Breeding & CRISPR

Domesticated wheat first appeared in the Fertile Crescent around 10 000 BP as an allotetraploid Triticum turgidum ssp. dicoccum (emmer). The modern bread-wheat genome (Triticum aestivum, AABBDD, 1C ≈ 16 Gbp) arose ~8 000 BP through an incidental hybridisation of tetraploid wheat with the goatgrass Aegilops tauschii (DD). For most of human history, improvement was mass-selection: farmers saved the heaviest heads from the best fields.

Scientific plant breeding began with Mendel’s rediscovery (1900), Rollin Thatcher’s wheat crosses (Minnesota, 1910s), Sir Rowland Biffen’s Mendelian demonstration of rust resistance (Cambridge, 1905), and the work of Nikolai Vavilov at the All-Union Institute of Plant Industry (Leningrad). Vavilov’s expeditions across five continents (1916–1940) assembled the largest pre-war germplasm collection and defined the “centres of origin” now codified in international gene-bank networks (CIMMYT, ICARDA, VIR, NSGC).

The pivotal 20th-century breakthrough was the Green Revolution. Norman Borlaug, based at the Rockefeller Cooperative Wheat Research Program in Mexico, crossed Japanese semidwarf cultivars (Norin 10, carrying Rht-B1band Rht-D1b) with tall, rust-susceptible Mexican spring wheats, producing varieties such as Pitic 62, Penjamo 62, and ultimately the photoperiod-insensitive Sonora 63/64. By 1970 these had transformed Mexico from a wheat importer to an exporter, doubled Indian and Pakistani yields, averted projected mass famines, and earned Borlaug the Nobel Peace Prize.

The Green Revolution depended on one molecular trick: the uncoupling of plant height from fertility. Rht-B1 (on chromosome 4BS) and Rht-D1 (4DS) encode DELLA proteins, GRAS-family transcriptional repressors that sit at the heart of gibberellin (GA) signalling. In wildtype plants, the GA receptor GID1 binds GA and recruits DELLA to the E3 ubiquitin ligase SCF^SLY1, targeting DELLA for proteasomal degradation. Without DELLA, PIF-family transcription factors activate cell-elongation and biosynthesis genes.

The Rht-B1b and Rht-D1b alleles carry a Q64-to-stop point mutation in the DELLA / TVHYNP N-terminal domain, creating a truncated protein that GID1 cannot recognise. The plant therefore behaves as though it is GA-deficient: shorter internodes, thicker stems, and a dramatic rise in harvest index.

\[ \text{DELLA}^{wt} + \text{GA-GID1} \;\xrightarrow{\text{SCF}^{SLY1}}\; \text{26S proteasome} \;\Longrightarrow\; \text{growth} \]

\[ \text{DELLA}^{Rht\text{-}B1b}\;\;\text{(GA-insensitive)}\;\Longrightarrow\;\text{persistent repression of growth genes} \]

At constant biomass, reducing plant height from ~140 cm to ~70 cm approximately doubles the harvest index (grain:total dry mass) from 0.25 to 0.50. It also eliminates lodging at high fertiliser rates — the barrier that previously capped the N response curve. Without Rht, the nitrogen-rich wheats of the 1960s would simply have fallen over before heading.

By the late 1980s restriction fragment length polymorphism (RFLP), then simple sequence repeat (SSR/microsatellite) markers, and finally diversity array technology (DArT) and genotyping-by-sequencing (GBS) allowed breeders to tag individual loci for indirect selection. The workflow:

• Bi-parental mapping population: cross two contrasting parents, produce ~200 recombinant-inbred lines (RILs) by single-seed descent through ~F6.
• Phenotype the RILs in replicated trials (yield, protein, rust score, 1000-kernel weight).
• Genotype with ~10 000–50 000 SNPs (Illumina 90K, Axiom 660K/35K, GBS).
• QTL scan: interval mapping or composite interval mapping (Zeng 1994) scores LOD support for a QTL at each marker position; MIM or Bayesian MCMC when many QTL are suspected.
• Validate the QTL in an independent population before deploying as a marker-assisted selection (MAS) tag.

