SEARCH

SEARCH BY CITATION

Keywords:

  • beneficial mutation;
  • effective population size;
  • experimental evolution;
  • limits to selection;
  • response to selection;
  • Saccharomyces paradoxus

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The purpose of this experiment was to find out how a population becomes adapted to extremely stressful conditions as its environment deteriorates. We created a deteriorating environment for experimental selection lines of yeast by a stepwise increase in the concentration of salt in the growth medium. After each step, we tested the ability of the lines to grow at a high concentration of salt near the lethal limit for the ancestral strain. We found that mutations enhancing growth in this highly stressful environment began to spread at intermediate salt concentrations. The degree of enhancement was related to effective population size by a power law with a small exponent. The effect size of these mutations also increased with the population size in a similar fashion. From these results, we interpret adaptation to lethal stress as an indirect response to selection for resistance to previous lower levels of stress in a deteriorating environment. This suggests that the pattern of genetic correlation between successively higher levels of stress is an important factor in facilitating evolutionary rescue.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Whether or not a population adapts to a new stress through natural selection depends on the severity of the stress, the rate at which it is applied and the availability of genetic variation in stress response (Lynch & Lande, 1993; Gomulkiewicz & Holt, 1995; Bell & Collins, 2008). If a lethal stress is abruptly applied, then the population will become extinct before any adaptation can occur. If the stress is applied more gradually, through a series of sublethal episodes of gradually increasing severity, then the population may survive. This reflects a general principle that adaptation is seldom if ever perfectly specific. The physiological systems that enable a species to survive when temperature is raised by 10 °C, say, are likely to enable it to survive almost equally well if the temperature is raised by 11 °C. If phenotypic effects were not transferable, at least in part, from one set of conditions to a similar set of conditions, then adaptation to a changing environment would seldom occur, because each trivial shift in conditions would require the selection of a completely different phenotype. Hence, the mutations that are ultimately responsible for the survival of a population exposed to a stress that would have been lethal to the ancestral lineage must often appear and spread during previous exposure to severe but sublethal stress.

A very simple situation is presented by a genetically uniform asexual population exposed to a toxic stress. If the stress is imposed abruptly it extinguishes the population, but if the population is propagated by serial transfer into successively higher concentrations of the toxin, beginning at a harmless level, a fully adapted population may eventually evolve. This design has been used effectively by experimental evolution experiments to investigate the evolution of resistance to toxins such as antibiotics (Perron et al., 2006) and nucleic acid-binding agents (Saffhill et al., 1970). As the stress tolerated by the evolved population is lethal to the ancestor, a beneficial mutation, or series of mutations, must have spread during culture at intermediate concentrations. The likelihood that this will happen depends on the mutation supply rate, the effect of beneficial mutations on fitness, and the correlation between the effect of a mutation at sublethal and lethal concentrations.

If the mutation rate is taken to be fixed, the mutation supply rate is determined by population size. In a serial transfer experiment, this varies continually, and the appropriate average, or effective population size, is the harmonic mean (see Lenski et al., 1991). For a population that doubles in size every generation for G generations, the harmonic mean size is {1/[(1/N0) + (1/2N0) + (1/4N0) + … + (1/2GN0)]}/G. If G is fairly large the harmonic mean is nearly (2/N0)/G, and as NG = 2GN0 the effective size is given by Ne ≈ ½N0 log2 (NG/N0), where N0 is the initial (inoculum) census size and NG is the final (transfer) size. Hence, Ne is proportional to N0 but only to the logarithm of NG.

If the conditions of growth remain constant for a sufficiently long period of time then any population will attain the same level of adaptation, regardless of its size, provided that mutations have independent effects on fitness. However, we are concerned with the case where conditions are changing, such that there is a relatively short period of time for adaptation to occur. The level of adaptation that is attained will then depend on the probability that at least one beneficial mutation will appear and spread before conditions become lethal, and this depends on the mutation supply rate, and therefore on the effective population size. If beneficial mutations were very rare, such that each is always fixed before the next arises, then the rate (or probability) of adaptation will be proportional to Ne. This is the conventional model of evolution in asexual populations. If beneficial mutations are more frequent, on the other hand, several mutations may be spreading at the same time, and in an asexual population adaptation will be retarded, because they collectively raise the mean fitness of the population and thereby reduce the intensity of selection. Hence, the rate of adaptation will in general increase less than proportionately with Ne because of this process of clonal interference (Gerrish & Lenski, 1998).

