Timmons and Fire [4] first described a method for RNAi in which bacteria expressing dsRNA are fed to C. elegans. A fragment corresponding to the gene of interest is cloned into a feeding vector (L4440) between two T7 promoters in inverted orientation and is transformed into a bacterial strain carrying IPTG-inducible expression of T7 polymerase [4]. Recently, Timmons and Fire also showed that use of an E. coli strain (HT115(DE3)), which lacks double-strand-specific RNase III, improves the ability to produce RNAi phenotypes by feeding (L. Timmons and A. Fire, personal communication; and Figure 1).

To determine feeding conditions that maximize observable phenotypes, we started with the existing L4440 vector and strain HT115(DE3) and varied a number of parameters that could affect the efficiency of RNAi. We chose two initial test genes that were easy to assay: gpb-1, for which mutants are embryonic lethal, and unc-22, which results in a post-embryonic uncoordinated phenotype (Unc), as determined by deletion mutants and by RNAi [4,7]. We first tested different methods of induction with isopropylthiogalactoside (IPTG) to see if this would affect the RNAi phenotypes observed. Uninduced bacteria produced no phenotypes, but, somewhat unexpectedly, each presumably stronger method of induction resulted in a lower penetrance of phenotypes, culminating in 0% phenotype from overnight induction in culture (Table 1). The best induction method was to grow bacteria in culture without induction, to seed these bacteria onto plates containing IPTG, and then to incubate overnight at room temperature; with this method, gpb-1 produced 100% dead embryos and unc-22 produced 99% Uncs. To further test this new induction method, we fed two more genes, par-1 and par-3, mutations in either gene result in embryonic lethality [8,9]. Using our optimized induction conditions, feeding par-1 and par-3 resulted in 100% and 96% dead embryos, respectively, similar to the results obtained with null mutants (Table 1).

To determine the optimal concentration of IPTG for producing RNAi phenotypes, we titrated the IPTG concentration from 10 mM to 1 pM plus no IPTG. We found that 1 mM IPTG gave us the highest penetrance of phenotypes (Table 2). In addition, we tested the effect of seeded bacterial density and growth phase on the ability to generate phenotypes; for both par-1 and unc-22, saturated cultures produced phenotypes as well as log-phase bacteria (see the Materials and methods section). We also compared feeding at 15°C versus 22°C and found that although there is some gene-specific variation in RNAi effectiveness between these two temperatures, there is no generalizable difference (data not shown).

Finally, we tested the effect of length of feeding time on the penetrance of RNAi phenotypes observed. Many genes were not effectively silenced after 24 hours at 22°C, but were after 48 hours (see the Materials and methods section and Figure 2). In general, it appears that allowing worms to ingest dsRNA-expressing bacteria for longer periods of time increases the efficiency of RNAi by feeding. However, feeding worms beginning earlier than the L4 stage did not improve the penetrance of RNAi in the progeny (data not shown). After 48 hours, L4-stage hermaphrodites become older adults, and many stop laying fertilized eggs; thus we have routinely used the longest feeding time possible (36-40 hours at 22°C), which ensures that these worms will continue to lay fertilized eggs during the time window from which progeny are assayed (the subsequent 24 hours). We found that 72 hours at 15°C gives a similar level of RNAi inhibition for most genes (data not shown).

Many genes have pleiotropic effects in vivo but have one dominant mutant phenotype that masks other informative phenotypes; thus, we decided to see if we could elicit such masked phenotypes by titrating the concentration of IPTG and, by extension, the degree of RNAi. We titrated the IPTG concentration from 1 pM to 1 mM in three log increments; for comparison, we also tested no IPTG (uninduced) and 10 mM IPTG. We analyzed four genes that could possibly be titrated to other distinct phenotypes: unc-37, which has a known allelic series with strong alleles resulting in embryonic lethality and weak alleles resulting in an uncoordinated (Unc) phenotype [10]; hlh-2, which is embryonic lethal by RNAi injection and is expressed in some neural precursors which eventually form the ventral nerve cord ([11] and M. Krause, personal communication); mei-1, which is required for meiotic spindle formation and is embryonic lethal by RNAi injection, but for which a weak mutant allele produces a high-incidence-of-males (Him) phenotype [12,13]; and rba-2, a very strong embryonic lethal gene by RNAi injection which is also involved in repression of vulval cell fates [14].

