Production of a high-efficiency TILLING population through polyploidization

Abstract

Targeting Induced Local Lesions in Genomes (TILLING) provides a nontransgenic method for reverse genetics that is widely applicable, even in species where other functional resources are missing or expensive to build. The efficiency of TILLING, however, is greatly facilitated by high mutation density. Species vary in the number of mutations induced by comparable mutagenic treatments, suggesting that genetic background may affect the response. Allopolyploid species have often yielded higher mutation density than diploids. To examine the effect of ploidy, we autotetraploidized the Arabidopsis (Arabidopsis thaliana) ecotype Columbia, whose diploid has been used for TILLING extensively, and mutagenized it with 50 mm ethylmethane sulfonate. While the same treatment sterilized diploid Columbia, the tetraploid M1 plants produced good seed. To determine the mutation density, we searched 528 individuals for induced mutations in 15 genes for which few or no knockout alleles were previously available. We constructed tridimensional pools from the genomic DNA of M2 plants, amplified target DNA, and subjected them to Illumina sequencing. The results were analyzed with an improved version of the mutation detection software CAMBa that accepts any pooling scheme. This small population provided a rich resource with approximately 25 mutations per queried 1.5-kb fragment, including on average four severe missense and 1.3 truncation mutations. The overall mutation density of 19.4 mutations Mb(-1) is 4 times that achieved in the corresponding diploid accession, indicating that genomic redundancy engenders tolerance to high mutation density. Polyploidization of diploids will allow the production of small populations, such as less than 2,000, that provide allelic series from knockout to mild loss of function for virtually all genes.

Figure 1.

Effect of mutation rate and ploidy on functional discovery through TILLING. A, The probability of getting at least one severe missense mutation (assumed to be 15% of all mutations) and one knockout (KO; 5% of all mutations) is plotted versus the total mutation number identified in the coding region of a gene (redrawn from Henikoff et al. [2004]). B, The relationship between mutant yield and mutation rate is illustrated by the number of individuals in a population required to yield a given number of mutations (in A) in a 1-kb fragment. For example, considering mutations in a 1-kb coding region, the hatched blue stripe highlights how a population with a mutation rate of 2 mutations Mb^–1 requires screening more than 15,000 individuals for a 0.8 confidence of obtaining at least a single knockout. The same result can be obtained with 768 individuals of a population that has a mutation rate of 40 mutations Mb^–1, such as hexaploid wheat. The number of mutations expected for a given population and mutation density should be scaled according to the gene size. For example, for a 2-kb coding region and a mutation density of 2 mutations Mb^–1, 60 mutations are expected. A population of 15,000 individuals would yield an approximately 95% chance of at least one knockout. C, Published mutation rates in TILLING populations of different species organized according to ploidy. The vertical bar connects instances of the same species. Ai, Peanut (Knoll et al., 2011); AtC, Arabidopsis Col-0 (Greene et al., 2003); AtL, Arabidopsis Ler (Martín et al., 2009), Bn, canola (Harloff et al., 2012); Br, B. rapa (Stephenson et al., 2010); Gm, soybean (Cooper et al., 2008); Hv, barley (Hordeum vulgare; Talamè et al., 2008); Mt, Medicago truncatula (Le Signor et al., 2009); Os, rice (Oryza sativa; Till et al., 2007); Ps, pea (Pisum sativum; Dalmais et al., 2008); Ta, wheat (6x; Slade et al., 2005; Uauy et al., 2009); Td, durum wheat (Triticum durum [4x]; Slade et al., 2005; Uauy et al., 2009); Zm, maize (Till et al., 2004b). [See online article for color version of this figure.]

Figure 2.

Embryo lethality in mutagenized diploid and autotetraploid Arabidopsis. Two genetically identical strains of Col-0, but differing in ploidy, were treated as seed with the chemical mutagen EMS. The effect of the treatment was monitored by observing the development of M2 seed in the siliques of the M1 plants. Mutagenesis-induced changes affect one allele in any target gene: the corresponding genotype is Aa in the diploid and AAAa (simplex) in the tetraploid. M1 individuals are chimerae because the mature embryo shoot meristem is a multicellular structure and each cell is mutagenized independently. Genetic evidence indicates that the same cell lineage forms male and female gametes in a single flower. The occurrence of failed seed in a 1:3 (failed:live) ratio in the 30 mm EMS-treated diploid can be attributed to homozygous recessive mutants in genes required for embryo development and formed by selfed Aa lineages of the M1 plant. Diploid Arabidopsis was effectively sterilized by 50 mm EMS, which only marginally affected seed set in the autotetraploid. In tetraploids, the probability of a homozygous recessive embryo (aaaa) produced from an AAAa flower lineage is minuscule (1:575; Table III). Therefore, the dead seed progeny of the polyploid M2 is not readily explained by mutations in genes where the wild-type allele has a dominant action. More likely, the dead seed in polyploids are the result of mutations, or perhaps chromosomal aberrations, that have a lethal dosage effect. [See online article for color version of this figure.]

