TILLING by Sequencing (TbyS) for targeted genome mutagenesis in crops

The need to identify novel genetic variation in crops

To feed the increasing global population, it will be necessary to sustain our current growth in agricultural productivity over the course of the coming century whilst reducing environmental and climatic footprints (Godfray and Garnett 2014). The availability of genome sequences for a wide range of crop plant species and their close relatives has generated a vast pool of data that needs to be harnessed for sustained and improved genetic gain (Onda and Mochida 2016). Such genomic data provides a framework that can be used to identify genes and alleles in crop species that are causal or contributory for important traits such as yield, disease resistance and nutritional composition (Wang et al. 2012). However, many crop breeding programmes continue to suffer from a lack of genetic diversity and/or useful traits within the primary breeding pools of many crop species (Cooper et al. 2001; Tang et al. 2010). Such constraints can arise due to genetic bottlenecks encountered during domestication and the focus on elite germplasm within the twentieth century breeding programmes (Gross and Olsen 2010; Meyer and Purugganan 2013; Till et al. 2003; Kovach and McCouch 2008; Tang et al. 2010).

Historically, the role of mutagenesis breeding as an approach has been to extend the low-genetic diversity of many primary genepools of crops (Konzak et al. 1976). Such mutagenesis approaches have typically used radiation or chemical mutagens to introduce additional genetic polymorphisms into crop plant genomes, and thereby supplement the available natural variation within the crop primary genepool. Whilst random mutagenesis approaches are powerful for forward genetics to identify easily scorable phenotypes within large mutagenised populations, genome-wide mutagenesis approaches on their own do not allow the identification of the underlying causal mutated genes (Murphy 2007). TILLING was developed as a widely applicable reverse-genetics strategy for mutagenising target genes of interest in a manner which allowed the plants carrying mutations in the target gene to be identified from within a large mutagenised population of plants (Slade and Knauf 2005; Till et al. 2003; Wang et al. 2010; Wang et al. 2012; Triques et al. 2007a). Figure 1a provides a summary of a traditional TILLING workflow.

Open image in new window

TILLING programmes have been used to generate and identify crop germplasm containing altered sequences in genes associated with important agronomic traits (see Table 2.2. in (McKeown et al. 2013b)). In some cases, this has provided insights into the biology of species with complex or poorly sequenced genomes. For example, novel variation in wheat genes involved in starch synthesis pathways was generated by TILLING of the starch synthase II gene Sgp-1. Similarly, 246 alleles of the three Waxy gene homologues were generated, where lines mutated in at least two of these homologues displayed a reduced amylose phenotype in the endosperm, as determined by iodine staining (Sestili et al. 2010; Slade et al. 2005).

Despite the potential for TILLING to induce novel variation in genes, there are also limitations in such random mutagenesis approaches. Many of the allelic variants induced and identified during TILLING are considered to be phenotypically neutral, which can be due to the occurrence of synonymous mutations or due to genetic redundancy (particularly in polyploid genomes). In the accession C24 of Arabidopsis thaliana, analysis of mutations in four genes targeted by TILLING indicated that 69.6% of changes were missense and 29.0% sense, with 1.4% being nonsense mutations (Lai et al. 2012). Such nonsense mutations can introduce premature stop codons, and where they affect, N-terminal regions are likely to generate knock-outs and are the class of mutations typically sought in many TILLING programmes. However, depending on the gene target, such complete (amorphic) knockouts also carry a potential risk of being deleterious to the organism. Undoubtedly, more subtle mutations (e.g. hypomorphic, hypermorphic, antimorphic and neomorphic alleles), conferring potentially valuable novel phenotypes caused by changes to the function of a gene, also exist within mutagenised populations developed during a TILLING programme. Indeed, nonsense mutations in specific genes can be rare, requiring the use of larger mutagenised populations and more high-throughput screening (Wang et al. 2012).

Harnessing ‘next-generation’ sequencing for TILLING

The difficulty of identifying suitably mutagenised lines within TILLING populations is compounded by a second key limitation facing current processes: namely, the requirement for what has been termed an ‘ad hoc investment of time and resources’ in the procedure (Henry et al. 2014). This limitation will become ever-more pertinent as TILLING is applied to more genomically complex species, species with long lifecycles or complex reproductive modes, or where wider ranges of phenotypes are sought although progress has been achieved in this regard in the application of TILLING to challenging dioecious perennials such as hops and cassava (reviewed in Wang et al. 2012 and dubbed ‘VeggieTILLING’).

