The Princeton Guide to Evolution

III.6

Responses to Selection: Experimental Populations

Graham Bell

OUTLINE

1. Will adaptation evolve?

2. How fast will adaptation evolve?

3. Does sex accelerate adaptation?

4. Is adaptation gradual or saltational?

5. What is the limit to adaptation?

6. Is adaptation based on gain or loss of function?

7. Is adaptation repeatable?

8. Is adaptation predictable?

9. Is adaptation reversible?

10. How do ancestry, selection, and chance contribute to adaptation?

11. How can selection maintain diversity?

12. What limits the extent of specialization?

Evolutionary biology has been an observational and comparative science for most of its history because “natural selection always acts with extreme slowness” (Darwin 1859, p. 121) and therefore produces adaptation “by minute steps, which, if useful, are augmented in the course of innumerable generations” (Weismann 1909, p. 24). Artificial selection in crop plants and domestic animals was from the first used to justify the general principle of modification, but the deliberate choice of breeding individuals by human agency made it only a simile for natural selection. A century passed before the invention of the chemostat led to the realization that microbial cultures, with their huge populations and rapid turnover, could act as time machines enabling us to investigate evolutionary change through natural selection in real time. One path led to a series of brilliant biochemical studies showing how individual enzymes and whole metabolic pathways could evolve (Mortlock 1984). Another led to investigations of more general evolutionary processes (Dykhuizen 1990), and this research has continued to the present day with experiments of increasing scope and power. The promise of experimental evolution is to provide decisive tests of specific hypotheses about adaptation, the core process of evolution. Questions such as what would happen if the tape of life were replayed or how is sex maintained have elicited endless debate, yet they can now be settled by experiments using laboratory microcosms. In this brief note—by no means a review of the field (see Bell 2008)—I shall list a dozen basic questions about adaptation and describe briefly how they have been illuminated by selection experiments in the laboratory. Most of them involve the very simplest scenario of asexual unicellular microbes growing in homogeneous culture medium—a small, cloudy tube of bacteria, yeast, or algae.

GLOSSARY

Adaptation. In the sense used here, a change in the genetic composition of the population caused by natural selection and resulting in elevated fitness in defined conditions.

Beneficial Mutation. Any genetic change that causes elevated fitness and thereby contributes to adaptation.

Evolutionary Rescue. The survival of a population in conditions lethal to its ancestor as the consequence of adaptation through natural selection.

Experimental Evolution. The study of natural selection under controlled conditions, usually in the laboratory.

Microbe. A unicellular organism, usually a bacterium, yeast, or alga.

Selection Experiment. The repeated propagation of a population in controlled conditions with the object of discovering how it adapts to those conditions through natural selection.

1. WILL ADAPTATION EVOLVE?

Adaptation can be demonstrated in the laboratory simply by subjecting a microbial culture to a stress such as high temperature or a toxic chemical. As a sample of the culture is repeatedly transferred to fresh medium, any variant type that is resistant to the stress will tend to spread, replacing more sensitive types, thus raising the average rate of growth. This process is bound to occur when there is heritable variation in fitness, and is driven by differences in relative fitness among types within the population, whether these types are already present when the stress is applied or arise afterward by mutation and recombination. It is not bound to produce permanent results, however, because the stress may be so severe that even the most resistant types are incapable of sustained growth. The population will then diminish, transfer after transfer, until eventually it becomes extinct.

Hence, we can recognize three levels of stress with different evolutionary outcomes. If only a mild stress is applied, the population will continue to have a positive rate of growth. The population may well evolve, because the new conditions cause shifts in relative fitness, but its persistence does not depend on a higher level of adaptation. On the other hand, if a very severe stress is applied, the population is certain to become extinct sooner or later. At intermediate levels of stress, the population at first declines but may recover later if resistant types with positive rates of growth have spread. The signature of evolutionary rescue is the U-shaped trajectory of abundance as collapse is followed by recovery. The boundary between the zones of recovery and extinction is set by variation in absolute rather than relative fitness, which depends on the presence of genotypes with positive rates of growth, which is more likely in larger populations because these encompass a broader range of variation. This has been demonstrated by culturing yeast populations of different sizes in medium with high concentrations of salt and showing that rescue occurs in those sufficiently large to include one or two resistant cells.