All four genes were 100% embryonic lethal at 1 mM IPTG (Table 2). For the first three genes, decreasing the IPTG concentration to 1 µM reduced the embryonic lethality sufficiently to expose high levels of a secondary phenotype. For unc-37 and hlh-2, 100% of the surviving worms were Unc. For mei-1, a Him phenotype was seen: 8% of the surviving worms were male, which is significantly greater than the 0.5% which normally arise by non-disjunction in a wild type hermaphrodite culture. As we further decreased the IPTG concentration to 1 pM, these levels of embryonic lethality tapered off, as did the penetrance of secondary phenotypes. With no IPTG, all the worms fed these genes had wild-type progeny. For rba-2, however, embryonic lethality was observed at all concentrations of IPTG, including 56% lethality without any IPTG. Therefore, these bacteria are likely to express a low level of T7 polymerase in the absence of induction, which is sufficient to produce an RNAi phenotype for some genes. From these data, we conclude that it is possible to titrate RNAi phenotypes by feeding bacteria induced with different concentrations of IPTG, which results in a series of hypomorphic alleles analogous to an allelic series of mutants.

We also tested RNAi by feeding for two genes simultaneously. When we fed one embryonic lethal gene and one non-lethal gene, we found a reduced penetrance of embryonic lethality compared to feeding the lethal gene alone: feeding gpb-1 and unc-22 together reduced the embryonic lethality to 50% (versus 100% for gpb-1 alone) and feeding par-3 and unc-22 together reduced the lethality to 85% (versus 96% for par-3 alone). Furthermore, in both cases the resulting progeny also failed to display the unc-22 phenotype (data not shown). Thus, it appears that feeding two genes greatly reduces the strength of phenotype produced by either.

In a separate experiment, we diluted the unc-37 HT115 strain 1:1 with E. coli that does not produce dsRNA (OP50, a strain commonly used for growing C. elegans). In this case, the phenotype changed dramatically from 100% to 0% embryonic lethality, with 100% of progeny displaying post-embryonic phenotypes (uncoordinated, rolling, and body morphology defects), which is similar to the phenotypes obtained by inducing the bacteria with 1 fM IPTG (data not shown). Thus, diluting feeding bacteria with bacteria not expressing the dsRNA can also be used to generate weak hypomorphic phenotypes.

To determine which tissues are susceptible to RNAi by feeding, we fed bacteria expressing green fluorescent protein (GFP) dsRNA to hermaphrodite worms with a transgenic GFP reporter gene expressed in all somatic tissues (egl-27::gfp; Figure 2a) [15]. After being fed for 24 hours at 15°C, GFP expression was markedly reduced compared to similarly treated unfed worms (n = 28; compare Figure 2b and a). After 48 hours, with the exception of the nervous system, GFP was not detectable in somatic tissues; furthermore, neural GFP was dramatically reduced in 91% of worms (n = 22; Figure 2c). A similar level of inhibition of GFP expression was observed after 72 and 96 hours of feeding (Figure 2d,e). Although GFP fluorescence in fed worms was abolished or severely reduced, GFP was sometimes expressed at high levels in late-stage embryos derived from these worms (data not shown). This suggests that some zygotic embryonic transcripts may be difficult to silence, possibly because a continuous supply of dsRNA cannot be provided through the eggshell. In summary, RNAi by feeding efficiently silences genes in most C. elegans somatic tissues; however, the nervous system has a delayed and somewhat less robust response to RNAi compared to other tissues.

To test the strength of RNAi by feeding using the above optimized protocol, we compared it to RNAi by injection. We first tested a set of 14 known maternal-effect embryonic lethal genes (gpb-1, par-1, par-2, par-3, par-6, cyk-1, skn-1, dnc-1, bir-1, pal-1, dif-1, plk-1, dhc-1, and mex-3) to compare the lethality obtained with both methods (see [16,17,18] for a review of maternal-effect genes). All genes tested were 100% embryonic lethal by both feeding and injection except for par-3, which resulted in 97% dead embryos by feeding but 100% by injection, and cyk-1, which resulted in 55% dead embryos by feeding but 100% by injection (Table 3). For some genes, for example par-1, RNAi by feeding (n = 32 fed worms) and by injection (n = 9) both resulted in 100% embryonic lethality of the progeny in all cases. Whereas RNAi of par-3 resulted in 100% embryonic lethality for all injected worms (n = 9), worms subjected to RNAi of par-3 by feeding resulted in 100% dead embryos in 16/24 cases but lower levels of lethality (on average 88%) in 8/24 cases (data not shown). The overall comparison shows that RNAi by feeding is of similar strength to RNAi by injection for maternal-effect embryonic lethal genes. For some genes, however, RNAi by feeding appears to be somewhat more variable than RNAi by injection.

To further confirm that RNAi by feeding is comparable in effectiveness to injection for analyzing maternal-effect genes, we made four-dimensional time-lapse video recordings of developing embryos whose mothers were fed with dsRNA corresponding to the 10 above genes with a mutant phenotype detectable by the third cell division (gpb-1, par-1, par-2, par-3, par-6, cyk-1, dnc-1, bir-1, plk-1, and dhc-1,). In each case, the known null phenotypes were obtained (Figure 3, and data not shown; for cyk-1, one of two embryos recorded had the known phenotype and the other survived). For example, par-2(RNAi) by either method yielded spindle orientation defects, and bir-1(RNAi) by either method yielded embryos with cytokinesis defects (Figure 3) [19,20].

