Overcoming barriers to the registration of new plant varieties under the DUS system

The study addresses whether the current DUS (Distinctness, Uniformity, Stability) testing system—central to granting Plant Variety Rights—is adequate in the modern era. DUS testing has long relied on a limited set of morphological traits, whereas crop breeding has advanced and the number of candidate varieties has increased. The authors outline multiple pressures on the DUS system: shrinking combinatorial space of morphological traits as more varieties are released; many DUS traits having low heritability and strong environmental sensitivity; challenges applying DUS to minor crops and outbreeding or hybrid systems; and ambiguity in defining essentially derived varieties (EDVs). They propose to systematically evaluate the UK barley DUS system using a large historical panel to quantify deficiencies and to explore genomics-enabled alternatives that could make DUS testing more robust, reproducible, and scalable.

Previous attempts to augment DUS with molecular markers date back to minisatellites in soft fruits (1990). Subsequent proposals include small sets of SSRs/SNPs (e.g., 28 SSRs in maize, 25 SNPs in barley, 5 SSRs in rice) and larger SNP arrays (e.g., 3,072 in maize, 6,000 in soybean). UPOV guidance currently restricts marker use to cases with perfect correspondence to DUS traits, which does not reflect advances in genotyping throughput and understanding of trait genetics. Prior studies in barley have used limited marker numbers and reported modest correlations between molecular distances and DUS trait distances. Speed breeding combined with DUS has been proposed to accelerate timelines, but changes in plant development may affect trait expression and require validation. Widely available SNP arrays in major crops (barley 50k, wheat 90k, maize 600k) and GBS offer scalable genotyping resources that could underpin a genomic DUS framework. The literature also highlights unresolved EDV definitions and legal complexities, suggesting potential benefits from objective genomic criteria.

Data and materials: The analysis focused on 805 UK barley varieties (432 spring, 372 winter, 5 alternative; 4 removed for missing AFP numbers) genotyped with 43,799 SNPs on the barley 50k iSelect array; after removing markers with missing data, 40,078 SNPs remained. DUS data comprised 28 traits (seasonal type plus 27 morphological traits) from NIAB and SASA public sources, supplemented with Cockram et al. NIAB data (more complete) served as the primary dataset; SASA was used for cross-organization comparisons. Trait scoring followed APHA scales (21 traits on 1–9 or subsets; 7 binary). Two traits were non-segregating in spring and one in winter. Missingness ranged 0–78% (only 5 traits >10%). The merged trait and marker dataset included 710 varieties. Analyses: - Cross-organization consistency: For 395 varieties scored by both NIAB and SASA, trait score discrepancies were computed as absolute differences (binary traits recoded to 1/2 for comparability). Variety-level consistency was the proportion of exact matches across the 28 traits. - Trait combinatorial space over time: Manhattan distances among DUS trait profiles were computed and a rolling mean (window of 20 varieties ordered by AFP submission date) assessed diversity trends separately for spring and winter barley. - Heritability estimation: Univariate genomic REML mixed models (sommer::mmer) partitioned phenotypic variance into additive genetic and residual components using a genomic relationship matrix. Fixed effects included intercept, year of national listing, and seasonal type (dropped in season-specific analyses). Heritabilities (h²) were estimated for combined (n=710), spring (n=370), and winter (n=335) datasets. - Yield data and correlations: For spring barley, dry matter yield from 509 varieties (1948–2019) was modeled with a mixed-effects model (lme4), accounting for management (pre-1983 vs fungicide-treated post-1983) and interactions; BLUEs were extracted (emmeans). Bivariate mixed models (sommer::mmer) for 192 spring varieties estimated genetic (ρ_g) and phenotypic (ρ_p) correlations between each DUS trait and yield using multivariate random effects with the genomic relationship matrix. - GWAS: Genome-wide association scans for each DUS trait were run in combined (n=710), spring (n=370), and winter (n=335) datasets using mixed models with polygenic background (sommer). Significance used FDR<0.05 (qvalue). For traits with very strong peaks (−log10 p>10), a conditional GWAS including the top marker as a fixed effect was performed to reveal masked loci and control for LD artifacts. - Evaluation of small DUS marker set: A proposed 45-marker DUS panel (Owens et al.) was evaluated using 39 usable markers. For 212 varieties with known intercross parent pairs, progeny marker haplotypes were simulated and exact marker matches were assessed between simulated progeny and their real sibling variety, parents, and among simulated progeny. Bootstrapping quantified match rates; LD structure and number of segregating markers per cross were noted. - Marker number required: Random subsets of markers (from small to large, increasing in log10 steps of 0.1) were sampled to compute Manhattan distances; correlations with DUS-trait Manhattan distances were calculated for all traits and for high-h² (≥0.5) vs low-h² (<0.5) traits. - Genomic DUS demonstration: Using haploid-coded SNPs (0/1), pairwise Manhattan distances were computed among all 805 varieties, within and across seasonal types. Two spring parent pairs were chosen: Propino×Quench (low parental distance 0.20) and Riviera×Cooper (high distance 0.59). From each, 1,000 F6 and 1,000 BC1S4 progeny were simulated. For each simulated progeny, the minimum distance to the 805-variety reference panel was recorded. An illustrative distinctness threshold of 0.05 was applied to estimate rejection rates. The approach was also used to flag potential EDV-like pairs below the threshold. - Time and cost appraisal: Comparative assessment of time and costs for morphological DUS (field/glasshouse, 1–2 years; ~£1040 per candidate), speed DUS (conceptual; requires trait validation), trait-specific KASP markers (~£11 per 100 markers), and genomic SNP arrays (~£40 for >40,000 markers on barley 50k).

