To determine whether TILLING represents an effective reverse genetic strategy for C. elegans we generated and arrayed an EMS-mutagenised population of approximately 1500 individual animals (see below) and screened for mutations in 10 genes varying in size from 788 base pairs (bp) to 9112 bp. A region of approximately 1500 bp from each gene was examined and a total of 71 novel mutations were identified by TILLING, thus providing multiple mutant alleles for each gene (Table 1). Some of the mutations we identified are predicted to be silent, either because they are in non-coding DNA or because they affect the third bp of a codon which does not change the amino acid encoded by that codon. However, 59% of the mutations we identified are missense alleles resulting in a change in one of the amino acids in the protein product of the gene, and 3% are nonsense alleles resulting from the insertion of a premature stop codon into the coding region of the gene, or in the elimination of a conserved splice junction site. These data demonstrate the efficacy of TILLING in C. elegans.

A C. elegans population was constructed for this TILLING project using the mutagen EMS (Figure 2). EMS was chosen because it has been shown to be an effective mutagen in this species and because it is known to generate primarily single bp mutations which can be identified using TILLING 233031323334.

A survey of Wormbase 7 was performed in order to examine the type of molecular lesion induced by EMS in C. elegans to ensure that the majority of these mutations are, indeed, small intergenic lesions of the type that can be identified by TILLING (see Additional File 1 for a list of alleles). Two hundred and thirteen alleles from 51 different genes whose molecular sequence was known were selected randomly from the database. All of the mutations were reported to be identified in screens of EMS-treated animals. Ninety three percent of the 213 alleles examined from Wormbase were found to be single bp mutations. Eighty seven percent were G/C-to-A/T transitions, six percent were other single bp mutations, and seven percent of the mutations reported were deletions that ranged in size from 88 bp to 2.3 kb.

In our TILLING experiment, 68 of the 71 independent mutations identified (96%) were G/C-to-A/T transitions and the remaining three mutations were A/T-to-T/A or G/C-to-T/A transversions (Table 2). This percentage of G/C-to-A/T transitions is significantly higher than that found in forward genetic screens and probably reflects the fact that many EMS-induced lesions are not identified using classical genetic screens because they do not have an effect on phenotype. The frequency of transversions seen with TILLING in our screen is similar to the frequency found in forward genetic screens using EMS in C. elegans (6%), but significantly lower than the number of transversions mutations reported in a small Drosophila TILLING project (16%) 29, and significantly higher than the number found in Arabidopsis where greater than 99% of mutations sequenced were G/C-to-A/T transitions 23.

Three of the mutant strains we generated in this study, CN579, CN1162 and CN1574 carried two point mutations within the target gene. In each case, one of the second-site mutations was in non-coding DNA and so presumably would not affect the phenotype of the animals carrying the linked mutation. Second-site mutations have been reported in forward genetic screens of EMS-treated worms as well. A case in point is the molecular analysis of unc-52 mutations and intragenic suppressor alleles 35. Two of the 19 mutations sequenced in the unc-52 study carried a second site mutation less than 400 bp upstream of the primary mutation. One of these second-site mutations was a single bp transition, and the other was a 311 bp deletion. In both cases, the second-site mutations were not found to affect the phenotype of the animals carrying them. Second-site mutations have also been found in EMS screens of yeast 33 and Drosophila 34, and so presumably reflect some common DNA repair mechanism that is induced upon EMS damage.

Deletions are another type of mutation that has been reported in forward genetic screens using EMS in several different species 303234. In C. elegans approximately 7% of all mutations from forward genetic screens of EMS-treated animals are deletions (Additional File 1). We did not identify any deletions in this TILLING project but we are confident (based on previous studies 2336) that if EMS does generate this type of mutation in C. elegans, TILLING will be able to detect these events as well as the single base pair changes that are more common. The reason that deletions have been identified more frequently in forward genetic screens than in our reverse genetic screen is probably because these mutations are much more likely to produce a phenotype than the single bp mutations more commonly produced by EMS.

