Exploring the potential of naked barley to manage deoxynivalenol accumulation from Fusarium head blight

Abstract

Developing and deploying multi-use naked barley varieties that accumulate lower levels of deoxynivalenol (DON) resulting from Fusarium head blight (FHB) disease could help farmers mitigate economic risks and expand market possibilities. Previous work established that a substantial amount of DON in covered barley accumulates in the hull and can be removed by pearling. We studied a diverse panel of 244 naked barley lines to determine if there is a genetic variation for the distribution of DON in Fusarium-infected spikes. We evaluated the panel genotyped with a 50K Barley SNP array for disease severity, toxin accumulation, and other agronomic traits in two FHB disease nurseries from 2020 to 2021. Harvested naked barley spikes separated into hull and kernel fractions revealed that 11.92–70.02% of the total toxins were localized in the hull. Single-SNP and haplotype-based genome-wide association studies (GWAS) were carried out using mixed linear models. Based on the single-SNP GWAS, 132 marker-trait associations, localized into 13 quantitative trait loci, were identified for all the traits except FHB severity. Haplotype-based GWAS found two haplotype-trait associations each for DON in hull and plant height, three for DON in rachis, and five for heading date. Notably, markers and haplotypes associated with later heading were also linked to higher levels of the toxin in the hull but not the kernel. Moderate-to-high predictive abilities suggest that genomic selection could be used to develop improved naked barley cultivars with a lower risk of DON contamination.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00122-025-05118-0.

Key message

Loss of hulls in a diverse naked barley panel resulted in a 32% reduction in mycotoxin contamination. The panel’s genetic variation can be exploited to develop improved cultivars through breeding.

Introduction

Barley (Hordeum vulgare L.) is considered one of the most important global staple crops. It had a world production volume of 155 million metric tons in 2022, ranking fourth after maize, wheat, and rice (FAOSTAT 2024). According to a report published by the Agricultural Marketing Resource Center in 2022, over 70% of barley grown in the USA, particularly the covered varieties, is used for malt production, while the rest is used for animal feed and food for human consumption. A variation of cultivated barley referred to as naked barley evolved about 8,000 years ago, much later than covered barley (Helback 1959), and has since been used for culinary purposes due to its dietary fiber content and ease of processing. One major unique feature of naked barley is that the hulls comprising the outer lemma and inner palea thresh freely from the caryopsis at harvest. The loss of hulls in this type of barley resulted from a spontaneous mutation event that occurred at the nud locus on the long arm of chromosome 7H (Franckowiack and Koshi 1997). A positional cloning study showed that the nud locus contains a gene encoding an ethylene response factor family transcription factor responsible for plant growth and development as well as stress responses (Nakano et al. 2006; Taketa et al. 2008).

Naked barley is characterized by lower yields than covered barley (Choo et al. 2001; Grewal et al. 2008; Griffey et al. 2010; Khanal et al. 2021; Meints & Hayes 2019; Thomason et al. 2009), and it is traditionally not preferred for malt production because the hulls in covered barley act as a filter in the lautering stage during the brewing process (Meints and Hayes 2019). As a result, it is underutilized in many parts of the world. However, it has the potential to be exploited in breeding for multiple end-uses. For instance, naked barley contains higher amounts of protein, total beta-glucan, extractable beta-glucan, starch, and fat compared to covered barley, which makes it appealing for human food (Oscarsson et al. 1996; Griffey et al. 2010; Desta et al. 2024). Also, higher levels of malt extract, the lack of hulls, which contain tannins and other polyphenols in naked barley, and advances in brewing technology can result in the production of higher quality malt (Edney et al. 2000; Agu et al. 2009). The absence of tough hulls in the naked barley used for animal feeds increases digestibility, especially for monogastric animals such as pigs and poultry (Bhatty 1986). In addition, naked barley may offer a possibility to manage mycotoxin contamination that results from Fusarium head blight (FHB) disease.