Clonal interference also leads us to expect that successful beneficial mutations will have larger effects on fitness in bigger populations. This is because the greater number of mutations simultaneously present will have a greater range of variation and thus a larger maximum value of fitness. The beneficial mutations of largest effect are the most likely to be fixed, so that adaptation in larger populations is likely to be attributable to mutations of larger effect.

The genetic correlation of fitness is related to the difference in mean fitness caused by altering a single environmental factor or set of factors (Bell, 1992). The correlation between sublethal and lethal conditions will therefore increase with the severity of the sublethal effect. We may expect, therefore, that mutations conferring resistance to lethal stress will appear and spread at intermediate levels of stress, rather than in the mildest conditions.

There have been several previous attempts to evaluate these ideas experimentally (see the review by De Visser & Rozen, 2005). In particular, De Visser et al. (1999) created a range of mutation supply rates by manipulating inoculum size and mutation rate (through the use of isogenic mutator strains). They found that the rate of adaptation of glucose-limited batch cultures of Escherichi coli was an increasing but non-linear function of the mutation supply rate for poorly adapted ancestors, whereas no non-linear trend was found for well-adapted ancestors. They argued that a greater range of beneficial mutations is available when the ancestor is poorly adapted, and thus a greater scope for clonal interference. Miralles et al. (2000) likewise found that the rate of adaptation of an RNA virus grown in tissue culture increased with effective population size. Burch & Chao (1999) found that beneficial mutations fixed in evolving populations of virus ϕ6 had larger effect size in larger populations. The same pattern was reported for populations of E. coli evolving in glucose-limited minimal medium (De Visser & Rozen, 2005).

We have extended these studies by using yeast cultures grown at high concentrations of salt (NaCl). Unlike previous studies, our cultures were exposed to highly stressful conditions and assayed in conditions lethal to the ancestor. We have also used a more finely calibrated series of population sizes than in other reports, allowing the effect of clonal interference to be precisely quantified. Previous work with this system has shown that the probability of adaptation to abrupt severe stress is directly related to inoculum size (Bell & Gonzalez, 2009). This demonstrated the evolutionary rescue of populations initially possessing a low level of variation in stress resistance. The experiment we describe here demonstrates evolutionary rescue attributable to novel beneficial mutations in a deteriorating environment.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Strains and media

Our base population was a single colony of the reference strain CBS432 of Saccharomyces paradoxus, the wild sister species of domesticated yeast, Saccharomyces cerevisiae. It is a wild type isolated recently from oak bark and carries no auxotrophic markers. The culture medium for all experiments was yeast peptone dextrose (YPD medium: 20 g peptone, 10 g yeast extract, 20 g dextrose, to 1 L with distilled water) supplemented with a specified concentration of NaCl.

Pilot experiment

We performed a pilot experiment to identify a salt concentration that would elicit adaptation within the duration of the main experiment. We grew 60 replicate lines in microwell plates at a range of concentrations of NaCl: 0, 2, 5, 10, 20, 40, 80, 120 and 150 g L−1. The lines were acclimated by two transfers in YPD. Thereafter, the cultures were transferred (10 uL into 150 uL medium) and optical density (OD) measured after 48 h growth at 28 °C at a specified salt concentration. Nine transfers were completed in this way. Two complete replicates (blocks) of this experiment were performed.

Main experiment

We used eight treatments to manipulate culture volume: 20 uL (wells in 384-well plate); 50 uL (wells in 96-well plate); 250 uL (wells in 48-well plate); 1.25 mL (wells in 24-well plate); 4 mL (wells in 6-well plate); 10 mL (ditto); 30 mL (glass flask); and 75 mL (ditto). At each level, the inoculum volume was 5% of culture volume, so that lines experienced the same conditions throughout growth.