2. HOW FAST WILL ADAPTATION EVOLVE?

When a population is stressed, its mean fitness at first declines and then increases as it becomes better adapted through natural selection. The rate at which it becomes adapted depends solely on the amount of variation in fitness. If this is small, there is little difference between the least fit and the most fit types in the population, so that selection will be weak and adaptation slow. Conversely, a large amount of variation in fitness implies strong selection and rapid adaptation (see chapter III.5). For the population to become permanently modified, this variation must be heritable, of course, leading to the conclusion that the rate of adaptation is equal to the heritable variance of fitness (Fisher 1930).

This classical result is always valid for infinitely large populations. It will not always accurately predict the rate of adaptation of finite populations, however, because the most fit types may be absent, so that adaptation is slow until they are generated by mutation or recombination. In this case, adaptation will occur in two stages: a waiting stage, before the first high-fitness type appears, and an establishment phase, during which it spreads to fixation. The dynamics of the establishment phase are indeed governed entirely by the difference in fitness between the superior type and the average of the population, and are (almost) independent of population size. The length of the waiting period, on the other hand, is inversely proportional to the rate of supply of beneficial mutations, which is higher in larger populations: more beneficial mutations are likely to occur within a given span of time in larger than in smaller populations. Hence, the rate of adaptation should increase with population size. This can be tested experimentally by allowing cultures of different volume to evolve under stress: after a prescribed number of transfers, the larger cultures, containing more cells, are found to be better adapted. The relationship seems to be a power law, with each doubling of population size leading to the same small fractional increment in the rate of adaptation.

3. DOES SEX ACCELERATE ADAPTATION?

The function of sex has long been vigorously disputed by evolutionary biologists (see chapter III.9). The prevailing hypothesis (although by no means the only one) is that the fusion of gametes followed by the recombination of their genomes generates genetic variation, with the effect of accelerating adaptation, a view dating back to Weismann. In the simplest terms, beneficial mutations arising in different lineages can be brought together by fusion and packaged together in the same lineage by recombination. The evolutionary effect of sex is best studied in eukaryotic microbes such as yeasts and unicellular algae, because they all have periods of purely vegetative growth between sexual episodes. In organisms like these, where sex can be controlled by environmental or genetic manipulations, it is possible to set up populations that are equivalent in every way except that some go through a sexual cycle of fusion and recombination from time to time, whereas others are perennially asexual. Experiments like this have shown that when sex is induced in a population growing in a novel and stressful environment, its immediate effects are a drop in average fitness, relative to a comparable asexual line, accompanied by an increase in the variance of fitness. The increased variation generates stronger selection, eventually driving the fitness of the sexual line above that of the asexual line. Over a period of several hundred generations, sexual populations thereby adapt faster than purely asexual populations in stressful environments; moreover, this effect depends on population size, because it is only in large populations that two or more beneficial mutations are likely to occur at the same time, and therefore only in large populations that fusion and recombination are likely to accelerate adaptation. Consequently, sex accelerates adaptation much more effectively in large than in small populations.

4. IS ADAPTATION GRADUAL OR SALTATIONAL?

The classical account of evolution is gradualist: populations become adapted through the substitution of beneficial alleles of small effect at many loci over long periods of time (see chapter V.12). Generally speaking, this is not borne out by laboratory experiments, in which adaptation can occur very rapidly through the substitution of a few alleles of large effect. It is true that most of the beneficial alleles that appear by mutation when a population first experiences stressful conditions have rather small effects on fitness. This can be demonstrated by isolating mutants resistant to an antibiotic and then measuring their fitness in the absence of the antibiotic. This provides the distribution of fitness of new mutations at the time when they first arise. A few of these mutants are fitter than their ancestor, but most are only slightly superior, and very few are much more fit. If such beneficial mutations are allowed to spread, however, and collected only when they have become fixed, a very different picture emerges: the bulk of these fixed beneficial mutations have large effects, often amounting to a doubling of fitness. The reason is that beneficial mutations that increase fitness only slightly, although they may be very numerous, are likely to be overtaken by the much faster spread of mutations that greatly increase fitness, despite their rarity. The rapid spread of large-effect beneficial mutations is often observed in laboratory experiments. This provides a concrete alternative to the gradualist interpretation of adaptation, especially when populations are severely stressed.