Fusarium head blight, also known as head scab, is an economically devastating disease affecting the production of small grain cereals, including barley. The disease occurs across Asia, Australia, South America, and North America, with a production loss of $220 million in the Midwest region of the USA between 1998 and 2000 (Nganje et al. 2004). The reduction in yield, end-use quality, and grade as a result of FHB infestation varies depending on the inoculum source, susceptibility of the cultivar planted, and market class (Schöneberg et al. 2018). Deoxynivalenol (DON), one of the most common trichothecene mycotoxins produced in grains during the infection process, is toxic to humans and animals (Desjardins and Proctor 2007). Moreover, grains with detectable levels of DON are more likely to be discounted at the point of sale or rejected entirely (Choo 2006). In wheat, advisory levels set by the US Food and drug administration (2010) are as follows: 1 ppm of DON in finished products for human consumption, 10 ppm in grains and grain by-products, 30 ppm in distillers’ grains, brewers’ grains, and gluten feeds and meals derived from grains, and 5 ppm in grains and grain by-products used for animal feed. The disease is caused by several Fusarium spp., with F. graminearum being the most common and virulent in the USA (Salas et al. 1999; McMullen et al. 2012).

Approaches to managing FHB have focused on reducing the concentration of DON in the harvested grain. This is especially important for organic producers who do not have access to synthetic fungicides to help control FHB or DON (Wegulo et al. 2015). Previous studies have shown that a considerable proportion of DON accumulates in the hull of naked barley. Buerstmayr et al. (2004) artificially inoculated 143 spring barley lines with a virulent Fusarium isolate. They compared the responses of naked and covered genotypes to FHB and found that a naked barley genotype named “HOR 1867” was moderately resistant to FHB and had the lowest DON concentration in the kernel compared to covered barley genotypes. A similar study comparing the mean DON accumulation in 29 pairs of covered and naked barley genotypes with similar Area Under Disease Progress Curve scores found a 16% reduction of DON in naked barley compared to covered barley (Legzdina and Buerstmayr 2004). These studies suggest that the naked phenotype is responsible for lowering DON concentration in the grain; however, a genetic study of the distribution of DON in the hull and kernel using naked barley germplasm is needed to confirm the role of the hull in toxin accumulation and determine whether there is genetic variation for these traits.

The genetics of FHB resistance are very complex. A wide variety of studies have shown moderate to high correlations between FHB severity and DON concentration in infected grains (Ma et al. 2000; Urrea et al. 2002; Mesfin et al. 2003; Huang et al. 2018; Jayathissa et al. 2024). On the other hand, FHB severity and DON accumulation are often negatively correlated with days to heading (Mesfin et al. 2003) and height (Buerstmayr et al. 2004; Huang et al. 2018). This implies that late heading and taller varieties may escape disease infection. FHB resistance is controlled by multiple small-effect quantitative trait loci (QTLs), and its expression can be modified by genotype-by-environment interactions and correlated traits (Stack 2003; Sallam et al. 2024). Also, many studies have reported the co-localization of QTLs for height, heading date, FHB resistance, and lower DON accumulation (de la Pena et al. 1999; Horsley et al. 2006; Massman et al. 2011; Mesfin et al. 2003; Ogrodowicz et al. 2020). The identification of true FHB-resistant and lower DON accumulation QTLs is often confounded because of the co-segregation of these correlated traits (Sallam et al. 2024). Unfavorable correlations between agronomic traits and DON concentration have slowed progress in improving resistance to FHB in barley.

Given the substantial challenges to improving FHB resistance in barley and the relatively unexplored potential of naked barley for various end-uses, we explored the potential of breeding naked barley to manage DON contamination. We evaluated a subset of the organic naked barley diversity (DIV) panel, phenotyped in FHB nurseries at two locations over 2 years, and also genotyped it using the Infinium 50K Barley SNP array. We conducted single-SNP and haplotype-based GWAS to identify markers and/or haplotype blocks in the genome that define the genetic architecture of morphological, disease, and mycotoxin traits.

Materials and methods

Plant materials

The full organic naked barley diversity (DIV) panel consists of 350 lines that were sourced by the Oregon State University (OSU) barley breeding program from Africa, Asia, Europe, the USA, and Canada (Kunze et al. 2024; Massman et al. 2023). The panel is comprised of lines varying for spike morphology (two-rowed and six-rowed) and vernalization requirements (winter, spring, and facultative) and also contains a mixture of landraces, experimental germplasm, doubled haploids, and released cultivars. A subset of the panel that does not require vernalization was used in the study. This facultative subset (F_DIV) had 244 barley lines that included 117 two-rowed and 127 six-rowed (Table S1).