The dose of EMS used for our TILLING project (0.025M) was lower than that used in many forward genetic screens because studies have shown that this lower dose simplifies the identification of mutant phenotypes caused by the gene of interest while limiting confounding background phenotypes or lethality 37. In two strains, however (CN843 and CN1643), we identified mutations in two different genes in the same strain. While this might seem to indicate a very high overall mutation frequency, we do not believe that this is the case since the mutagenised animals seem healthy and fertile, and since the overall mutation rate was calculated to be one mutation every 293 kb (71 mutations in 14225 bp of DNA from 1464 animals). This is not significantly higher than the rate of one mutation in 300 kb seen for TILLING in Arabidopsis 23, and is lower than the rate of one mutation in 156 kb reported for Drosophila 29. Hence, the 0.025M dose of EMS appears to be an adequate dose for TILLING based on comparison with these other systems in which this technique is currently being used.

The spectrum of mutations identified through forward and reverse screens using EMS, although similar at the level of DNA sequence is much different when the effects of the mutations are compared. Of the 194 single bp mutations we analysed from Wormbase (Additional File 1) 50% resulted in missense alleles and the remaining 50% in nonsense or splice junction mutations. In our TILLING screen, because the selection of mutants was not based on phenotype, 38% of the mutations we identified are predicted to be silent. These include mutations in introns and intergenic regions and mutations that alter the third bp of a codon such that it still encodes the wildtype amino acid. The majority of the mutations we identified (59%) were missense alleles that alter the amino acid sequence of the protein encoded by the target gene (Table 2). Of the 42 missense mutations identified in our screen, 17 of these may not have a significant effect on phenotype since the amino acid mutated was replaced by an amino acid of similar charge and polarity, but the 25 remaining missense mutations are predicted to significantly affect the structure of the protein product of the target gene by changing the charge or hydrophobicity of this region of the protein.

Two of the mutations identified in our TILLING screen, mel-32 C05D11.11(vc68) and cki-2 T05A6.2(vc39), are predicted to result in a complete loss-of-function, or null, phenotype because they truncate the protein product of the gene. One of these introduces a premature stop codon into the third exon of gene mel-32 C05D11.11, and the other is a splice junction mutation that eliminates the splice donor site of first intron of the gene cki-2 T05A6.2 (Figure 3). The proportion of putative null mutations identified in our screen was 3% which is not significantly different than the frequency of 2% seen in the Drosophila TILLING study published 29 or than the 5% reported from the much larger Arabidopsis TILLNG project 23. This frequency is higher than would be expected using other chemical mutagens in C. elegans such as ENU 38 which cause a different spectrum of mutagenic events that are more likely to result in missense than nonsense mutations.

A frozen library of approximately 1500 individual EMS-treated lines of C. elegans was constructed for this study (Figure 2), and DNA was isolated, purified, and arrayed in pools of eight as has been shown to work for TILLING in other diploid species such as Arabidopsis thaliana 22, Lotus japonicas 24, Zea maize 27 and Brassica oleracea [Gilchrist and Haughn, unpublished]. Purification of the DNA was found to be necessary both because the TILLING reactions did not work on unpurified DNA samples and because accurate quantitation of the samples is essential before pooling so that DNA from mutant animals is sufficiently represented in the 8-fold pools 27. Both 4-fold and 12-fold pools were tested in C. elegans in order to confirm that 8-fold pooling would be efficient (see Materials and Methods for details). With 4-fold pooling all of the mutant bands were detected, as they were with the 8-fold pools, whereas with 12-fold pooling only a subset of the mutant bands were seen on the TILLING gel. Although 10-fold pools might be possible in C. elegans given that the genome size of this organism is slightly smaller than Arabidopsis, the construction of libraries for screening in 96 or 384-well plates dictates that pooling is only efficient in multiples of eight or 12, and since mutations are missed with 12-fold pooling in C. elegans, libraries constructed for this TILLING study were pooled in 8-fold aliquots.

Our first library consisted of 696 rather than the planned 768 (8 × 96) mutagenised worms because the DNA quality in 72 of the 768 lines established was too poor to be used for TILLING. Only 8-fold column pools were constructed for this library and when a mutation was detected in a column pool well, each of the eight individuals that made up that pool was examined by TILLING in order to determine which strain carried the mutation. For the second library of 768 animals, however, both 8-fold row pools and 8-fold column pools were constructed and screened. The DNA from individual worms was arrayed in 12, eight-by-eight grids so as to simplify pooling in either direction. Then the DNA was pooled by combining samples from one row or one column into a single pool, resulting in a total of 96 pools. The identity of the strain carrying a mutation detected on the TILLING gels was computed automatically by cross-referencing the data from the row and column pools.