5. WHAT IS THE LIMIT TO ADAPTATION?

If conditions remain unchanged, the first few mutations to be fixed may well increase fitness substantially, but the supply of these large-effect mutations is sure to be limited, and as the supply is depleted, only those of more modest effect remain available for selection. Consequently, the rate of adaptation will tend to diminish over time. This pattern has been demonstrated by long-term serial-transfer experiments with E. coli in which replicate lines have been maintained in a simple glucose medium for tens of thousands of generations. These increased rapidly in fitness over the first 2000 generations, but the rate of adaptation was clearly decreasing throughout this period, and was much slower in the subsequent 20,000 generations. This illustrates a process whereby adaptation is driven by mutations of smaller and smaller effect as time goes by. Nevertheless, although adaptation decelerated, it did not stop completely: some improvement is still being made after 50,000 generations in culture. Hence, there may be no definite limit to adaptation, but rather a continuously diminishing response, with alleles of diminishing effect being substituted at longer and longer intervals. Moreover, the supply of large-effect mutations, while it will become very meager, may not vanish completely. A very rare mutation conferring the ability to utilize citrate as a carbon substrate appeared in the long-term lines after about 35,000 generations and resulted in a large increase in fitness. It is not safe to conclude that a population has lost all capacity to adapt, even after a very long period of observation. In practice, of course, conditions will seldom remain constant for tens of thousands of generations, so the prolonged improvement of lines in uniform laboratory conditions implies that natural populations may be actively and often rather rapidly adapting most of the time.

6. IS ADAPTATION BASED ON GAIN OR LOSS OF FUNCTION?

Fitness is increasing over time in any population adapting to novel conditions, and in this sense adaptation always involves a gain of function. At a genetic or biochemical level, however, this is often based on a loss of function, in the sense that a particular chemical reaction can no longer be carried out. This may occur because a reaction that promotes growth in most conditions becomes unnecessary in highly simplified laboratory culture, and variants that lose the ability to conduct it gain an advantage because of some economy in time or materials. The uptake of glucose by bacteria, for example, is regulated by a series of genes that switch on transport systems when glucose is available and switch them off when it is not. When experimental lines are cultured in glucose medium, mutations that cause defects in this regulatory system and leave glucose transport permanently switched on are often among the first to spread. This increases fitness, in laboratory conditions where glucose is continuously supplied, through the degradation of a coordinated biochemical pathway. Once these loss-of-function mutations have become established, however, the stage is set for gain-of-function mutations to spread. These are mutations that alter, rather than degrade, a reaction so as to create a new biochemical ability. The experimental evolution of amide utilization illustrates this sequence. The ancestral strain expresses an amidase enzyme enabling it to use acetamide, the simple two-carbon amide responsible for the characteristic odor of mouse cages. The amidase is hardly able to process the four-carbon amide butyramide at all, however, so that if butyramide is supplied as the sole carbon source, growth is very slow. Among the first beneficial mutations to appear are alleles conferring faster growth simply by expressing very large quantities of the amidase, for example, by loss-of-function mutations in the genes that regulate its expression or by increasing the number of copies of the structural gene. Once these have become fixed, further adaptation is based on a gain-of-function mutation in the gene encoding the amidase that alters its structure so that it processes butyramide more efficiently. The ability to utilize larger and more complex amides can subsequently evolve in a similar fashion, through the successive modification of the amidase by gain-of-function mutations to increase its activity on particular substrates. The evolution of new metabolic capabilities by bacteria often begins with exaptation, the use of an inefficient enzyme normally responsible for degrading some other substrate; this is followed by deregulation and amplification to increase the supply of this inefficient enzyme and culminates in the modification of the enzyme to produce a new and more efficient version. The exact course of this exaptation-deregulation-amplification-modification (EDAM) process varies from case to case, but the transition from loss-of-function to gain-of-function mutations during the course of adaptation may be quite general.

7. IS ADAPTATION REPEATABLE?

The spread of loss-of-function mutations affecting glucose uptake in bacterial cultures growing in glucose-limited medium has been repeatedly observed and can be confidently expected whenever such experiments are conducted. The specific mutations involved, however, differ from case to case. There are several genes in the outer and inner membranes of the cell that contribute to the regulation of glucose uptake, and any of them may be affected; moreover, the mutations themselves may be substitutions of single nucleotides, or frameshifts, or short insertions or deletions. There are many other examples of similar changes occurring in response to the same agent of selection. The long-term E. coli lines, for example, have consistently lost the ability to utilize ribose as a substrate for growth. In other cases, however, the same agent of selection leads to different outcomes in replicate selection lines. Bacteria cultured on a range of different substrates will often become adapted to each of them, evolving higher rates of growth; at the level of fitness, adaptation is quite highly repeatable. Whether the same underlying genetic changes are responsible in each case can be evaluated by culturing all replicate lines that have adapted to a single particular substrate on each of the other substrates: they will show the same pattern of growth on these exotic substrates if they have acquired the same genetic changes, but different patterns of growth otherwise. In one extensive experiment involving nearly a hundred different substrates, there was little correlation in most cases between lines cultured on the same substrate, indicating a low level of repeatability. The cause may be the historical nature of adaptation, arising from the stochastic appearance of mutations and the interaction between their effects.