Field experiment

The F_DIV panel lines were planted in two irrigated FHB nursery experimental fields: the University of Minnesota Northwest Research and Outreach Center at Crookston, MN (47.7746° N, 96.6094° W) and the University of Minnesota Agricultural Experiment Station, Saint Paul, MN (44.9537° N, 93.0900° W), during the spring seasons in 2020 and 2021. The trials were planted using an augmented design with incomplete blocks ranging from two to eight blocks. Six checks, including ‘CDC Clear’, ‘Quest’ (Smith et al. 2013), ‘Stander’ (Rasmusson et al. 1993), ‘Lacey’ (Rasmusson et al. 2001), ‘Robust’ (Rasmusson & Wilcoxson 1983), and ‘Full Pint’ were randomly assigned to each block in 2020 trials, while four checks, (excluding ‘Robust’ and ‘Full Pint’) were used in 2021 trials. In the same trials, a few covered barley lines were included to complete the experimental layout and maintain uniformity. However, only naked barley lines were used in the second stage of the phenotypic data analysis.

The seeds of test lines and checks were planted in April in single-row plots of length 1.3 m spaced 30 cm apart at the two locations. The two FHB nurseries were artificially inoculated and mist-irrigated as follows. In St. Paul, plants were sprayed after full spike emergence with a suspension consisting of 1 L of inoculum from 30 Fusarium graminearum isolates mixed with 7 L of distilled water and 20 ml of Tween® detergent at a concentration of 100,000 macroconidia per ml using a CO₂ pressurized backpack sprayer. Plots were inoculated when at least 90% of the heads had emerged from the boot and sprayed again 3 days later (Mesfin et al. 2003). Mist irrigation was applied immediately after inoculation for 30 min and then at regular intervals from late afternoon until early morning. The irrigation provided a conducive environment for infection and disease development, and continued until the hard dough stage of kernel development. In Crookston, plants were inoculated using a grain-spawn technique. Autoclaved maize and barley seeds colonized with five isolates of F. graminearum were scattered over the barley plots at a rate of 50 g of colonized grain m⁻² for 4 consecutive weeks, starting from 1 week before the anthesis of the earliest maturing line (Horsley et al. 2006). Overhead mist irrigation started 2 weeks before anthesis and continued until the hard dough stage of maturity. Harvesting was done by hand-sickling in July and August for Saint Paul and Crookston, respectively. All field trial management was carried out following barley technical recommendations and standard agricultural practices.

Phenotyping

Plants were evaluated for heading date, height, and FHB severity at the two nurseries for 2 years. The heading date is the number of days between the planting date and the date when at least 50% of the plant heads emerge from the boot. Plant height, measured in centimeters, was taken using a ruler placed on the soil surface to the tip of the spike, excluding awns, by randomly selecting plants from the middle of the plot. FHB severity was assessed in St. Paul in 2020 and 2021 and in Crookston in 2020. For the FHB severity assessments, infected kernels were identified by their discoloration, which was either chlorotic or necrotic. Each infected kernel on a two-rowed head and every three infected kernels on a six-rowed head were counted as 5% infection. Ten heads from each plot were scored on a scale of 0, 5, 10, 15, 25, 50, 75, and 100% according to Capettini et al. (2003), and then the values were averaged as the FHB severity.

Spikes harvested in paper bags were placed in a drying oven for 3 days at 95°F. Once dry, 20 spikes were de-stemmed and de-awned with scissors, then gently threshed using a belt thresher with a mesh bag attached to capture the lighter material. The heavy and light fractions of the threshed samples were each separated into rachis, hull, and kernel using a Carter-Day Fractioning Aspirator. The samples of each tissue type were then sorted by hand to place all material in the correct labeled envelope. The mass of each sample was measured in grams before grinding with a Retsch ZM 200 grinder at 10,000 rpm. The concentration of DON in the kernel, rachis, and hull was determined on the ground samples by gas chromatography–mass spectrometry and expressed in ppm according to the procedures of Mirocha et al. (1998) and Fuentes et al. (2005).

The traits—hull mass proportion, hull DON proportion, and hull-kernel DON ratio were estimated as follows:

$$\text{Hull mass proportion}=\frac{\text{Mass of hulls}}{\text{Mass of hulls}+\text{Mass of kernels}}$$

$$\text{Hull DON proportion}=\frac{\text{Mass of hulls}\times \text{Concentration of DON in hulls}}{\text{(Mass of hulls}\times \text{Concentration of DON in hulls)}+\text{(Mass of kernels}\times \text{Concentration of DON in kernels)}}$$

$$\text{Hull}-\text{Kernel DON ratio}=\frac{\text{Concentration of DON in hulls}}{\text{Concentration of DON in kernels}}$$