To compare the sensitivity of feeding versus injection in detecting phenotypes from randomly selected genes, we tested 86 genes from the middle of C. elegans chromosome I on consecutive cosmids from K04G2 to R05D11. Of these genes, 12 yielded a phenotype by injection and 13 by feeding (Table 4). These data suggest that feeding and injection are similarly sensitive for detecting genes that give an RNAi phenotype. The number of phenotypes detected in this experiment is, however, too small to allow a fair comparison of the strengths of the two methods.

To obtain a quantitative measure of the efficiency of feeding versus injection for RNAi, we compared the two methods on a larger data set. Fraser et al. have constructed an RNAi feeding library for C. elegans chromosome I [21]. After performing RNAi by feeding on the first 1,200 predicted genes on chromosome I, phenotypes were identified for 168 predicted genes. We then performed RNAi by injection on these genes and compared the phenotypes to those obtained by feeding (Figure 4; see the Materials and methods section for scoring criteria) We reasoned that using a set of genes initially identified by feeding for this comparison would be valid, as in our previous comparison of feeding and injection, feeding successfully detected all those genes with RNAi phenotypes detected by injection.

Of the 168 genes, 123 were determined to be embryonic lethal by either method, of which feeding detected 97% and injection 91%. The embryonic lethality was of equal penetrance using either method for 77% of these genes, was stronger by feeding for 15% of the genes, and was stronger by injection for 8% of the genes. Of 52 genes giving a sterile phenotype by either method, 96% were identified by feeding and 48% by injection; 44% were detected by both methods, 52% only by feeding, and 4% only by injection. And, finally, of 154 genes giving a post-embryonic phenotype by either method, 96% were identified by feeding and 64% by injection; 60% were detected by both methods, 36% only by feeding, and 4% only by injection. In addition to the above, we also injected 30 random genes that gave no phenotype by feeding, and none gave a detectable phenotype by injection (data not shown). Thus, from this expanded data set, we conclude that RNAi by feeding is roughly equivalent to injection for detecting embryonic lethality, and is somewhat better for detecting genes causing sterility or other post-embryonic phenotypes.

Standard methods were used for culturing C. elegans on NGM (nematode growth medium) [30]. Fragments designated for RNAi were obtained by polymerase chain reaction (PCR) from genomic DNA and were cloned into the L4440 feeding vector (pPD129.36) [4]; all fragments were between 500 and 2,700 base pairs (bp) in length. The resulting plasmids were transformed into the HT115(DE3) RNase Ill-deficient E. coli strain, which was previously shown by Timmons and Fire to be beneficial for RNAi by feeding (L. Timmons and A. Fire, personal communication). The HT115 genotype is as follows: (F^-, mcrA, mcrB, IN(rrnD-rrnE)1, lambda^-, rnc14::Tn10(DE3 lysogen:lacUV5 promoter-T7 polymerase)). The RNase III gene is disrupted by a Tn10 transposon carrying a tetracycline-resistance marker. Inclusion of tetracycline in feeding plates or in bacterial cultures used for feeding in many cases resulted in a weaker RNAi effect (data not shown), perhaps because the cultures grew very poorly, and thus it was not included in our feeding experiments; however, bacteria were selected on tetracycline plates before feeding.

The following primer pairs were used for PCR amplification: gpb-1 (5'-ATGAGCGAACTTGACCAAC-3' and 5'-TTAATTCCAGATCTTGAGG-3'), par-1 (5'-CAAAGCACGTGATAACCGG-3' and 5'-TTGGTGGCTCAATAAATGGC-3'), par-3 (5'-TTTGGCTTCACTGTGACCG-3' and 5'-TGATGTGCTGTGGATCAGC-3'), par-2 (5'-GCCGTCGCCCACTGTCG-3' and 5'-CCGGCTCCAGAGTGTCC-3'), cyk-1 (5'-GAAGAACAGCTGACCAGCG-3' and 5'-GACGATTCAATGCAATGATGG-3'), skn-1 (5'-CTGCCGAAGAGAATGCTCG-3' and 5'-GTTTGGTACAACTTCTGTTGG-3'), dnc-1 (5'-TCTCCACTTTCTACTACAGC-3' and 5'-TGTTCTTGGAAGCCAGCG-3'), bir-1 (5'-ATGGCACCCGGGACCAA-3' and 5'-TTATTTGCCGCGGCGGC-3'), pal-1 (5'-GGGGTACCCCAATGTCGGTCGATGTCAAGTCG-3' and 5'-CATGCCATGGCATGGTACTTATAGCCGAATCTTCTG-3'), dif-1 (5'-ACGCATTGAAATGTCGGACG-3' and 5'-TTGCAGGGAAAGCACGGAG-3'), and plk-1 (5'-GACAAGGATCGTGGGACC-3' and 5'-AGCACAGCAACTTGGTGG-3'). Primer pairs for mex-3, dhc-1, par-6, unc-37, hlh-2, mei-1 and rba-2, the 88 genes on cosmids K04G2 to R05D11, and the 1,200 genes used for large-scale comparison of feeding and injection were obtained as part of the Research Genetics C. elegans GenePairs collection. Fragments corresponding to unc-22 and GFP were generated from L4440-based vectors containing those inserts (pLT 61.1 and pPD128.110, respectively) [4].

