We have used an antisense strategy to effectively disrupt the expression of two genes encoding myofilament proteins present in C. elegans body wall muscles. DNA segments from the unc-22 and unc-54 genes have been placed in reverse orientation in vectors designed to produce RNA in body wall muscles. When the resulting plasmids are injected into oocytes, progeny with defects in muscle function are produced. These animals have phenotypes consistent with reduction and/or elimination of function of the gene to which antisense RNA has been produced: twitching and disorganization of muscle filaments for the unc-22 antisense constructs and lack of muscle tone, slow movement, and egg laying defects for the unc-54 antisense constructs. A fraction of the affected animals transmit the defective-muscle trait to subsequent generations. In these cases the transforming DNA is present at high copy number and cosegregates with the observed muscle defects. We have examined several of the unc-22 antisense plasmid transformed lines to determine the mechanistic basis for the observed phenotypes. The RNA product of the endogenous unc-22 locus is present at normal levels and this RNA is properly spliced in the region homologous to the antisense RNA. No evidence for modification of this RNA by deamination of adenosine to inosine was found. In affected animals the level of protein product from the endogenous unc-22 locus is greatly reduced. Antisense RNA produced from the transforming DNA was detected and was much more abundant than ‘sense’ RNA from the endogenous locus. These data suggest that the observed phenotypes result from interference with a late step in gene expression, such as transport into the cytoplasm or translation.
The ability to disrupt expression of specific genes can provide a valuable tool to relate cloned genes to their underlying functions. The most elegant methods for gene disruption utilize homologous recombination between injected DNA and the corresponding chromosomal locus; at present such methods are only available in a limited set of experimental systems. An alternative approach has been proposed whereby expression of specific genes could be effectively disrupted by the presence of excess quantities of negative-stranded nucleic acid (Izant and Weintraub, 1984). Antisense inhibition has been tried for several different systems and applications; varying degrees of success have been reported (for review see Takayama and Inouye, 1990). We have used a set of muscle expression vectors in Caenorhabditis elegans to produce antisense RNA directed to genes encoding abundant muscle filament proteins. We show that production of antisense RNA homologous to a specific gene leads to a decrease in expression of the endogenous chromosomal locus and to structural, functional and biochemical defects in muscle that parallel the effects of known mutations in the corresponding chromosomal locus.
Materials and methods
Selection for unc-22 disruption in transformed lines
Oocyte injections were performed as described (Fire, 1986). Following injection, animals were grown at 20°C or 25°C and progeny examined for any evident alterations from the wildtype. For detection of weak twitching phenotypes, we used the procedure of Moerman and Baillie (1979). The animals were washed into 1 % nicotine (in H2O) as late L4 larvae or young adults, and placed in an empty Petri plate. Wild-type animals hypercontract in this assay while the unc-22 hypomorphs (animals with reduced unc-22 activity) are evident by their inability to hypercontract and their characteristic twitching. Free nicotine is extremely toxic both as a vapor and if absorbed through skin, and should only be used in well-ventilated containment conditions.
The strong twitching phenotypes could be easily seen without the nicotine selection. We transferred the injected adults to fresh seeded Petri plates at 1 day intervals; this allows slower growing or odd progeny to be easily identified. Among the progeny are twitching animals, which are particularly evident after the animals have been adults for 1 –2 days.
Construction and growth of plasmid DNAs were as described (Fire et al. 1990). DNAs in TE [10 mM Tris –HCl pH 7.4, 1 HIM EDTA] were diluted directly into injection buffer (Fire, 1986) and used for injection. Linear DNAs to be injected were cleaved with the appropriate restriction enzyme and then checked by gel electrophoresis to confirm complete digestion. Sequences and restriction maps have been compiled using the program DNA Strider (Marck, 1988).