Phenotypic data analysis

The observed trait distribution for each environment was visualized using the ggplot2 package (Wickham 2016) in R software version 4.1.1 (R Development Core Team 2021) before formal analysis. Phenotypic data were subjected to a two-stage analysis using the asreml R package (Gilmour et al. 1995; Möhring and Piepho 2009). Outliers, defined as observations with a standardized residual absolute value greater than three, were removed. The first stage accounted for the trial design using a linear mixed model to obtain the best linear unbiased estimates (BLUEs) for each line in each environment. Generally, lines and blocks were considered fixed and random, respectively. For a few traits, residuals were spatially correlated in row and column directions if the Akaike information criterion (AIC) value was lower than the value obtained from the base model. The base model for the incomplete block design is as follows:

$$y_{\mathit{ij}}=\mu +\beta +G_I+{\epsilon }_{\mathit{ij}}$$

where y_ij is the vector of phenotype data, μ is the grand mean, β is the block effect, G_i is the line effect, and ε_ij is the residual term.

For the second stage analysis, the response variable was the BLUEs obtained in stage 1. Covered checks and other lines not included in the F_DIV were excluded from the analysis at this stage. The mixed model fitted to account for environmental variation is written below:

$$y_{\mathit{ij}}=\mu +G_i+E_i+{\text{GE}}_{\mathit{ij}}+{\epsilon }_{\mathit{ij}}$$

where y_ij is the BLUE of the ith line in the jth environment, μ is the overall mean, G_i is the fixed or random effect of the ith line, E_j is the fixed effect of the jth environment, GE_ij is the random effect of the ith line in the jth environment with GE ~ N (0, σ²_GE) where σ²_GE is the variance of the line by environment effect, and ε_ij is the residual error.

Broad-sense heritability (H²) was estimated as:

$$H^2={\sigma }_g^2/\left({\sigma }_g^2+{\sigma }_{ge/e}^2+{\sigma }_{e/er}^2\right)$$

where σ²g is the genotypic variance, σ²g $e$ is the genotype x environment variance, σ² $e$ is the residual variance, e is the number of environments, which is 4 (calculated as 2 locations multiplied by 2 years), and r is the number of replications across environments. The variances were extracted when line and line x environment were fitted as random effects in the second stage analysis.

To assess the relationship among traits, pairwise correlation analysis was carried out using the cor function of the stats R package. The significance test of the correlation values at p < 0.05 and visualization of the correlation matrix were done using the corrplot R package. The predictions when lines were fixed were used for the correlation analysis.

Genotyping and quality control

Leaf tissues of the lines in the panel were collected from fresh young plants and were genotyped with the Infinium 50K Barley SNP array (Massman et al. 2023). Out of the 44,040 SNP markers, 962 (~ 2%) markers were removed because they did not have physical positions on the Morex V3 reference genome. After removing monomorphic markers (~ 7%) and filtering markers having a minor allele frequency of less than 0.05, greater than 20% missing markers and individuals, and observed heterozygosity greater than 0.35, a total of 34,137 polymorphic markers were retained. Then, the missing marker data were imputed using each marker’s mean. To reduce redundancy in the set of markers used for subsequent analyses, they were further pruned to 15,042 at an r² > 0.9, a window size of 50, and an overlap of 10 between consecutive windows. Filtering, imputation, and pruning were done using the ASR genomics R package.

Linkage disequilibrium and population structure

Linkage disequilibrium (LD) analysis was performed by computing squared allele frequency correlation (r²) for all pairs of markers at a sliding window size of 50 in TASSEL (Trait Analysis by Association, Evolution and Linkage) v5.2.93 (Bradbury et al. 2007). Marker positions were then used to investigate LD decay across the entire genome. The LD decay distance was determined by fitting a non-linear regression model developed by Hill & Weir (1988) and modified by Remington et al. (2001). The half-decay distance was estimated based on the physical distance at which the LD value between two loci drops to half of its maximum value.

The presence of sub-groups within a population and unequal distribution of alleles between these groups can result in spurious associations (Flint-Garcia et al. 2005). Hence, it is important to account for population structure in GWAS analysis if present. To describe the diversity between pre-defined groups, discriminant analysis of principal components (DAPC) was done as implemented in the adegenet R package version 1.4 (Jombart and Ahmed 2011). Before that, the optimal number of genetic clusters that served as pre-defined groups in the DAPC was determined using the average silhouette width values based on K-Means clustering. This was done using the fviz_nbclust function in the factoextra R package. To capture a substantial portion of the group separation in the panel subset, the maximum number of linear discriminants (number of clusters—1) was retained.

