilitates survival in uncertain conditions [51–53] because it enhances the likelihood of an adaptive phenotype to be expressed [50,51,54].
Several factors influencing this switching strategy have been described and include the ability of directly sensing the environment, the environmental cues involved and the cost–benefit balance . Additionally, the rate of switch between different phenotypic states is an important parameter and needs to be tuned to the rhythm of changes in environmental conditions [55,56]. To demonstrate the importance of this parameter, Acar et al.  engineered strains of S. cerevisiae to measure the advantage of adjusting the rate of stochastic switching in the galactose utilization pathway. The authors designed a system using the bistable galactose pathway that expresses two phenotypic states, ON and OFF, to which they coupled a gene that allows for positive and negative selection. The rate at which the expression of the gene switches from one state to another was adjusted externally, which allows for the measurement of the fitness of each strategy. By monitoring the growth of cells in changing environments of different switching-frequencies, the authors showed that fast-switchers recover faster after environmental change but show a lower growth rate during steady state than low-switchers . In addition to demonstrating predictions made from theoretical models, this study is a spectacular example of the utilization of cell engineering as a powerful tool to elucidate ecological and evolutionary questions.
Although stochastic gene expression is the designated primary driver for phenotypic switch, several experimental studies point out more complex underlying mechanisms, as reviewed in Norman et al. . Complex regulatory pathways may allow for a more long-lived stochastic phenotype than noisy gene expression alone . The emergence of bet-hedging strategies is often reported to be regulated by feedback loops. For instance, Lactococcus lactis and E. coli appear to use similar bet-hedging strategies: upon a shift from glucose to, respectively, cellulobiose or gluconeogenic substrate, the initial population is separated into two phenotypes, growing and non-growing cells [58–60]. In E. coli, the Cra-FBP flux sensor measures the carbon uptake flux and is involved in phenotypic diversification. If cells achieve a certain level of substrate uptake flux, they will grow on the gluconeogenic substrate, otherwise they will reach a dormant state without growth . In addition to the metabolic state, some studies mention the possible role of epigenetics in cell destiny [58,61].
Another strategy that may allow survival during sudden environmental changes is persister cells, which is well documented in numerous bacterial species including pathogenic species such as E. coli , Myobacterium tuberculosis  and Pseudomonas aeruginosa . Bacterial persistence consists of non-growing cells in a dormant state within an isogenic population and that are particularity tolerant to drug treatment [60,65–67]. Persister cells are characterized by a cessation of most cellular process with a low level of translation . Their tolerance to antibiotics could be the consequence of the inactive pathways of the drug targets  because cells remain sensitive to antibiotic upon being regrown. The molecular bases of persistence has been elucidated in some cases: the action of toxin–antitoxin (TA) modules appear to play a role, for example the HipA toxin and mRNAs endonucleases encoded by TA loci inhibit protein synthesis [68,69]. Several other persistence genes have been identified as playing a role in the formation of these dormant cells and they are involved in various processes such as carbon, amino acid or lipid metabolism . However, understanding their emergence and evolution remains a challenge especially in the context of pathogenic species. Together, strategies of random phenotype switching, bet-hedging and bacterial persistence are allowing for long-lasting stages in responses to changing environments and thus could be considered a generalist strategy as a subpopulation is able to survive in diverse conditions.
6. The race against extinction
The study systems described above consider environmental changes that reduce fitness but do not lead to population extinction if adaptive mutations fail to rapidly occur and fix. Environmental changes may be abrupt and severe, in which case populations are likely to drastically reduce in size and face extinction [70,71]. Adaptive evolution leading to population recovery is termed evolutionary rescue [72–74]. This ‘race’ against extinction implies that adaptive genotypes rise rapidly in frequency from standing genetic variation available prior to the environmental change or from de novo mutations [75–78]. The respective influence of mutation acquisition de novo or from standing genetic variation on rapid adaptive evolution and their long-term effects remain to be investigated with models and experimental studies [79,80]. Another key factor that may influence rescue probability is the rate of environmental change. Lindsey et al.  showed that for E. coli under rifampicin antibiotic treatment, the mutational paths followed are different between abrupt and gradual stressful changes , suggesting that the response to selection may vary at the molecular level.
Insights into the molecular bases of evolutionary rescue would help understand whether it involves specific types of adaptive molecular changes and how the genomic architecture of the traits involved may interact with each of the previous described factors. The genetic bases of evolutionary rescue have, however, not been studied extensively apart from a few specific contexts such as antibiotic resistance. For example, rpoB mutations in E. coli are known to occur in response to mild rifampicin exposure and evolutionary rescue under antibiotic stress , suggesting the possibility that the same mechanisms are involved because they rely on changes in the same genes.
One general mechanism of evolutionary rescue that was shown to be important over the past decade is aneuploidy, particularly in fungi [82–84]. Changes in chromosome copy number appear to be a rapid mechanism of adaptation in the context of abrupt stress, but also seem to impose a cost such that it is rapidly lost when the stress is removed or replaced by other mutations under continuous selection . In yeast, cells maintained in minimal medium and heat stress evolve chromosome III trisomy and this additional chromosome is lost when the stress is removed. However, under continuous selection, cells evolve gene expression change on chromosome III, which allows higher fitness without the cost of aneuploidy . This observation illustrates that evolutionary rescue could proceed through chromosomal changes, allowing population survival in the short term but requiring additional changes in the long term. Aneuploidy could illustrate the differences between the molecular bases of evolutionary rescue and adaptation, where rescue requires a fast response to face extinction; when population survival is not threatened, the most advantageous mutations overall may be favoured by selection according to their net benefit.
