One of the main goals of evolutionary biology is to understand the process leading to the observed patterns of phenotypic diversity. Natural selection, historical events and chance have been identified as factors shaping diversity at different scales, from local adaptation to speciation
[1, 2]. These evolutionary processes are not mutually exclusive and often contribute together to the pattern of differentiation. While natural selection leads to a deterministic adaptation to environmental conditions, historical factors and chance can produce different outcomes despite similar environmental conditions. The idea of contingency playing a role in the evolution and generation of biological diversity was actually central in Darwin’s work and a key point differentiating his theory from the ones of his contemporaries. Chance plays a role both in the initial generation of diversity, i.e. mutation, and in the maintenance or elimination of the diversity in the population, i.e. genetic drift. History might play a role if initial differences in the phenotype and/or the genotype affect adaptation.
In this context, as outlined by Travisano et al. and Blount et al., we will define historical contingency as an evolutionary situation where the initial phenotype and/or genotype influences the response to a selection episode and the evolutionary outcome in terms of phenotype and/or genotype. There are various approaches for studying the importance of historical contingency, the choice of which strongly depends on the possibility of performing experimental evolution. First, for large complex organisms, a number of studies have characterized groups (from population to species) adapting in parallel to similar environments, in the context of adaptive radiation. In this class of studies, related species (e.g., anole lizards
[4, 5], mosquitofishes
, or orb-weaving spiders
) or populations (e.g., of mosquitofish
 or of a freshwater isopod
) are characterized at the morphological or behavioral level. The differences found are attributed to the difference in selection pressures, the replicate population and the interaction between these two factors. Some studies control for the phylogeny, so that they can distinguish between the influence of past evolutionary events and specificities of each replicate in determining the current pattern of differentiation. The broad picture coming out of these studies is that the environment is the first determinant of the phenotype but historical events usually also have significant effects, although of lower magnitude.
For organisms amenable to experimental evolution, replaying part of an evolution experiment can also test the importance of historical contingency. This is possible by building “fossil records” along the experimental evolution, as in the case of viruses
 or bacteria
[3, 11]. In this last study, by generation 31,500 of the long-term experiment evolution
, one of the 12 lines of E. coli B became able to metabolize citrate. Blount et al. determined that the probability of this evolutionary event is much higher when experimental evolution is replayed from samples frozen shortly before its initial appearance. This result highlights the importance of historical contingency and favors the hypothesis that new traits emerge by the occurrence of a series of mutations in a specific order rather than being the result of a unique rare mutation
An alternative version of the “replay experiment” is to evolve in a common environment lines or populations that initially differ for known phenotypic and/or genotypic characteristics. The initial diversity can be generated by a strong founder effect in complex organism such as anole lizards
, a previous phase of experimental evolution (e.g.,
[1, 14, 15]) or be present in a mutant or isolate collection (e.g.,
[16–19]). A variation of this general protocol is to start the experimental evolution with identical populations, place them in a similar environment but vary the conditions of adaptation (e.g.,
), the size of the transfer bottleneck or the presence of a mutagen. The lesson learned from these studies is that the influence of historical contingencies on evolution strongly depends on the trait measured. Historical contingency tends to have a lower impact on traits that determine fitness than on those that have weak impact on fitness
[1, 14]. Moreover, a form of historical contingency is systematically found in studies that analyze DNA sequences
[15, 16, 20].
If past historical events constitute evolutionary constraint, identifying its impact on viral evolution is of great importance for understanding the evolution of host-usage dynamics by multi-host viruses or the emergence of escape mutants that persist in the absence of antiviral treatments. Among the few studies addressing the question of historical contingency in virus evolution, Burch and Chao identified two populations of different evolvability during fitness recovery after a mutation accumulation experiment in the bacteriophage ϕ6
: one was climbing a fitness peak whereas the other was at the top of a second, less optimal, fitness peak
. In a second study, Herrera et al. explored the role of contingency in the coevolutionary process between cells and Foot-and-mouth disease virus during persistent infections
. Independently evolved lineages that started with the same original viral and cell clones, fixed the same mutations and showed a strong role for historical contingency: the presence of a given pair of mutations in early stages of the coevolutionary process determined the subsequent fixation of other mutations. Finally, in the Rice yellow mottle virus (RYMV), it has been demonstrated that the different resistance-breaking mutations of isolates from different cultivars or species cannot be explained by a classical arms race between host and pathogen but result from epistasis between a previously polymorphic site and the site conferring the resistance breaking phenotype
In the present study, we used populations of Tobacco etch virus (TEV) generated by Bedhomme et al. to assess the importance of historical contingency in the evolution of this virus. The initial experiment was designed to analyze the adaptation of TEV to different host species and to contrast generalist and specialist strategies in the context of adaptation to new hosts
. Starting from a single infectious clone, we derived two types of evolutionary histories: (1) viral populations transmitted on the same host; and (2) viral populations transmitted on alternate hosts
. The first case was expected to select for specialists whereas the second should have favored generalists. The phenotypic characterization of the evolved lineages allowed us to identify a pattern of higher infectivity and virulence on host(s) present during experimental evolution, which indicates the existence of local adaptation in the majority of the host × evolutionary history combinations
. Local adaptation comes, in some of the cases, at a cost on alternative hosts. We did not find any specific characteristics for the alternate-host infecting lineages. Moreover, we did not find a cost for being a generalist. The full-genome consensus sequences of the evolved lineages revealed the fixation of some host-specific mutations but a low level of parallel evolution
. These independently evolved lineages, characterized for some phenotypic traits and for their full-genome consensus sequence, constituted an ideal material to investigate historical contingency: we have been able to ask if their initial characteristics affected their phenotype and genotype after a new phase of evolution in a common host. Moreover, the common host chosen for this second phase of evolution was Nicotiana tabacum, to which the ancestral infectious clone is presumably adapted
. Consequently, evolving all the differentiated lineages on N. tabacum constitutes a reverse evolution experiment.
TEV genome is characterized by pervasive epistasis and in particular by a high frequency of reciprocal sign epistasis
. This is predicted to produce a highly rugged adaptive landscape, in which many adaptive pathways are inaccessible
[27, 28]. Moreover, it is known that the sign and the magnitude of epistasis between mutations vary from one host to another for TEV
. Such epistasis suggests an important role of historical contingency in TEV, at least at the genotypic level. We made the following predictions: (1) if historical contingency plays a role in phenotypic evolution, the phenotypes at the end of the “common environment” phase will not be the same for all lineages and will depend on the phenotypes at the beginning of this phase and (2) if historical contingency plays a role in genotypic evolution, the level of sequence convergence will be low and the genetic difference between lineages that differ in evolutionary history before the “common-environment” phase will be higher than between the lineages of identical evolutionary history. In terms of adaptive landscape, a significant historical contingency would imply that N. tabacum represents an environment with multiple fitness peaks, whereas a lack of historical contingency would suggest that the ancestral host represents an environment with a single accessible peak