Different viewpoints of epistasis. Coat color variation in mammals has long served as one of most fruitful examples of genetics, with over 120 loci and 800 alleles described in mice alone101. The coat color genetic pathway can be used to illustrate different usages of the term “epistasis”. In the original sense, from Bateson, epistasis arises when the effects of alleles at one locus are blocked by the presence of a specific allele at another locus. (a) For example, a cross between an agouti and extension (now called the melancortin 1 receptor or Mc1r) double heterozygotes yields the non-Mendelian segregation ratio of 9:4:3 (instead of 9:3:3:1), with the excess of dusty yellow extension offspring suggesting that the Mc1r locus operates downstream of agouti (an example of recessive epistasis). Crosses with other mutants can be used to further order the genetic pathway, relying on combinations of knockout mutants to generate compositional epistasis (Box 1). (b) The outcomes of this same cross can be illustrated in a 3×3 genotype interaction table that is common in population genetics. (c) The biochemistry of this pathway within the melanocyte has since been fully worked out. Here, activation of Mc1r turns on the production of the (dark) eumelanin, as opposed to (the yellow/red) pheomelanin. Agouti acts an antagonist to Mc1r such that periodic activation of agouti leads to banding of color on individual hairs. Disruption of other loci, such as the classic albino (Try) locus, destroys function of the entire pathway. The specific interactions between the proteins in this pathway are representative of functional epistasis. (d) Recently, Steiner et al.102 have used this pathway to probe natural variation between dark forest mice and light colored beach mice. They find that the adaptive transition from dark to light that accompanied the movement of mice from the forest to the beach is accomplished by an interaction between structural changes to the agouti locus and regulatory changes to the Mc1r locus. The specific effects of these interactions can be roughly quantified by averaging the effects of these markers overall all sampled genetic backgrounds to give an estimate of the statistical epistasis between these loci. (e) Interestingly, the pattern of epistasis for these loci in nature is reversed from the standard cross, presumably because the Mc1r allele in the beach mice has partial function and is therefore still susceptible to suppression from agouti. This later observation is a clear illustration that epistasis is a property of specific alleles, rather than a particular locus in general. Pictures for (a, b) reprinted with permission from Ref 103. Images are representative of similar phenotypes. (c) is modified from Ref 104). (d) is from Ref 102. Images for (e) kindly supplied by Hopi Hoekstra.

Reconstructing genetic pathways using epistasis analysis. (a) A classical example of epistasis using flower color in sweet peas. Combinations of mutations at two loci generating enzymes responsible for processing anthocyanin operate within a single biochemical pathway. Picture reprinted from Ref 105. (b) Construction of the epistatic network of genes underlying sensitivity to the mutagen methyl methanesulfanate (MMS) in yeast. The fitness of all 650 combinations of 26 genes known to be sensitive to MMS were measured quantitatively; combinations showing departures from multiplicative interactions are connected in the network. Arrows represent five interaction subtypes: coequal (green), partial masking (light blue), masking (dark blue), partial suppression (purple), suppression (pink). Reprinted from Ref 34.

Epistasis in complex traits. (a) More than forty generations of selection on body size in chickens lead to a large difference among lines. (b) QTL mapping of an F2 cross between the divergent body size lines revealed a network of six genes that explains 45% of the among line difference. Five of these loci are only show their effects via epistasis with the sixth (Growth9). (c) An example of one of the genotype/phenotype relationship between two of these genes. Note the large difference in the double Growth12/Growth9 high line homozygotes. (d) Epistasis in the autoimmune reaction thought to underlie many cases of multiple sclerosis. Mice that had been transformed to produce human T-cell antigen receptors, are then also subsequently transformed to express the human major histocompatibility (MHC) locus DRB5*0101 (DR2a), the MHC locus DRB1*1501 (DR2b), or both. The incidence of disease is much lower when both loci are present, even though the DR2a locus does not independently influence disease onset. (e) These two loci encode antigen binding proteins that are expressed on the surface of an antigen presenting cell (APC). A possible model for the epistatic effect is that the overall antigenic effect is maintained by a balance between T-cell proliferation induced by DR2b and T-cell apoptosis induced by DR2a. The complex relationship between these two might explain the complex pattern of onset and periodic nature of some multiple sclerosis cases. (a-c) from Ref 46. (d) is based on data from Ref 67. (e) is adapted from Ref 68.

Four different views of the generation of epistasis under natural selection. Each figure shows a hypothetical adaptive landscape in which the mean fitness of a population is a function of underlying variation at two or more loci. Lines display points of equal fitness, with warmer colors indicating areas of higher fitness. (a) Under Fisherian Adaptation77, adaptive evolution is seen as a hill climbing process in which new beneficial mutations that increase fitness build upon substitutions that are already fixed within the population. Here, epistasis is a byproduct of the historical contingency built into the evolutionary process (both from the random effects of mutations and shifts in the adaptive optimum), but at any given time point the average or additive effects of a particular mutation are responsible for the actual evolutionary change. (b) In Wright’s Shifting Balance Process106, epistasis generates multiple adaptive peaks in the fitness landscape. If stable through time, these valleys can only be traversed via genetic drift or tight linkage. Here, epistasis plays a fundamental role in determining the direction and likelihood of any particular evolutionary change. (c) In Gavrelet’s Holy Landscape model107, adaptive landscapes are rugged as in Wright’s case, but the multidimensional nature of the fitness landscape leads to ridges of nearly equal fitness that populations move along either via directional selection or genetic drift. This represented metaphorically here by a ridge of nearly equal fitness whose endpoints are spatially separated by a deep valley. In multidimensional space, this might be something more like a tunnel connecting regions of high fitness and avoiding complex adaptive valleys. Here, epistasis will be strongly apparent after populations diverge, and may impact the direction of evolutionary change, but does not generate a barrier to change as in Wright’s model.