Mixed stage (non-starved) populations of animals were floated in [IM sucrose, 1 mM EDTA] (to remove bacteria and agarose) and washed repeatedly in [10mM NaEDTA pH7.5, 100 mM NaCl] before freezing at –80°C in aliquots containing 100 μl of packed worms. To extract an aliquot, 800 μl of 4 M guanidine isothiocyanate (Chirgwin et al. 1979) and 600 μl of acid washed sand were added and the samples shaken vigorously for 5 min; SDS was added to 17 mM and the samples were extracted three times with phenol –CHCl3 and once with CHCl3, added to 3.5 ml of [4 M NH4AC, 4MM EDTA], and precipitated by the addition of 10.5ml of ethanol. Samples were resuspended in 300 μl of [40 mM Tris pH7.5, 1 mM EDTA] and reprecipitated twice from 1M NH4AC by addition of 2.5 volumes of ethanol. This yields a preparation of total C. elegans nucleic acid, which is then treated with DNAase I (RNAase free; Boehringer-Mannheim) to give a ‘total RNA’ preparation that has not been preselected for single-stranded character, size, or poly(A) content.
Primed cDNA was prepared by mixing 10 ng of oligonucleotide primer with RNA from 2.5 μl of packed worms in 8.0 μl of TE8.4 [10mM Tris –HCl pH 8.4, 1 mM EDTA], heating to 72°C for 5 min, mixing with 2.0 μl of EB1 [0.25 M Tris – HCl pH 8.4, 0.25 M KC1], cooling slowly (approx. 1°C per min) to 48°C, and mixing with 15 μl of EB2 [50mM Tris –HCl pH8.4, 50mM KC1, 8 mM MgCl2, 5 mM DTT]. AMV reverse transcriptase was then added (20 units to start reactions +10 units added 30 min later; total incubation 70min, 48°C).
Injection of some unc-22 DNA segments causes a twitching phenotype
Initial suggestions that antisense inhibition might be possible in C. elegans grew out of attempts to develop a homologous recombination system for gene knockout. The unc-22 gene seemed particularly suitable for these experiments: mutants with reduced unc-22 activity are unable to sustain muscle contraction and therefore twitch (Brenner, 1974; Moerman and Baillie, 1979). The twitching phenotype is a relatively specific indicator for disruption of unc-22 function. Over 500 mutants exhibiting the distinctive ‘twitcher’ phenotype have been characterized; all except for three are unc-22 mutations (the three exceptions are rare non-null alleles of the gene lev-11I:Moerman and Baillie, 1979; Lewis et al. 1980; Moerman and Waterston, 1984; R. Waterston; D. Clark and D. Baillie, personal communications). When animals are examined in the presence of nicotine, the twitching phenotype becomes a very sensitive assay for partial reduction in unc-22 activity (Moerman and Baillie, 1979). Table 1 shows the relationship between dosage and phenotype for unc-22 (from Moerman, 1980). Animals with a 30 –50% decrease in unc-22 expression can reproducibly be identified in large populations (>106 animals) using the nicotine screen (Moerman and Waterston. 1984).
In an attempt to disrupt the endogenous unc-22 gene by homologous recombination, we injected a variety of unc-22 plasmid and lambda DNA clones into wild-type animals and selected in the next generation for the effective disruption of one of the copies of unc-22. Each of the unc-22 clones injected contained only part (3.6 –15 kb) of the unc-22 gene (Fig. 1). From the viewpoint of achieving homologous gene disruption, these experiments were unsuccessful: there were no cases of homologous recombination. There was, however, an unexpectedly high frequency of twitching animals derived from the unc-22 DNA injections. In most cases, the twitching trait could be transmitted to progeny, so that lines could be established. These lines fell into two classes. (1) Three bona fide unc-22 alleles which appear to be small deletions in the gene. The deleted segments did not correlate with the injected DNAs, and we assume that the relatively high frequency of ‘spontaneous’ unc-22 alleles that was observed in these experiments results from generally mutagenic effects of the injected DNA and/or the microinjection process. (2) A large number of lines (37 total) in which the endogenous unc-22 loci were apparently unaltered, but that contained several hundred copies of the injected DNA in a large tandem array (Fig. 2). The inheritance of the twitching phenotype in these lines indicates that the twitching is due to extrachromosomal elements. Different twitching lines have transmission frequencies of 5 % to 95 % per generation. No homozygous lines could be obtained; some progeny in each generation are completely wildtype (i.e. no extrachromosomal unc-22 DNA, no alterations in the unc-22 locus and no twitching). Animals were separated into twitching and nontwitching classes and stained with dye (DAPI or Hoechst 33258) to label the DNA. The twitching phenotype is correlated with the presence of extra-chromosomal ‘spots’ of staining material present in metaphase nuclei (data not shown). Similar spots are observed with other extrachromosomal arrays and in cases of genetically isolated free duplications (Herman et al. 1976; Stinchcomb et al. 1985). The twitching lines can apparently be maintained indefinitely (at least 20 generations) by continually choosing twitching animals.
