We began our analysis with 26 genes, 23 PAL-1 targets and three genes likely to have genetic interactions with PAL-1 targets. Mutations in 15 of these genes have been isolated, and appropriate strains were obtained from the Caenorhabditis Genetics Center (Table 1). To screen for synthetic lethal interactions, RNAi of the 13 nonessential genes was performed and embryonic lethality scored in each of the mutant strains as well as wild type to generate a synthetic lethal matrix (Figure 1 and Additional data file 1). However, RNAi of a given gene could not be used if it results in 100% lethality in wild type (pal-1, elt-1, unc-62, and lin-26), and so the matrix is not entirely symmetric. To reduce the penetrance of lethality following RNAi for these four genes we used a lower concentration of double-stranded RNA (dsRNA), but the results were variable (data not shown), and we decided that it was an unreliable approach. We also tried to assemble the matrix using exclusively RNAi, but we found that soaking worms in two different dsRNA molecules targeting two different genes at once to be unacceptably inefficient (Additional data file 2).

Given the effects of mutation alone without RNAi, and the effects of RNAi in wild type, it may not be obvious from the raw data which disruptions result in significantly more lethality when paired than when alone. Measurements of lethality were thus duplicated and a t-test was used to assign statistical significance to each of the interactions tested (Figure 1 and Additional data file 1). Seven genetic interactions were identified (P < 0.001) among six genes out of 195 total interactions tested. In addition to the genes shown in Figure 1, RNAi of the remaining genes for which no mutant was available was performed in all 16 strains. Because synthetic lethality was measured only a single time for each of these possible interactions, the data could not be subjected to statistical analysis. However, RNAi of these additional genes did not appear to elevate lethality in any of the genetic backgrounds tested (data not shown).

The T-box family of proteins regulates diverse aspects of animal development, including cell fate specification, differentiation, and morphogenesis 17. The C. elegans genome contains 21 of these animal-specific genes, compared to 32 in C. briggsae and eight in D. melanogaster 1819, suggesting duplication or retention within nematodes. tbx-37 and tbx-38 have been shown to redundantly control embryonic patterning of the C. elegans AB lineage 20, and our systematic approach identified tbx-8 and tbx-9 as a synthetic lethal pair. tbx-8 and tbx-9 are expressed in nearly identical temporal and spatial patterns in the early embryo, with the strongest expression detected in dorsal epidermal cells and posterior expression depending on pal-1 function 132122. tbx-8(ok656), tbx-8(RNAi), and tbx-9(RNAi) each result in a low frequency of embryonic lethality (less than 5%) with 1-10% of hatching larvae displaying posterior morphological defects (Figure 2) 2122. However, 50% of tbx-8(ok656); tbx-9(RNAi) embryos fail to hatch and those that do display severe morphological defects and die as larvae (Figure 2). Phenotypic analysis demonstrates that tbx-8 and tbx-9 are required for intercalation of dorsal epidermal cells thus explaining the morphological defects observed following their disruption 22.

The overlapping function of tbx-8 and tbx-9 is not the residual result of a recent gene duplication event. The C. briggsae genome contains a pair of syntenic orthologs (by best reciprocal blastp match) 18, and as in C. elegans, the C. briggsae orthologs have overlapping function during morphogenesis (Table 2), indicating that the functional relationship between this pair of genes has been selectively maintained for over 100 million years. This observation suggests that the overlapping functions of other pairs of T-box genes, which are likely to be common throughout the Metazoa, may also be conserved.

It has been suggested that modules of genes comprise units of biological function 23. Because they operate in concerted fashion, genes in a module are expected to have increased likelihood of being co-expressed and sharing physical and genetic interactions relative to a random set of genes. Clustering a large synthetic lethal matrix in S. cerevisiae identified multiple functional modules 9, and we identify a module in our matrix around the hlh-1 gene (Figure 1). We detect five genetic interactions (P < 0.001) out of six tested among hlh-1, hnd-1, and unc-120. All three genes encode transcription factors expressed exclusively in the muscle progenitor cells of the early embryo 132425. Disruption of function by mutation or RNAi of any one of the three genes results in a low frequency of embryonic lethality with embryos arresting paralyzed at the two-fold stage (Pat) (Figure 3). Embryos require muscle contraction to elongate and pass from the two-fold to three-fold embryonic stage, and the Pat phenotype results from a complete failure of muscle differentiation 2627. A large fraction of animals that do hatch after disruption of hlh-1, unc-120, or hnd-1 show a less severe phenotype (uncoordinated, dumpy larvae) likely reflecting partial muscle function 2627 (Figure 3). However, disruption of function of any two of the three genes significantly elevates the frequency of Pat embryos, with disruption of hlh-1 and unc-120 resulting in 100% Pat embryos (Figure 3). The fact that hlh-1, unc-120, and hnd-1 have identical individual and synthetic phenotypes, in addition to their identical early embryonic expression patterns, indicate that they comprise a muscle-differentiation module. Although fewer genes define this module than those identified in larger screens in yeast 9, the specificity of its defining phenotype indicates that hlh-1, unc-120, and hnd-1 function in concert, and it is likely that future synthetic screens will identify additional members of the module.

If both RNAi and genetic disruptions of gene function are null, the synthetic lethal matrix should be symmetric. However, we can only test the pair of reciprocal interactions when mutations of both genes are available and RNAi of neither gene results in 100% lethality in wild type. Although there is some symmetry in the matrix around hlh-1 (Figure 1), the interactions detected are not all reciprocal (Figure 3). The efficiency of RNAi varies from gene to gene and often fails to result in null phenotypes, suggesting a significant number of false negatives among the interactions tested and explaining the imperfect symmetry among the interactions detected. The pattern of interactions defining the muscle module suggests that hlh-1 is the most essential of the three genes, or that its function is the most potent, and hnd-1 the least (Figure 3), a model that accounts for the reported lack of genetic interaction between heterozygous hlh-1 and homozygous hnd-1 mutations 25.

The functional importance of the muscle module we identified is underscored by its conservation. Well characterized genetic interactions exist among the vertebrate homologs of hlh-1 - the myogenic basic helix-loop-helix (bHLH) regulatory factors MyoD, myf5, myogenin, and MRF4 28. In addition, interactions between vertebrate myogenic bHLH genes and homologs of unc-120 (the MEF2 group of MADS-box regulators) have been described 29. Interactions among and between vertebrate myogenic bHLH and MADS-box regulators is likely to be the result of cooperative activation of common target genes as well as their ability to activate each other transcriptionally 282930. It is possible that hlh-1, hnd-1, and unc-120, like their vertebrate counterparts, display modular genetic interactions because they share common target genes, and they may also activate each other transcriptionally.