The repeatability of adaptation can be assessed at a genomic level by obtaining complete sequences and expression profiles for replicate lines under uniform selection. In the long-term E. coli lines, similar changes in expression evolved in a limited suite of about 50 genes, although the details often differed between lines, for example, by the production of similar changes in gene expression by different regulatory mutations. Likewise, a small group of genes consistently had modified sequences in the lines, although the particular mutations that had occurred varied from line to line. At present, it seems that microbial adaptation in uniform laboratory conditions often involves a few themes and many variations. The few themes are the major genes where beneficial mutations can occur; the variations are the many possible alleles of these genes.

8. IS ADAPTATION PREDICTABLE?

The spread of loss-of-function mutations in glucose transport systems is not only repeatable but also predictable from first principles. In other cases, such as loss of ribose metabolism in the long-term E. coli lines, the event is repeatable, but the reason for it is not understood. Where genetic changes are tightly coupled to fitness, the course of adaptation is usually rather highly predictable. Lactose metabolism in E. coli depends on three processes acting in succession: diffusion through pores in the cell wall, active transport across the cell membrane by a permease, and hydrolysis by β-galactosidase in the cytoplasm to split the molecule into glucose and galactose. The effect of altering the activity of the permease or β-galactosidase on the overall flux through this pathway can be calculated from biochemical principles and compared with the observed relative fitness of a series of mutants in lactose-limited medium. Since the flux through the pathway is equivalent to fitness when lactose is the sole carbon source, the biochemical and evolutionary estimates should coincide, and experiments show that they are indeed very highly correlated.

A more extensive attempt was made to predict evolutionary change in the genome of phage T4. This small genome is better understood than any other, with each of its 288 genes being well characterized at a molecular level; hence, it should be possible to predict the genetic basis of adaptation to a stress such as high temperature with a high degree of confidence. When this was tested by experiment, about one-half of all changes were correctly predicted, at least to some extent, such as the gene or region of the gene in which they should occur, whereas the other half were unexpected despite our deep functional knowledge of the genome. This can be seen as a glass half full: the demonstrable possibility of predicting a large proportion of evolutionary changes in this small genome clearly foreshadows eventual success in predicting the response of larger genomes in cellular organisms.

9. IS ADAPTATION REVERSIBLE?

A population that becomes adapted to a novel environment will usually recover a high level of adaptedness to ancestral conditions once these are again imposed. Any change in the morphological, physiological, and developmental features that contribute to adaptation, however, may not be as readily reversible (see chapter VI.12). This follows from the historical nature of evolutionary change, with each step in the adaptive walk arising stochastically and depending for its success in some degree on the modifications that have evolved previously. The simplest and most dramatic experiment is to delete an essential gene and investigate how a population can recover the ability to grow. If the structural gene encoding β-galactosidase is deleted in E. coli, the population is unable to grow on medium with lactose as the sole carbon source. Supplementing normal growth medium with lactose, however, permits the cells to grow while creating strong selection for any mutants able to utilize lactose once all other substrates have been used up. This procedure eventually leads to the evolution of populations able to grow slowly on lactose alone. This is not attributable to the restoration of the original β-galactosidase gene, of course, which would involve a very long and improbable series of mutations; rather, a different gene, present in the ancestor and producing an enzyme able to hydrolyze lactose at a very low rate, has become modified and appropriately regulated through a series of five beneficial mutations that together confer the ability to grow on lactose alone.

The irreversibility of evolution is not only an academic issue. It has often been suggested that the resistance that often evolves in bacterial populations to any given antibiotic would disappear if the antibiotic were withdrawn completely, because of the resurgence of the original susceptible type. Experiments have shown, however, that although resistant populations readily become adapted to an antibiotic-free environment, they usually do not lose their resistance, often because compensatory mutations, occurring at loci not concerned with resistance, arise and spread by restoring normal rates of growth without affecting resistance to the antibiotic. An expanded program of laboratory selection experiments to investigate the evolutionary dynamics of antibiotic resistance might make an important contribution to public health.

