Experimental Evolution with Caenorhabditis Nematodes

Abstract

The hermaphroditic nematode Caenorhabditis elegans has been one of the primary model systems in biology since the 1970s, but only within the last two decades has this nematode also become a useful model for experimental evolution. Here, we outline the goals and major foci of experimental evolution with C. elegans and related species, such as C. briggsae and C. remanei, by discussing the principles of experimental design, and highlighting the strengths and limitations of Caenorhabditis as model systems. We then review three exemplars of Caenorhabditis experimental evolution studies, underlining representative evolution experiments that have addressed the: (1) maintenance of genetic variation; (2) role of natural selection during transitions from outcrossing to selfing, as well as the maintenance of mixed breeding modes during evolution; and (3) evolution of phenotypic plasticity and its role in adaptation to variable environments, including host-pathogen coevolution. We conclude by suggesting some future directions for which experimental evolution with Caenorhabditis would be particularly informative.

Keywords: C. briggsae; C. elegans; C. remanei; WormBook; adaptation; domestication; experimental design; laboratory selection experiments; mutation accumulation; reproduction systems; self-fertilization; standing genetic variation.

Figure 1

Phenotypic divergence and differentiation in EE. A typical evolution experiment involves manipulation of environmental conditions and letting populations diverge from the ancestral population in trait value. To show that divergence is genetically based, one needs to account for confounding environmental effects (including transgenerational carryover effects) by assaying ancestral and derived populations after some number of generations of common “garden” culture. “Control” replicate populations are usually kept in the environment to which the ancestral population has adapted to, that is, the domestication environment, and run alongside those in the new environment such that the mean differentiation between treatments is interpreted to result from selection. It is usually assumed that control populations are under no selection, but it is perhaps more realistic to think that they are under stabilizing selection for intermediate trait values. In mutation accumulation (MA) experiments, since selection efficiency is low, a domestication stage prior to subjecting populations to new environmental conditions is not necessary. Any differentiation among replicates within each treatment is presumed to result from genetic drift and idiosyncratic selection (e.g., placement within incubators, the experimenter doing the culturing, etc.). We illustrate divergence and differentiation of a quantitative trait with individual-based simulations (http://datadryad.org/, doi: 10.5061/dryad.bg08n). The plots in the figure show these simulations of divergence and differentiation for 20 populations reproducing exclusively by selfing evolving under genetic drift and no selection (gray lines), or genetic drift and selection (black lines). With lower initial heritability it takes longer for selection to take populations to the new phenotypic optimum (scenario 1; left), and the higher the initial heritability the faster the initial response (scenario 2; middle). In natural populations, lower heritability is typical of behavioral traits, while high heritability is typical of morphological traits (Lynch and Walsh 1998). The phenotypic “optimum” is shown by a dashed line. The right plot shows the fitness function employed for the simulations, with the circles illustrating the mean evolution of the selected populations starting with high heritability. Further details about the simulation can be found in Figure 2.

