• Open Access

TILLING in extremis


  • Trevor L. Wang,

    Corresponding author
    1. John Innes Centre, Norwich Research Park, Norwich, UK
    2. RevGenUK, John Innes Centre, Norwich Research Park, Norwich, UK
      (Tel +44 1603 450283; fax +44 1603 450014; email trevor.wang@jic.ac.uk)
    Search for more papers by this author
  • Cristobal Uauy,

    1. John Innes Centre, Norwich Research Park, Norwich, UK
    2. National Institute of Agricultural Botany, Cambridge, UK
    Search for more papers by this author
  • Fran Robson,

    1. John Innes Centre, Norwich Research Park, Norwich, UK
    2. RevGenUK, John Innes Centre, Norwich Research Park, Norwich, UK
    Search for more papers by this author
  • Brad Till

    1. Plant Breeding and Genetics Laboratory, Joint FAO/IAEA Division of Nuclear Techniques in Food and Agriculture, International Atomic Energy Agency, Vienna International Centre, Vienna, Austria
    Search for more papers by this author

(Tel +44 1603 450283; fax +44 1603 450014; email trevor.wang@jic.ac.uk)


Targeting induced local lesions in genomes (TILLING), initially a functional genomics tool in model plants, has been extended to many plant species and become of paramount importance to reverse genetics in crops species. Because it is readily applicable to most plants, it remains a dominant non-transgenic method for obtaining mutations in known genes. The process has seen many technological changes over the last 10 years; a major recent change has been the application of next-generation sequencing (NGS) to the process, which permits multiplexing of gene targets and genomes. NGS will ultimately lead to TILLING becoming an in silico procedure. We review here the history and technology in brief, but focus more importantly on recent developments in polyploids, vegetatively propagated crops and the future of TILLING for plant breeding.

Introduction: a brief history of TILLING

The advent of targeting induced local lesions in genomes (TILLING) a little over 10 years ago led to a sea change in our ability to obtain specific mutants in plants and especially crops; no longer were plant biologists restricted to transformable species or often fickle forward genetic screens. This together with the almost universal applicability of the process to both model plant and crop species meant that TILLING was adopted by many researchers worldwide. In many instances, their resources were opened to the community either as a service or research collaboration. Although TILLING has found its forte in plants, it is one of those rare examples of a development in plants that has also been adopted in animals, such as the models zebrafish (Moens et al., 2008) and Drosphila (Cooper et al., 2008a,b). TILLING has been reviewed previously in general terms (Gilchrist and Haughn, 2010; Henikoff and Comai, 2003; Henikoff et al., 2004; Kurowska et al., 2011; Rashid et al., 2011), as a method (Comai and Henikoff, 2006; Till et al., 2003a) or by plant families (e.g. cereals, Weil, 2009; legumes, Tadege et al., 2009). Here we outline the development and application of the technology and focus on the less well-studied areas of TILLING in polyploids and in vegetatively propagated crops. We also indulge ourselves in elaborating some pros, cons and approaches to the process, and we explore the future of TILLING.

Arguably, the best technological innovations in science are driven by their biology and TILLING has been no exception to this. It is a beautiful example of a well-established, almost traditional, mutation breeding method meeting modern technology. Steve Henikoff’s group at the Fred Hutchinson Cancer Research Center in Seattle were driven to the method by their need to identify mutants for a DNA methyltransferase (chromotransferase) from Arabidopsis thaliana that they were studying following the failure of other reverse genetics methods (McCallum et al., 2000). In this initial work, they used a chromatographic method to separate heteroduplexes between a wild-type and mutant DNA strand, but recognized swiftly that this could never be used in a high-throughput manner. The lack of throughput led them to the use of an alternative method based on the labelling of individual PCR products in a manner suitable for the recognition by high-throughput instruments, namely DNA sequencers. LI-COR® DNA analyzers (Colbert et al., 2001) were chosen because they were gel-based and this instrument remains the most widely used technology, although several alternatives exist (e.g. Le Signor et al., 2009). Many of these, however, do not improve on the method finally decided on by Henikoff and co-workers (Till et al., 2006c) and are more the product of convenience, that is, the instrument was available locally. The group’s work led to the first high-throughput TILLING platform and community service (Arabidopsis TILLING project, ATP; Till et al., 2003b). Many of the bioinformatics tools required for the process were also produced by the group and remain in use today (e.g. CODDLE, http://www.proweb.org/coddle/). The group then set up numerous collaborations with other groups to apply the methods to additional species by running pilot experiments. Following this, they changed their name to the Seattle TILLING Project (STP), which continued to operate until quite recently (http://tilling.fhcrc.org/). It is perhaps a tribute to the original team that their openness, sharing of informatics tools and willingness to train others clearly contributed to the widespread adoption of the technique despite several attempts to protect the technology by others. There are now numerous publications on TILLING in a wide range of plants (see Table S1 for a list of those from the last 5 years, and Kurowska et al., 2011).

TechTILLING: the TILLING process

The TILLING technique has been described on numerous occasions since the seminal publications on the topic (e.g. Till et al., 2003a,b, 2006c) and hence will only be covered briefly here. It is readily combined with forward genetic screens and this is represented by the cartoon in Figure 1, which was used on the first platform developed outside Seattle, for the model legume, Lotus japonicus (Perry et al., 2003, 2009). This particular platform subsequently expanded to cover the other model legume, Medicago truncatula and a model for brassica crops, Brassica rapa. It operates now as RevGenUK at the John Innes Centre, Norwich (http://revgenuk.jic.ac.uk).