Single colonies of HT115 bacteria containing cloned L4440 plasmids were picked and grown in culture in LB with 50 µg/ml ampicillin (Amp), except where indicated. The following methods were used to induce expression of dsRNA:

L4-stage hermaphrodite worms were placed onto NGM plates containing seeded bacteria expressing dsRNA for each gene and were incubated for 36-40 h at 22°C or for 72 h at 15°C. Then, three worms were independently replica plated onto plates seeded with the same bacteria and were allowed to lay eggs for 24 h at 22°C before being removed. Progeny were scored for embryonic lethality after a further 24 h at 22°C, and post-embryonic phenotypes were scored blindly by two independent observers at the end of four successive 12-h intervals. Progeny laid on the first plate were also scored for post-embryonic phenotypes. A gene was found to be positive for a given phenotype if it could be observed in at least two of three worms or their progeny in at least two independent feeding experiments. Feeding times shorter than 36-40 h at 22°C or 72 h at 15°C were not always sufficient to produce a strong RNAi effect; for example, par-2 and par-3 were 53% and 23% embryonic lethal, respectively, after feeding for 24 h but were both 100% embryonic lethal after feeding for at least 36 h at 22°C (our unpublished results).

Single colonies of HT115 bacteria containing cloned L4440 plasmids with fragments corresponding to unc-37, hlh-2, mei-1 or rba-2 were treated as described in the optimal induction method above, except that for each gene, NGM plates were used with the following IPTG concentrations: 0, 1 pM, 1 nM, 1 µM, 1 mM, and 10 mM. Worms were incubated for 72 h at 15°C before being replica plated and scored.

Worms containing an extrachromosomal array expressing egl-27::gfp, a ubiquitous somatic GFP reporter, were fed bacteria expressing GFP dsRNA for either 24, 48, 72, or 96 h at 15°C. Those worms, as well as control worms fed bacteria not expressing GFP dsRNA, were photographed under identical conditions. Figure 2 illustrates the GFP expression of a representative worm from each time point. Many worms from the 72 h and 96 h time points had almost no detectable GFP expression in any tissue.

Injections were performed as in [1]. Templates for dsRNA synthesis were made by PCR on L4440-based feeding constructs using T7 primer (5'-CGTAATACGACTCACTATAG-3'). Sense and antisense RNAs were synthesized in a single reaction in vitro using a T7 polymerase-based kit (Promega). Double-stranding was achieved by incubation at 72°C for 10 min, and the sizes of dsRNA products were verified by electrophoresis. dsRNA was injected at a concentration of 0.5-1.0 mg/ml into one or both gonad arms (we and others have found that injection into one or both gonad arms produces equivalent effects). Injected worms were allowed to recover at 22°C for 24 h post-injection, then were replica plated and allowed to lay eggs for 24 h. Injected worms and their progeny were scored as previously described for RNAi by feeding.

Mothers fed dsRNA were dissected in egg buffer (118 mM And, 40 mM KCl, 3 mM CaCl₂, 3 mM MgCl₂, 5 mM HEPES pH 7.2) on a coverslip to release young embryos. The cover-slip was inverted onto a 3% agar pad, which was then sealed with petroleum jelly. A series of 12 focal planes was recorded every 30 sec for 1 h using Openlab software (Improvision) controlling either a Zeiss Axioplan 2 or Leica DMBRE microscope.

RNAi was performed by feeding at 15°C using the optimized method described above or by injection. Embryonic lethality was scored by estimating the percentage of dead embryos to the nearest 10% among the offspring of the three worms replica plated for each gene. For embryonic lethality, feeding and injection were considered to be of equal strength if the percentages of dead embryos were within 10%, and one method was considered stronger than the other if the percentage of dead embryos was at least 20% greater than that obtained by the other method. Fed or injected worms were considered sterile if they had fewer than 10 progeny on average, as wild-type worms under similar conditions typically have more than 50 progeny. Post-embryonic phenotypes were scored if more than 10% of the progeny had a given phenotype. For sterility and post-embryonic phenotypes, feeding and injection were considered to be of equal strength if both resulted in that phenotype, and one method was considered stronger if it resulted in that phenotype but the other method did not.