10. HOW DO ANCESTRY, SELECTION, AND CHANCE CONTRIBUTE TO ADAPTATION?

The issues of repeatability, predictability, and reversibility are tied to the contributions of three processes to overall evolutionary change. The first is phylogenetic: a particular feature was inherited from more or less remote ancestors. The second is adaptive: it evolved through natural selection because it confers higher fitness. The third is neutral: once having arisen by chance, it persists because it has no appreciable effect on fitness. The contribution made by each of these three factors to biological diversity has been strenuously debated, for example, between those who believe that almost all features, from the most fundamental aspects of development to the most trivial details of morphology, have been precisely sculpted by natural selection and those who maintain that chance and history are often responsible. All three may affect any particular feature, although it is usually difficult to evaluate their relative contributions; in experimental situations, however, evolutionary change can be unambiguously partitioned between history, selection, and chance. In one very elegant experiment, E. coli lines that had become adapted to glucose medium were switched to medium in which maltose was the only carbon source. Some were able to grow well, whereas others grew only feebly. After several hundred generations their evolved ability to grow on maltose could be partitioned among the three fundamental processes. Selection will cause a general increase in growth in all the lines; if history has an effect, however, the differences among the ancestral lines will be retained among their descendants; and finally, the divergence of replicate lines founded from the same ancestral line is attributable to chance. In this case, almost all the variation among the maltose lines was attributable to selection, with only very small contributions from chance and history. This finding is consistent with those of most experiments in which exposure to a new environment elicits strong selection, leading to a rapid and consistent increase in fitness that largely effaces any historical signal. This does not necessarily apply to features other than fitness. In the maltose experiment, for example, cell size also changed over time, but history, selection. and chance contributed more or less equally to the final state of the lines. It is likely generally true that selection is almost exclusively responsible for overall adaptedness, whereas the morphological and physiological features that underlie adaptation have more complex roots.

11. HOW CAN SELECTION MAINTAIN DIVERSITY?

Selection drives the spread of highly adapted types and is therefore likely to lead to the fixation of the best type and the elimination of all others. In practice, populations are often rather diverse, and there has been much debate about whether diversity is attributable to forces such as mutation and recombination running counter to selection, or whether selection itself often acts to preserve diversity (see chapter III.3). In general, selection has the property of preserving diversity when rare types consistently have an advantage that is lost once they become more abundant. Although this may seem an unusual and onerous requirement, it is likely to occur whenever a range of growth conditions is available, as in most natural habitats. Simple media with a single limiting substrate such as glucose are laboratory artifacts: bacteria outside the lab grow in very complex media with many carbon sources, none of which is limiting. This creates an opportunity for specialization, because a genotype able to utilize any substrate with exceptional efficiency will tend to spread. If it depletes this substrate as it grows, however, the genotype necessarily limits the extent of its own spread. It will have high fitness when it is rare, when its preferred substrate is present at relatively high concentration by virtue of the scarcity of types that can consume it efficiently, but low fitness once it has become abundant, when its own consumption has removed the substrate from the medium. Hence, bacterial populations cultured in complex media become more diverse than comparable lines cultured in simpler media. Hence, populations growing in complex medium do not become dominated by a single generalist type with maximal growth on all substrates; nor do they consist of narrow specialists, each restricted to a single substrate. Rather, the outcome of selection in complex media is a set of imperfect overlapping generalists, each superior on a restricted range of substrates. This may be a plausible explanation of the broad but not unlimited metabolic diversity of natural bacterial communities.

The term substrate can be extended to denote any aspect of the conditions in which an organism is growing. Even the simplest laboratory microcosm then substantiates some irreducible level of complexity. A simple glass vial, for example, does not supply a perfectly homogeneous environment, because conditions at the surface will differ from those below. A culture of the soil bacterium Pseudomonas fluorescens in a glass vial that is left to stand on the bench for a day or two develops a thick mat at the surface, where the cells are stuck together by cellulose-like fibrils. These cells have evolved from the normal cells growing in the body of the medium through loss-of-function mutations in the operon responsible for the synthesis of the polymer. They have an advantage because they monopolize the supply of oxygen diffusing into the medium at the surface, but this advantage is balanced by the lower availability of nutrients in this zone. A third type may also appear at the bottom of the vial as an extreme low-oxygen specialist. By isolating these types and competing them against one another, it can be shown that each type is superior so long as it is rare, but loses this advantage once it has become sufficiently abundant. Hence, each is able to invade a culture dominated by the other but cannot completely replace it, and diversity is actively maintained by selection. This divergent selection is ultimately based on the oxygen gradient initially set up by purely physical forces, and if this gradient is destroyed by shaking the vial, diversity does not evolve. Nevertheless, the initial gradient is greatly exaggerated by the growth of the mat-forming type, so the environmental complexity sustaining diversity is itself reinforced by the evolution of specialized types.