Figure 1.

 A summary of the TILLING process embracing forward genetic screening. The pure seed line (a) is treated with EMS at an appropriate concentration (determined by testing small batches of seed with a range of concentrations), then (b) seeds are sown, grown (M1 plant generation) and harvested as individuals for M2 seed. This seed is then re-sown as families (12 seed usually) and the plants screened for segregating phenotypes. A single fertile plant from each family (c) is used to collect leaves for DNA extraction and (d) the seed harvested and stored appropriately. Siblings can be screened for a variety of phenotypes by eye or by high-throughput (bio)chemical screens (e.g. Takos et al., 2010; Vriet et al., 2010). Seeds from sibling plants bearing useful phenotypes are also stored. Individual DNAs are pooled (e) in microtitre plates and pools used for PCR amplification using a mixture of labelled and unlabelled primers. Products are heated, cooled and (f) the annealed products cut (g) with CEL1 enzyme. The resultant products are purified and run on gels (h) or capillaries to detect the mismatched products. Individuals are then selected from the pools (i) by re-sequencing and their seed sown to identify mutant plants. Reproduced with the permission of Jillian Perry.

As mentioned earlier, TILLING can be carried out using different technologies, but the central principle is the same; the detection of rare genetic mutations in pooled DNA samples from large mutant populations in a first round of screening, followed by de-convolution of the pool to identify the individual carrying the mutation (Figure 1). The majority of current TILLING protocols utilize two different and sequential techniques—one to screen the pools and another to identify the individual. In their seminal work, McCallum et al. (2000) used denaturing (D) HPLC (the Transgenomic® WAVE System; Jones et al., 1999) from pooled DNA samples to detect mutations in small PCR fragments that had been denatured by heating and subsequently cooled to promote the formation of heteroduplexes. PCR and sequencing of these fragments from individual DNA samples then identified both the individual mutant and the exact position and nature of the mutation. While DHPLC has largely been superseded by higher throughput and mainly gel-based techniques using fluorescent PCR primers for visualization, conventional sequencing remains the most common method to confirm the putative mutations detected in the first round.

TILLING progressed rapidly with the advent of mutation detection by single-strand-specific endonucleases that cleave DNA at the site of mismatched nucleotides in otherwise double-stranded DNA (Colbert et al., 2001). The mismatches form a tiny single-stranded ‘bulge’ that is recognized and nicked by these enzymes, and the resulting cleaved PCR fragments are generally separated by denaturing PAGE. The ATP published a protocol that has been widely adopted as the most efficient and cost-effective method (Till et al., 2003a). They used a nuclease extracted from celery, CEL1, to cleave at the site of induced mutations in heteroduplexed PCR fragments amplified with fluorescently labelled primers, and the LI-COR® DNA analyzer denaturing polyacrylamide gel electrophoresis system to separate the resulting fragments. That even crude celery juice extracts (CJE) can substitute for the more expensive purified CEL1 enzyme has made this protocol accessible to many laboratories. Other enzymes have also been successfully used (e.g. ENDO1, Triques et al., 2008).

The Medicago Grain Legumes Integrated Project (http://www.pcgin.org/GLIP/pubrep.pdf) used the published LI-COR® method to identify mutations in some of their target genes, but also developed a capillary sequencing method for the remaining gene targets (Le Signor et al., 2009). In this method, fluorescently labelled and heteroduplexed PCR products from pooled DNA samples are cleaved using CJE and separated on an Applied Biosystems™ 3730xl DNA sequencer (Life Technologies, Applied Biosystems, Paisley, Scotland). Potential mutations are identified as anomalous ‘empty’ peaks on overlaid chromatograms using their GeneMapper® software. The plant carrying the mutation is identified after conventional PCR and sequencing of each individual DNA sample from the relevant pool using the software MutationSurveyor® (SoftGenetics LLC, State College, PA), which can analyse multiple mutations simultaneously. This technique has been adopted by the RevGenUK TILLING platform for all its current populations (see Table 1 and references therein).

Table 1.   Distribution of mutation types identified from TILLING assays in Lotus japonicus, Medicago truncatula and Brassica rapa mutant populations*
 Number of fragments TILLedMutation type
Mis-senseTruncationSilent/non-codingTotal mutations
  1. *Data are from populations used for reverse screens only; data from forward screened populations are excluded as they will be biased towards deleterious mutations.

  2. Premature stop or splice junction.

  3. 78 fragments TILLed by RevGenUK, 84 fragments TILLed by the Lotus TILLING project (GENPOP, Perry et al., 2009).

  4. §Percentage of each mutation type compared to total mutations by species.

  5. Twelve fragments TILLed by RevGenUK, 66 fragments TILLed by the European Grain Legumes Integrated (GLIP) Project (Le Signor et al., 2009).

  6. **Twenty-nine fragments TILLed by RevGenUK, eight fragments TILLed by Stephenson et al. (2010).

  7. ††This figure represents an underestimate of the mutation load of B. rapa; for practical reasons, only a subset of the potential mutations was confirmed.

  8. ‡‡Average percentage across the three populations.