- DUS trait inconsistencies across organizations/environments: About two-thirds of trait scores were consistent between NIAB and SASA. Variety-level differences were generally small but pervasive (mean absolute score difference 0.55, sd 0.28; n=395; two-sided t-test p<0.05). Systematic biases were noted for trait 6 (higher at NIAB) and trait 25 (lower at NIAB), consistent with environmental effects. Only 3 of 395 varieties had no inconsistencies. - Low heritability of many traits: 15 of 28 traits had h²<0.50 (combined analysis), implying poor replicability across environments. The proportion of score differences between NIAB and SASA was negatively correlated with heritability (r≈−0.67 overall; −0.61 spring; −0.59 winter), indicating environmental sensitivity undermines DUS reliability. - Shrinking trait combinatorial space: Rolling Manhattan distances of DUS profiles decreased over time (especially in spring barley), showing convergence of morphological profiles and reduced ability to distinguish new varieties based on traits alone. - Genetic and phenotypic correlations constrain DUS space: In spring barley, 12 of 21 evaluable DUS traits showed non-zero genetic correlations with yield. Notably, traits 10 (ear attitude) and 11 (plant length) were negatively correlated with yield, favoring semi-dwarf, erect-ear ideotypes among high-yielding lines and further restricting distinct morphological combinations as yield improves. - GWAS reveals mixed architecture: 32 significant loci (FDR<0.05) across 14 traits were detected (30 in combined, 12 spring-only, 16 winter-only). Several traits are controlled by major loci, some fixed within seasonal pools (e.g., Ant1, Ant2 for anthocyanin traits; Vrs1/Vrs2/Vrs3 for row number and spikelet development; Nud for husk; Cly1 for lodicules; Ppd-H2, Vrn-H1/H2 for seasonal type). However, half of the traits lacked major-effect loci, making trait-perfect markers impractical under current UPOV guidance. - Small DUS marker set is inadequate: Of the 39 evaluated markers, only 4–22 segregated per parent pair, with LD reducing informativeness. In newer varieties, high exact-match rates were observed among simulated progeny, their real sibling variety, and parents (e.g., for LG Goddess and its parents, exact matches around 7–8% of simulated progeny). 88.4% of simulated progeny had >1% probability of matching others, indicating poor discriminative power as diversity narrows. - Required marker density: Correlation between marker-based and DUS-trait Manhattan distances plateaued with ~500–1000 markers (max ~0.60 overall; higher when using only high-h² traits). Variance of distance distributions stabilized similarly, indicating that <500 markers are insufficient for robust separation. - Genomic distance distributions and thresholds: Among 805 varieties, within-season Manhattan distances ranged 0.04–0.69 (spring) and 0.04–0.87 (winter); between-season distances 0.44–0.97. Using an illustrative distinctness threshold of 0.05, simulated progeny rejection rates depended on parental distance: for low-distance parents (Propino×Quench), 13.0% of F6 and 59.6% of BC1S4 progeny failed distinctness; for high-distance parents (Riviera×Cooper), 0% of F6 and 4.9% of BC1S4 progeny failed. Four real variety pairs in the reference panel fell below 0.05, consistent with indistinguishability previously noted by protein electrophoresis or microsatellites. - Time and cost advantages: Genomic DUS offers rapid turnaround (days to weeks) at low cost (~£40 for >40k SNPs) compared to morphological DUS (~£1040; 1–2 years). Trait-specific KASP panels are inexpensive (~£11/100 markers) but insufficient for comprehensive DUS. Speed DUS still requires trait validation due to developmental effects.