Haplotype block construction

Markers obtained after quality control were grouped by chromosome before constructing haplotypes needed for the haplotype-based mapping approach. Haplotype blocks were identified separately for each chromosome using the default parameters, which are based on the confidence interval algorithm developed by Gabriel et al. (2002) in Haploview version 4.2 (Barrett et al. 2005). Briefly, D¹ (a normalized measure of allelic association) between markers at 95% confidence bounds was estimated, and each comparison was categorized as "strong LD", "inconclusive", or "strong recombination". A block is created if 95% of informative comparisons are "strong LD". The pairs of markers were considered in strong LD if the upper boundary confidence interval of D¹ was ≥ 0.98 and the lower boundary was ≥ 0.7. All rare haplotypes (frequency ≤ 5%) were removed before converting each haplotype block into numbered alleles. The conversion was done using an R script (R Development Core Team 2021) where alleles for each haplotype block were numbered in order of increasing frequency (1, 2, 3, …, n).

Single-SNP and haplotype-based GWAS

Genotype–phenotype associations were analyzed for the ten traits using single-SNP and haplotype-based GWAS. The first approach was based on associations between each trait's BLUEs and individual SNP markers. The analysis was performed using a mixed linear model (MLM) that accounts for population structure and kinship implemented in the GWASpoly R package (Rosyara et al. 2016). The mathematical expression of the model is:

$$y=\alpha s+Q\delta +Za+e$$

where y is a vector of predictions obtained from the second stage analysis when lines were fixed; α is the vector of slope associated with the marker matrix s; δ is the vector of fixed population effect (Q), which is a matrix of q vectors describing population structure; a is the vector of random additive effects from incidence matrix Z with a ~ N (0, Kσ²_A); and e is the vector of random residual effects with e ~ N (0, Iσ²_e). K is the realized additive relationship matrix derived from the markers.

The Q matrix used in the model is the membership probabilities obtained from the discriminant analysis of principal components. Manhattan and Q-Q plots were generated using the ggplot2 R package (Wickham 2016). Considering a p-value threshold of 0.05, a Bonferroni-corrected − log₁₀(p) of 5.19 was used as the threshold to declare a significant marker-trait association in the GWASpoly R package (Rosyara et al. 2016). This multiple-testing method is less stringent because it uses an effective number of markers estimated based on the LD between markers (Moskvina and Schmidt 2008). Potential QTL regions associated with traits of interest were identified from markers above the Bonferroni-corrected threshold. These regions were defined by using the local LD half-decay physical distance as a benchmark, extending in both directions from the position of the most significant marker. This marker was referred to as the tag-SNP because it represented each QTL region. A region on a chromosome with only one significant marker was also considered a QTL. A linear regression model was fitted to estimate the phenotypic variance explained by each tag-SNP.

Haplotype-based GWAS was performed in TASSEL version 3 (Bradbury et al. 2007) using the compressed mixed linear model with population structure (Q) and kinship (K) as fixed and random variates, respectively. A Q matrix was constructed from a principal component analysis using centered and scaled marker scores. The number of principal components included in the model was determined by using a scree plot. The kinship matrix was estimated by using VanRaden’s approach (VanRaden 2008). The Bonferroni value was calculated as 0.05/number of haplotype blocks, and only haplotypes above this threshold were declared significant. Manhattan and Q-Q plots were also plotted as described above. The phenotypic variance explained by the most significant haplotype blocks was estimated as previously described.

Genomic‑enabled prediction models

Genomic predictions were performed using the RR-BLUP model as implemented in the rrBLUP R package, utilizing the same set of GWAS markers (Endelman 2011; R Development Core Team 2021). A five-fold cross-validation (CV) procedure replicated ten times was used to assess the model's predictive ability. In each replicate of a five-fold CV, individuals were randomly divided into five folds; in this case, the number of lines in the subset panel was not divisible by five, so one-fold had 48 lines and the others 49. Predictive ability was assessed using Pearson's correlation between the genomic estimated breeding values and the BLUEs for each line. This was chosen over predictive accuracy as a criterion to compare models, since there are possibilities of estimating and interpreting heritability poorly.