L. japonicus 162834 (51.6%)§72 (4.5%)711 (43.9%)1617
M. truncatula 78469 (65.1%)37 (5.1%)214 (29.7%)720
B. rapa 37**578 (56.9%)44 (4.3%)394 (38.8%)1016††
Totals2771881 (57.9%)‡‡153 (4.6%)1319 (37.5%)3353

Techniques that do not rely on mismatch cleavage by endonucleases have also been developed, such as high-resolution melting analysis (Gady et al., 2009; Parry et al., 2009; Ririe et al., 1997; Zhou et al., 2004). This technique exploits the denaturing dynamics of double-stranded DNA after PCR amplification and heteroduplex formation. Single-base mismatches in the mutant pools are detected by a slight shift in their melting temperature, with respect to wild-type homoduplexes, and can be assayed using a variety of fluorescent dyes and instruments. This method is especially suited for target genes with small exons separated by large introns as the detection sensitivity is limited to amplicons of <450 bp.

Next-generation sequencing (NGS), which is based on completely different (non-Sanger) technologies, has helped to bypass some of the challenges presented in more conventional TILLING approaches such as mutation detection in pools deeper than eight individuals. TILLING by Sequencing (TbyS; Tsai et al., 2011) has been demonstrated for rice and wheat using Illumina® (Illumina Inc., San Diego, CA) sequencing of target genes amplified from multidimensionally pooled DNA templates. Each pool of PCR products is barcoded by the addition of a unique DNA adapter allowing stacking and processing of multiple genes from different species. After sequencing, samples are deconvoluted using a bioinformatics pipeline, which can classify reads based on the adapter sequence (Missirian et al., 2011). Mutations are identified by comparison with a wild-type reference sequence, thus circumventing the requirement for detection by enzymatic cleavage or PCR-based melting techniques. The 3D pooling strategy enables the detection of an individual mutant plant and the molecular identity of the mutation without the need for extra deconvolution of pools and more sequencing steps. This technique still relies on PCR amplification of individual genes of interest, however, and as such remains a targeted approach. Should sequencing costs continue to decrease at their current rate, untargeted methods can be adopted. This is discussed in the final section of this review.

MutTILLING: mutagenesis-based reverse genetics

The original concept of TILLING described the use of mutagens to randomly induce genomic changes followed by high-throughput recovery of lesions for reverse genetics applications (McCallum et al., 2000). As such, a variety of mutagens can be considered for TILLING. Different mutagens can produce different spectrums of mutations, allowing a more tailored approach to achieve desired alleles. For example, while MNU treatment of rice and soybean resulted in ca. 90% GC-AT transition changes, a combination of sodium azide and MNU resulted in ca. 20% AT-GC changes (Cooper et al., 2008c; Suzuki et al., 2008; Till et al., 2007). Physical mutagens can also be considered. Gamma irradiation, long a mainstay for mutation breeding, produced ca. 30% small indels of only a few base pairs, suggesting that this mutagen may be efficient for generating knockouts of single genes (Sato et al., 2006). Fast neutrons have also been used for reverse genetics (Li et al., 2001). This mutagen produces primarily deletions on the hundreds to thousands of base pairs scale, making it markedly different from the point mutations or small indels so far recovered from chemical or gamma mutagenesis. This process, therefore, has been coined deletion-based (de-) TILLING (Rogers et al., 2009). The method was devised for detecting specifically sized (small) deletions in large amplicons from genes to obtain knockout mutations for the model legume, M. truncatula. Fast neutrons produce a range of different-sized deletions and so the trick here was to find the right sized deletion (in their case around 2.5 kb) to knockout a single gene. Theoretically, tandem duplications can also be detected, constituting a major advantage of the technique over traditional TILLING or insertional mutagenesis. Mutation discovery is quite different to that used for SNPs and small indels. Large populations of plants at high pooling depths are screened using two rounds of nested PCRs in which a third ‘poison’ primer is included. This primer is within the deleted region and amplifies a ‘suppressor’ product. The suppressor, being smaller, is amplified from the WT more efficiently, but the amplicon cannot act as a template in the second round, which allows the mutant product to be amplified very competitively. Although a relatively cheap identification method, it is not easily adapted to be a high-throughput process. Rogers et al. (2009) used it to identify mutations in several genes involved in the root nodule symbiosis.

EcoTILLING: assessing diversity

TILLING is a very quick way to examine the allelic diversity within a species for a particular gene. This process was introduced by Comai et al. (2004) using A. thaliana and given the name ecoTILLING. It basically requires DNA from a species diversity set rather than from mutagenized plants and these DNAs are mixed with a reference DNA to identify polymorphisms using the TILLING process. The authors determined also that ecoTILLING at that time had a distinct advantage over re-sequencing if there were many fewer haplotypes than individuals in the set. EcoTILLING was used successfully to examine virus resistance in melon and Capsicum (Ibiza et al., 2010; Nieto et al., 2007), diversity in poplar (Gilchrist et al., 2006), polyphenol oxidase in tea (Jin et al., 2010) as well as mining different species of brassica (Wang et al., 2010a). It is an especially useful method of screening for genetic variation in species where classical genetic analysis is difficult if not impossible (Gilchrist et al., 2006) and has even been extended to humans (Till et al., 2006b).