Twitching lines can be derived from a variety of unc-22 clones with no common region of overlap (Fig. 1). Lines have been divided into categories based on the severity of the twitching phenotype. The ‘weakly’ twitching lines can only be scored in nicotine, while the ‘strongly’ twitching strains exhibit clear twitching behavior even in the absence of nicotine. Not every unc-22 clone has the ability to cause twitching when injected to generate the long tandem arrays. Indeed, the cloned DNAs that were used in these experiments apparently generate different spectra of twitching levels: some constructs give rise primarily to weakly twitching lines and some to strongly twitching lines. The ability to induce twitching behavior is specific to unc-22 DNA: strains with high copy numbers of recombinant DNAs with other C. elegans genes have not exhibited twitching behavior (Stinchcomb et al. 1985; Jefferson et al. 1987; Way and Chalfie, 1988).
We considered several possible ways in which these arrays could cause a twitching phenotype.
Aberrant protein products might be produced from array transcripts which interfere with muscle function to cause twitching.
Some level of indiscriminate transcription from the arrays might produce antisense RNA molecules that would specifically interfere with expression of the endogenous unc-22 gene.
Regulatory sites contained within the injected DNA might compete with the endogenous unc-22 gene for specific factors required for expression.
Chromosomal pairing between the endogenous loci and the large number of extrachromosomal copies of unc-22 might disrupt normal unc-22 gene expression.
The latter two possibilities seemed less likely, given the observed patterns of active and inactive constructs. Protein or antisense RNA effects would both require transcription of the injected DNA segments in muscle. Note that the 5 ’ end of unc-22 is not included in any of the injected segments. Any expression of the arrays must therefore be promoted from ectopic sites in vector DNA or from within the unc-22 gene. The pattern of active and inactive constructs could reflect differences in orientation and placement of the inserted unc-22 sequences relative to the corresponding bacterial vectors. Since we could not detect any abundantly produced protein product from the arrays (data not shown), we suspected that the above phenomena could be due to antisense interference.
Deliberate disruption of gene activity by an ‘antisense’ strategy
In order to test directly for the ability of antisense RNA to disrupt gene expression in C. elegans muscle, we designed a series of expression constructs predicted to produce RNA complementary to the unc-22 message. For these ‘antisense’ constructions, we used an expression vector (pPD 12.01: Fig. 3 and Fire and Harrison, unpublished) containing the promoter, enhancer and 3 ’ nontranslated elements from the gene encoding the major myosin heavy chain isoform expressed in body wall muscle (Epstein et al. 1974). Since unc-22 is also expressed in body wall muscle, this vector with an appropriate unc-22 insert should produce an unc-22 antisense transcript.
When the expression vector pPD 12.01 (without insert) was injected into animals, no unusual phenotypes were observed among the progeny (Table 2). Injection of pPD12.01 derivatives carrying either of two different unc-22 segments in antisense orientation leads to the production of F1 animals with strong twitching behavior (Table 2). Approximately half of these F! twitchers give some twitching F2 progeny, and in each of these cases, the twitching phenotype segregates in further generations as a dominant genetic element. The twitching phenotype varies somewhat between strains, but tends to be a much more severe phenotype than with the lines derived after injection of unc-22 DNA fragments cloned into vectors without muscle expression signals (e.g. those described in Fig. 1). In some of the lines made with the pPD 12.01 driven antisense constructs, very severely affected animals are observed. These have phenotypes similar to the null phenotype for the unc-22 locus (Moerman, 1980): highly disorganized muscle tissue with weak twitching and very little movement or egg laying in adults
For most of these lines, the dominant twitching behaviour segregates as an extrachromosomal genetic element similar to lines described in Fig. 1 (i.e. a dominant, heritable twitch inducing ‘locus’ with transmission between 5% and 95% per generation). In all lines examined (10), the injected DNA is present in a high copy number array (>100 copies per nucleus). In one case the dominant twitching phenotype is caused by a long array which has integrated into one of the chromosomes, generating a semi-dominant twitch inducing locus that can be maintained in a homozygous stock.
