Gene function prediction from congruent synthetic lethal interactions in yeast

Ping Ye^1,2,a, Brian D Peyser^3,4,a, Xuewen Pan^2,4, Jef D Boeke^2,4, Forrest A Spencer³ & Joel S Bader^1,2

^aThese authors contributed equally to this work.

Article highlights

Synopsis

With the completion of the genome sequence for human and model organisms, the next phase is to understand how genes and gene products function together in pathways. Assays for physical interactions between proteins reveal how protein subunits assemble into larger machines and how protein–protein interactions provide the mechanism for regulation. Distinct from assays for physical interactions are assays for genetic interactions. Two genes have a genetic interaction if a double mutant (including null alleles, other mutant alleles, and dosage-dependent effects) has a phenotype distinct from the phenotype of the individual single mutants. Unlike a physical interaction, however, a genetic interaction does not provide direct evidence for the pathway wiring underlying the observation.

This manuscript describes a method for reverse-engineering a pathway wiring diagram underneath genetic interaction data and applies the method to high-throughput screens in yeast, which has 6000 total genes and 5000 non-essential genes. While each of the 5000 non-essential gene deletions yields a viable phenotype, pairwise deletions of non-essential genes may be lethal. A lethal pairwise mutation is termed a synthetic lethal genetic interaction and indicates how gene functions buffer or compensate each other.

The metaphor we employ for an essential biological process is an electric circuit where nodes represent genes or gene products and wires represent physical interactions between biomolecules (Figure 1). Deleting a gene corresponds to cutting the wires it connects. Robustness arises from multiple pathway branches connected in parallel. If one branch is cut, current still flows, but if both are cut the process fails and the cell dies. In this picture, synthetic lethal interactions should be observed between pathway branches, but not within branches. Synthetic lethal interactions between pathways are orthogonal to physical interactions within pathways. Two genes that share synthetic lethal interaction partners are therefore likely to function within the same pathway branch. The genes that share synthetic lethal partners, termed congruent genes, should have greater functional similarity than genes with direct synthetic lethal interactions.

Figure 1

Congruent synthetic lethal (SL) interactions are consistent with functional pathway membership. (A) A simplified synthetic lethality pathway model. Black arrows indicate the schematic flow of a process, with essential genes (red circles) connected by non-essential genes (black circles) organized into two parallel pathway branches (black dashed lines). If at least one of the pathway branches is required for viability, SL interactions (red lines) will be observed between the pathway branches but not within a pathway branch. In this picture, deleting any component of a pathway branch destroys its activity. (B) Directly observed SL genetic interactions bridge pathway branches. The table indicates that SL interactions will be observed between components of the two pathway branches, whereas no interactions will be observed within a branch. (C) Functional associations inferred from the congruence score (blue lines) join the components of a pathway branch. The table indicates raw number of SL interaction partners shared by a pair of genes and its conversion to the congruence score, calculated as the -log₁₀P-value for partner sharing. The congruence connections are orthogonal to the direct SL interactions and align with pathway membership.

Full figure and legend (219K)Figures & Tables index

The congruence score provides a numerical ranking of the degree of partner sharing. It is defined as the -log₁₀ of the P-value for the number of shared genetic interaction partners of two genes. We find that genes with significant congruence score have more similar database annotations than genes with direct synthetic lethal interactions. Products of congruent genes are also more likely to have direct physical interactions or to share protein complex membership than products of synthetic lethal genes.

We conducted unbiased, genome-scale tests of the concept of congruence by identifying landmark genes whose mutants have a distinct phenotype, ranking the rest of the genome by congruence to the landmarks, and scoring the phenotypes of the mutants in rank order (Figure 3). Genes congruent to known members of the dynein–dynactin spindle orientation pathway exhibit a nuclear migration defect rate that increases with increasing congruence score (Figure 3A). One of these genes is YLL049w, an uncharacterized ORF. Pathway membership for YLL049w is further defined by the observation that the temperature dependence of its defect rate matches JNM1, a component of dynactin, rather than KIP2, a kinesin-like motor protein involved in delivering dynein to the cell cortex. We have also independently validated a physical interaction between Yll049w and Jnm1p. Taken together, these data indicate a role for YLL049w in a dynactin-related activity within the dynein–dynactin spindle orientation pathway.

Figure 3

The congruence score but not the number of synthetic lethal interactions predicts numeric phenotypes for deletion mutants. (A) Null mutants of 59 genes with congruence score 4 for six landmark genes (DYN1, ARP1, DYN2, JNM1, NUM1, and NIP100) known to be required for robust nuclear migration were measured for percent abnormal nuclear migration at 13°C. Each mutant's nuclear migration defect is plotted by congruence score to each landmark gene (congruence score range is labeled) and by average congruence score (dots). (B) Null mutants of 31 genes with congruence score 4 for landmark gene CIN1 known to be required for benomyl resistance were tested for benomyl sensitivity at concentration 5 g/ml. The fraction of benomyl-sensitive null mutants is plotted with each congruence score cutoff. (C, D) Null mutants of 451 candidate benomyl-resistant genes are ranked based on their average congruence score or number of synthetic lethal interactions with seven landmark genes (CIN1, YML094C-A, PAC10, PFD1, GIM3, TUB3, and GIM5) known to be required for benomyl resistance (Pan et al, 2004). The LD₅₀ benomyl concentration is defined by the lowest benomyl concentration when the control/experimental hybridization signal concentration 2. The red mark represents the median LD₅₀ benomyl concentration.

Full figure and legend (116K)Figures & Tables index

In a second test, we identified genes congruent to CIN1, a microtubule biogenesis gene, whose deletion mutant confers sensitivity to the anti-microtubule drug benomyl. Genes congruent to CIN1 are enriched for benomyl sensitivity (Figure 3B). Furthermore, the quantitative LD50 benomyl concentration is correlated with the congruence score to seven benomyl-sensitive landmarks (Figure 3C). Finally, a predictor based on the number of direct synthetic lethal interactions with benomyl-sensitivity landmarks, rather than on the congruence score, fails to predict benomyl sensitivity (Figure 3D). This result is consistent with a metric that successfully identifies within-pathway gene pairs, which are expected to exhibit more phenotypic similarity than between-pathway gene pairs.

Beyond providing novel evidence for the function of Yll049wp, this work is significant in providing a framework for interpreting the results of genetic interaction screens. Networks are becoming increasingly popular models for visualizing and analyzing biological information. Genetic interaction screens of knockout alleles yield pairwise relationships between genes. While it is tempting to use genetic interactions as evidence for network edges, we show that a more powerful interpretation is to infer edges that are orthogonal to the direct genetic interactions. While the methods have been developed and applied to yeast knockout screens, they should be applicable to reduction-of-function screens using RNA interference in higher eukaryotes and metazoans.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.