ProTILLING: pros and cons of TILLING

Many plant researchers consider insertion mutants as their first port of call and this is for a number of reasons: they are openly available in those species with genomic resources (both dicotyledonous and monocotyledonous) including several crops; they are often relatively cheap to obtain; the whole process is often an in silico exercise; they produce non-functional alleles in most instances. In contrast, only ca. 5% of mutant alleles recovered in TILLING screens of ethyl methane sulphonate (EMS) mutagenized populations are truncations (Greene et al., 2003; Perry et al., 2009), and the mutation discovery process itself is laborious and relatively expensive ($1500–$5000 USD, including materials, labour and overheads). On the other hand, insertional mutagenesis requires an efficient transformation system and cannot therefore be readily carried out in most plant species. Moreover, it is often not saturable and therefore requires large population sizes to have a high probability of identifying an insertion in the target gene. This feature makes it best suited for species with small genomes because the development of the populations is expensive and requires complex identification strategies. Although recent rapid sequencing technologies have largely overcome the latter, insertional mutagenesis is still not a valid alternative for many crop species.

To TILL or not to TILL

The mutagens used for traditional plant breeding work are readily applicable to most plants, so the application of TILLING has been wide ranging and especially useful where limited genomic resources are available. Populations from chemical mutagenesis can also be used for forward screens. In the current research climate, especially in Europe, TILLING has the distinct advantage of being a non-genetic manipulation (GM) technology and hence most appropriate for crop species. It is also an advantageous approach in developing countries that as yet lack the necessary regulatory infrastructure for work on genetically modified organism (GMOs). In many cases, desired mutant alleles can be easily introgressed into different genetic backgrounds or the mutagenesis itself can be developed in advanced genetic material. This argues for exploiting economies of scale and developing centralized facilities that serve entire research communities. In reality, however, such facilities require a significant investment in resources (human and financial) and it is often extremely difficult to make a single crop TILLING service economically attractive to researchers and be self-sustaining. Few laboratories, therefore, are able to undertake the process. An important aspect to this is that most funding sources tend to be national or regional and many plant research communities are international.

Ten years after its inception, the Seattle TILLING Project has now come to an end and much of the work has been transferred to the group of Dr. Luca Comai in UC Davis (http://tilling.ucdavis.edu/index.php/Main_Page) where TbyS predominates. For those contemplating setting up their own resource by conventional TILLING, given the constraints mentioned earlier, a guide was produced recently as a by-product of a workshop in Beijing to help researchers decide whether TILLING was appropriate for their purposes (Wang et al., 2010b). The guide gives some ‘do’s and don’ts’ for budding TILLers and also helps with deciding the approach to use. Part of that guide, the ‘decision tree’ is show in Figure 2.

Figure 2.

 A flow chart for decision-making in the TILLING process. Reproduced with permission from Wang et al. (2010b).

Allelic variation

The high mutation frequency from chemical mutagens such as EMS used in TILLING often worries researchers who want to assign phenotypes quickly. Misassignment is not really a problem where few genes are involved in the phenotype, but this is frequently not the case in plants. A simple solution to this comes from the ability of the technique to reveal multiple deleterious mutations as both homozygotes and heterozygotes. One can either look for segregation of phenotypes following selfing of a heterozygous plant or one can carry out a single cross between homozygous alleles, where maintenance of the phenotype in the F1 indicates you have the correct phenotype (Henikoff et al., 2004). Alternatively, multiple independent alleles for a given gene can also provide support to assign phenotype to a specific gene.

For structure–function studies, TILLING is invaluable because it generates allelic series often including null alleles, albeit at low frequency. This is especially important if a total loss of function is lethal. In this instance, knockout resources will not recover mutants, whereas TILLING will do so through the recovery of heterozygotes and weaker alleles. In L. japonicus where reverse genetic resources have been limited, TILLING has been used extensively to investigate the root nodule symbiosis especially signalling pathways. It has been used to determine residues affecting, for example, changes in receptor activity (Perry et al., 2009). Another recent example of such studies was the work of the Stitt group in Golm (Hädrich et al., 2011) where they showed the effects of subtle changes in ADPglucose pyrophosphorylase activity on the rate of starch synthesis. In platforms where there have been sufficient mutations discovered, it is normal to obtain them in ratios of two heterozygous to one homozygous mutant (Greene et al., 2003), although L. japonicus may be an exception to this with many more heterozygous lines being isolated (Perry et al., 2009). Of the mutations obtained from screening a typical EMS mutagenized population, ca. 50% are silent (either in introns or causing no change in amino acid) and ca. 5% lead to a non-sense transcript. Perry et al. (2009) also observed in symbiosis mutants of L. japonicus that there was a bias in functionally defective alleles towards glycine substitutions. Cumulative data from nearly 300 TILLed targets covering the populations used by RevGenUK produce a value of 4.6% across coding and non-coding regions, which is close to the theoretical maximum value (5.2%) from nucleotide sequence for G/C to A/T transitions (Table 1). Given the relatively small data sets that are available, however, and the current bias towards screening coding sequences, many deviations from the expected random distribution of mutations may be difficult to interpret. The advent of whole genome sequencing for TILLING should mitigate this issue and allow investigation into local biases in mutation accumulation that could reveal, for example, selective pressures on protein folding and gene regulation.

