The loss of function phenotypes generated by RNAi, like those generated by classic genetics, are highly dependent on the genetic background - many genes have very different RNAi phenotypes in a wild-type worms from those seen in animals mutant for a specific gene. Such genetic interactions can provide insight into how genes are organised into pathways. To begin to map out genetic interactions in the signal transduction and transcriptional networks that underpin C. elegans development, we used RNAi to individually target approximately 1,700 genes and compare the phenotypes generated in wild-type animals with the phenotypes in each of about 40 mutant strains; each strain carries a mutation in a key signalling component or chromatin regulator (B.L, C.C, J.T, A.F and A.G.F, manuscript submitted). The approximately 1,700 genes targeted encode the great majority of genes involved in signal transduction and transcriptional regulation, as annotated in Kamath et al. 4 (Additional data file 1). During this screening, we noticed that the RNAi phenotypes of many genes that had weak phenotypes in wild-type animals were greatly enhanced in the strain lin-35(n745); this carries a putative null mutation in the p105Rb ortholog, lin-35 8. In this strain, the sterility and/or embryonic lethality of approximately 30% of all genes that had a weak phenotype in wild-type worms were enhanced (78 genes; Additional data file 2). Furthermore, 35 genes that had no detectable phenotype in wild-type worms had strong phenotypes in lin-35 mutants (Additional data file 2). In particular, many RNAi clones that result in partial F1 embryonic lethality in wild-type worms have complete P0 sterility or growth arrest in lin-35(n745) worms, suggesting a more rapid and complete inhibition of gene expression in the absence of lin-35 function.

The difference in RNAi phenotype for any gene that we observe in lin-35(n745) compared with wild-type could formally result either from an increase in RNAi sensitivity in the mutant or through some more complex genetic interactions (for example, through genetic buffering between lin-35 and a target gene). We believe the principal effect is through an increase in RNAi sensitivity for four reasons.

First, for genes that have a nonviable RNAi phenotype in lin-35(n745), the genetic null allele is also always nonviable, when known (35 genes) (Additional data file 3), suggesting that the stronger phenotype represents a near-null state.

Second, for genes whose genetic nulls are viable but have distinct postembryonic phenotypes (for example, uncoordinated, dumpy), we also detect enhanced similar postembryonic phenotypes by RNAi (Table 1); this group of genes includes genes that affect the neuronal system, which is largely refractory to RNAi in wild-type animals 4.

Thirdly, we tested whether inactivation of lin-35 increases gene silencing resulting from expression of an endogenously transcribed dsRNA. To do this, we took advantage of a system in which worms express GFP exclusively in the hypodermal seam cells, along with a dsRNA targeting the green fluorescent protein (GFP) mRNA 9. In wild-type animals, where RNAi works with normal efficiency, there is a low level of GFP fluorescence in the seam cells due to targeting by the co-expressed dsRNA (54% of worms have GFP expression visible in their midbody seam cells; Figure 1b, Table 2). If RNAi is used to target genes required for RNAi, however, this reduces GFP knock-down, and there is an observed increase in GFP levels (for example, for rde-4, 67% of worms have GFP expression visible in their midbody seam cells; Figure 1c, Table 2); conversely, targeting genes whose loss increases RNAi efficiency results in a further reduction of GFP expression (for example, for eri-1, 13% of worms have GFP expression visible in their midbody seam cells; Figure 1d, Table 2; p < 0.001, Chi squared test). We found that targeting lin-35 causes a strong enhancement of GFP silencing in the seam cells (1% of worms have GFP expression visible in their midbody seam cells; Figure 1e, Table 2; p < 0.001). When combined with the enhanced RNAi phenotypes described above, this result is consistent with a model in which inactivation of lin-35 enhances the efficiency of RNAi. In addition, since in this system the dsRNA is expressed in the same cells in which the targeting occurs, we conclude that inactivation of lin-35 must enhance the cellular process of RNAi-induced gene silencing, rather than just altering the uptake or systemic transport of dsRNA. Taken together, these results indicate that mutations in lin-35 cause an increase in the effectiveness of RNAi and that this results in stronger and more penetrant RNAi phenotypes for many genes, making lin-35(n745) an invaluable research tool. We note that similar findings were reported by the Ruvkun lab while this manuscript was in preparation 10.

