How genes interact – the phenomenon termed “epistasis” – is complex, yet must be understood to accurately interpret the contribution of individual genes to the development and behavior of the organism. Two groups -- Chou et al. and Khan et al. -- recently reported in Science the results of their remarkably similar experiments; they isolated individual, beneficial mutations in bacteria then tested how the mutations interact when possessed by the same bacterium.
Previous studies (cited by Chou) suggested that two deleterious mutations in the same pathway are generally less-than-additive but were greater-than-additive when they were in parallel pathways, which seems intuitive. Interactions between mutations in single genes were shown to depend on the background (other genes). However, epistasis among beneficial mutations in different genes was “unexplored”, and the focus of these new experiments. Evolution in laboratory conditions is initially rapid but quickly slows, which fits a model of mutually antagonistic beneficial mutations. Khan states the deceleration is due to either (1) negative epistasis or (2) because the most beneficial would “tend to be incorporated earlier owing to their faster spread and greater success in the face of competing beneficial mutations” (which sounds suspiciously convenient – the best arrive early – but they give a reference: Gerrish 1998 so you can look it up).
Khan grew E. coli with a glucose supplement for 20,000 generations, when they sequenced a clone and identified 45 mutations. They say other beneficial mutations arose but were lost due to “interference” with more-beneficial mutations (24, 26) (which seems to answer their question), or because they were “less able to evolve than the eventual winners” (33). They took the first 5 “that fixed … and whose spread coincided with the period of fastest adaptation” (arbitrary? How many generations was that, 200? 2,000?), which were, in order of appearance: rbs operon, topA, spoT, glmUS promoter, and pykF. Together, these 5 mutations increased fitness ~30%, accounting for ~80% of the fitness increase over the full 20,000 generations.
Then they produced 32 populations of E. coli, one for each possible combination of the 5 genes (Fig. 1 from Khan, copied here, shows the ancestral genotype at the top and the 32 combinations of mutations, with increasing fitness downward). They found that although each combination improved fitness, improvement was less than expected from a “multiplicative null model” (in which the individual fitness effects are multiplied).
Similarly, Chou selected a bacterium to grow with methanol as its sole carbon source and then replaced a key metabolic pathway with a less efficient pathway from a different bacterium. The resulting bacterium grew only one-third as fast as the original. Eight separate populations improved the efficiency over 600 generations. In the fittest strain, 9 mutations were identified; 3 were clearly related to the pathway while 6 others were deemed “unlikely” to contribute to fitness. To determine the interactions among these 3 genes, they constructed 16 strains, one with each combinations of these 3 mutations plus the WT allele. They also observed that combination strains were much less fit than expected if they were acting independently (our old favorite, the multiplicative null model).
These findings should influence our expectations in genetic studies in humans; e.g., two disease-risk alleles may have little additive or even negative effects (ergo protective?), contradicting the simple expectations of most biologists. [Chou distinguished beneficial from detrimental effects.] I wonder if these experiments might be misleading because they first fixed the individual mutants, thereby eliminating mutants that interact positively during selection.
The same issue has a commentary.