Several properties of the transformed lines are consistent with an antisense mechanism as the cause for the twitching phenotypes. One prediction would be that disrupting expression of a specific gene by antisense should result in a specific reduction of the corresponding protein product. To assay levels of unc-22 and unc-54 protein products we stained animals with antisera specific for either of the two proteins (Miller et al. 1983; Moerman et al. 1988), comparing twitching animals from the transgenic strains with their nontwitching siblings and wild-type animals. Body wall muscle in the control animals stains very strongly with both antisera. The strongly twitching animals show dramatically reduced unc-22 staining in body wall muscle, but show no such reduction in the unc-54 product (Fig. 4 and data not shown). The reduction in unc-22 staining represents at least a ten-fold difference. Less dramatic reductions in unc-22 staining were seen upon staining of the less severe twitching lines (data not shown). The reductions in unc-22 staining are not due simply to disorganization of the muscle cells: unc-54 null mutant animals with very disorganized muscle structure stain intensely with antibodies to unc-22 protein (Moerman et al. 1988; A.F. unpublished results).
Faint residual staining with the unc-22 antibody is seen in muscle from the strongly twitching transgenic lines. This staining could be due to a residual level of unc-22 product in these lines, although residual staining with this antiserum has also been observed upon staining of deletions of the endogenous unc-22 locus (Moerman et al. 1988).
The antibody used to localize unc-22 product also stains a pharyngeal protein, possible an unc-22 homologue encoded by a separate gene (Moerman et al. 1988). The pharyngeal staining is not affected in the transgenic twitching lines. This would be expected from the tissue specificity (non-pharyngeal muscle) of both unc-22 and the expression vector pPD 12.01.
Effects of injecting unc-54 antisense constructs
The phenotype caused by the expression driven antisense constructs is dependent on the source of the inverted segment. If a segment from the unc-54 gene is inserted in reverse orientation into pPD 12.01 and the resulting plasmid (pPD10.35) is injected into wild-type animals, uncoordinated animals are produced (Table 2). These animals do not twitch but have a lack of muscle tone, consistent with a decrease in unc-54 gene function. The most severely affected animals have a paralyzed phenotype virtually identical to the unc-54 null mutant phenotype (Brenner, 1974; Epstein et al. 1974). The majority of the uncoordinated animals derived from pPD 10.35 injection are less severely affected. Levels of gene expression in individual cells have been assayed by staining of whole animals with antisera (Miller et al. 1983; Moerman et al. 1988) to the unc-22, unc-54, and myo-3 protein products [myo-3 encodes the minor myosin heavy chain isoform (mhcA) present in body wall muscle; Miller et al. 1986]. The disorganized cells stain strongly with antibodies to myo-3 and unc-22 products but exhibit very little staining with antibodies to unc-54 product, while the well-organized cells yield essentially a wild-type pattern of staining with all three antibodies (data not shown).
In order to test the orientation dependence of the putative ‘antisense’ phenomenon, we have used the ‘sense’ equivalents of each of the plasmids as shown in Table 2. In the case of the unc-54 insertion, the uncoordinated phenotype is apparently dependent on the orientation of the inserted segment. This strongly argues against a direct titration of DNA binding factors as the cause for the uncoordinated phenotype, since the two plasmids pPD10.34 and pPD10.35 have the same sequences in different orientations. For the unc-22 constructs, no strong dependence on the orientation of the inserted segment was observed. Immunofluorescent staining (not shown; performed as described in Fig. 4) revealed dramatically reduced unc-22 protein levels in lines transformed with the two unc-22 ‘sense’ constructs. For this reason, as well as the results described in Fig. 1, we believe that the effects of unc-22 ‘sense’ constructs actually result from antisense RNA. Indiscriminate transcription from the extrachromosomal arrays apparently occurs (Jefferson et al. 1987; also see legend to Fig. 6C). In addition, the expression vector pPD 12.01 carries a muscle specific enhancer element that can act in either orientation and can enhance expression from both correct and ectopic initiation sites (Fire and Harrison, unpublished data). Thus it seems conceivable even with a ‘sense’ construct that sufficient antisense RNA could be made from the large arrays to interfere with expression of the endogenous unc-22 gene.