From model to crop

The initial TILLING resources were developed in model plants, first A. thaliana and then L. japonicus (McCallum et al., 2000; Perry et al., 2003; Till et al., 2006a). There is no doubt, however, that the major advantage of TILLING is that it utilizes established mutagens and non-GM methods. This makes it especially attractive to plant breeders given that the often laborious, relatively low-throughput transgenic approaches require extensive regulatory oversight. Following the development of the technology in model species, therefore, TILLING has spread rapidly to many agricultural and horticultural crops both major and minor (see Table S1). There are now several indications that it will emerge as a major plant breeding tool with some good examples of direct applications to breeding—virus resistance in tomato (Piron et al., 2010), virus resistance and ripening in melon (Dahmani-Mardas et al., 2010; Nieto et al., 2007), starch in potato (Muth et al., 2008) and wheat (Sestili et al., 2010; Uauy et al., 2009), natural products in sorghum (Blomstedt et al., 2012) and reduced nornicotine content in tobacco (Julio et al., 2008). As with conventional breeding, the germplasm that is produced using mutagenesis still requires a long lead time to a final variety. In addition, the mutagenesis process behind TILLING creates mostly loss of function alleles, thereby limiting the range of phenotypes produced and potentially the number of genes that can be targeted for crop improvement. However, the number of crop species where TILLING has been developed has increased rapidly over the last 5 years, indicating its attraction as a tool for breeding.

PolyTILLING: TILLING in polyploids

Many plants of agronomic importance correspond to polyploid species, including oilseed rape (B. napus), cotton (Gossypium hirsutum) and both bread and durum wheat (Triticum aestivum and T. turgidum). In these species, most genes are represented by multiple homoeologous copies, which share a high sequence identity. This redundancy limits the use of forward genetic screens based on phenotype as the effects of single gene knockouts are frequently masked by functional complementation with homoeologous genes present in the other genomes (Lawrence and Pikaard, 2003).

Despite this apparent drawback, the multiple copies of each gene make polyploid species very well suited for TILLING as they can tolerate very high mutation densities. This is well exemplified by the multiple TILLING populations, which have been published in recent years. The average mutation densities from diploid species are one mutation per 380 kb, whereas this drops dramatically to one mutation per 49 kb in tetraploids (durum wheat and tobacco) and one mutation per 32 kb in hexaploids (bread wheat and oat) (Figure 3, Table S1). There are two exceptions in diploid TILLING populations, which have unusually high mutation frequencies; B. rapa (one mutation per 60 kb) and A. thaliana cv Landsberg erecta (one mutation per 89 kb) which are highlighted in Figure 3 (yellow circles). Although B. rapa is diploid, it went through an ancient triplication event, which results in many genes being represented by three paralogous copies. Therefore, some level of functional redundancy still exists (Stephenson et al., 2010). In the Landsberg erecta population (diploid Arabidopsis), the mutation frequency was maximized by selecting M1 plants with very low seed fertility (Martin et al., 2009). This differs from tetraploid and hexaploid species where populations with higher mutation frequencies rarely exhibit sterility or apparent mutant phenotypes (Uauy et al., 2009).

Figure 3.

 The mutation rate in published TILLING populations based on the species ploidy level. Yellow circles represent the two diploid populations with highest mutation frequency that are discussed in the text. A detailed list of the populations included in the figure is presented in Table S2.

This high mutation frequency observed in polyploid TILLING populations facilitates the identification of large allelic series in target genes screening a relatively small number of individuals. For example, by screening 2000 individuals in a population with one mutation per 32 kb, we would expect to recover ca. 81 mutant alleles, whereas this number drops considerably to only seven mutant alleles in a typical diploid population (one mutation per 380 kb). To recover 65 mutant alleles in the same diploid population, one would need to screen 19 000 individuals. Ultimately, these factors directly determine the probability of identifying truncation (premature stop codon or splice junction mutant, Figure 4a) or deleterious mis-sense mutations (Figure 4b) within the population. Following the earlier example, the probability of identifying at least one truncation by screening 2000 individuals should be 98% in the hexaploid and 91% in the tetraploid population, whereas it is only 27% in the diploid.

Figure 4.

 Probability of identifying non-functional alleles by TILLING in different population sizes by ploidy. (a) the probability of finding at least one truncation or (b) one deleterious mis-sense mutation in TILLING populations of different ploidy. The average mutation rates are based on the published populations from Figure 3. Graphs are based on the expectation that 4.5% of mutations will lead to a truncation (either stop codon or splice junction mutation; see Table 1) and that 20% of all mutations will be deleterious mis-sense mutations (Henikoff et al., 2004). From screening small genic fragments of 1–1.5 kb, it was observed that mutations are distributed randomly across euchromatic regions and that this is used to generate the mutation density estimates based on total base pairs screened and mutations recovered (Greene et al., 2003).

Increasing the probability of identifying at least one truncation mutation per gene target is very important in polyploid species because phenotypic effects in a single mutant may be masked by wild-type homoeologues present in another genome. This implies that in most cases, it is necessary to combine mutations by developing double or triple mutants through genetic crosses. If truncation mutants are identified for each homoeologue, this is a straight-forward endeavour. On the other hand, if multiple mis-sense mutants need to be combined across genomes, the crossing scheme becomes lengthy and/or risky. For example, if three mis-sense mutations are chosen for each of two homoeologues, then nine double-mutants need to be developed. If only one mis-sense is chosen from each genome, then the double-mutant may not yield a phenotype. This could be due either to one of the mis-sense mutations not being deleterious or to the fact that the candidate gene does not affect the expected phenotype. This exemplifies the value of maximizing the probability of identifying truncation mutations in polyploid species, at least for an initial evaluation of gene function prior to more nuanced studies with mis-sense changes.