Results

Phenotypic variation and broad-sense heritability estimates

We observed substantial variation among the lines in the F_DIV panel for all the traits, of which half followed a normal distribution (Fig. 1). Traits such as DON in hull, DON in kernel, DON in rachis, hull-kernel DON ratio, and hull mass had noticeably higher mean values compared to their medians. With most observations on the left side of the histogram, the central tendency measures indicated that the BLUEs are slightly or moderately skewed to the right. The heritability estimates and the proportion of variance components associated with the traits are shown in Fig. 1 and Fig. S1, respectively. The heritability estimates ranged from 0 (hull mass prop; trait excluded from downstream analyses) to 0.92 (height). Height and heading date exhibited high heritabilities as expected, while derived traits like the proportion of DON in the hull and the ratio of the hull to kernel for DON were low. Low heritabilities were primarily due to higher residual variance. The trait, hull mass proportion, with a heritability estimate of 0, was excluded from further analyses.

Fig. 1.

Distribution of BLUEs for the ten traits in the F_DIV panel comprising 244 lines. The values in the red box are the broad-sense heritability estimates. DON in hull, DON in Kernel, and DON in rachis were measured in parts per million (ppm), FHB severity in percentage (%), heading date in days after planting, height in centimeters (cm), hull mass, and kernel mass in grams (g) (color figure online)

Distribution of DON levels in the kernel, hull, and rachis

The total amount of DON in only hulls as well as intact grains (hulls and kernel) was significantly higher in six-rowed lines (56.09 µg, 152.17 µg) compared to two-rowed lines (27.57 µg, 86.27 µg), representing a 103% and 76% increase, respectively (Fig. 2). The mean toxin retained in the hulls after threshing ranged from approximately 31% (two-rowed) to 33% (six-rowed). Overall, the hulls contained an average of 32% of the total amount of DON in the intact grains using the BLUEs obtained across environments for all the lines (Fig. 2). Moreover, 63% and 97% of the lines in the F_DIV panel had higher concentrations of DON in the rachis than in the hull and kernel, respectively.

Fig. 2.

Mean total DON in the hulls and intact grains in the 117 two-rowed lines, 127 six-rowed lines, and 244 lines in the F_DIV panel across the combined environments. This was calculated by multiplying the DON concentration (ppm) in the tissue by the mass (g) of the tissue from a sample of 20 spikes. The error bars indicate significant differences in mean total DON between hulls and intact grains in two-rowed and six-rowed lines. The standard error was calculated as the standard deviation divided by the square root of the sample size

Phenotypic correlation among traits

Pairwise phenotypic correlations among the ten traits across the combined environments are shown in Fig. 3. The magnitude of correlation coefficient values ranged from − 0.42 for DON in kernel and hull-kernel DON ratio to 0.80 for DON in hull and DON in rachis. Low positive relationships were observed between FHB severity and DON in the hull, rachis, and kernel. Heading date was positively correlated with DON in hull but not DON in the kernel or rachis. Also, we observed a negative correlation between height and DON in the kernel. Kernel mass, a yield-related trait measured in this study, was found to be positively correlated with heading date and height. This observation and the negative correlation between heading date and DON in the kernel indicate that higher yields and lower DON accumulation in the kernel are associated with late heading and taller plants.

Fig. 3.

Pearson’s correlation coefficient values among the BLUEs of the ten traits across combined environments (color figure online)

Distribution of SNP markers, LD decay, and the genetic structure of the F_DIV panel

After performing quality control on the marker data, a subset of 15,042 (~ 34%) high-quality SNP markers was retained and distributed across seven chromosomes. Across chromosomes, there was an average of 2149 markers, with the least (1540) on chromosome 4H and the most (2838) on chromosome 5H (Fig. S2a). These markers have a density of 1 SNP per 339 Kb. On average, minor allele frequency was 27.8%, indicating the enrichment of common alleles in the panel. The percentages of heterozygosity and missing markers were 1% and 1.2%, respectively, across all the chromosomes. Of all the pairwise LD comparisons, only 76% were significant at p-value < 0.05, with an average r² value of 0.18. The average half-decay distance, which indicates the distance at which LD drops by half (r² = 0.23), was approximately 730 Kb within the panel subset (Fig. S2b).