Figure 2

Detecting phenotypic differentiation and divergence. The statistical power to detect divergence or differentiation depends on many variables. Here, we consider how the number of replicate populations and sample size affect the power to detect responses to selection on a quantitative trait when reproduction occurs exclusively by selfing. We rely on the numerical simulation model presented in http://datadryad.org/, doi: 10.5061/dryad.bg08n; the reader can explore it to generate trait value trajectories and power curves as a function of standing and mutational genetic variance, population size and number of offspring, truncation and Gaussian selection, sample sizes, etc. The analysis we show is based on the simulated evolutionary trajectories of Figure 1. Individual fitness was defined as: w = exp{−[offspring_trait_value − optimum]^2/[2*(intensity^2)]} (Figure 1, right plot), with the new fixed phenotypic optimum set to ∼1 SD from the ancestral phenotypic distribution, and an intensity resulting in an initial linear selection gradient of 0.15. These are realistic numbers for natural populations (Kingsolver et al. 2001), but in the laboratory selection may be more substantial. The trait value of each hermaphrodite in the simulations is defined by its breeding value plus a stochastic component, assumed both to follow Gaussian distributions. Each hermaphrodite produced a fixed number of 10 offspring, with each one of them being represented in the next generation with a probability given by the fitness function, while keeping population size constant at 1000 individuals. This number was the approximately the effective population size of EE done by Teotónio et al. (2012), employing nonoverlapping and constant adult densities of 10⁴ (Chelo and Teotónio 2013). Many studies, however, employ overlapping generations, and do not test for the stability of age-structure during EE. Even though densities may be large in these studies, it is unclear which effective population sizes are realized (Box 1). For all plots, segregation/mutational heritability was of ∼0.15% at each generation, in line with Caenorhabditis MA estimates, e.g., (Salomon et al. 2009). But note that trait heritability will change during EE depending on how selection and genetic drift influence standing levels of genetic variation. Once EE is done, testing for divergence or differentiation requires that replicate populations within each treatment show random effects due to genetic drift and uncontrolled selection. When employing frequentist statistical modeling, taking replicate populations as a fixed factor is incorrect since the degrees of freedom over which the significance of evolutionary responses are tested will be inflated. Using a linear mixed effects model, we tested for differentiation of unselected and selected populations after 50 generations by sampling 10 or 50 individuals from each of 2, 5, or 15 replicate populations. This model assumes that the heterogeneity among replicates is the same between treatments, which may not be true. Left panels show the probability of detecting differentiation (the cumulative density distribution of the simulations) as a function of significance level, for a trait that has a starting heritability of 5% (scenario 1 in Figure 1) or 30% (scenario 2 in Figure 1). Right panels show similar power curves when testing for divergence, where EE populations at generation 50 are compared to the ancestral state. The null hypothesis is that there is no divergence or differentiation (the identity dashed line). This analysis shows that detecting divergence is much easier than differentiation, and that a moderate to high number of replicate populations and high sample sizes are necessary to be confident that divergence or differentiation are due to selection (power of at least 80% at the significance level of 0.05; vertical line). Sampling several time points during EE will generally increase power, although it is often unclear how to model for trait autocorrelations across generations and if linear or nonlinear trajectories are expected (see Figure 1).

Figure 3

Measuring natural selection and genetic drift. Fitness can be estimated in Caenorhabditis EE with competitions between ancestral and derived populations with a “tester” strain that is distinct in morphology, for example by the expression of a green fluorescent protein (GFP) (Morran et al. 2009a; Theologidis et al. 2014). EE populations and tester are usually placed at equal frequencies, and their frequency is followed for one generation in the same environmental conditions as those during EE. Their relative proportion after the competition can then be taken as an estimate of relative fitness, assuming that there is no assortative mating. This way of estimating fitness is appropriate for populations cultured under nonoverlapping generations and stable densities. Fitness components can be measured in a similar fashion. For example, to estimate male fitness, EE males can compete with tester GFP males for the fertilization of tester females (these can be fog or fem mutants, for example), with male fitness being the relative offspring proportion of wild-type vs. tester GFPs (Teotónio et al. 2012). For Caenorhabditis EE with overlapping generations, and when a stable age-structure has been reached, then relative fitness can be estimated as the solution of Σe^−rx l(x)m(x) = 1 for r, the “Malthusian” intrinsic growth rate parameter (Roff 2008). Here l(x) is the proportion of individuals surviving to age x, and m(x) the fecundity at age x, see e.g., Estes et al. (2011) for an application of this model. For populations that have not reached stable age distribution r may be taken as absolute “Darwinian” fitness, by measuring the total number of offspring produced during an individual’s lifetime. In population genetics studies, the relative fitness of particular alleles at a given locus can be measured as the log ratio of their frequency change across generations, a coefficient that is usually called “selection coefficient”: s = {ln[pG/(1−pG)] − ln[p0/(1−p0)]}/G; where pG and p0 are the frequencies of the allele in question after G generations, and at the beginning of the competition G = 0, respectively. Similar but more complicated expressions can be used when there is density- or frequency-dependent selection (Chevin 2011). In genome-wide studies, neutral polymorphisms (SNPs, for example), linked to the putative selected alleles, but unlinked to each other, can be employed to measure selection. The plots illustrate such examples: left, frequencies of the allele of interest; right, selection coefficients, with one SEM between the four replicates/neutral SNPs. Note that, with frequency-dependence, selection coefficients may change sign during the competitions (gray). The variance in allele frequency dynamics with EE can also be (retrospectively) used to estimate the effective population size, and thus the extent of genetic drift and inbreeding (Goldringer and Bataillon 2004; Chelo and Teotónio 2013). Typically, the effects of genetic drift at the genotype or phenotype levels are only accounted for as the random variation observed between replicate populations subject to the same treatment. One exception is when genetic drift can be thought of as a branching process and the growth rates of alternative types (be it alleles, genotypes, phenotypes) are independent. This occurs when mutants appear in very low numbers and invade a “resident” population composed of alternative types. Under discrete and nonoverlapping generations, with stable densities, and with successful offspring distributions following Poisson distributions, the proportion of invasions that are not successful is an estimate of the extent of genetic drift (Haldane 1927; Kimura 1957). Even for strongly selected mutants genetic drift can be estimated as the proportion of mutant invasions as: (1 – probability of extinction) = 1−e ^−2sn(Ne/N), where s is the selection coefficient, n the number of mutants invading, and N_e and N the effective and census population size, respectively. These considerations are important since transitions in character states (for example between outcrossing and selfing) are ruled first by genetic drift, and only later by selection.