An additional consequence of the high mutation frequency in polyploids is that each individual carries an extremely high mutation load. In wheat, it is predicted that any given individual will carry between 260 000 (tetraploid) and 415 000 (hexaploid) mutations (Parry et al., 2009; Uauy et al., 2009). This suggests that each individual is expected to have thousands of mis-sense and hundreds of truncations mutations when gene space, GC content and mutation spectrum are taken into account. If the mutant is to be used for breeding, these background mutations can be effectively removed through backcrossing and selecting the mutation at each generation using SNP-based markers such as KASPar (Allen et al., 2011). If the objective is to assign function to a specific gene, then the use of homozygous sibling lines segregating for wild-type and mutant alleles provides a very powerful means to reduce interference from background mutations (Perry et al., 2005).

The TILLING protocol in polyploid species is essentially the same as for any diploid, requiring a unique PCR product for CEL1/CJE heteroduplex digestion. The presence of multiple copies of each gene implies that generating this homoeologue-specific PCR product is more cumbersome than in diploids. For each target gene, the sequences of all homoeologues have to be obtained and annotated to allow an efficient primer design. This task can be time-consuming and is further aggravated by the fact that in most cases, the publicly available genomic sequence of polyploids is limited or incomplete. Several strategies can be used to obtain homoeologue-specific sequence, such as sequencing diploid progenitors, BAC clones, etc. With these sequences, genome-specific primers can be designed by exploiting unique indels or inter-homoeologue SNPs (reviewed in Uauy et al., 2009) and several rounds of PCR optimization are usually required before the TILLING screen can begin.

The majority of the traditional screening approaches require the use of fluorescently labelled primers, which can significantly reduce the genome specificity of primers. This is especially relevant in polyploid TILLING where the prolonged primer design and optimization process can cause further delay. As a consequence of this, many alternative screening methods have emerged for polyTILLING as a way to circumvent the use of labelled primers. Modifications include the use of ethidium bromide to detect CEL1/CJE cleavage products resolved by PAGE (Uauy et al., 2009) or agarose gels (Dong et al., 2009a) and high-resolution melt analysis of amplicons (Dong et al., 2009b; Zhou et al., 2004). The complex primer design process has meant that many polyploid species do not feature in centralized TILLING services because the majority of these use laser detection of fluorescently labelled PCR products. However, this situation is rapidly changing with the advent of NGS technologies which we discuss further later. The ultra low-cost strategies such as the agarose or PAGE-based platforms will still remain, however, to ensure projects can continue in the absence of sophisticated technology.

VeggieTILLING: TILLING in vegetatively propagated plants

Many potential bottlenecks in a successful TILLING project have been addressed through methodological developments. A range of mutation discovery methods have now been developed that balance equipment cost, reliability and accuracy, enabling TILLING by both large facilities and small groups. A suite of protocols have been developed for inducing mutations in plant seeds (Table S1; Kurowska et al., 2011). While some seed-propagated crops are recalcitrant to the accumulation of mutations, suitable mutation densities for many species have been achieved using alternative chemical mutagens, or altering mutagenesis conditions (e.g. Cooper et al., 2008c; Till et al., 2007). Perhaps one of the last and potentially largest bottlenecks left in the application of TILLING across the plant kingdom can be found in genetic constraints inherent to specific species or cultivars. Here, we describe approaches to overcome constraints in vegetatively propagated species.

Vegetative propagation of plant materials occurs via successive rounds of mitotic cellular division. Examples of vegetatively propagated crops include apples, bananas, citrus, cassava, grapevine, hops and potatoes, and thus represent crops important for global economies and also for food security in developing nations. The switch from sexual to vegetative propagation can occur naturally by providing competitive advantage, or can be driven by the actions of man. For example, it is hypothesized that a reduction in the number of flowers can result in increased yields in tuber crops like potatoes (Jansky and Thompson, 1990). Vegetatively propagated species can be further categorized as either maintaining the ability for sexual reproduction (facultative) or having lost that ability (obligate). Natural vegetative propagation occurs through a variety of mechanisms including the production of rhizomes and stolons. These have been exploited by farmers for crop production. In vitro techniques such as shoot and root culture, callus culture, cell suspensions and somatic embryogenesis have been developed for many plant species that allow the rapid clonal amplification of plants for research or industrial purposes (Jones et al., 1988). While mutagenesis of tissue culture material has been described, adoption of TILLING in vegetatively propagated crops promises to provide new inroads for functional genomics and breeding (Mba et al., 2009).

As in sexually propagated species, the first stage in developing a TILLING platform for vegetatively propagated plants is the generation of a population with a suitable density of induced mutations and devoid of chimeric sectors. Chimeric sectors arise because the majority of in vitro culture techniques rely on propagation of multicellular tissues. The process of mutagenesis is random, causing different mutations to accumulate in different cells. Mutagenesis itself is rather straightforward: isolated tissues are soaked in chemical mutagen for a prescribed amount of time, or exposed to a pre-determined dosage of ionizing radiation (e.g. Mba et al., 2010). The process of dissolving chimeras typically involves successive rounds of tissue culture whereby the number of unique genotypes in meristematic cells is gradually reduced (van Harten, 1998). Using this approach, the Plant Breeding and Genetics Laboratory of the FAO/IAEA Joint Programme is currently developing TILLING platforms for vegetatively propagated banana.

