One of the most direct approaches to elucidating the role of any particular gene is to characterize its loss-of-function phenotype. Loss-of-function phenotypes have now been analyzed for almost all of the predicted genes of Saccharomyces cerevisiae
], Caenorhabditis elegans
], and Drosophila melanogaster
], and there are ongoing efforts to make comprehensive collections of mouse knockouts. In all, this gives us an unprecedented level of insight into eukaryotic gene function. However, the loss-of-function phenotype of any individual gene is highly dependent on the genetic context; specifically, variations in the activities of other genes will affect this phenotype (for review [4
]). If changes in the activity of one gene affect the loss-of-function phenotype of a second gene, then these two genes are said to interact genetically. Genetic interactions can be used to identify novel components of molecular pathways and can reveal the redundancy that underlies the robustness of genetic networks. Thus, although analyzing the loss-of-function phenotypes of all genes in a wild-type animal is a major advance, an understanding of how each phenotype is modulated by the activities of other genes will prove to be just as critical.
Recently, genetic interactions in S. cerevisiae
were investigated in a systematic manner using matings within a comprehensive collection of mutant strains. Pair-wise matings have identified over 4500 genetic interactions, demonstrating the extensive degree of redundancy in yeast [5
]. However, this approach is not currently feasible in any animal. No complete collection of mutant strains exists, and even if such strains were all available, large-scale matings are far more laborious in animals than in yeast, and so alternative strategies are needed.
One underlying cause of genetic redundancy may be gene duplication. Duplicated genes that retain at least partially overlapping functions can confer robustness to mutation in the other copy [7
]. However, there is still much debate about whether redundancy of duplicated genes can be evolutionary selected [9
]. Theoretical models have been proposed to explain the evolutionary stability of redundancy [12
], and indirect experimental evidence for the redundant functions of duplicated genes comes from the analysis of loss-of-function phenotypes of single genes; in both yeast and worms, inactivation of a duplicated gene is less likely to result in a nonviable phenotype than inactivation of a single copy gene [2
]. However, there are strong biases in the types of genes that are duplicated in genomes, which complicates the interpretation of these results [16
], and no attempt has yet been made to examine the extent of redundancy between duplicated genes in vivo
directly and systematically.
RNA-mediated interference (RNAi) is a powerful tool for studying the loss-of-function phenotypes of genes. In particular, in C. elegans
, RNAi by bacterial feeding has been used for genome-wide screens because it allows high-throughput (HTP) and low-cost analysis of the loss-of-function phenotypes of genes in vivo
]. However, RNAi has only been used extensively to target single genes. To study genetic redundancy systematically and to identify genetic interactions using RNAi, it is critical to establish and validate robust methods for simultaneously targeting multiple genes by RNAi using bacterial feeding ('combinatorial RNAi'). In the present report we show that by using combinatorial RNAi by bacterial feeding we can identify the majority of a testset of previously described genetic interactions. We used this technique to provide the first large-scale analysis of the redundant functions of duplicated genes in any organism, and we found that many duplicate gene pairs can retain redundant functions for more than 80 million years of evolution.