Experimental determination of fitness interactions in Drosophila melanogaster by the method of marginal populations.

I N the last decade the theoretical consequences of the joint effects of recombination rate and fitness interactions between loci have been studied by KIMURA (1956, 1965) , KOJIMA (1959‘ a,b), LEWONTIN and KOJIMA (1960), LEWONTIN (1964 a,b.c), and FELSENSTEIN (1965). Several observations support the relevance of this body of theory to natural and laboratory populations. Non-random associaticins of inversions in natural populations may be the result of fitness interaction between the inversions (LEVITAN, 1958; LEVITAN and SALZANO 1959; STALKER 1960; WHITE, LEWONTIN and ANDREW 1963). Viability interactions between inversions have been demonstrated experimentally ( SPASSKY, DOBZHANSKY and ANDERSON 1965) and in nature (LEWONTIN and WHITE 1960). CANNON (1963) reported an increase in linkage disequilibrium for markers introduced into :laboratory populations of Drosophila melanogaster. This suggests, among several hypotheses, fitness interactions between the marked loci. The non-random associaiion of shell color and banding pattern in Cepaea nemoralis (CAIN and SHEPPARD, 1954)-characters controlled primarily by two linked loci (CAIN, KING and SHEPPARD 1960)-indicates a fitness interaction underlaid by differential predation. A survey for fitness interactions, which implies estimation of a genotypic fitness array of a population segregating for two or more loci, has not been conducted on successive generations of laboratory populations. This may be due to technical difficulties, uiz. the geonietrical increase of the number of different genotypic classes to be scored and fiitnesses to be estimated with each additional segregating locus, the estimation of linkage disequilibrium, the sparse representation of some genotypic classes leading to unreliable estimates of their fitnesses, and the lack of independence between estimates. However, fitness estimation in the one-locus “marginal” populations described below yields information about the presence or absence of fitness interactions in multilocus populations. At the same time, the problem of estimation of linkage disequilibrium is removed and the other difficulties of multilocus fitness estimation are diminished. There is a pitfall in fitness estimation with one-locus populations that is per-