In theory, the method used for screening library #1 only required an average of 120 reactions: one 8-fold column pool, plus eight individuals for each of three mutations detected (24 additional reactions). However, the 24 reactions done to determine which individual from a pool carried the mutation had to be set up manually and often needed to be repeated in order to ensure that all of the reactions worked. Thus, this pooling strategy was more time-consuming and often required almost as many TILLING reactions as the screening of both row and column pools simultaneously. In addition, the row and column pool strategy allowed false positive bands to be excluded from our study since a mutation was never followed unless it appeared in at least one channel in both row and column pool gels.

Mutations identified through TILLING are randomly distributed in the genome, thus making it possible to target genes of any size and at any location 23. The average size of the region that is tested in one TILLING screen is usually about 1500 bp and this is the size that was used for most genes in this study (Table 1). For the gene C05C10.5 whose genomic sequence is only 788 bp we designed primers that amplified a region of approximately 1200 bp to avoid screening excess intergenic DNA upstream or downstream of the locus where mutations would have a higher probability of being silent. In addition, for gene mus-81 C43E11.2, the primer sets that we designed to amplify a product of 1500 bp gave multiple amplification bands when used with our standard PCR conditions. Thus we were forced to use a primer set that amplified a smaller region in order to obtain a single, clean PCR product from this locus.

For genes that are much larger than 1500 bp, two approaches have been used for TILLING in other systems. The web-based programme, CODDLE 39 was originally designed for use in the Arabidopsis thaliana TILLING project and assists researchers in designing primers to select regions of a gene of interest that are most likely to provide loss-of-function or deleterious alleles. Researchers also have the option of requesting that CODDLE design primers that will amplify a fragment within a specific region of the gene in which they are most interested, for example, a specific domain that is known to interact with another gene or protein. A different approach was used for a recent Drosophila melanogaster TILLING project 29. In this study, multiple primers were designed to amplify overlapping fragments of a gene so that the entire gene could be screened by TILLING. For our TILLING project in C. elegans, as mentioned previously, the average amplicon size was 1500 bp and primers were designed to amplify the region of the gene predicted by CODDLE to be most susceptible to EMS-induced mutations. Primer sequences used in this study are listed in Table 3. For C. elegans, where the average gene size is 3000 bp and introns are generally small 40, TILLING should prove to be even more efficient than in other species where larger genes and intervening sequences are the norm. Indeed, 62% of mutations recovered in this C. elegans screen are predicted to have an effect on the protein product (Table 2), although long-term studies are required to determine exactly how many of these mutations will result in a mutant phenotype.

Each F₁ mutagenised line was frozen individually in order to ensure that identified mutations could be recovered even if a sample was frozen and thawed multiple times (see Figure 2). Except in the case of deleterious mutations that affect viability or reproduction, each mutant allele that is heterozygous in the F₁ parent is expected to be present in 3/4 of the progeny of this individual. If the mutant allele is recessive lethal, then the frequency of progeny carrying this allele should be 2/3. In this study we were able to recover individual descendants of F₂ worms carrying the mutant alleles in 19 of the 20 strains thawed for testing. The remaining strain proved to be a false positive isolated from our screening of our first library. The probability of this type of error occurring with our current methodology of screening both row and column pools is very low.

There are several methods that can be used to follow the segregation of point mutations in F₂ and subsequent populations. We used three different strategies for comparison in this study: TILLING, cleaved amplified polymorphic sequences (CAPS) 41, and direct sequencing. TILLING was the most expensive and labour-intensive method because of the requirement to purify the extracted DNA for PCR and because detection of homozygous individuals necessitated duplexing the DNA from the putative mutant with wildtype DNA (since fragments are only cleaved with CJE if there is a mismatch in the DNA). In addition, multiple rounds of TILLING were sometimes necessary if one of the two reactions (duplexed and non-duplexed DNA) failed, because in such a case the zygosity of the animal in question could not be determined. CAPS was tried if the PARSESNP programme (42 indicated one or more restriction enzyme polymorphisms between the mutant and wildtype sequences (Figure 4). Sixty of the 71 mutations we identified (85%) in this screen are of this type. For the samples where there were no restriction enzyme polymorphisms either TILLING or direct sequencing was used to detect mutant individuals because of time constraints, although studies have shown that the derived cleaved amplified polymorphic sequences (dCAPS) technique works well and would be a less expensive method for following mutations in the long-term 43.