Twelve replicate lines for each treatment were maintained as shaken cultures at 28 °C and transferred every 24 h. They were transferred five times in YPD supplemented with an increasing concentration of NaCl: 0, 5, 10, 15, 20, 25, 35 and 45 g L−1, for a total of 40 transfers. Hence, each line passed through the same series of salt concentrations, representing a deteriorating environment in which each line experienced a gradual increase in stress. The final level represents a concentration at which the pilot experiment had identified a steep decline in culture OD.

After the final transfer at each concentration, four representative lines were tested by thin spreads on YPD agar plates supplemented with 80 g L−1 NaCl. Individual cells on the agar surface are clearly visible under the dissecting microscope. After 4 days growth, we scored the number of doublings for 200–300 microcolonies for each line. At the end of the experiment, with medium containing 45 g L−1 NaCl, we carried out this assay for the remaining eight replicate lines for each culture volume. Average values differed between the two assays, presumably because of some unidentified change in the lab environment, so the four lines assayed earlier were assayed again. This gave consistent results, except that one set of four lines, all with a culture volume of 1.25 mL, scarcely grew at all, and was omitted from subsequent analyses. The average number of doublings was our basic estimate of the ability to grow at 80 g L−1 NaCl.

The average number of doublings was calculated over all categories of microcolony size (i.e. 1, 2, 4, 8 … cells). Any difference between treatments could be attributed to two causes: (i) an increase in the frequency of colonies of some given size, caused by an increase in the rate of substitution of beneficial mutations of given effect; or (ii) a shift in the frequency distribution towards larger colonies, caused by an increase in the average effect of beneficial mutations. To detect any change in effect size, we picked the four largest colonies from each line from 80 g L−1 NaCl plates after 9 days’ growth and re-spread to give 5–20 cells per plate to minimize nutrient competition between growing colonies. Colony diameter after 9 days’ growth was used to express the size of the effect on fitness at 80 g L−1 NaCl of the beneficial mutations appearing in each of the treatments.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Response to salt

Growth over a range of salt concentrations over eight transfers in the pilot experiment is shown in Fig. 1. OD decreased linearly from 0 to 40 g, but within this range cultures could be transferred successfully and reached final densities in excess of 107 cells mL−1. At 120 g L−1 NaCl, there is barely detectable growth that improves slightly over time; at 150 g L−1 NaCl, there is no detectable growth. Growth at 80 g L−1 NaCl would barely support dilution by transfer at first, but improved substantially during the course of the experiment. We interpreted this as adaptation through natural selection to a stress that is lethal or nearly lethal to the ancestor. We therefore chose 80 g L−1 NaCl as the test environment for the main experiment, being assured that beneficial mutations of large effect could be expected to occur and spread.

image

Figure 1.  Time course of growth at different concentrations of salt. Two replicate experiments each with 60 replicate lines are represented at each transfer. The first transfer into each treatment is affected by carry-over from prior acclimation in yeast peptone dextrose (YPD) and is not shown. The dashed lines are the upper and lower 95% confidence limits of uninoculated wells.

Download figure to PowerPoint

Adaptation to high salt concentration

The mean growth of the lines on 80 g L−1 NaCl YPD plates is shown in Fig. 2. Until salt concentration reached 35 g L−1, there was very little growth when the cultures were tested on plates with 80 g L−1 NaCl, and no large microcolonies (more than eight doublings) were found. Large microcolonies first appeared in some of the lines with culture volumes of 4 mL or more at 35 g L−1. They were much more abundant at 45 g L−1, at which point there was a sharp increase in mean doubling rate (Fig. 2). The selection experiment was therefore terminated at this point, when growth of the lines was compared among population size treatments. As the rate was so low for other conditions, the mean doubling rate of the 45 g L−1 assay is essentially the response to selection at this point.

image

Figure 2.  Growth at 80 g L−1 NaCl of lines in relation to their previous exposure to NaCl. When cultured at less than 15 g L−1 NaCl, no growth at 80 g L−1 was found; a few resistant colonies appeared on 80 g L−1 plates as the concentration was raised from 15 to 35 g L−1; there was a large increase in the frequency of resistant colonies once the lines had adapted to 45 g L−1 NaCl.

