Epistasis III

In many ways, the two different definitions of epistasis in the previous posts are a bit like blind men describing different parts of an elephant.  Statistical epistasis at the population genetic level can be caused by genes whose products directly interact, as in biochemical genetic epistasis, but it can also be cause by indirect interactions between gene products (i.e. one or more intervening molecules in a developmental or metabolic network).  Many traits, including human disease states, are affected by epistasis among multiple genes.  These types of higher order interactions are difficult to characterize and understand.
There are other complications.  From a population genetic standpoint, many compensatory epistatic situations, such as the parallel pathway example given in the previous post, many not "read" as epistasis statistically.  Genes can be viewed as adding pluses (+) and minuses (-) to an overall trait value, which can be obtained by adding up the sum of the + and - alleles. A mutation that turns a + into a - can be compensated for by a second mutation that changes a - to a +.  If their effects simply add to a sum, then this is not considered statistical epistasis, it is simply "additivity". Additive gene actions formed the original basis for the field of quantitative genetics, and such additive effects are still believed to provide the basis of most adaptive evolution and responses to artificial selection (as in agricultural and domestic animal and plant breeding).  However, quantitative genetic experiments have uncovered a great deal of statistical epistasis in the "genetic architecture" of trait differences between different strains of the same species and between different species.  It now appears that epistatic gene action is just as important as additive gene action in these architectures.  Understanding the molecular mechanisms underlying these gene interactions is a large task that still mostly lies ahead.