Results from direct sequencing of DNA from F₂ descendants of mutant animals showed that the mutation of interest was present in an average of two out of three of the thawed progeny (43 out of 64 samples), although this varied from a low of one out of eight with F25H2.13(vc9), and a high of 13 out of 16 with htp-3 F57C9.5(vc2). Both of these alleles with non-typical segregation patterns are missense mutations and, although the vc2 allele does not have an obvious phenotype in homozygous mutants, we speculate that it may be incompatible with the wildtype allele of this locus since heterozygous individuals carrying both mutant and wildtype alleles together were never recovered. The high frequency of homozygous vc2 mutants compared to wildtype individuals may just be a statistical anomaly or may indicate that the mutant protein confers some type of fitness advantage upon animals carrying it.

Two other mutations, F25H2.13(vc10) and mel-32 C05D11.11(vc11), were apparently homozygous in the parent F₁ strains in which these alleles were detected because all of the thawed progeny from the F₂ plates were found to be homozygous for these mutations. All other mutations were heterozygous in the F₁ generation, as expected, because EMS-induced damage in C. elegans usually occurs in either the P₀ egg or the P₀ sperm before fertilization. Neither of the homozygous mutations was present as a background mutation in the wildtype strain used for mutagenesis, or in any of the other mutant strains whose DNA was sequenced. The F₁ homozygosity of vc10 and vc11 may be the result of an early mitotic crossover or other heritable event that occurred in the developing zygote rather than in the maternal germ cells, during or after EMS mutagenesis. The fact that these mutations were detectable at all using TILLING is curious since a DNA mismatch is needed for cleavage with CJE. In the pooled samples, adequate wildtype DNA from other samples would have been amplified and paired with the mutant DNA for cleavage to occur, but when testing the DNA from the individual F₁ animals no wildtype DNA should have been present since we did not detect any wildtype progeny from these animals. It is possible that only the germ cells of the F₁ animal carried the mutations identified in our screen and that wildtype DNA from the somatic cells of this animal was sufficient to allow for cleavage with CJE when amplified using PCR. It is also possible that some wildtype DNA contamination was present in these reactions and was amplified along with the homozygous DNA from the mutant. The fact that the cleavage bands were very faint on the TILLING gels is consistent with either of these ideas.

An additional polymorphism in gene F25H2.13 was identified when sequencing other alleles of this gene and found to be homozygous in all TILLING strains and in the N2 P₀ strain utilised for mutagenesis in this study, making it clear that this strain is different from the N2 strain used in the C. elegans genome sequencing project. Although this polymorphism induces an amino acid change in the protein sequence, it has no obvious effect on gene function since animals carrying the polymorphism are seemingly wildtype.

Some of the loci chosen as candidates for this pilot project were genes that are thought to play roles in chromosome segregation, recombination or genome maintenance. In some cases, RNAi constructs of these genes had been shown to induce varying phenotypes, and we reasoned that null and missense alleles of these targets would allow us to better identify the true function of these genes. In other cases, the genes we targeted had no known RNAi phenotype, but had been implicated in meiotic functions through two-hybrid or bioinformatics studies.

Examination of the sequence of the 71 alterations shown in Table 2 revealed that 27 of the changes resulted in no amino acid change (silent mutations) and these strains are unlikely to have visible phenotypes. Of the remaining 44 mutations resulting in amino acid changes, 24 affected charged or conserved residues. These are the alleles that are most likely to affect the functioning of the gene product and thus most likely to have phenotypic changes. We have done preliminary phenotypic analysis on 25 of the TILLING alleles we identified and observed phenotypes for 16 of these alleles. Although further studies are needed to confirm these data, the TILLING alleles reported here clearly allow characterisation of these genes in a manner that was not previously possible.