TILLING approaches have recently been undergoing a range of modifications to allow integration with high-throughput sequencing technologies. Such changes have been accompanied by improvements to pooling strategies, usually involving use of robotics platforms. As DNA sequencing has become cheaper and faster due to the advent of so-called next-generation sequencing (NGS) technologies, opportunities have arisen for incorporating such efficiencies into TILLING programmes. NGS platforms such as those driven by Illumina sequencing technology have allowed SNP (single-nucleotide polymorphism) analysis across multidimensional pools and have opened the door to more efficient strategies for mutant detection.

Development of robust, standardised pipelines for TbyS

TbyS can be defined as the application of next-generation sequencing supported by bioinformatics tools applied to the field of TILLING. One approach for focussing high-throughput sequencing on regions (i.e. exons) of the genome most likely to alter protein expression and protein-function dependent phenotypes is to use exome capture, in which mutagenised DNA is hybridised to oligos corresponding to the expressed transcriptome of the organism prior to sequencing. This procedure has been combined with TbyS in the case of wheat, rice and African rice, Oryza glaberrima (Henry et al. 2014). The exome-captured portions of these genomes represent the regions of the genome which are transcribed into protein-coding mRNAs. Such targeted-reduced-representation approaches do not generate data on the majority non-coding regions of the genome which do not code for proteins. Exome-captured portions of many plant genomes can potentially be multiplexed up to 30-fold and sequenced with a standard Illumina platform. Exome-captured approaches provide cost-benefit advantages which are especially significant for larger crop genomes which contain extensive tracts of non-coding repetitive sequences. For smaller crop genomes, such as rice, the cost benefits of exome sequencing are less evident, although this can be partially compensated for by increasing the representation of the exons most likely to undergo a loss-of-function mutation (Henry et al. 2014).

All TbyS requires robust bioinformatic protocols for handling the datasets produced by these high-throughput workflows. In many cases, this has been achieved by developing ‘in house’ bioinformatic workflows, e.g. using bespoke Python scripts. For instance, the analysis of mutagenised wheat and rice populations by (Tsai et al. 2011) was performed using Coverage Aware Mutation calling using Bayesian analysis—CAMBa (Missirian et al. 2011). Development of bespoke bioinformatic pipelines allows an integrated approach to TbyS workflow development. For example, CAMBa takes account of both the pooling setup and sequencing coverage levels when calculating mutations and noise probabilities, maximising the reliability of its outputs. CAMBa also takes account of the barcoding of multiplexed samples to identify the sub-pools in which mutations of interest have occurred (Tsai et al. 2011). Similarly, other groups have employed a pipeline called Mutations and Polymorphisms Surveyor, or MAPS (Henry et al. 2014). MAPS is of particular utility in the analysis of polyploid genomes as it uses each sample as a control against every other, allowing it to distinguish between allelic variants due to divergence between homeologues and induced mutations (Henry et al. 2014).

However, the need to develop such ‘in house’ solutions to bioinformatic analysis of TILLING populations also poses challenges in terms of bioinformatic expertise required to establish and sustain the bioinformatics pipeline. In the absence of community standardisation, bespoke approaches to bioinformatics pipelines run the risk of reduced compatibility between TbyS datasets. Hence, greater standardisation of TbyS bioinformatics analysis pipelines is desirable.

The complete datasets produced by TbyS experiments from different labs should ideally be archived for use by the wider scientific community, subject to commercial interests and according to best practice for data archiving (Sulpice and McKeown 2015). To meet this need, some existing databases have been developed by labs in UC Davis (http://comailab.genomecenter.ucdavis.edu/index.php/TILLING_by_Sequencing) and INRA (http://www-urgv.versailles.inra.fr/TILLING/TILLING-by-sequencing.htm) for understanding and interpreting the pools of data produced by these approaches.

