Precision plant breeding with CRISPR

A Cas nuclease cuts where a guide RNA directs it, and the plant repairs the break. Left to itself, that repair is end-joining, which is error-prone and leaves short insertions or deletions — usually a frameshift, usually a dead gene. This single fact explains most of what crop editing has and has not delivered.

Loss of function is the natural target

If the desired phenotype is the absence of a protein, editing is direct, fast and precise. Knocking out a MLO susceptibility gene gives powdery mildew resistance, demonstrated in tomato and then in hexaploid wheat where all three homoeologous copies were disrupted at once — a job conventional breeding does badly because the copies segregate together. Removing a polyphenol oxidase gives non-browning flesh in potato and mushroom. Disabling a starch-branching or fatty-acid desaturase gene changes composition. Editing a promoter so a pathogen effector can no longer induce a sugar transporter gives bacterial blight resistance in rice.

The common shape is that the trait was one gene, and the useful allele was its absence. Where the wanted change is a new function or a specific substitution, homology-directed repair is needed, and in plants it competes badly against end-joining and requires delivering a template to the same nucleus at the same time. Base editors and prime editors were developed to route around this — installing defined substitutions without a double-strand break — and both work in cereals, prime editing so far at low efficiency.

The polygenic wall

Yield, drought performance and most agronomic value are not single genes; they are hundreds of loci of small effect, and there is no set of edits that adds up to them. Multiplexing does not solve this, because the problem is not the number of cuts but that no one knows which sites to cut or what each contributes.

There is a real and more modest use of editing on quantitative traits, and it is worth stating precisely. Rather than trying to install an optimum, a promoter or other cis-regulatory region of a known developmental gene is cut repeatedly to generate a series of alleles with graded expression, and the resulting range is then phenotyped and selected in the ordinary way. This has produced usable variation in tomato inflorescence architecture and fruit size. Editing here is a source of variation, feeding the selection machinery described under genetic engineering and breeding — not a substitute for it.

Delivery is the actual bottleneck

In most crops the limiting step is not the nuclease. It is getting the reagent into a cell and getting a whole plant back out. Transformation by Agrobacterium or particle bombardment, followed by regeneration through tissue culture, is strongly genotype-dependent: a few model varieties are tractable and most elite lines are recalcitrant. The standard workaround — edit an amenable genotype and introgress the edit — reintroduces years of backcrossing and the linkage drag that editing was supposed to avoid.

Morphogenic regulators are the main advance against this. Transient expression of developmental genes such as Baby boom and Wuschel2, and separately chimeric growth-regulating factor fusions, substantially widens the range of genotypes that will form transformable, regenerable tissue in maize, sorghum and wheat. Approaches that bypass tissue culture altogether — viral delivery of guides, editing via a haploid inducer — are progressing but are not yet general.