Download figure to PowerPoint

Effect of population size on doubling rate

Mean doubling rate D increased consistently with culture volume, and thus with effective population size Ne. The relationship is much better fitted by log-transforming both axes (r2 = 0.62; Fig. 3) than by transforming Ne alone (r2 = 0.48), D alone (r2 = 0.33) or neither (r2 = 0.38). Hence, the relationship between response to selection and effective population size is best represented by the power law inline image with β << 1. The appearance of this relationship on arithmetic axes is that doubling rate at first increases sharply with Ne, but then becomes almost (but not quite) asymptotic.

image

Figure 3.  Increase in doubling rate at 80 g L−1 NaCl in cultures of different volume at 45 g L−1 NaCl. Each point is a mean for each of 12 replicate lines, based on 200–300 colonies per line scored after the growth of 4 days. The regression is log doublings = 0.17 log Ne − 1.20, r2 = 0.62 (d.f. = 90, P < 0.001).

Note: Four aberrant plates from the fourth smallest treatment were excluded; if they are included, r2 drops to 0.39 (d.f. = 94, P < 0.001).

Download figure to PowerPoint

Effect of population size on effect size

The size of the largest colonies isolated on 80 g L−1 plates also increases with population size and is well described (r2 = 0.88 for treatment means) by a power law with exponent β << 1 (Fig. 4). To check that colony size was not a physiological response to culture volume, we grew colonies from small-volume lines in both small and large volumes and did the same for colonies from large-volume lines. The difference in colony size between small-volume and large-volume treatments was maintained regardless of the assay culture volume (data not shown), so we concluded it was a true effect of treatment.

image

Figure 4.  Increase in colony size at 80 g L−1 NaCl in cultures of different volume at 45 g L−1 NaCl. Colony size is diameter in mm after 9 days of growth. Each point is the mean over lines, with error bars +1 SD. The regression equation is log colony size = 0.15 log Ne − 0.53, r2 = 0.88 (d.f. = 6, P < 0.01).

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Selection of beneficial mutations in finite asexual populations

The mutation supply rate is the product of the effective population size, the genome size (1.25 Mb) and the fundamental rate of mutation (about 4 × 10−10 bp−1 per replication: Lang & Murray, 2008; Lynch et al., 2008). This ranges from log NeU = 1.5 for the 20 uL lines to log NeU = 5.1 for the 75 mL lines. We define the effective mutation supply rate as the overall rate of mutation of genes that govern the physiological response to a given stress. This is equivalent to NeγU, where γ is the fraction of the genome in which beneficial mutations conferring resistance to the given stress can occur. These genes can be most readily identified by a substantial impairment of growth specific to the given stress when they are deleted, if we allow that loci liable to loss-of-function mutations are likewise liable to gain-of-function mutations. Warringer et al. (2003) found that 488 deletion strains were significantly impaired in salt medium, although the effect was slight in most cases. Giaever et al. (2002) found that 62 strains had severely reduced competitive ability in salt medium, constituting about 1% of the genome. This is perhaps the most appropriate estimate of the number of loci potentially affecting adaptation to high concentrations of salt, although it is an underestimate because single-gene deletion does not produce a phenotype for duplicated genes, such as the ATPases involved in Na+ efflux. More generally, the number of conditionally lethal deletions specific to an arbitrary physical factor in yeast appears to be on average about 20–40 (G. Bell, in preparation), based largely on studies involving media supplemented with small molecules. This suggests that typical values for γ lie in the range 0.003–0.01.