Finally, although inactivation of LIN-35 results in RNAi hypersensitivity, it is possible that some of the genes with an enhanced phenotype in lin-35(n745) animals could represent genetic interactions between lin-35 and a target gene via a mechanism that is independent of the RNAi hypersensitivity of this strain. To directly identify these genes, we took advantage of a strain carrying a mutation in a lin-35 pathway gene that does not show an increased sensitivity to RNAi. In both mammals and worms, LIN-35/Rb proteins are proposed to function by directly binding E2F family proteins 1112. The strain efl-1(se1) 13 carries a weakloss-of-function mutation in the worm E2F family gene efl-1, which is known to function with lin-35 in regulating cell-cycle progression 14, as well as development of the vulva 12 and pharynx 15. efl-1(se1) animals do not show an increased sensitivity to RNAi, as judged by testing genes with an enhanced RNAi phenotype in rrf-3(pk1426) animals, or by inhibiting expression of efl-1 in the RNAi reporter strain GR1401. Thus, to identify genes that interact genetically with the lin-35 pathway, we tested whether genes that have an enhanced RNAi phenotype in lin-35(n745) animals, but not in rrf-3(pk1426) animals, also had enhanced RNAi phenotypes in efl-1(se1) animals (Additional data file 2). We found three genes that fulfilled these criteria (Table 3). The first of these genes is pha-1, which has previously been identified as genetically interacting with lin-35 and efl-1 15, so validating the success of our approach. The other two genes represent novel lin-35 pathway genetic interaction partners: dpy-22 is predicted to encode a component of the mediator complex that, like LIN-35 and EFL-1, probably also functions in chromatin remodelling 16, and Y106G6E.6 encodes a Casein Kinase I family member. Intriguingly, targeting Y106G6E.6 by RNAi results in abnormalities in early embryonic polarity (C Panbianco and J Ahringer, personal communication); strong reduction of efl-1 function has previously been shown to affect embryonic polarity 13. EFL-1 affects embryonic polarity at least in part through regulation of MAP kinase activity in the oocyte 13 and our data thus suggest that LIN-35, EFL-1, and Y106G6E.6 cooperate in some way to regulate MAPK activity in the C. elegans oocyte. There is no previously published functional association between p105Rb, E2F and a CKI family member and this underlies the strength of genetic interaction mapping as a way to reveal gene function.

We compared the RNAi sensitivity of animals carrying strong loss-of-function mutations in the two previously described genes that are known to negatively regulate RNAi in C. elegans, rrf-3 or eri-1, to that of lin-35(n745) animals. rrf-3(pk1426)5 and eri-1(mg366)6 enhanced the RNAi phenotypes of 70 and 69 of 1,749 genes tested, respectively, compared to 113 genes enhanced by lin-35(n745) (Figure 1a; Additional data file 2). Every gene displaying an increased phenotype with rrf-3(pk1426) or eri-1(mg366) also has an increased RNAi phenotype with lin-35(n745). In addition, many genes that have enhanced RNAi phenotypes in rrf-3(pk1426) or eri-1(mg366) have even stronger phenotypes in lin-35(n745).

Although the RNAi clones that we tested in each of the four strains represented a functionally biased set of genes, we also found very similar results when using random RNAi clones targeting genes with many diverse functions. In addition to the approximately 1,800 RNAi clones originally screened, we also screened the first 682 RNAi clones targeting genes on C. elegans chromosome III. These genes have very diverse molecular functions (Additional data file 4) and we found that 42 of these clones also had RNAi phenotypes that were stronger in lin-35(n745) than in rrf-3(pk1426) worms (Additional data file 5). In addition, it is not just the number of genes with enhanced RNAi phenotypes that is greater in lin-35 than in the other strains; the strengths of the RNAi phenotypes are also enhanced. For example, 11 of the genes we tested from chromosome III had an RNAi phenotype in rrf-3 worms that was further enhanced in lin-35 worms (Additional data file 5).

These results show that lin-35(n745) worms are more sensitive to RNAi than any previously described single mutant strain and are an ideal strain for new RNAi-based screens. This is a key finding - merely finding another hypersensitive strain is not a particularly useful research tool unless it is an improvement on the previously identified strains. Our ranking of the three strains is based on the use of a large set of test genes, and thus our conclusion is robust and not a curiosity of a few atypical RNAi phenotypes. We note, however, that Wang et al. 10 also provide evidence that a lin-35(n745); eri-1(mg366) double mutant strain may display a further enhancement in RNAi sensitivity to lin-35(n745), suggesting that these two genes may partially function in parallel.

