Strengths and weaknesses of experimental evolution.

A recent article in TREE [1] presented a heavily documented review of experimental evolution, addressing three major issues: main questions and applications; experimental design and study systems; and caveats and limitations. Here, we comment on the latter issue. The soundness of a scientific method depends on not only its adequacy relative to the specific goals, but also the relative suitability of alternative methods. Without intending to be exhaustive, we focus on other methods that are frequently used in evolutionary biology, and compare them with, and contrast them to, experimental evolution. We briefly analyze the objectives and relative limitations of each, as we feel that this analysis was lacking in the original article [1]. The most widely used method in evolutionary biology, ever since Darwin, is the comparative method. By comparing traits between extant populations, the general goal of this method is to infer the evolutionary history underlying present diversity. Hence, it is widely used to trace phylogenies, for which it is the only methodology available. It also serves to infer the microevolutionary processes that have shaped current natural populations, based on the patterns observed. However, knowledge of the ancestral state of populations, required to infer such processes, is usually missing. Hence, this method often uses present populations as surrogates of the ancestral state of others. However, this relies on several assumptions, such as simple evolutionary history, repeatability of evolution, and so on [2]. Experimental evolution is free of such assumptions, as it is evolutionary biology in its most empirical sense, enabling researchers to follow microevolutionary processes directly and to establish a causal link between patterns and processes. Our own work helps illustrate how experimental evolution can unravel the pitfalls of a comparative approach. Magalhães et al. [3] showed that a genetic tradeoff may be incorrectly inferred by direct comparison of the performance of populations of spider mites adapting to different hosts, whereas the experimental evolution analysis, taking into account the common ancestral state, showed a positive covariance. Matos and collaborators showed that laboratory evolution in Drosophila subobscura varies across foundations, particularly for life-history traits that are less relevant to fitness [4]. The direct evolutionary trajectories observed differ from those inferred using different populations as ‘surrogates’ for evolutionary states across generations [5,6]. Another goal in evolutionary biology is to characterize the evolutionary potential of populations. One frequent approach is to use inbred lines in sexual populations, assuming that the range of values across lines is a good representation of the standing genetic variation of outbred populations. However, inbreeding artifacts may lead to overwhelmingly positive genetic correlations between life-history traits that are absent in outbred populations. The study of mutants and how they differ from the ‘wild state’ is another approach used. However, mutants often have low fitness, and selection acts against them, in practice suppressing their role in evolution. The study of the evolution of aging provides a good illustration of the contrasting outcomes of using such approaches. Disparities were reported using inbred lines, mutants, and experimental evolution, with only the latter providing evidence for Hamilton’s analysis [7]. Characterizing the evolutionary potential of a population is also done using additive genetic variance–covariance (G) matrices between fitness-related traits. This tool may accurately predict the short-term evolution of a population. However, there is a long-standing debate on the stability of G-matrices, which is required for longer-term predictions [8]. Experimental evolution is a fundamental, complementary tool, as it can test the stability of G-matrices by following the evolutionary trajectory of populations over a longer period of time. Unfortunately, to our knowledge, no study has yet provided robust conclusions on this matter. Recently, genome scans have been used to compare populations exposed to different selection pressures, as well as to characterize their evolutionary potential. Such scans are valuable, especially when combined with experimental evolution, thus enabling researchers to trace the genome-wide real-time evolution of replicated populations [9–11]. Finally, modeling is a heuristic tool in evolutionary biology, particularly for exploring complex scenarios, such as evolution in heterogeneous environments. Again, experimental evolution may be an added value, by providing empirical data to test and adjust such models [12]. Experimental evolution does have its own limitations. However, it is still one of the most powerful tools in evolutionary biology, especially when combined with other approaches. Its most important goal is to establish links between microevolutionary processes and patterns. This in turn helps disentangling the evolutionary and genetic mechanisms underlying adaptation and diversity, the main goal of evolutionary biology ever since Darwin.