This study provides an experimental test of a quantitative optimality model in a phage system. Although optimality models are widely used to explain nature, there have been few attempts to actually evolve organisms to an optimum. Is it possible to predict phenotypic evolution using a simple model based largely on natural selection?
Optimality models may fail for a variety of reasons. An evolutionary constraint might prevent any adaptation at all from the starting, non-optimal state. For example, in a study of optimal foraging evolution, comparisons between ΦX174 and the related phage G4 demonstrated that selection would favor indiscriminate foraging under certain conditions, yet an absolute genetic constraint prevented adaptation of G4 to infect a new host . However, another optimal foraging study using T7 showed that adaptation was possible for that phage . Another form of constraint would be genes with pleiotropic effects on traits not included in the model. In Φ6, expansions in host range were often associated with decreased attachment to the original host  while in T7 a similar mechanism caused a failure of some optimal foraging predictions . There is thus precedence for constraint on adaptation of simple phenotypes in phage genomes, but the extent of constraint is variable.
Phage lysis interval, and thus lysis time, is likely to be under strong and fluctuating selection in the wild, thus providing an ecologically important model system for the evolution of a life history trait. The lysis time optimality model is based on a trade-off between progeny number and generation time. It predicts how long a phage should wait after eclipse to lyse its host based on the growth rate of the phage, assuming that phage progeny accumulate linearly within the host after eclipse. The model has been previously tested in three phages: λ, T4 and T7. In all of those phages, lysis is thought to require a 2-component lysis mechanism; a lysin that destroys the cell wall and a holin that destroys the cell membrane allowing the lysin to access the cell wall. The holin triggers lysis with great precision in at least some phages .
The optimality model was tested in phage λ and T4 by competing lysis mutants with wild-type phages. In both cases, as predicted by the model, different environments selected for different mutants. [15, 16, 25]. In T7, quantitative optimality predictions were tested by allowing lysis to evolve without previously introduced genetic changes known to affect lysis. Adaptation had mixed success; selection for late lysis did not generate the optimal phenotype. Model failure was attributed to a violation of the model assumption that phage particles increase linearly inside the cell if the normal lysis mechanism is suppressed .
This prior work motivated the work here. How would the model fare in a one-component system such as ΦX174 and ST-1? ΦX174 satisfies the model assumption of linear accumulation . In addition, mutations in the lysis gene (E) have been found that hasten lysis time under some conditions . There was thus no a priori reason that the mean lysis interval for the population could not evolve to the optimum.
Under selection for a shorter lysis interval (according to the optimality model) ST-1 evolved earlier lysis by 4 min (although 2 min of this was caused by earlier eclipse), while ΦX174 did not; no changes in the ΦX174 lysis gene evolved in any adaptation to normal hosts. In none of the adaptations on either phage did lysis interval reach the predicted optimum. Quantitatively optimal evolution is not expected in a single step, because a large change is not likely to carry the phenotype directly to the optimum [47, 48]. However, it is surprising how far from the optimum the phages in this study remained after adaptations involving a number of genetic changes.
While precise control of the phenotype appears difficult, it is clear that rapid evolution is possible in a different part of the phenotype space. When ΦX174+ is grown on hosts lacking the cofactor slyD, it does not lyse detectably, yet the adapted phage attains an lysis interval on the slyD-defective hosts indistinguishable from that of a virus adapted to normal hosts. This is similar to the pattern seen in T7, which can recover from complete genetic ablation of lysis but cannot reach the predicted optimum [17, 49]. Interestingly, despite the ability of ΦX174 to recover lysis on slyD-defective hosts, genetic details are relevant here as well; the coding mutations seen in our adaptations were the same as those previously found in another study , demonstrating a limited number of paths to maximal recovery and thus the possibility for constraint.
It could be that the phages would have reached the optimum if allowed to adapt for longer. However, repeated evolution of ΦX174 to a common phenotypic endpoint and the observation that phenotypic evolution was rapid over other areas of phenotype space suggest that this was not the primary limit on adaptation.
Instead, constraint on lysis interval, either absolute or due to pleiotropic effects on other phenotypes, seems the most likely explanation for failure to reach the predicted optimum. This may be caused by the overlap of gene E with gene D, which is highly conserved , or by the relatively crude ΦX174 lysis mechanism . If evolution were possible but the quantitative optimality prediction itself was inaccurate, introducing major changes that would alter the optimum would lead to a different results, yet this had no effect.
A large part of the ST-1 lysis time adaptation was driven by changes in eclipse time. Phenotypes that are outside the scope of a model may frequently play a role in evolution. When optimality models are tested in nature, researchers often form conclusions after considering only the life history trait thought to be under selection. However, most life history traits, like lysis time, result from the interplay of a number of factors, which may mask the nature of adaptation.
This complexity is also evident on the genetic level. For example, ST-1 evolved to lyse closer to the optimum (shortening both its post-infection lysis time and its lysis interval) without changes in gene E, the only gene implicated in the widely accepted 1-component model for lysis in ΦX174-like phages . Other genes (A, F and H) that did evolve in ST-1 may play some heretofore unsuspected role in mediating lysis. However, it is far more likely that, as previously observed in T7 phage, genes with no direct role in cell lysis affect lysis timing by altering timing of lysis gene expression or by indirect effects on other stages of the life cycle rather than lysis mechanisms per se [17, 20, 49]. Similar interactions may make optimality predictions in other organisms more difficult.