For unknown reasons, many neuronally expressed genes appear largely refractory to RNAi in wild-type worms, precluding reverse genetic analyses 4. We generated strong phenotypes for several neuronally expressed genes in lin-35(n745) animals (Table 1), suggesting RNAi-based screens for neuronal functions might be feasible in this strain. To test further for enhanced RNAi sensitivity in the nervous system of lin-35(n745) animals, we focused on genes expressed in the six touch receptor neurons of C. elegans. These neurons sense gentle touch to the body, and several mechanosensory abnormal (mec) genes have been identified that are needed for their development or function 1718. Although RNAi has been detected in these neurons when dsRNA is injected into animals 19, it is not seen when dsRNA is delivered by feeding in wild-type animals (AC, C Keller, and MC, unpublished data), rendering high-throughput RNAi screens impractical.

We tested the touch sensitivity of wild-type and lin-35(n745) animals fed on bacteria targeting eight mec genes (mec-2, mec-3, mec-4, mec-8, mec-9, mec-10, mec-12 and mec-18) and two unrelated genes (gfp and sym-1). In wild-type worms, none of the bacterial strains caused touch insensitivity - that is, the Mec phenotype - either in adults that had fed on the bacteria throughout their entire larval development or in their progeny (n > 30 for each). Thus, if bacterial-mediated RNAi is having an effect in the touch neurons of wild-type animals, the effect is too small to generate a detectable phenotype. In contrast, in parallel experiments, lin-35 adults that had been fed with bacteria targeting mec-2, mec-3, mec-4, mec-9 and mec-18 throughout their larval development were touch insensitive, although the animals displayed the Mec phenotype with differences in penetrance and expressivity. Penetrance ranged from 47% (mec-9) to 83% (mec-2). Bacteria expressing mec-2, mec-3, and mec-4 dsRNA consistently gave a highly penetrant phenotype with strong expressivity (that is, the animals had a touch insensitivity similar to animals with null alleles). Bacteria making dsRNA for mec-12 produced a highly penetrant phenotype (63%) with intermediate strength (the animals responded to a few touches). mec-18 bacteria produced less consistent but easily detectable results; in some experiments the penetrance was high (60%) and expressivity strong, whereas in others the penetrance was lower (45%) and the expressivity intermediate. Bacteria producing mec-9 dsRNA gave the weakest positive results with penetrance of 47% and intermediate expressivity. These weaker effects seen with mec-9, mec-12 and mec-18 may be a consequence of the high expression of these genes in the touch neurons 20, which might overwhelm the RNAi machinery. Animals fed on bacteria targeting mec-8 or mec-10 were indistinguishable from those fed on bacteria for the gfp and sym-1 controls. Although negative RNAi results are difficult to interpret, genetic experiments 18 indicate that the amount of mec-8 activity produced in the embryo is sufficient for subsequent adult touch sensitivity, and elimination of mec-10 has only a slight effect on touch sensitivity (R O'Hagan, M Goodman, and MC, unpublished data).

These data indicate that neuronally expressed genes are effectively targeted by bacterial-mediated RNAi in the lin-35(n745) strain, thus providing a very useful tool to study gene function in these cells. These results also point to the expression and function of lin-35 in post-mitotic neurons.

In addition to demonstrating the usefulness of the lin-35(n745) strain for generating enhanced RNAi phenotypes, we wished to explore the connection between lin-35 and RNAi. lin-35 encodes the C. elegans ortholog of the human tumour suppressor gene p105Rb and is, therefore, presumed to act as a chromatin regulator. Thus, while rrf-3 and eri-1 encode proteins that are intimately connected with dsRNA synthesis and turnover, no clear mechanistic link is known between lin-35 and RNAi, making the connection between chromatin remodelling and RNAi an intriguing question. lin-35 functions in the synthetic Multivulva (synMuv) B pathway that is redundantly required with the synMuv A pathway to antagonise the outcome of Ras signalling in the specification of vulval cell lineages 8. Although some synMuv genes are of unknown molecular function, several synMuv B genes encode the worm orthologs of components of p105Rb transcriptional repressor complexes identified in mammals and flies 11212223. If the chromatin remodelling function of LIN-35 is important for its effect on RNAi, one would anticipate that strains carrying mutations in othersynMuv B genes would also be hypersensitive to RNAi. To test the RNAi sensitivity of synMuv strains, we used the subset of bacterial feeding clones that gave an enhanced RNAi phenotype in rrf-3(pk1426) animals. We tested these clones for enhanced RNAi phenotypes in each of the synMuv strains compared to in wild-type worms. We found that strains carrying inactivating mutations in the synMuv B genes lin-15B (Figure 1a, Table 4), dpl-1, and lin-9 (Table 4) also enhanced the RNAi phenotypes of the majority of these genes. In addition, Sieburth et al. 24 have shown that an eri-1; lin-15B double mutant is also hypersensitive to neuronal RNAi phenotypes. In contrast, strong loss-of-function mutations in two other synMuv B genes, lin-36 and tam-1, or the synMuv A gene lin-15A did not enhance any RNAi phenotypes for these genes (Table 4). We conclude that a subset of synMuv B genes negatively regulate RNAi, which we refer to as synMuv B(R) genes. Wang et al.10 obtained similar results.

