The existence of stresses that alleviate average mutation effects could imply that there is average epistasis between mutations. Average absolute growth rates of the parental strain (with no mutations) and of the mutant library (defined as having an average of 1 unit of mutation per mutant in the library) are shown under favorable conditions (black) and under trimethoprim stress (gray). Linear extrapolation (dashed) of the data, assuming an absence of epistasis, would lead to intersection of the lines. Such an intersection seems unrealistic, however, as it would imply an increase of the average absolute growth rate under stress. To avoid intersection at least one of the lines must bend, which would reflect the existence of average epistatic interactions between mutations. Note that the fact that our library may contain a variable number of mutations per genome does not affect the argument presented above.
Kishony and Leibler Journal of Biology 2003 2:14 doi:10.1186/1475-4924-2-14