Classical wheat QTL intervals include the Rht loci, Vrn-1/Vrn-2/Vrn-3(vernalisation), Ppd-D1 (photoperiod), Glu-1/Glu-3 (HMW/LMW glutenins),GPC-B1 (grain protein, NAM-B1, Uauy 2006), Yr5/Yr15/Yr36(stripe rust), and Lr34/Yr18/Sr57/Pm38 (multipathogen adult-plant resistance).

\[ \text{LOD}(x) = \log_{10}\!\frac{L(\text{QTL at } x \mid \text{data})}{L(\text{no QTL} \mid \text{data})} \]

Genome-wide association studies (GWAS) extended the concept to diversity panels of unrelated lines, trading off some resolution for mapping precision on the historical recombination events accumulated over thousands of generations. The wheat 15K and Axiom 660K arrays are the common workhorses; imputation to the IWGSC RefSeq v2.1 reference has brought marker density effectively to whole-genome resolution.

Meuwissen, Hayes & Goddard (Genetics, 2001) proposed that if the marker density is high enough that every QTL is in linkage disequilibrium with at least one marker, one could skip QTL discovery entirely and instead fit a statistical model that estimates an effect for every marker simultaneously. The candidates’ genomic estimated breeding values (GEBVs) are then predicted as a weighted sum of all marker scores.

In its simplest form, rrBLUP solves a ridge regression:

\[ \hat{\boldsymbol\beta} \;=\; \bigl(X^\top X + \lambda I\bigr)^{-1} X^\top y,\qquad \lambda = \frac{\sigma_e^2}{\sigma_g^2 / m} \]

where X is the centred marker matrix, y the phenotype vector, and the shrinkage λ is set by the ratio of residual to genomic variance. Equivalent mixed-model formulations (GBLUP) replace the marker matrix with a genomic relationship matrix G = XX⊤/m. Bayesian variants (BayesA, BayesB, BayesCπ, BayesLASSO, BayesR) allow different shrinkage priors for different loci, which helps when a few major-effect QTL dominate.

Typical prediction accuracies (Pearson r between GEBV and realised breeding value in a validation set) for wheat are 0.4–0.7 for yield, 0.6–0.8 for plant height and heading date, and 0.5–0.7 for adult-plant rust resistance (Rutkoski 2014, Crossa 2017). Crucially, GS replaces generations of field trialling with a DNA test on seedlings, cutting the breeding cycle from ~8 to ~3–4 years and approximately doubling the genetic gain per unit time (Heffner 2010).

Doubled-haploid (DH) technology fixes homozygosity in one generation rather than the six or seven produced by single-seed descent. The two routes in wheat are:

• Wheat × maize (Laurie & Bennett 1988): pollination with maize pollen yields haploid embryos because the maize chromosomes are eliminated during the first few mitoses. Embryos are rescued on tissue culture and the remaining haploid wheat genome is diploidised with colchicine.
• Microspore culture: anthers or isolated microspores are induced to form haploid embryoids, then doubled with antimitotic agents.

DH lines reach F-infinity homozygosity in ~12 months rather than ~6 years. Combined with marker-assisted backcrossing they make it feasible to stack 4–6 QTL in a fully homozygous target background within a public-breeder timeframe.

Speed breeding (Watson et al., Nat. Plants, 2018; Ghosh et al. 2018) exploits wheat’s long-day photoperiod requirement to accelerate the generation cycle. Under continuous 22 h of LED illumination at ~450 µmol m⁻² s ⁻¹ and well-regulated temperature (22 °C day / 17 °C night), spring wheat progresses from sowing to first fully mature spikes in ~70 days, halving the classical ~140-day cycle. Combined with early-harvest of immature seed and in-vitro embryo rescue, breeders can obtain six generations per year instead of the typical two.

Speed breeding synergises with genomic selection: GEBVs are calculated on seedlings, the best lines are fast-tracked, and fixed homozygous progeny are available for field-trialling within 2–3 years of the initial cross. Ghosh 2018 estimated the approach compresses the decade-long CIMMYT public breeding cycle by 30–50%.

Precise genome editing in wheat is complicated by homoeology. For every functional gene there are typically three near-identical copies — one each on the A, B and D subgenomes — and altering phenotype often requires editing all three homoeologs simultaneously. This is one of the central bottlenecks that delayed wheat biotechnology relative to maize or rice.

The first major CRISPR wheat demonstration was Shan 2013 (Nat. Biotechnol.), who edited TaMLO with TALENs and then with SpCas9. The sustained effort of Wang 2014 produced a triple-homoeolog tamlo-aabbdd knockout resistant to powdery mildew, but the knockouts were accompanied by yield penalties from pleiotropic growth defects. Li 2022 (Nature) solved this with tamlo-R32: a precise 304 kb deletion that preserved disease resistance while up-regulating a neighbouring sugar transporter, restoring yield to elite levels.