TILLING in polyploid genomes

Most crop plants have polyploid genomes (wheat, potato, brassicas, okra and many fruit crops), thus the use of high-throughput TILLING in polyploids is of particular interest in the field of crop improvement (Tsai et al. 2013). TILLING by Sequencing in polyploid organisms like Camelina sativa is being utilised to generate lines with improved oil quality and other non-nutritive traits (Anishkumar et al., unpublished). The fact that polyploids have a higher gene dosage than diploids poses challenges in this regard, due to the risk of phenotypes being masked by complementation from another homeologue.

An approach for TILLING autopolyploid A. thaliana has been proposed as a strategy for generating a higher mutational load in TILLING populations (Tsai et al. 2013). This mutant population was raised by treating the wild type Col-0 (2n = 10) and autotetraploid Col-4X (4n = 20) with about 30 mM and 50 mM EMS. A three-dimensional pooling method was again utilised and the analysis carried out with a modified version of CAMBa, adapted for the pooling scheme (Tsai et al. 2013). TILLING by Sequencing resulted in 19.4 mutations per megabase in the 15 genes which were analysed.

TILLING-based technology and the rise of genome editing technologies

There is rapid growth underway in the development and use of genome editing (GE) technologies to directly modify nucleotides or sequence tracts within a locus, e.g. to introduce premature stop codons in the 5′-region, or other targeted changes to an open reading frame (ORF) (Quetier 2016). Although several such methodologies have been proposed for use in crops and other plants (ZFN, TALENs), major interest has focussed on the Streptococcus-derived CRISPR/Cas9 system (Belhaj et al. 2013; Wang et al. 2014a; Cermak et al. 2015; Svitashev et al. 2015; Zhu et al. 2016).

The major advantage of genome editing approaches is that desired specific mutation(s) or changes can be introduced into the genome, including the possibility to target multiple genes in diploid genomes, or indeed homeologs in polyploid genomes (Lowder et al. 2015; Shan et al. 2013; Upadhyay et al. 2013; Wang et al. 2014b; Belhaj et al. 2015; Gil-Humanes et al. 2016; Morineau et al. 2016). Genome editing by CRISPR/Cas9 is also possible in perennial species such as poplar (Fan et al. 2015) and has been applied to plant pathogens such as Phytophthora species (Fang and Tyler 2015).

CRISPR/Cas9 approaches were initially developed using easily identified target genes (such as the tomato ARGONAUTE7 (SlAGO7) which produce a wiry-leaved phenotype when mutated (Brooks et al. 2014) to demonstrate proof of concept prior to moving on to targets of research or agricultural interest. The use of CRISPR/Cas9 systems for genome editing is now rapidly spreading across plant research (Puchta 2016), with a wide range of applications emerging including gene replacement (Li et al. 2016), promoter swaps (Shi et al. 2016), metabolic engineering (Alagoz et al. 2016), gene activation/repression (Seth and Harish 2016) and QTL (Quantitative Trait Loci) editing (Shen et al. 2016), targeted to modify a wide range of traits (e.g. nitrogen fixation) (Wang et al. 2016), plant virus resistance (Chandrasekaran et al. 2016) and thermosensitive male sterility (Zhou et al. 2016). CRISPR/Cas9 has also been proposed as a means of targeting non-protein coding genes such as lncRNA and miRNAs (Basak and Nithin 2015), and is also subject to a range of improvements to accelerate the speed and reliability of locating homozygous mutants such as systems specifically expressed in germline tissue (Wang et al. 2015; Mao et al. 2016).

To move beyond discovery science, a key challenge for genome editing (and TILLING approaches) based on candidate genes (or more precisely candidate SNPs) is some prior predictive knowledge of what phenotypic changes can be elicited by specific nucleotide changes to a particular gene or locus. For instance, genome editing or TILLING approaches may be suitable for engineering point mutations conferring single-locus heterosis, as we have previously suggested (McKeown et al. 2013a). Both genome editing and TILLING in their current forms can (in principle) be applied to generate allelic series of the same gene to determine whether different variants of the same gene can display different phenotypic effects.