The population structure analysis using the silhouette clustering method revealed the presence of four subpopulations in the panel subset (Fig. S3a). The first two principal components (PCs) explained 19.5% of the total phenotypic variation, while PCs 3–10 explained about 25% (Fig. S3b, c). The first and second linear discriminants retained in DAPC captured 48% and 34%, respectively, of the total variation, indicating that they effectively capture the majority of the differences between the subpopulations (Fig. S3d). In subpopulation one, there were 45 lines; in subpopulation two, 65 lines; in subpopulation three, 82 lines; and 52 lines in subpopulation four. About 93% of the lines in subpopulation one were collected from the Germplasm resources information network (GRIN) and the OSU colored barley breeding experiment. Subpopulation two included lines from Washington State University (~ 54%) and double-haploids from OSU (~ 23%). Subpopulation three primarily consisted of doubled-haploid lines from OSU (~ 87%), while subpopulation four was a mix of collections from North America (GRIN and OSU), Europe, and Asia.

Single-SNP GWAS

We identified a total of 132 marker-trait associations (MTAs) for all the traits except FHB severity in the combined environment analysis (Fig. 4, Table S2). These MTAs involved 83 markers, which were distributed across chromosomes 2H, 3H, 4H, 5H, and 7H, with about 72% of the unique markers localized on chromosome 2H. As shown by the Q-Q plots, the deviations in the distribution of observed p-values from the diagonal line are also indicative of true associations for traits with signals (Fig. 4). The MTAs were grouped into 13 QTLs based on the confidence interval defined as the local LD half-decay distance from the marker with the strongest association (Table 1). One QTL each was uncovered for DON in hull, DON in kernel, heading date, hull DON proportion, hull-kernel DON ratio, kernel mass; two each for DON in rachis and hull mass; and three for height. The R² explained by the tag-SNPs ranged from 1.11% (hull mass) to 27.97% (DON in rachis) (Table 1). There were overlaps of QTL regions for DON in hull, DON in rachis, heading date, height, hull DON proportion, and hull-kernel DON ratio on chromosome 2H, and also for height and kernel mass on chromosome 3H.

Fig. 4.

Single-SNP GWAS Manhattan plots for ten traits in the F_DIV panel evaluated in four environments. The inset on the Manhattan plot is the Q-Q plot with expected p-values on the x-axis and observed p-values on the y-axis. The red-dotted line represents the GWAS Bonferroni-corrected threshold of 5.19 (p < 0.05). The red solid circles represent markers above the threshold. Other shapes in black color on the Manhattan plots for DON in hull, DON in rachis, heading date, and height represent markers in the haplotype blocks declared significant from the haplotype-based GWAS (Bonferroni threshold = 4.76, p < 0.05)

Table 1.

QTL regions identified from the marker-trait associations and the summary statistics of the Tag-SNP

			Tag SNPb
Traits	QTL region (Mb)a	Chrom	Name	Major allele	Minor allele	MAF	− log10(p)c	Effectd	R2 (%)e	Other markers in the QTL region
DON in hull	26.15–35.33	2H	JHI-Hv50k-2016–75226	B	A	0.17	7.13	− 6.02	11.50	JHI-Hv50k-2016–73689
DON in Kernel	NA	3H	BOPA2_12_30924	A	B	0.27	5.24	1.87	14.96	NA
DON in rachis	27.75–28.45	2H	SCRI_RS_173164	B	A	0.15	7.82	− 9.97	27.97	JHI-Hv50k-2016–74351
	NA	5H	JHI-Hv50k-2016–310062	B	A	0.20	6.38	− 8.31	4.45	NA
Heading date	24.31–27.99	2H	JHI-Hv50k-2016–73689	A	B	0.39	19.58	− 3.81	21.19	JHI-Hv50k-2016–72946, JHI-Hv50k-2016–73122, JHI-Hv50k-2016–73545, JHI-Hv50k-2016–73501, JHI-Hv50k-2016–73487, BK_15, JHI-Hv50k-2016–73422, JHI-Hv50k-2016–73615, JHI-Hv50k-2016–73617, JHI-Hv50k-2016–73673, JHI-Hv50k-2016–73689, JHI-Hv50k-2016–73712, JHI-Hv50k-2016–73697, JHI-Hv50k-2016–73723, JHI-Hv50k-2016–73930, SCRI_RS_143250, SCRI_RS_186387, JHI-Hv50k-2016–74094, JHI-Hv50k-2016–74101, JHI-Hv50k-2016–74210
Height	24.67–27.63	2H	JHI-Hv50k-2016–73689	A	B	0.39	7.91	− 4.59	2.33	JHI-Hv50k-2016–73487, JHI-Hv50k-2016–73422, JHI-Hv50k-2016–73617, JHI-Hv50k-2016–73712, JHI-Hv50k-2016–73930, SCRI_RS_143250
	565.17–572.99	3H	JHI-Hv50k-2016–206416	A	B	0.14	6.68	− 5.90	23.54	BOPA2_12_10188, BOPA2_12_10205
	NA	4H	JHI-Hv50k-2016–230344	B	A	0.20	5.75	− 6.76	5.85	NA
Hull DON proportion	14.30–38.00	2H	JHI-Hv50k-2016–73689	A	B	0.39	11.55	− 0.07	15.30	JHI-Hv50k-2016–73501, JHI-Hv50k-2016–73617, JHI-Hv50k-2016–73673, JHI-Hv50k-2016–73723, SCRI_RS_186387, JHI-Hv50k-2016–74284, BOPA1_2646-1277, SCRI_RS_239882
Hull mass	NA	2H	JHI-Hv50k-2016–73689	A	B	0.39	5.76	− 0.28	1.11	NA
	12.81–23.45	7H	JHI-Hv50k-2016–451563	B	A	0.24	6.55	− 0.39	16.78	SCRI_RS_126944, SCRI_RS_222330, JHI-Hv50k-2016–451807
Hull-kernel DON ratio	22.05–30.69	2H	JHI-Hv50k-2016–73723	A	B	0.32	8.28	− 0.84	15.40	JHI-Hv50k-2016–72435, BOPA2_12_10777, JHI-Hv50k-2016–73617, JHI-Hv50k-2016–73673, JHI-Hv50k-2016–73689, SCRI_RS_143250, SCRI_RS_186387, JHI-Hv50k-2016–74094, JHI-Hv50k-2016–74101, JHI-Hv50k-2016–74210, JHI-Hv50k-2016–74284, BOPA1_2646-1277, SCRI_RS_239882
Kernel mass	566.81–571.35	3H	JHI-Hv50k-2016–206416	A	B	0.14	9.04	− 2.40	4.15	JHI-Hv50k-2016–205861, BOPA2_12_10188, JHI-Hv50k-2016–206312, SCRI_RS_146012, JHI-Hv50k-2016–206631