Delivery routes in wheat include:

• Biolistic transformation of Cas9 ribonucleoproteins or DNA into immature-embryo calli.
• Agrobacterium-mediated transformation of the spring cultivar Fielder or Bobwhite.
• Haploid-inducer (HI-Edit): edit the inducer line once, then cross to any elite recipient.
• Plant virus delivery: barley stripe mosaic virus (BSMV) vectors carrying guide RNAs enable transgene-free somatic editing.

Beyond nuclease editing, base editors (cytosine deaminase CBE, adenine deaminase ABE; Komor 2016, Gaudelli 2017) and prime editors (Anzalone 2019) now allow single- nucleotide edits without double-strand breaks, greatly expanding the catalogue of accessible alleles.

Modern hexaploid wheat arose from just a handful of Aegilops tauschiiindividuals ~8 000 BP, which means the D-genome diversity in cultivated wheat is a tiny fraction of what the species contains in the wild. Synthetic hexaploid wheats (SHWs, Mujeeb-Kazi 1996) recreate the original polyploidisation by crossing a modern tetraploid durum (AABB) with a wild Ae. tauschii (DD) accession, then doubling with colchicine. CIMMYT’s SHW-derived cultivars (e.g., Chuanmai series, Weebill) contribute drought and rust tolerance that had been lost at domestication.

Distant wild relatives serve as a reservoir of novel R-genes and stress-tolerance alleles. Alien introgressions of broad utility include:

• Sr24, Sr25, Sr26, Sr43 from Thinopyrum / Agropyron relatives.
• Sr31, Lr26, Yr9, Pm8 from rye’s 1BL/1RS translocation.
• Lr19, Sr25 from Th. ponticum.
• Lr47 from Aegilops speltoides.

The advent of chromosome-scale genome assemblies for Ae. tauschii, wild emmer (Avni 2017), einkorn (Ahmed 2023) and the global wheat pan-genome (Walkowiak 2020, 16 reference assemblies) now allows breeders to mine specific haplotypes from thousands of accessions without leaving the desktop.

The same edited allele can be regulated very differently depending on jurisdiction. As of 2026:

• USA — USDA APHIS SECURE rule: site-directed nuclease (SDN-1) edits indistinguishable from natural mutations are exempt from GMO regulation. Corteva/Syngenta editing pipelines routinely reach commercial release in ~5 years.
• Argentina, Brazil, Japan, Australia — similar “case-by-case” product-based frameworks.
• EU — the 2018 Court of Justice ruling (C-528/16) classified all gene-edited plants as GMOs under Directive 2001/18. In 2024 the European Commission proposed a new category of “NGT-1 plants” indistinguishable from conventional breeding, to be exempted; at the time of writing, final trilogue negotiations are ongoing.
• China — commercial approvals since 2023 under a dedicated gene-edited crop regulation.

Regulatory asymmetry matters: an edited wheat released in Argentina or the US may still face export-market frictions in EU-dependent supply chains, and this risk currently shapes the commercial investment decisions of multinational breeders.

Four trajectories will shape the next decade of wheat improvement:

• Pan-genome-aware breeding: the Walkowiak 2020 pan-genome and its subsequent expansions reveal that each elite cultivar carries presence/absence variation for thousands of genes. Tools like PanBLUP and haplotype-aware GS move beyond SNP-only models to exploit this variation directly.
• De novo domestication: Li 2018 (tomato), Yu 2021 (wild rice) showed that CRISPR editing of 4–6 domestication genes can convert a wild species to a cultivated phenotype in one generation. Th. intermedium (Kernza, perennial wheatgrass) and Agropyron species are candidate substrates.
• Hybrid wheat — long attempted, now plausible with CRISPR-engineered male sterility systems (SeedLink, Croptimize). If breakeven hybrid-seed economics can be achieved, ~15% heterosis could translate to an additional 100–150 Mt of global wheat production.
• Machine-learning phenomics: UAV hyperspectral + deep-learning canopy segmentation routinely substitutes for hand-scoring disease ratings, stand establishment, and physiological traits, raising phenotypic throughput by orders of magnitude for GS training populations.