N2 wild type hermaphrodites were exposed to 0.025 M EMS for 4 h (as in Brenner, 1974 51, but at a lower EMS concentration). After exposure to mutagen, the worms were washed and allowed to recover for 1–2 h before young gravid hermaphrodites were picked to fresh plates (10 per plate). After 24 h, the mutagenised P₀ animals were transferred to fresh plates for a second 24 h brood. F₁ progeny were picked 1 per plate and allowed to self-fertilize. The F₁ progeny of mutagenised worms were set up individually rather than in pooled populations so as to ultimately simplify the isolation of F₃ animals carrying any mutations. This was considerably more work than would have been required with pooled F₁ samples but, because worm libraries were frozen, the extra work was a one time occurrence and greatly simplified identification of thawed mutants. When the mutagenised worm populations had exhausted the bacteria on the plate, worms were washed off each plate with 2 ml of M9 buffer. One third of the population from each F₁ line was used for DNA and the other two thirds was frozen for future use as follows: from each plate, 0.5 ml of worms were mixed with 0.5 ml of 2X freeze solution (100 mM NaCl, 50 mM KPO₄ pH 6.0, 0.3 mM MgSO_4, 30% glycerol) and frozen at – 80°C to generate a frozen library stock.

The remaining 1 ml of worms in M9 buffer for each line was centrifuged at 14,000 g to pellet worms. Excess buffer was aspirated leaving approximately 50 μl of buffer and worms in each tube. 50 μl of worm lysis buffer (50 mM KCl, 10 mM Tris (pH 8.3), 2.5 mM MgCl2, 0.45% Nonidet P-40, 0.45% Tween 20, 0.01% (w/v) gelatin, and 10 mg/ml proteinase K) was added to each tube which was then frozen at -80°C for at least 1 h. DNA was extracted by proteinase K lysis at 57°C for 4 h with occasional vortexing, and then 100 μl of phenol:chloroform:isoamyl alcohol (24:24:1) solution was added to the crude DNA extracts and tubes were vortexed for 5 minutes. Phases were separated by centrifugation at 14,000 g for 5 minutes and then aqueous layer was removed to a fresh tube containing 100 μl of chloroform, vortexed for 5 minutes and the phases separated by centrifugation at 14,000 g. The aqueous layer was again removed to a fresh tube containing 400 μl of isopropanol, and then the DNA was precipitated by centrifugation at 14,000 g for 10 minutes. DNA pellets were washed once with 70 % ethanol, resuspended in 50 to 100 μl ddH₂O and quantified using a ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). Samples were diluted to 1 ng/ul in 10 mM Tris, 1 mM EDTA pH 7.4, and 1 ml aliquots were distributed in plates of 64 samples (8 rows by 8 columns) before pooling. The DNA samples were then pooled 8-fold, in both column and row directions, and then distributed into 96-well plates of 8-fold column pools and 96-well plates of 8-fold row pools for each library of 768 individuals.

Primers were designed, using the web-based programme CODDLE 39 and selecting "EMS (not TILLING)" as the Mutation Method since a C. elegans option was not available and we did not know how similar the spectrum of mutations caused by EMS was in C. elegans compared to other organisms. Primers for C05C10.5 were designed to amplify a fragment of approximately 1200 bp since the genomic size of this gene is only 788 bp. The other primer sets were designed to amplify fragments as close to 1500 bp as possible, given the structure of the DNA in the region. Primers were purchased from MWG Biotech, Inc. (High Point, NC, USA), suspended to a concentration of 100 uM in 10 mM Tris, 1 mM EDTA pH 7.4 and used at a final concentration of 0.2 mM in a mixture of 3:2 (labeled:unlabeled) for the forward (IRD700-labeled) primers and 4:1 (labeled:unlabeled) for the reverse (IRD800-labeled) primers as per Colbert et al., 2001 52. PCR was also performed according to Colbert et al.,2001 52: 10 ul PCR reactions with 2.5 ng – 5 ng of genomic DNA were used for amplification in 96-well or 384-well PCR plates using ExTaq polymerase (Takara Bio Inc, Japan), but with 0.6 times the recommended concentration of ExTaq buffer and 2 mM MgCl₂. PCR cycles were as follows: 95°C for 2 min; eight cycles of [94°C for 20 sec, 73°C for 30 sec (decrementing 1°C per cycle), 72°C for 1 min]; 45 cycles of: [94°C for 20 sec, 65°C for 30 sec, and 72°C for 1 min]; 72°C for 5 min; 99°C for 10 min (denaturation and inactivation of taq enzyme); and 70 cycles of 20 sec at 70°C (decrementing 0.3°C per cycle for random reannealing to allow hybridisation of mutant and wildtype molecules), hold at 4°C.