In addition to increasing sensitivity to RNAi, inactivation of the genes rrf-3 or eri-1 also results in the silencing of somatically expressed transgene tandem arrays via an RNAi-dependent mechanism 9. Consistent with their having roles as negative regulators of the RNAi pathway, inactivation of synMuv B(R) genes also results in somatic transgene silencing 25 (data summarised in Tables 4 and 5 and Figure 2b,c). In contrast, inactivation of other synMuv B or synMuv A genes does not result in somatic transgene silencing 25 (Table 4). Somatic transgene silencing in animals with inactivated synMuv B(R) genes can be suppressed by inactivation of components of the RNAi machinery (Figure 2e,f,h,i, Table 5). Thus we conclude that inactivation of synMuv B(R) genes induces somatic transgene silencing as a result of an increase in RNAi. We also note, however, that mutations in at least one other synMuv B gene, tam-1, can enhance somatic transgene silencing 25, but without any observable effect on RNAi sensitivity (Table 4). This suggests that other genes may be able to enhance transgene silencing independently of the RNAi pathway via an unknown mechanism. Indeed, we found that RNAi of dcr-1 does not suppress transgene silencing in a strain carrying a mutation in tam-1 (reduced transgene silencing was seen in 0 of >300 tam-1(cc567) 25 animals tested (Additional data file 6)). The general picture is clear, however: the subset of synMuv B genes that affects RNAi sensitivity is very similar to the subset that alters transgene silencing, suggesting that these form a genetically distinct group of synMuv B genes.

Additional evidence suggests this subclassification of synMuv B genes is functionally relevant. Inactivation of a subset of synMuv B genes results in ectopic expression of a lag-2::gfp reporter gene 26. Strikingly, all of the synMuv B genes that we found to be negative regulators of the RNAi pathway and negative regulators of somatic transgene silencing also negatively regulate lag-2::gfp expression 26 (Table 4). This result suggests a similar synMuv B(R) pathway may regulate both the RNAi pathway and correct expression of this transgene, and supports the classification of synMuv B genes into at least two distinct functional subsets.

Combined mutations in both a synMuv A and a synMuv B gene (for example, in the lin-15A;B (n765) strain) results in the development of ectopic vulvae (the so-called multivulva or Muv phenotype). In an RNAi screen for genes that can suppress the Muv phenotype of a lin-15A;B(n765) strain, we identified two genes, mes-4 and zfp-1, that have been previously identified as necessary for both RNAi and somatic transgene silencing (Table 6; C.C and A.G.F, unpublished data) 927. These genes both have clear human orthologs, and are both predicted to encode components of chromatin modifying complexes: mes-4 encodes a putative trithoraxgroup histone methyltransferase protein with a SET domain and three PHD finger domains and is orthologous to human WHSC1L1, and zfp-1 encodes a PHD finger domain protein and is orthologous to human MLLT10/AF10, a fusion partner of MLL (a mes-4 related gene) in acute leukaemia 28. The requirement of mes-4 and zfp-1 for RNAi and their ability to suppress mutations in synMuvA;B mutants, suggests that a common mechanism may underlie the role of synMuv B genes in these two processes (Figure 3).

The precise molecular functions of the synMuv B(R) genes and of mes-4 and zfp-1 in vulva development and in RNAi are unknown; however many of these genes are predicted by sequence homology to regulate chromatin structure. One intriguing possibility is that a key function of the synMuv B(R) genes during vulva development may be to repress RNAi. The Muv phenotype might thus be due in part to alterations in RNAi-related processes. We investigated this in two complementary ways. Firstly, if the sole effect of synMuv B(R) genes on vulval development was through their effect on RNAi sensitivity, then other genes that similarly increase RNAi sensitivity should act as synMuv B genes. However, while targeting lin-35 by RNAi produces a strong Muv phenotype in a lin-15A mutant animal (as expected given its synMuv B activity), targeting eri-1 has no similar effect. Secondly, inactivation of the synMuv B(R) genes enhances RNAi and in the absence of synMuv A activity leads to multivulval development; to determine if these two functions were causally related, we asked whether inactivation of other genes that are essential for RNAi (rde-1, rde-4, rde-5, mut-7 or mut-16) suppresses the Muv phenotype of lin-15A;B(n765) - they do not. Hence, we find that genes that enhance RNAi do not all act as synMuv B genes and, conversely, that the RNAi machinery is not necessary for the synMuv phenotype. Thus, alterations in the efficacy of RNAi cannot alone account for the action of lin-35 in vulval development, although it may contribute to lin-35's role.