At present, useful genetic variation within the secondary and tertiary genepools of crop wild relatives remains difficult to incorporate into breeding programmes and breeding lines due to crossability challenges and also linkage drag of multiple genes with genes (or alleles) of interest that confer useful traits (Warschefsky et al. 2014). Where specific mutations in orthologous genes are known in wild relatives that confer phenotypes that would be useful in the primary genepool breeding materials in principle, it could be swifter to introduce such polymorphisms to the primary genepool by targeted mutagenesis (TILLING or genome editing). However, apart from some pioneering approaches that have demonstrated such concepts using tomato wild relatives (Soyk et al. 2016), the current reality is that most candidate mutations in wild relatives are insufficiently functionally characterised to facilitate targeted mutagenesis strategies to recapitulate wild alleles in the primary genepool (Brozynska et al. 2016).

At present, an important commercial (as opposed to fundamental scientific) consideration when comparing TILLING approaches to genome editing approaches are the following: (1) the freedom to operate and costs to access the proprietary technology and (2) the regulatory status of products derived from genome editing technologies (such as CRISPR/Cas9) (Voytas and Gao 2014; Wolt et al. 2015; Morris and Spillane 2008; Waltz 2016; Sprink et al. 2016a; Ricroch et al. 2016b; Straubeta and Lahaye 2013; Uluisik et al. 2016; Pyott et al. 2016; Spillane and Swanson 2002).

Both TILLING and CRISPR/Cas9 methods for targeted mutagenesis in plants are subject to a range of patents, which could impact on freedom to operate in relation to commercialisation of products derived from such mutagenesis approaches. A recent analysis of intellectual property rights in plant breeding (focused on European IPR databases) identified 38 mutagenesis patents, of which 20 related to mutation detection by sequencing (van de Wiel et al. 2016). In the case of CRISPR/Cas9, there is a flurry of patenting activity given the pervasive impact that the CRISPR/Cas9 technology is expected to have across all biological domains (Sheridan 2014; Sherkow 2015). Given the financial stakes, initial patent submissions on CRISPR/Cas9 methodologies have become the subject of legal disputes and challenges in both the US and European patent jurisdictions (Ledford 2016). Since the first patents were filed on CRISPR/Cas9 in 2004/2005, there has been a rapid rise in patent filings (especially since 2012), with over 600 inventions filed by 2015 (Egelie et al. 2016). Patent families are sets of patents taken in different countries to protect a single invention. In this regard, the Swiss consulting firm IPStudies has estimated that there are over 860 CRISPR patent families worldwide, with a new one currently being added every day (Ledford 2016). As the patent landscape for CRISPR/Cas9 has not yet stabilised, this poses challenges in the near-term for licencing of CRISPR/Cas9 technologies for generation of commercial plant products.

A second commercially relevant point of comparison between TILLING and CRISPR/Cas9 approaches relates to the regulatory process relating to, and the status of, improved plant products from each approach. In the case of TILLING, the mutant plant lines generated by chemical or radiation mutagenesis are currently not regulated as genetically modified organisms (GMOs) in most jurisdictions. Given the cost, time and resource burdens associated with the regulatory procedures relating to plant GMOs, this is a significant commercial consideration. For instance, a 2011 consultancy study for CropLife International which surveyed plant biotechnology companies estimated a total cost of US$ 136 million for the discovery, development and authorisation approval of a plant biotechnology trait. Within this cost, an estimated US$35.1 million (~26%) of the costs was for meeting the regulatory requirements for commercial authorisation of a GM crop trait (McDougall 2011).

In the case of CRISPR/Cas9, even though the mutants generated can be scientifically indistinguishable from those generated from TILLING mutagenesis, it is unclear in a number of jurisdictions, particularly in European community (Hartung and Schiemann 2014; Smyth 2016; Abbott 2015), whether CRISPR/Cas9 genome-edited plants will be subject to regulation as GMOs, as some form of ‘GMO-lite’ bespoke genome editing legislation, or as ‘substantially equivalent’ to lines generated by traditional mutagenesis techniques (including TILLING) that are considered by some regulatory regimes as non-GMO. Notwithstanding the need for innovations to advance sustainable intensification of agriculture (Ricroch et al. 2015), the extent of deployment of genome-edited plants will be influenced by whether CRISPR/Cas9-genome editing of plants (in different jurisdictions) will be subject to a process or product based regulatory lens (Morris and Spillane 2008; Ricroch et al. 2016a; Sprink et al. 2016b; Huang et al. 2016).