12. WHAT LIMITS THE EXTENT OF SPECIALIZATION?

Selection in complex environments can maintain a range of specialized types, but not an unlimited range. The extent of an experimental adaptive radiation is governed by two features of adaptation: functional interference, and degradation through disuse. Functional interference arises when adaptation to one way of life necessarily leads to loss of adaptation to another. The long-term E. coli lines maintained on glucose, for example, lost the ability to utilize a range of other carbon sources. In simple growth media, a universal source of interference is the contrast between rapid but wasteful use of resources and slower but more efficient use. Laboratory cultures are often taken over at first by types that process the limiting resource very rapidly, because in removing it from the medium they deny it to their competitors. They do so at the expense of discarding incompletely metabolized molecules that can be used as substrates by more frugal types as these substances accumulate in the medium. This “cross-feeding” arising from the biological modification of an initially simple medium often leads to the evolution of complex bacterial communities in laboratory microcosms. It is ultimately based on the irreconcilable demands of different metabolic processes, especially the conflict between fermentation and respiration as energy-producing pathways.

A specialized type restricted to a particular way of life derives no benefit from being adapted to others. Loss-of-function mutations in the genes that are required only for these other ways of life are then neutral and will tend to accumulate without limit (see chapter III.14). A population that has become adapted to a novel way of life may then be severely impaired if ancestral conditions of growth are restored. For example, green algae possess a carbon-concentrating mechanism that increases the efficiency of photosynthesis by transporting carbon dioxide to the site in the chloroplast where the initial reactions of photosynthesis occur. This is an expensive process that is switched on only when the external concentration of carbon dioxide is low. If lines are grown at elevated concentrations of carbon dioxide, diffusion alone is sufficient to maintain high internal concentrations and the carbon-concentrating mechanism is unnecessary. Consequently, when these lines have been propagated for a few hundred generations, their carbon-concentrating mechanism becomes degraded, and they are unable to function normally when grown at normal atmospheric levels of carbon dioxide.

Adaptation to environments that vary in time is quite different: any type unable to survive conditions of growth that recur from time to time will become extinct, and more generally selection will favor generalists able to grow moderately well in all conditions. If specialization is limited by interference between different specialized functions, then generalists will be less well adapted to given conditions than the corresponding specialist, whereas degradation through disuse should be halted by the recurrent change in conditions, and broad generalists with no impairment of function may evolve. When bacterial populations are cultured with fluctuating temperature, or algae are exposed to alternating light and dark, the usual outcome is the evolution of generalists, often with fitness in either environment comparable with that of the corresponding specialists, suggesting that degradation through disuse is a frequent consequence of specialization.

This article outlines only a few of the simplest issues tackled by experimental evolution without referring to more complex themes such as metabolic pathways, social behavior, multispecies communities, host-pathogen dynamics, sexual selection, speciation, and multicellular organisms. With the aid of the appropriate model system, there are few fundamental questions in evolutionary biology that cannot be investigated by selection experiments. It may be objected that studying highly simplified laboratory systems may not be relevant to the behavior of populations embedded in highly diverse communities of interacting organisms living in a complex and shifting environment. It is certainly true that particular adaptations, at all scales, from photosynthesis to webbed feet, will require particular explanations, which in many cases cannot be tested experimentally. The justification for experimental evolution, however, is that a distinctive category of evolutionary mechanisms is operating on all characters in all organisms—indeed, in all self-replicating entities—and that it can be elucidated by experiment in much the same way the very different mechanisms governing physiological or developmental processes are being elucidated. We have now acquired a useful understanding of the way these mechanisms operate and the outcomes they produce in simple laboratory systems. The study of more complex systems is only just beginning, and field experiments have seldom even been attempted. Nevertheless, it is clear that the future of the experimental research program lies in greater realism, so that a steadily broader range of evolutionary phenomena will become explicable in terms of clear, testable mechanisms.