Wang et al. 10 suggest that the enhanced RNAi seen in synMuv B mutants may result from the misexpression of germline genes in somatic cells. Although this may contribute to the enhanced somatic RNAi seen in synMuv B strains, we found that lin-35(n745) animals also showed enhanced germline RNAi phenotypes (>50 genes gave strong sterility in lin-35(n745) but not in wild-type worms) (Additional data file 2). Although some of these sterile phenotypes may result from defects in somatic cells, a subset of these genes has been previously shown to function within the germline itself. In C. elegans, the Notch and MAP kinase pathways are both required within the germline for correct germline development 2930, and we found that four genes that function in these pathways also show strongly enhanced RNAi-induced sterility in lin-35(n745) worms (the genes glp-1, lag-1, let-60 and lin-45; Additional data file 3). Since the enhanced sterility seen with these genes must result from enhanced gene silencing within the germline itself, these data demonstrate that RNAi is also enhanced in the germline of lin-35(n745) worms, and that somatic misexpression of germline genes does not alone account for the enhanced RNAi seen in synMuv B mutants. We favour a model in which the synMuv B(R) genes and mes-4/zfp-1 act antagonistically to regulate the expression of a common set of target genes. These targets could include genes that are required for vulval development and genes required for RNAi, or the genes targeted by RNAi themselves. The antagonism may involve the direct repression of mes-4 and zfp-1 by the synMuv B(R) genes, or the antagonistic action of mes-4/zfp-1 and the synMuv B(R) genes on a common set of target genes (Figure 3). Alternatively, MES-4/ZFP-1 and the synMuv B(R) gene products may antagonise each other's functions by competing for a common set of co-factors.

All of the RNAi feeding experiments described in this manuscript were performed in liquid culture by adding synchronised L1 stage worms, unless otherwise indicated. A total of 1,868 bacterial RNAi feeding strains from the Ahringer library 4 targeting 1,749 genes were tested with each worm strain. The vast majority of these genes are those annotated as 'signalling', 'chromatin', or 'transcription factors' in reference 4 (Additional data file 1); feeding approximately 75% of these bacterial strains gave no visible RNAi phenotypes in wild-type worms 4. Bacterial RNAi feeding strains were grown overnight at 37°C in 400 μl 2TY plus 100 μg/ml ampicillin, induced with 4 mM IPTG (isopropyl-beta-D-thiogalactopyranoside) at 37°C for 1 hour, and resuspended in 400 μl NGM(Nematode Growth Medium) plus 4 mM IPTG plus 100 μg/mlampicillin. Approximately 10 L1 stage worms were dispensed to each well of a 96-well flat bottomed tissue culture plate together with 40 μl of resuspended bacterial culture. The plates were incubated with shaking at 20°C for four days, and embryonic lethal, sterile, growth defect and post-embryonic phenotypes were scored on a dissecting microscope. All RNAi feeding experiments were performed in quadruplicate and the phenotypes observed in each strain were directly compared to those seen in N2 worms grown in parallel.

Worm strain JR667 expresses GFP specifically in the hypodermal seam cells from an integrated tandemly repeated array of the construct wIs51, which contains the scm::GFP reporter. Worm strain GR1401 expresses the same integrated GFP transgene, as well as a dsRNA that targets GFP mRNA for degradation, also specifically in the hypodermal seam cells of the worm 9. RNAi experiments were performed on six-well plates seeded with the indicated bacterial RNAi feeding strain (grown overnight as described above). Approximately 10 L1 stage worms were added to each well and incubated at 25°C for four days, unless otherwise indicated. The progeny worms were washed off the plates, paralysed in 100 mM levimasole and GFP expression was visualised using an Olympus IX81 microscope. Control experiments used a feeding strain that does not target any C. elegans gene (constructed using primers sjjY95B8A_84.g, defined in 31). All other RNAi assays were performed exactly as described in 4.