Genetic evidence for an antisense mechanism
To confirm that the disruption of expression requires homology between ‘sense’ RNAs from the endogenous gene and antisense RNA products of the arrays, we made use of an in-frame deletion mutation unc-22(st528), which lacks a 1964 bp region of the unc-22 gene but has unc-22 function close to normal (Kiff et al. 1988). A 1427 bp fragment contained entirely within this 1964 base deletion fragment was cloned in reverse orientation into the muscle expression vector pPD21.36 to make plasmid pPD34.147 (pPD21.36 is similar to pPD12.01, but devoid of unc-54 coding sequences; Fig. 5). When injected into N2, this plasmid yields many twitchers (41 twitching progeny from 16 injected adults), while no twitchers have been derived after many injections into st528 (zero from 19 injected adults). When arrays of pPD34.147 from two strongly twitching lines derived from N2 injections were moved genetically into an st528 background, the twitching phenotype was lost. These experiments confirm that homology between the inverted region and the wildtype chromosome is necessary for the observed disruption of gene function. This is consistent either with an antisense mechanism or with a requirement for pairing between the array and the endogenous gene at the DNA level. A simple pairing model is incompatible both with the observed pattern of active and inactive constructs (Fig. 1 and Table 2) and with the lack of a change in endogenous mRNA level (see below) in strains with disrupted gene function.
Analysis of sense and antisense RNA
To analyze RNA structures and levels in a uniform population of animals, we have taken advantage of a strain (PD68) in which a long tandem array containing the unc-22 antisense plasmid pPD 10.46 has integrated, leading to a stable, strong, twitcher phenotype. Two techniques for RNA analysis have been used: polymerase chain reaction (PCR) of reverse transcriptase products (Saiki et al. 1988; Frohman et al. 1988) and quantitative RNAase protection (Melton et al. 1984). In these experiments, care was taken to avoid any selection step that would distort the true ratios of different RNA species that might be present. Thus total RNA was prepared without selection for polyadenylation or single stranded character (see Materials and methods).
Results of the PCR confirm the presence of both sense and antisense RNAs (Fig. 6). The PCR products were cloned and sequenced to confirm their identity. Plasmid pPD 10.46 contains the first four introns of the unc-54 gene upstream of the unc-22 insertion point. The fourth intron lies within the region amplified by the primers AF55 and AF56. From gel analysis of PCR products and sequencing of clones, this intron appears to be quantitatively spliced out in the transcript derived from pPD 10.46. Likewise an intron from unc-22 is present in the segment of sense RNA amplified from the endogenous unc-22 locus. This intron is also precisely removed in PCR products from sense RNA in both N2 and PD68 animals. The spliced nature of the PCR products rules out the possibility that the PCR signals are due to contaminating DNA.
A dsRNA unwinding enzyme has been detected in Xenopus embryos (Bass and Weintraub, 1987; Rebigliati and Melton, 1987). This enzyme acts by catalyzing conversion of adenines to inosines in double stranded regions of RNA (Bass and Weintraub, 1988; Wagner and Nishikura, 1988). Similar activities have been found in a variety of mammalian cell types (Wagner et al. 1989) and in mixed stage embryos from C. elegans (M. Krause, personal communication). The adenine to inosine transitions in the RNA could be detected as complementary T to C transitions in cDNA clones (Kimelman and Kirschner, 1989). In order to test for such an activity in C. elegans muscle, we have sequenced segments of five different cDNA clones derived from the sense transcript of the endogenous unc-22 locus in the region homologous to the antisense RNA. Our PCR oligonucleotides were designed so that adenine to inosine transitions in the regions of homology would not affect the ability of the primers to bind RNA. This was done by choosing priming segments with few (0, 1 or 2) adenines in the coding strand, and making the primers 2 – 4-fold degenerate to account for possible mismatches at these positions caused by covalent modification of the RNA. Approximately 300 bases of each clone were examined, with no evident point changes, either by PCR errors or by covalent transitions in the original RNA sample.