One pivotal development regarding evolutionary rescue is the demonstration that initial genotypes determine the probability of rescue and the effects associated with the adaptive mutations. Different yeast species with similar past salt stress experience, for instance S. cerevisiae and S. paradoxus, display different probability of evolutionary rescue under lethal salt concentrations . Genetic background could, therefore, play an important role in determining which genes are mutated in the rescue process. Filteau et al.  performed an experiment to test this hypothesis. They used two distinct strains of S. cerevisiae that carry the same mutation that makes them sensitive to high temperatures. They evolved by rescue independent populations of the two genotypes at high temperature, measured the probability of rescue and identified the causal mutations. Their result shows that different genetic backgrounds largely display different rescue probabilities and are also rescued to a large extent by different molecular changes . Some gain of function mutations are specific to one genetic background while the losses of function of specific signalling pathways are specific to the other. Once again, the adaptive changes include aneuploidies, which also appear to evolve in a genetic background specific manner.
Overall, the few studies that have investigated the molecular changes associated with evolutionary rescue revealed that these changes often have large pleiotropic effects and may thus represent short-term solutions to an evolutionary challenge, as shown by ploidy changes. Another important observation is that rescue may take place following different evolutionary paths depending on the starting genotypes of the populations. The strong contribution of the starting genotype on the probability of rescue and on the molecular changes involved suggests that genetic variation in the potential to be rescued may exist in natural populations. The most important consequence of this finding is that the extrapolation of what is observed for a particular population or genotype to other populations is risky, for example, in the case of antibiotic resistance or rescue in the context of conservation genetics.
7. Conclusion and future perspectives
Although the studies that have elucidated the molecular bases by which adaptive mutations provide a fitness advantage or disadvantage are limited in number, they allow the identification of potential trends for future investigations. First, transcriptional regulators and signalling proteins are often involved in adaptations and the loss of plasticity caused by mutations in these regulators is often the cause of fitness trade-offs. Mutations in these regulators can also lead to cross-protection by affecting the same set of genes and physiological processes across different conditions. As adaptation to a condition could proceed through mutations in different genes within a given molecular function or pathway, another key observation is that different genes and different mutations within a gene often lead to various pleiotropic effects. In addition, these pleiotropic effects often occur in a genetic background-dependent manner. Finally, the study of rapid adaptation through evolutionary rescue suggests that the ability of a population to adapt and its mechanisms of adaptation may depend on its starting genotype.
Another challenge will be to determine whether short- and long-term adaptation draw from the same mechanisms. As discussed above, the most rapid changes may actually be more likely to show negative pleiotropic changes in other conditions, making them unlikely to contribute to adaptation in the long term. Gene loss or the loss of regulatory mechanisms is a good example. Gene or chromosomal region amplification and aneuploidy are also molecular mechanisms that have been described as fast and easy short-term adaptive solutions. They may allow a population to survive or grow in response to sudden abrupt change but may be too costly to prevail in the long term [5,22,87]. One can imagine that these simple adaptive changes will be replaced by more refined regulatory mechanisms in the long term.
One of the major investigations to come will be to better understand the molecular bases of second-order effects (background-dependent pleiotropy, environment-dependent signs of pleiotropy) that have been brought to light through the experiments reported above. For instance, why are mutations that inactivate the same molecular pathway neutral in some cases and deleterious in others? How and why mutations that lead to rescue in one condition affect the course of evolutionary rescue in other conditions?
A better understanding of the complex effects of environmental fluctuation will enable better prediction and manipulation of the course of evolution in applied contexts, for instance, to predict the outcomes of cancer drug and antibiotics treatments [88–90]. For example, it was shown that spatial heterogeneity in drug concentration in the tumour microenvironment is associated with metastatic cell motility along the gradient, which facilitates the emergence of acquired resistance to cancer therapy . Knowing that a given antibiotic resistance leads to trade-offs in resistance to other antibiotics has led to the use of drug combinations to slow the evolution of resistance [70,71,92]. The same principles apply to any other situation where humans impose a selective pressure on other organisms that will eventually adapt, for instance in agriculture and pest control . Knowing how adaptation to a first condition could lead to molecular changes that become liabilities in other conditions would allow the exploitation of these weaknesses for treatments or intervention. Understanding evolution at the molecular level is, therefore, crucial in many fields outside of evolutionary biology.
We declare we have no competing interests.
This work was supported by a NSERC discovery grant to C.R.L. and a PROTEO Scholarship to C.B. C.R.L. holds the Canada Research Chair in Evolutionary Cell Biology.
We are grateful to the members of the Landry laboratory and Nadia Aubin-Horth for helpful comments on earlier drafts. We thank the two anonymous reviewers whose comments helped improve and clarify this manuscript. We thank Dr Per Lundberg for the invitation to write this review.
- Received June 30, 2016.
- Accepted October 4, 2016.
- © 2016 The Author(s)
Published by the Royal Society. All rights reserved.