Defining genetic interaction

Abstract

Sometimes mutations in two genes produce a phenotype that is surprising in light of each mutation's individual effects. This phenomenon, which defines genetic interaction, can reveal functional relationships between genes and pathways. For example, double mutants with surprisingly slow growth define synergistic interactions that can identify compensatory pathways or protein complexes. Recent studies have used four mathematically distinct definitions of genetic interaction (here termed Product, Additive, Log, and Min). Whether this choice holds practical consequences has not been clear, because the definitions yield identical results under some conditions. Here, we show that the choice among alternative definitions can have profound consequences. Although 52% of known synergistic genetic interactions in Saccharomyces cerevisiae were inferred according to the Min definition, we find that both Product and Log definitions (shown here to be practically equivalent) are better than Min for identifying functional relationships. Additionally, we show that the Additive and Log definitions, each commonly used in population genetics, lead to differing conclusions related to the selective advantages of sexual reproduction.

Fig. 1.

Different definitions of genetic interaction lead to different distributions of ε (the deviation of the observed double-mutant fitness from expectation). (A) Distributions from all reproducibly measured pairs from Study J that involve genes with singly deleterious mutations show the Min definition to have a negative bias and clear differences from other definitions. (B) The subset of pairs from A involving genes with minor fitness effects shows no significant differences between definitions. (C) The subset of pairs from A involving genes with moderate fitness effects shows Min to have the most severe bias in ε. (D) The subset of pairs from A involving at least one extreme fitness defect exhibits a positive shift in bias in ε for definitions.

Fig. 2.

Comparison of Study S synthetic interactions with those of Study T and Study P. a–f contain a Venn diagram characterizing agreement between two sets of interactions, with the sum of numbers equaling the number of pairs tested in both compared studies. a–c compare synthetic genetic interactions derived from Study S using the Min definition (Study S_Min) with those of a Study T (12); (b) Study P_severe (17) (Study P considering only the two more severe synthetic interaction levels) and (c) Study P (17) (interactions at all levels of severity). Note that all 14 interactions of Study T and all eight of the Study P_severe interactions were confirmed, confirmation rates of 100% with a 95% C.I. of 76.7–100% and 63.1–100%, respectively. Furthermore, 53 of the 63 Study P_slight interactions were also identified by Study S_Min (confirmation rate 84.2% with a 95% C.I. of 72.7–92.1%). d–f compare synthetic interactions derived from Study S using the Product definition (Study S_Product) with those derived from (d) Study T, (e) Study P_severe, and (f) Study P. In these comparisons, 13 of 14 Study T interactions were confirmed by Study S_Product (confirmation rate 92.9%, with a 95% C.I. of 66.1–99.8%). By contrast, five of eight Study P_severe were confirmed (confirmation rate 62.5%, with a 95% C.I. of 24.5–91.5%), and only 21 of the 63 Study P_slight synthetic interactions were confirmed (confirmation rate 33.3%, with a 95% C.I. of 22–46.3%).