We used cDNA clones derived from the PCR reactions to quantify sense and antisense RNA. Radioactive RNA probes were synthesized in vitro from these clones, hybridized to the total RNA population following denaturation and treated with RNAase A (Melton et al. 1984). The protected products were analyzed on both native (agarose) and denaturing (acrylamide) gels. The abundance of the sense RNA is similar in wild-type and PD68 animals (Fig. 6C). The antisense RNA is seen in PD68 but not in N2. The antisense RNA in PD68 is approximately 16-fold more abundant than the sense RNA. Two other bands appear in examining the hybridizations. A short protection product (316nt), seen with RNA probe 2 and either input RNA, is due to partial overlap of the endogenous unc-54 mRNA with this probe (Fig. 6A). A second band appears in PD68 hybridized with any of the probes used and has subsequently been shown to result from transcription of plasmid vector (pEMBL) sequences in PD68. The abundance of transcripts from plasmid vector sequences is not surprising in the light of suggestions that high copy number arrays may be subject to some degree of spurious transcription.
Stinchcomb et al. (1985) showed that certain extra-chromosomal tandem arrays could be lost mitotically, yielding mosaic animals. This phenomenon is evident in some but not all of the extrachromosomal transformed lines described in this work, and is manifest in two classes of animals. Some animals from transformed strains show phenotypic mosaicism in muscle function: parts of the animal move normally while other parts show an uncoordinated phenotype. Mosaicism at the single cell level is evident when the mitotically unstable lines are examined by immunofluorescence or polarized light microscopy (some cells are normal while others show disorganized filament structure; data not shown). Mitotic loss of the extrachromosomal arrays is also evidenced by a class of animals that exhibit twitching behavior but do not pass the trait on to any of their progeny (loss of the array in the germline). The degree of mosaicism seems to be a property of the particular transformed line: some extrachromosomal arrays exhibit no detectable mitotic loss (these also tend to be the more stable arrays meiotically) while other lines exhibit high frequency loss (with a majority of affected animals being visibly mosaic).
Mechanism of antisense disruption of gene function in C. elegans muscle
Several different mechanisms have been proposed or demonstrated for antisense inhibition of gene expression in vivo or in vitro (Fig. 7). From our experiments, several of these models do not appear to be relevant to C. elegans muscle. The presence of the sense RNA in the disrupted lines at normal levels suggests that double-strand-specific RNAase mediated degradation and inhibition of transcription do not play a role in the observed disruption of expression. The ability of the RNA to properly splice in the region of presumed heteroduplex formation suggests that inhibition of splicing is likewise not a major factor, and in addition suggests that heteroduplexes between antisense and sense RNA take some time to form following transcription. Finally, the lack of observed transitions characteristic of the covalent unwinding activity described for Xenopus oocytes (Bass and Weintraub, 1988; Wagner et al. 1989) strongly argues against a cognate enzyme being effective in unwinding duplexes that form in C. elegans muscle. The latter result was somewhat surprising given that dsRNA unwindase activity has been observed in extracts of C. elegans embryos (M. Krause, personal communication). Perhaps the dsRNA unwindase activity is lost (or sequestered to the nucleus; see Bass and Weintraub, 1988) in mature muscles. In any case, the observations leave us with the suggestion that antisense RNA disrupts expression in C. elegans muscle by hybridization to the sense transcript followed by steric hindrance, blocking either a late processing step, RNA transport, or translation.