Given that γ ≈ 0.01 for salt, the effective mutation supply rate in our experiment ranged from NeγU < 1 for the 20 uL lines to NeγU >> 1 for the 75 mL lines. Only a small fraction of these mutations will be beneficial, however. This fraction has no definite value, because it depends on the state of adaptation of the population, being larger when the loss of fitness caused by a stress is greater. In our experiment, the number of exceptionally large microcolonies (more than eight doublings) increased sharply between the 10-mL and 30-mL treatments. If a single large-effect mutation out of the NeγU mutations expected to occur were responsible for all the large microcolonies observed in each line, the fraction of mutations that were substantially beneficial at 80 g L−1 salt would have been between 0.002 and 0.006. The most precise estimate of the beneficial mutation rate at a nucleotide level is the study of an RNA virus by Sanjuan et al. (2004), who found 2/29 viable single-nucleotide mutations were beneficial. They used a chimaeric ancestral sequence that had not been cultured long under laboratory conditions and may well have been poorly adapted. Other virus experiments, however, yield much lower estimates: the initial probability that a random mutation will increase fitness is about 0.0034 for phage ϕX174 adapting to high temperature (Bull et al., 2000) and about 0.0009 for RNA vesicular stomatitis virus (VSV) adapting to a new host cell type (Cuevas et al., 2002).

Gerrish & Lenski (1998) show that the rate of fitness improvement is expected to increase with population size following a power law up to a value of one beneficial mutation per generation in the population as a whole, after which it remains essentially constant. The estimates of mutational target size and the fraction of mutations that are beneficial we have given above suggest that the power law response continues to hold for supply rates of up to about 25 beneficial mutations per generation. The uncertainties involved in estimating this figure are considerable; nevertheless, a power law seems to be an adequate empirical description of the effect of population size on adaptation to extreme stress over the range covered by our experiment, roughly from Ne = 105 to Ne = 108.5.

Selection of genetic variation in finite sexual populations

The rate of adaptation in multicellular sexual eukaryotes is also likely to be affected by population size, perhaps largely through its effect on standing genetic variation rather than the mutation supply rate. The field has been reviewed by Willi et al. (2006). Weber (1990) and Weber & Diggins (1990) applied artificial selection to Drosophila populations and related the short-term response after 50 generations, presumably generated almost exclusively by the recombination of alleles present in the ancestral population, to population size. Their results, together with a few other studies, are well fitted by a logarithmic function (Bell, 2008), although a power law with β << 1 is nearly as successful (data not shown), and to this extent artificial selection in Drosophila populations follows a similar rule to natural selection in our yeast lines.

Evolutionary rescue

Our experiment suggests a ‘creeping barrage’ interpretation of adaptation to deteriorating conditions of growth: adaptedness continually advances from less stressful to more stressful conditions. At the current level of stress, mutations that are able to grow more rapidly than average will tend to spread. Some of these mutations will be fortuitously able to survive somewhat more stressful conditions, enabling the population to persist as conditions become worse. They then provide the basis for further adaptation to even more stressful conditions. This amounts to interpreting adaptation to extreme stress as an indirect response to selection in less stressful conditions. In general, the indirect response to selection R’ of a character y (e.g. growth in highly stressful conditions) will be equal to the direct response R of a character z (e.g. growth in less stressful conditions) depreciated by the genetic correlation of y with z and scaled by the ratio of their genetic standard deviations: R’ = ryzyz) R (see Falconer, 1981). The overall indirect response integrates the direct responses to all prior states of a deteriorating environment. Adaptation to extreme stress is thus prefigured by the cross-environmental genetic correlation of mutations selected in previous environments.

This interpretation emphasizes two features of a population experiencing a deteriorating environment. The first is the rate of deterioration itself, since if this is too rapid, mutations that enhance growth at both current and future levels of stress will not have sufficient time to spread. The rate of deterioration in the physical environment, at least, can often be measured straightforwardly. The second is the pattern of genetic correlation between current and future environments. Evolutionary rescue is much more likely when these correlations are positive and large, but in practice they are difficult to estimate in natural populations. Our experiment suggests that the crucial value for this correlation lies near the point where any further increase in stress causes a precipitous decline in growth. However, much more extensive work will be required to understand more clearly how prior adaptation facilitates adaptation to lethal stress.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

This work was funded by the Natural Sciences and Engineering Research Council of Canada. Saccharomyces paradoxus CBS432 was a gift from Vassiliki Koufopanou, Imperial College, London.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References