Figure 4

Artificial vs. natural selection. A good example of the use of artificial selection to test a specific genetic hypothesis is provided by the work of Azevedo et al. (2002), who selected for increased and decreased body size on replicated highly inbred (and presumably isogenic) populations of C. elegans of 48 generations. They were able to generate a rapid response to decreased body size but not to increased size. Because the populations were initiated without any standing variation, the response to selection must have been due to the contribution of novel mutations, indicating that the asymmetry in the response to selection is caused by an asymmetry in the distribution of mutational effects on body size—something that was verified via direct estimates of the mutational effect (Azevedo et al. 2002; Ostrow et al. 2007). These results suggest that either increasing body size is difficult because of a small set of genetic targets relative to decreasing body size, and/or increases in body size are constrained because of the pleiotropic effects of new mutations on other fitness components. The plots show figure illustrates a putative trade-off between body size and fecundity such that artificial selection (AS, line) for increased body size would lead to the fixation of deleterious mutations across replicate populations, as observed after stopping AS. Conversely, natural selection (NS, dashed lines) would result in a correlated decrease in body size. One important distinction between AS and NS is that the first usually involves “hard” truncation selection, where individuals below a certain trait value threshold do not contribute to the next generation.

Figure 5

Mutation accumulation (MA) experiments. Left, illustration of how an MA experiment is performed, adapted from Halligan and Keightley (2009) and Katju et al. (2015). G indicates the generation of EE, N the number of individuals allowed to reproduce within each MA line, and n the number of MA lines. We schematize a diploid chromosome devoid of genetic variation in the ancestor that accumulates and fixes different (independent) mutations within each MA line. The cryopreserved ancestor and/or control is usually phenotyped alongside the MA lines. Right, graphical representation of expected outcome of an MA experiment. The blue line shows the expected trajectory of evolution of fitness among MA lines, which is unknowable in practice. Since most mutations will be deleterious in a well-adapted population, it is expected that fitness will decline with MA. Trait X represents the expected average evolutionary trajectory of a trait positively correlated with fitness; size at maturity is a typical example. Trait Y represents a trait negatively correlated with fitness; time to maturity is a typical example. The orange line represents the genetic variance between the MA lines, which (ideally) is zero in the G0 common ancestor of the MA line. The per-generation increase in the genetic variance is the “mutational variance,” V_M.

Figure 6

Host–pathogen coevolution experiments. Here we show a schematic relationship of experimental treatments in a comprehensive host–pathogen coevolution experiment. Columns show host treatments, rows show pathogen treatments. Cells with shading represent evolving experimental populations; arrows represent the direction of evolutionary causation.