A rough estimate of the efficiency of antisense RNA action in vivo can be obtained from the phenotype and relative RNA abundances in line PD68(cc68). When homozygous, the cc68 locus leads to strong twitching, while cc68/ + heterozygotes have a weak but still visible twitching defect. From comparisons similar to those in Table 1, this places the activity in the heterozygote at approximately 25 –33% of wild-type. In the cc68 homozygote the ratio of antisense to sense RNA is approximately 16, which suggests that this ratio in the heterozygote would be at least 8. These data suggest that efficient interference with gene function requires a high concentration of the antisense RNA as well as a molar excess of antisense over sense RNA.
Prospects for general use of antisense in C elegans
In certain cases one is presented with a cloned gene that has been obtained through non-genetic criteria: by homology to another gene of interest or by the biochemical properties of the protein or its expression pattern. In these cases it is usually of great interest to determine the phenotype of null mutations in the gene. Protocols for mutating endogenous genes using homologous recombination with incoming DNA are an obvious technique of choice, but this technology is not yet available for C. elegans. A second strategy involves precise localization of the gene on the chromosome, followed by extensive analysis of mutations induced in the region; this strategy relies on some ability to predict the phenotype of such a mutation (Waterston, 1989).
The antisense strategy could provide an alternative to these for determining null phenotypes (Izant and Weintraub, 1984). In order to do this, a set of segments internal to the gene would be reversed and inserted into either a genomic copy or into a universal expression vector active in all tissue types (such a universal vector has not yet been described for C. elegans). The resulting constructs could then be injected into wild-type animals (with or without coselection for a second plasmid) and the phenotypes of the progeny carefully examined. As a guide, the frequencies of transformation observed with the antisense constructs directed to unc-22 and unc-54 are sufficiently high that any visible defects should readily be found.
Any phenotypes observed should be interpreted carefully. If long tandem arrays are obtained (and this is likely unless selected against, given their prevalence among C. elegans transformed lines) the possibilities of titration of DNA binding factors and aberrant peptide production must be considered. The production of antisense RNA could also have effects on homologous genes, and dsRNA could itself have physiological effects on certain cells (Farrel et al. 1977).
A second set of problems comes with the question of whether a gene has been completely knocked out. Obviously, biochemical criteria such as immunofluorescent staining are useful in this regard, but it is never possible to confirm biochemically complete loss of function. If the gene under study was required for embryonic viability, then any initial transformants with complete knockout due to antisense RNA would die as embryos, and these dead embryos could easily be missed among inviable progeny that are victims of the injection process itself. We noted with the putative unc-22 and unc-54 disruption experiments that a variety of severities were observed among the different affected animals ranging from a slightly hypomorphic phenotype to phenotypes close to those of well characterized deletions of the corresponding chromosomal loci. In addition, the transformed animals tend to be mosaics. We should thus expect to find some abnormal progeny in attempting to knock out any essential gene. These would in general be animals that are either mosaic for the transforming DNA or that are the result of only partial decrease in gene function. In initial experiments using constructs that should produce antisense RNA to the essential gene myo-3 (Waterston, 1989) we generated a number of strains that segregate animals with varying degrees of muscle disorganization and some inviable animals. Encouragingly, the phenotypes of the most severely affected animals from these strains are close to the null phenotype for myo-3 (data not shown).
Our current view of the use of antisense to study cloned C. elegans genes is that the strategy can best be used to yield hypotheses about null and/or hypomorphic phenotypes for a gene of interest. These hypotheses should then be useful for designing genetic screens for mutations, allowing the gene’s normal role to be addressed independently.
We are grateful to M. Krause, R. Waterston, D. Baillie, R. Barstead, G. Benian, S. Brenner, D. Clark, D. Dixon, I. Greenwald, J. Hodgkin, S. L’Hcrnault, P. Masson, A.-M. Murphy, S. McKnight, I. Mori, J. Priess, A. Rose, M. Sepanski, M. Shen, A. Smith and J. Sulston for their help over the course of this work.
This work was supported by the Medical Research Council and the Carnegie Institution of Washington and by grants from the National Institutes of Health (R01GM37706 to A.F.), the Helen Hay Whitney Foundation (to A.F.). A.F. is a Rita Allen Foundation Scholar.