Figure 3.

Mutations detected in the Ala-tRNA ligase gene (At1G50200). The change frequencies for the 24 sequenced pools are plotted versus the base position on the queried DNA fragment. The triplet pattern formed by three outliers, corresponding to a unique mutant individual shared by three pools, is evident in the G→C and C→T frequency tracks. No triplets (i.e. candidate mutations) are visible in the A→C and A→G tracks. This is consistent with the observation that EMS in Arabidopsis is specific for GC→AT base pair changes (Greene et al., 2003) and with the notion that the triplet pattern is not generated by random sequencing errors. Frequency plots for the remaining base changes were comparable to the latter two in pattern and in range (data not shown). Background noise consists of high-quality sequence changes and is characteristic of each change type: for example, higher in A→G than in A→C. SNP, Single-nucleotide polymorphism. [See online article for color version of this figure.]

Figure 4.

Discovery of mutations through Illumina sequencing of overlapping pools. A, The probability score F(t) is the mean centered log of the posterior probability “t” of the null hypothesis. Higher F(t) scores are desirable, as they correspond to lower probability of false discovery. The distribution of 413 mutation candidates is shown. The mutations were detected by TILLING 15 amplicons corresponding to 15 Arabidopsis genes in a population of 528 autotetraploid individuals. Changes detected by aligning Illumina reads to the queried sequence reference were subjected to CAMBa2 analysis to discover mutation candidates. The vertical gray line indicates the threshold of F(t) = 2, corresponding to an empirically determined FDR of 0.02 (see C) and to a statistical FDR of 0.005. B, Probability is independent of sequencing coverage. Each point corresponds to a putative mutation and displays F(t) versus mean coverage. The lack of correlation between F(t) and coverage (r² = 0.02) indicates that once sequencing coverage is above a critical threshold, discovery is reliable. C, Setting the threshold probability for mutation calling. To shorten the time required by the computationally complex step, half the data set (264 individuals) used in A was subjected to mutation detection using CAMBa2 allowing all potential pool overlaps, both true and false. Mutations were then divided according to the overlap (i.e. the shared individuals from the population) type and are displayed both as frequency histograms and as dithered dots. Those detected in true overlaps (had common individuals) were defined as plausible (i.e. true). Those detected in false overlaps (did not share individuals) were defined as implausible (i.e. false). “Double mutations” are consistent with two identical changes occurring in independent individuals and thus found in two sets of three correctly overlapping pools. The threshold at F(t) = 2 was arbitrarily chosen to provide a predicted FDR of 2% and corresponds to the F(t) value above most implausible mutations and below most plausible ones. Using that threshold, 384 mutations are called in the whole data set. [See online article for color version of this figure.]

Figure 5.

Diploidization strategies for mutations discovered in autotetraploids. Two alternatives are illustrated for a simplex autotetraploid mutant with genotype CCCc. A, Diploidization via triploid bridge (Henry et al., 2009). All progeny of cross 1 are expected to be triploid. Triploids transmit either one or two copies of each chromosome type (Table III). The progeny of cross 2 will fit in three classes: diploids (approximately 30%; Henry et al., 2009), aneuploids, and some triploids. B, Diploidization via genomic elimination achieved through the use of a haploid inducer variety, such as strain “Col-0 cenh3-1, GFP-tailswap” (Ravi and Chan, 2010). Cross 3 will produce tetraploids, diploids, and aneuploids. The direction of the cross influences the percentage of diploids produced. Furthermore, some plant varieties, such as Col-0, carry male-specific interploidy lethality factors (Dilkes et al., 2008). Reciprocal crosses should be tested for optimal results. c, Mutant allele; C, wild-type allele; 2X, diploid; 4X, autotetraploid. Brackets indicate gametic genotypes; thick-lettered genotypes in haploid inducers refer to alleles on genomes targeted for elimination. [See online article for color version of this figure.]