Large-scale genetic epistasis networks using RNAi (Nat Methods. 2011 Apr;8(4):341-6)

Wang X, White KP.

Pairwise quantitative genetic interactions are mapped by combinatorial RNA interference in metazoan cells

An epistatic interaction between two genes is observed if the combined effects of mutations in both genes deviate from the sum of each mutation’s individual effects. We need to understand epistatic genetic interactions if we want to solve the puzzle of genotype-phenotype relationships in complex human diseases. Advances in genetic tools and automated technology have enabled investigators to construct genome-scale genetic interaction maps in yeast and bacteria1. However, it has been a challenge to develop comparable methodologies in metazoan cells. In this issue of Nature Methods, groups led by Michael Boutros and Wolfgang Huber report their pilot experiments in systematically mapping pairwise genetic interactions in cultured Drosophila melanogaster cells by RNA interference (RNAi)2.

Genetic interactions underlie the robustness of biological systems and the complexity of common genetic disorders. So far, genetic interaction networks are mostly constructed from experimental data in yeast, taking advantage of the large collection of deletion strains, and the automated technology to sort and score double mutants3, 4, 5. In higher organisms and tissue culture cells, such large deletion collections do not exist, and the process to generate double mutants is much more labor-intensive and time-consuming. RNAi is the most practical way to simultaneously perturb multiple genes in these systems. Several groups have used RNAi to map genetic interactions based on semiquantitative readouts such as lethality and sterility1, 6, 7. But concerns about inherent noise in RNAi assays, caused by off-target effects and variable knockdown efficiencies, delayed the use of RNAi in large-scale mapping of genetic interactions based on more quantitative readouts.

Huber, Boutros and colleagues set out to overcome the limitations of using RNAi for quantitative genetic interaction mapping by combining a rigorous experimental design with robust statistical modeling2. To assay pairwise synthetic interactions between 93 genes, they designed two independent ds-RNAs to each gene and did reciprocal RNAi knockdowns with all four combinations in Drosophila Schneider S2 cells (Fig. 1). Therefore, they assayed each pair of genes in eight biological replicates. And within each biological replicate, they repeated the phenotypic measurements eight times. The measurements in both biological and technical replicates were highly reproducible, allowing calculation of reliable interaction scores for gene pairs from a multipli-cative model.

Figure 1: Workflow of a combinatorial RNAi screen for genetic interactions developed by Horn et al.2.

shRNA, short hairpin RNA.

Full size image (92 KB)

An intriguing aspect of this study is its multiparametric design. Instead of focusing on a specific pathway or a defined biological process, they assayed three phenotypes that are easily measured: cell number, nuclear area and fluorescence intensity of the nuclear area after Hoechst (DNA) staining. This is the first effort to quantify multiple complex phenotypic features at a large scale for genetic epistasis in metazoan cells. These readouts provided rich and nonredundant information on genetic interactions. Only 20% of all the 637 interactions discovered were common among phenotypes, and about 50% of the interactions were specific to a single phenotypic readout.

Although individual interactions are context dependent, the global interaction profiles can provide important aggregate information that reveals biologically relevant interactions. Horn et al.2 constructed the interaction profiles for each gene as the vector of its interaction scores with all other genes. Without any prior assumptions, genes that have similar functions were clustered robustly by their global interaction profiles, independent of the phenotypes from which these profiles were generated. This property was consistent with the finding in yeast studies that genes in the same biological pathway tend to have similar profiles of interactions4, 5.

Horn et al.2 use these global interaction profiles to predict functions of previously uncharacterized genes. They first trained a classifier on the combined multiparametric interaction profiles of known genes in two signaling pathways, Ras-MAPK and JNK pathways. The classifier correctly classified both positive and negative regulators of the Ras-MAPK pathway and also led to the discovery of a new positive regulator of the Ras-MAPK pathway, whose function is conserved between fruit flies and humans. Although the number of genes in each pathway studied is relatively small, the predictive power of the classifier is very strong, which bodes well for the application of the method to larger datasets as they become available.

This elegant study illustrated how combinatorial RNAi can be used to systematically assess pairwise genetic interactions of metazoan genes. Although the current study accessed signaling molecules at a medium scale, this method could be readily applied to different gene sets at variable scales by other investigators with the appropriate resources, such as small interfering RNA libraries, liquid-handling robotics and automated imaging systems. These epistasis mapping approaches also pave the way for studies in mammalian cells, which will require improvements in RNAi knockdown efficiencies and avoidance of off-target effects. But for now it will be interesting to see this approach expanded in Drosophila cells. For example, larger-scale mapping may reveal the extent to which interactions among the cellular networks derived from yeast studies are conserved in higher organisms. Additionally, chemical-gene interaction networks can be mapped, as has been demonstrated in yeast5, to allow discovery of targets for small-molecule drugs. A future challenge will be to develop complementary computational methods to extract more information from the interaction profiles, to integrate data from different experiments, and to guide experimental design using existing data. As all-by-all genetic interaction matrix mapping is still a daunting prospect, it will be very useful to select gene sets that are likely to provide the richest information based on prior information8. Horn et al.2 have cleared the way for large-scale studies of genetic interactions in metazoan cells.