Evolutionary paths of least resistance

When organisms face a challenging environment, they may respond by adapting along different potential evolutionary paths. Each potential path will likely correspond to a different biological mechanism for fitness increase, and which of these alternative mechanisms will evolve depends on a number of factors, including the relative fitness benefits and costs of a given mechanism, the number and type of mutations necessary to evolve the mechanism, and random chance. If one of these evolutionary paths is consistently chosen over the others, then we can think of it as the path of least resistance. In PNAS, Bratulic et al. (1) study the evolutionary paths of least resistance in Escherichia coli bacteria exposed to the antibiotic ampicillin. The authors study E. coli strains carrying a plasmid expressing TEM-1, a protein that confers resistance to ampicillin. The experimental strain has an elevated rate of translation errors, whereas a control strain has a normal translational error rate. The experiment is set up such that only the coding sequence of TEM-1 can evolve, but not the E. coli strains themselves nor the rest of the plasmid. Bratulic et al. find that under an elevated translational error rate, the path of least resistance is to evolve reduced translation-initiation efficiency for TEM-1 when ampicillin concentration is low, and robustness to translation errors when ampicillin concentration is high (Fig. 1). In contrast, a reduction in translational error rate, which could have evolved through modified synonymous choice, is not observed.

Bratulic et al.’s (1) study addresses two long-standing questions in molecular evolution: what are the primary costs of translation errors, and how can these costs be mitigated (2)? Every translation error that occurs has at least two potential consequences. First, it can de-activate the protein, thus reducing the amount of functional protein available to the cell. Second, it can cause the protein to misfold, and misfolded proteins frequently are cytotoxic and reduce organism fitness (3, 4). Which of these two costs is the greater has been subject to extensive debate. However, as we see in Bratulic et al.’s work (1), the answer may depend on the specific circumstances. Under low concentrations of ampicillin, when not a lot of functional TEM-1 is required for successful bacterial growth, TEM-1 evolves reduced translation-initiation efficiency. Thus, in this scenario, reducing the cost of misfolded TEM-1 is apparently more important than reducing the cost of low TEM-1 abundance. In contrast, under high concentrations of ampicillin, when a lot of functional TEM-1 is required, the TEM-1 gene instead evolves translational robustness (i.e., the ability to fold and function despite translation errors that may have occurred).

Under both scenarios, for low and high ampicillin concentrations, a third possibility would be a direct reduction in the translational error rate through the evolution of translational accuracy (5). Different synonymous codons in E. coli are translated with substantially different error rates (6), and thus one might expect that the TEM-1 genes would accumulate synonymous mutations reducing that error rate. However, Bratulic et al. (1) see little evidence for such synonymous evolution. The explanation is simple. Individual synonymous mutations protect only against translation errors at the specific site where they occur, and hence they individually provide only a small fitness increase. In contrast, a single mutation modifying the start codon or increasing robustness by increasing protein stability (7, 8) will protect the organism against the deleterious effects of translation errors along the entire TEM-1 gene, thus providing a large fitness benefit. Because the various protective mechanisms are not mutually exclusive, however, we can surmise that if the experiment was carried out for many more generations, translational accuracy would eventually evolve in addition to translational robustness and reduced expression level.

The results of Bratulic et al. (1) mirror those of an earlier study by Goldsmith and Tawfik (9), but there are several critical differences. Goldsmith and Tawfik set up a system in which TEM-1 was subjected to elevated transcriptional rather than translational errors. They observed adaptations that lead to increased expression level and, when expression level was held constant, increased thermostability (9). Because the errors were introduced at the transcription stage, there was no opportunity for the TEM-1 gene to reduce the error rate directly, through the evolution of an alternative and more accurate synonymous genetic encoding. Thus, in Goldsmith and Tawfik, only two adaptive paths were possible, and one (robustness) was observed only when the other (expression) was disabled. In Bratulic et al.’s (1) work there were three potential paths, and two of them were each found to be preferred under one of the two selection regimes considered.

The work by Bratulic et al. (1) raises two important questions. First, in the stringent selection regime, where a high abundance of functional TEM-1 is required for bacterial growth, is the primary fitness cost associated with insufficient abundance of functional TEM-1 or the presence of misfolded TEM-1 or both? The results of Goldsmith and Tawfik (9) suggest that misfolded TEM-1 presents a secondary fitness cost relative to lack of functional TEM-1, but a definite answer to this question remains outstanding. Interestingly, in the relaxed selection regime of Bratulic et al. (1), where large quantities of TEM-1 are not required, the predominant fitness cost is likely the presence of misfolded TEM-1. Second, would it be possible to set up an experiment where translational accuracy does evolve: that is, where one can actually see the replacement of nonpreferred with preferred codons? As Bratulic et al. (1) note, such adaptation can take a very long time (10) and may thus be out of reach for typical experimental evolution protocols. However, it might be possible to force the issue by taking away all evolutionary paths that offer less resistance. For example, if one started the experiment with a TEM-1 that was already highly stable and robust to amino acid substitutions, then significant further adaptations for increased translational robustness would not be possible, and the bacteria might instead accumulate adaptations for translational accuracy.

Bratulic et al.’s paper (1) adds to a growing body of evidence demonstrating that phenotypic mutations (i.e., mutations occurring solely in the expressed phenotype) may indirectly influence the evolutionary dynamics of the underlying genotype (2, 9, 11–14). In particular, genetic adaptations can protect organisms from the deleterious consequences of phenotypic mutations (1, 9, 12). This mechanism in turn opens the possibility that elevated phenotypic mutation rates may be beneficial for further adaptation (11, 14).