Edible bananas are triploid, parthenocarpic and represent an example of obligate vegetatively propagated crops. EcoTILLING studies with banana and accessions of the genus Musa showed efficient recovery of natural nucleotide polymorphisms in polyploids using a crude CJE and fluorescence detection on a LI-COR® DNA analyzer (Till et al., 2010). For banana TILLING, shoot apical meristems were isolated and treated with EMS (Figure 5). After recovery from mutagenesis, plantlets are allowed to grow for ca. 1 month. Meristems are then isolated and divided, effectively reducing the number of meristematic cells producing the next generation and thus reducing genotypic heterogeneity. Phenotypic evaluations suggest that chimeras can be greatly reduced or stabilized by the third vegetative cycle (Jain et al., 2011). The use of mutation discovery and TILLING allows careful evaluation of this process at the genotypic level. Studies of the density and spectrum of induced mutations in triploid banana support a model of mutations accumulating as expected when the data were compared to TILLING experiments in diploid and polyploid seed-propagated species (J. Jankowicz-Cieslak and B. Till, unpublished). This suggests that mitotically stable mutations can be induced and maintained through successive rounds of tissue culture. Investigations are underway to evaluate the rate of dissolution of chimeric sectors. It is expected that the act of mutagenesis itself can have deleterious effects on the fitness of cells to divide, allowing a competition between cells in the meristem and possibly providing a selective pressure for the rapid dissolution of chimeric sectors in totipotent cells (Klekowski and Kazarinovafukshansky, 1984; Figure 5).

Figure 5.

 A strategy for banana mutation induction and the dissolution of chimeric sectors via dissection of apical meristems. (a) A large population of identical plantlets are prepared through clonal micropropagation. (b) Apical meristems are isolated and soaked in chemical mutagen or treated with ionizing radiation. After mutagenesis, the resulting plantlet, including the shoot apical meristem, is chimeric with different cells comprising different genotypes (indicated as cells marked with different shading and patterning). (c) Meristem isolation and longitudinal cutting result in the production of sibling plants with reduced genotypic complexity in meristematic tissue. (d) successive rounds of isolation and cutting generate tissues devoid of chimeric sectors. (e) Tissue for the preparation of a TILLING DNA library is ideally collected in the first generation where plantlets are non-chimeric. This avoids the inefficiencies of sampling the same allele more than once in TILLING screens. (f) The act of mutagenesis may induce cell death, cell cycle arrest and other phenomenon that allow a single cell to outcompete other cells and eventually produce meristematic tissue of a single genotype without extensive tissue culture interventions.

While vegetatively propagated triploid bananas serve as an excellent tool to study the genetics of mutation induction in vegetatively propagated species, the lack of meiosis remains a major bottleneck. Induced mutations are heterozygous, meaning only dominant or hemizygous alleles will likely yield phenotypes or traits of interest. Because of the lack of independent assortment, it is expected that deleterious alleles will naturally accumulate over time, resulting in fewer copies of genes that maintain wild-type functionality (McKey et al., 2010). This would increase the likelihood that deleterious heterozygous alleles would have phenotypic consequences, but the extent of this effect is currently unknown. In addition, treatment with X-rays can induce mitotic recombination, or gene conversion, providing a means for generating homozygosity (Haendle, 1979). The efficacy of this approach is unknown and may be most suitable for organisms with low frequencies of natural heterozygous SNPs. A seemingly inescapable caveat for TILLING in such crops is that there are no methods to remove background mutations. Hence, point mutagens, such as EMS, that induce a high density of mutations may be inappropriate for functional genomics and cultivar development in triploid banana. A mutagen that induces a wider spectrum of alleles at a lower density, such as gamma irradiation, may prove most effective. An advantage of obligate vegetative propagation is that any induced trait is instantaneously fixed and can be immediately evaluated. Indeed, some success has been recorded for improving banana using gamma irradiation and traditional mutation breeding techniques (Jain et al., 2011). The mutated genes causative for the observed traits in these released varieties have yet to be isolated. As NGS becomes more cost-effective, it will be interesting to investigate the density and spectrum of induced alleles in these cultivars to learn more how mutagenesis can be optimized for obligate vegetatively propagated plants.

Tissue culture mutagenesis can also be considered in facultative vegetatively propagated species where seed propagation is possible. It can be an especially efficient approach where factors limit the efficiency of seed mutagenesis such as constraints on field space, limited growing seasons, slow production of seed or low fecundity. For example, a strategy can be employed where non-chimeric plants are rapidly produced and evaluated for induced heterozygous mutations in target genes of interest. Only plants harbouring wanted alleles need to be retained and transferred for seed propagation. This represents a small number of the total population. From mutation densities estimated using Arabidopsis TILLING, only 10–20 mutations in target regions of ∼1 to 1.5 kb are expected to be recovered when screening a population of 3000 individuals, and less than half of recovered alleles are predicted to be deleterious to protein function (Greene et al., 2003; Till et al., 2003b). Because most alleles will be recessive, lethality in the population should be low. Such an approach to limit the substantial labour investment of maintaining live populations has been described for Drosophila, zebrafish and rat (Bentley et al., 2000; Draper et al., 2004; Smits et al., 2004). To evaluate this approach in plants, work is underway to establish TILLING populations in cassava using nodal cuttings. Several mutant populations produced by treatment with gamma irradiation or EMS have been prepared and are currently being evaluated for mutation density and spectrum (B. Till, S. Bado, and J. Jankowicz-Cieslak, unpublished). An important component in using mutagenesis of multicellular tissue and in vitro propagation is the determination of the vegetative generation where plants first become genotypically homogeneous. Collecting tissue samples for the TILLING library too early would mean chimerism and potential sampling of mutations that may not be retained in the population. However, sampling at too late, a stage would result in the inclusion of clonal replicates and unnecessary oversampling of the same alleles that could result in a several fold reduction in throughput. Genotyping experiments are being carried out to determine the number of propagation cycles required for tissues to become genotypically homogeneous.