Findings demonstrate that morphology-based DUS is increasingly unreliable and inefficient: many traits are environmentally labile (low h²), cross-organization inconsistencies are common, and morphological diversity is converging due to selection for yield and other VCU traits, restricting the trait combinatorial space available for distinctness. GWAS shows that while some DUS traits map to major loci, many do not, limiting the feasibility of trait-perfect markers under current UPOV TGP/15 guidance. Small fixed marker sets also fail to uniquely identify varieties in a narrowing diversity context. The study shows that genome-wide SNP data can provide a robust, quantitative distinctness criterion via genomic distances, with practical thresholds to flag EDVs and to manage sameness among closely related lines. Genomic data also enable straightforward assessments of uniformity (distances among multiple seeds or pools) and stability (monitoring genomic heterogeneity over generations). Such a system would be scalable, faster, cheaper, and more reproducible across environments and jurisdictions, facilitating data-sharing across testing centers and reducing logistical constraints (e.g., seed movement limitations). It further offers an objective framework to clarify EDV boundaries and supports integration with modern breeding tools, including gene editing and improved seed certification practices.

Using a comprehensive UK barley panel, the authors show that the current DUS system based on morphological traits suffers from environmental inconsistency, low heritabilities, shrinking trait diversity, and inadequate marker alternatives. They propose a genomic DUS framework leveraging genome-wide markers to assess distinctness via distance thresholds and to evaluate uniformity and stability via within-variety genomic homogeneity. The approach addresses key shortcomings in time, cost, reproducibility, and legal clarity around EDVs. Future directions include: (i) establishing standardized, crop-specific genomic distance thresholds using large, curated reference panels; (ii) validating genomic DUS across additional crops, including outbreeders and hybrids; (iii) harmonizing protocols for sampling seeds and defining uniformity/stability metrics; (iv) integrating genomic DUS with regulatory and legal frameworks (UPOV) and seed certification systems; and (v) ensuring accessibility of genotyping platforms (arrays or GBS) for minor/orphan crops.

- Data coverage and completeness: Public DUS trait datasets (especially SASA) had higher missingness; the merged dataset may not include all UK barley varieties historically, potentially biasing diversity estimates. NIAB data were used as primary due to completeness. - Yield correlations limited to spring barley: Older winter barley yield data were unavailable; thus, genetic/phenotypic correlations with yield were analyzed only for spring barley (n=192), not necessarily generalizable to winter barley. - Arbitrary distinctness threshold: The 0.05 Manhattan distance threshold was illustrative; optimal thresholds require formal calibration and may be crop- and panel-specific. - Marker simulation assumptions: Evaluation of the small marker set used simulations and limited markers (39 usable) with LD and variable segregation; results may vary with different panels or marker selection. - GWAS power and LD: Sample sizes and LD structure, as well as traits fixed in seasonal pools, affected detection power and locus resolution; some candidate loci remain unverified. - Trait scope: Correlations were assessed with yield only; other VCU traits may further constrain DUS trait space but were not analyzed here. - Environment and scoring differences: Cross-organization inconsistencies partly reflect environmental effects and scorer differences; precise partitioning of these sources was not directly modeled across multi-environment trials for all traits.