Crude celery juice extract (CJE) was prepared as described by Till et al. 53. Briefly, 0.5 kg of celery was processed in a kitchen-quality juicer until liquefied. Tris HCl (pH 7.7) was added to 0.1 M along with Phenylmethylsulphonylfluoride (PMSF) to 100 mM. The solution was spun at 2600 G for 20 minutes and the supernatant removed, brought to 25% saturation in (NH₄)₂SO₄, mixed for 30 minutes at 4°C, and spun at 15,000 G for 40 minutes at 4°C. The supernatant was removed again and adjusted to 80% saturation in (NH₄)₂SO₄, mixed for 30 minutes at 4°C, and spun at 15,000 G for 1.5 hours at 4°C. The pellet from this cut was resuspended in 1/10 the starting volume of 0.1 M Tris HCl (pH 7.7), 100 mM PMSF. The suspension was dialysed against 8 L of the same buffer, four times, for one hour each time at 4°C using Spectrapore dialysis tubing (10,000 MW cut-off). Aliquots were stored at -70°C and were spun at approximately 2000 G for one minute before use to remove any tissue debris.

PCR products were digested with CJE by adding 20 μl of extract and buffer mix (100 mM MgSO4, 100 mM HEPES, 300 mM KCl, 0.02% Triton X-100, 0.002 mg/ml BSA, and 0.2% to 0.3% crude CJE) directly to the PCR reactions and incubating at 45°C for 15 minutes. Reactions were stopped by adding 2.5 μl of 0.5 M EDTA. The DNA was purified by passage through G50 Sephadex in 96-well Millipore Multiscreen^® filtration plates (Millipore Corporation, Billerica, MA) and concentrated for 30 minutes at 90°C before running on a 25 cm long LI-COR acrylamide gel with a 0.4 mm wide, 96-well sharkstooth comb. Analysis of the gel images was done using GelBuddy 54 to define lanes and estimate sizes of cleavage products. The correlation of row and pool columns indicated which individual F₁ worm carried the mutation. Most mutations were sequenced in both directions using either the same forward or reverse primers as for PCR or an internal primer designed for sequencing. Sequence analysis was performed using Sequencher 4.2 (Gene Codes Corporation, Ann Arbor, MI, USA) and the potential effect of the mutations was predicted using PARSESNP 42.

DNA from 4 strains previously shown to carry a single mutation each was screened to test the efficiency of different pooling depths. Four bands were expected in each of the two channels of the Li-cor sequencing image when amplified DNA from these individuals was run on a gel. With 4-fold pooling all eight mutant bands were detected (four in the 700 channel image and four in the 800 channel image), as they were with the 8-fold pools, whereas with 12-fold pooling, six of the eight mutant bands were seen on one test gel (three in the 700 channel image and three in the 800 channel image), and only three bands on the second gel (two in the 700 channel image and one in the 800 channel image). Based on these data and data from other studies, we conclude that 8-fold pooling was the best option for C. elegans.

Frozen worms were thawed and 20 individuals were plated 1 per 60 mm NGM plate and allowed to self-fertilize. Two approaches were taken to identify lines containing the target mutation. In the first approach, the 20 individuals from the strain carrying the mutation of interest were plated individually, allowed to grow until plates were starved, and DNA was prepared as described for the DNA library construction. In the second approach, individual animals from the strain carrying the mutation of interest were allowed to lay eggs for 2–3 days, and then the parent was picked into 5 μl of lysis buffer and lysed at 57 °C for 1 h followed by 95 °C for 15 m to inactivate the proteinase K. In both approaches the extracted DNA was PCR amplified and the mutation detected either by TILLING, or by direct sequencing of the amplified DNA, or by digestion of the PCR product with restriction enzymes resulted in a banding pattern different from wild type. Restriction enzyme site changes were found by PARSESNP to occur in 60 of 71 mutations.

If homozygous lines were not identified, 20 more individuals were set up from one line that had been shown to be heterozygous and the progeny from each of these lines was screened for evidence of deleterious mutations that might result in inviability. If sterile adults, dead embryos or larval lethals were observed, DNA was prepared from these and tested for the presence of the targeted mutation. Mutations that segregated with inviable phenotypes were balanced with genetic balancers to prevent the loss of the mutation.

Frequencies of EMS-induced mutations identified during different TILLING experiments have been based on different samples sizes in different species. When comparing our data with others we used an on-line Proportions Test 55 to compare the frequencies we observed with those reported in other species and determine whether or not observed differences were significant (at the 90% confidence level) or were simply likely to be the result of sample size differences.