The need to remove chimerism might be completely obviated through the development of protocols for cell culture mutagenesis. In this approach, an adult plant can be produced from a single cell (Taylor et al., 1996). This might enable the production of non-chimeric plants in the first vegetative generation after mutagenesis, allowing for rapid production of a TILLING population. However, aside from the extra efforts required to generate cell cultures, one caveat to this approach is the single-stranded nature of mutagenesis. After treatment with chemicals such as EMS, only one of the two strands of a DNA duplex is modified at a specific base, with mutations being fixed after one or more rounds of DNA replication. This suggests that single cell mutagenesis may result in a plant that is genotypically heterogeneous. It remains unclear whether low genotypic heterogeneity could be overcome through selective tissue sampling for the TILLING library to ensure that the mutations discovered are heritable. The feasibility of this approach is being explored for both cassava and banana through support for an International Atomic Energy Agency Coordinated Research Project (M. Dickmann, K. Danso, and J. Lopez, personal communication).

In silico TILLING: the future of mutation breeding?

Mutation breeding has been part of the agricultural landscape (literally) for more than 80 years. The advent of TILLING extended the range of plants that could succumb not only to reverse genetic approaches for gene function analysis, but also to targeted breeding. The development of TILLING has relied in large part on improvements in mutation discovery technology for high throughput and accurate recovery of induced mutations. NGS approaches are the latest advancement being applied to TILLING. Whole genome re-sequencing is already being used to access natural diversity in populations (e.g. Medicago HAPMAP http://medicagohapmap.org/; Arabidopsis 1001 genomes, Cao et al., 2011). The expanding data set of natural nucleotide variation will no doubt add new alleles to functional genomic and breeding projects, and in some cases may obviate the need for generating novel diversity through mutagenesis. However, because of inherent genetic constraints on populations, there will most likely remain a high demand for generating novel diversity through the use of physical and chemical mutagens.

Currently, whole genome sequencing costs are sufficiently prohibitive that pooling of targeted PCR amplicon approaches are being used (Tsai et al., 2011). We expect that as sequencing costs continue to decrease, this approach will be short-lived and re-sequencing of individual plant genomes will become economically practical. Using data from the National Human Genome Research Institute (NHGRI) (http://www.genome.gov/sequencingcosts/), we projected two cost scenarios for the future (Figure 6a). Projection A is based on the average decrease in sequencing costs per Mb from 2008 to 2011 (74% per year), and projection B is based on a slightly more conservative case where sequencing costs per Mb would decrease by 50% per year. Assuming 20× coverage per mutant individual, we calculated the costs of sequencing 5000 individuals from the M. truncatula (550 Mb), A. thaliana (135 Mb) and B. rapa (283 Mb) TILLING populations under the two cost scenarios (Figure 6b,c). We also included the cost for 2000 B. rapa mutants based on the normal population size, which is currently screened in this diploid with high mutation frequency. The current costs (late 2011) of sequencing all three populations are close to US$10M, but the costs should decrease significantly in the next few years so all individuals in the three populations could be sequenced for ca. $600 000 in late 2013 (Figure 6b) or late 2015 (Figure 6c). There will no doubt be a transition period where many laboratories choose lower cost and lower throughput for their own projects, but regardless of the exact timeframe, it seems inevitable that whole genome re-sequencing, or a heretofore undescribed technology for recovering allelic variation, promises to move TILLING into a largely in silico screening activity.

Figure 6.

 DNA sequencing costs per Mb in USD from July 2008 to July 2011 based on published data from the NHGRI, National Institutes of Health (http://www.genome.gov/sequencingcosts/, accessed 12 December 2011. Costs include those directly associated with sequencing up to the initial quality control and therefore do not include downstream data analysis (full details are provided on the website). (a) Two projections are included from July 2011 to July 2015 based on the average rate of decrease in costs from 2008 to 2011: (A) 74.4% projection; (B) a more conservative decrease in sequencing costs of 50% per year. Panels (b) and (c) represent the projected costs of sequencing 5000 TILLING individuals in three diploid species (Arabidopsis thaliana, Medicago truncatula and Brassica rapa) based on (b) the 2008–2011 average (inset is an enlargement of the graph over the last 2 years for clarity) and (c) the 50% reduction scenarios. A smaller population size of 2000 is also plotted for B. rapa.


The work of TLW, CU and FR was supported by the Biotechnology and Biological Sciences Research Council, UK (BBS/B/02401, BB/F010591/1 and BB/I000712/1). Funding for the work described on vegetatively propagated banana and cassava (BJT) was provided by the Food and Agriculture Organization of the United Nations and the International Atomic Energy Agency through their Joint FAO/IAEA Programme of Nuclear Techniques in Food and Agriculture. We thank Jillian Perry for providing Figure 1 and Helen Ghirardello for compiling Table S1.