Chrom = Chromosome number, MAF = Minor Allele Frequency, p = marker-trait association p-value

^aQTL region was determined using the local half LD physical distance with the position of the Tag-SNP as the benchmark, ^bTag-SNP is the SNP with the strongest association in the defined QTL region, ^C-log₁₀(marker-trait association p-value), ^dEffect size of Tag-SNP, ^ePhenotypic variation explained by the Tag-SNP

Thirty-two MTAs defined by 29 markers were detected in both Crookston and Saint Paul on chromosomes 2H (27) and 3H (2) (Table S3). Of these stable SNP markers, 26 were associated with a single trait and three with two traits. The highest number of stable MTAs was detected for heading date (27), followed by height (2), hull DON proportion, hull-kernel DON ratio, and kernel mass (1). Twenty-seven markers on chromosomes 2H (24) and 3H (3) were associated with two or more traits (Fig. S4). For example, the minor allele at marker JHI-Hv50k-2016–73689 on chromosome 2H was associated with a decrease in hull DON proportion, hull mass, hull-kernel DON ratio, heading date, DON in hull, height, and DON in rachis by 0.07, 0.28 g, 0.66, 3.81 days, 4.54 ppm, 4.59 cm, and 6.19 ppm units, respectively. Similarly, on chromosome 3H, the major allele at marker JHI-Hv50k-2016–201886 was associated with an increase in height (3.72 cm) and kernel mass (1.30 g).

Haplotype-based GWAS

Given that more than two alleles may be present at a locus in the diversity panel, a haplotype-based GWAS approach, which can capture joint effects of multiple SNPs, was also explored in this study. From the haplotype construction analysis, a total of 2869 haplotype blocks were established across the genome. The haplotype blocks covered a total genome length of 4194 Mb, with 516, 562, 588, 610, 621, 632, and 665 Mb covered by chromosomes 1H, 6H, 5H, 4H, 3H, 7H, and 2H, respectively (Table 2). The number of blocks distributed on each chromosome is proportional to the density of the markers, with the highest number of blocks obtained on chromosome 5H (548) and the lowest on chromosome 4H (297). Overall, the number of SNPs per haplotype block ranged from 2 to 14, with a total number of approximately 10,000 alleles across chromosomes (Table 2). There were two to eight alleles per haplotype block. Four haplotype blocks, three on chromosome 1H and one on chromosome 5H, each with only one allele, were excluded from the analysis.

Table 2.

Summary of 2,865 haplotype blocks (HB) constructed using 15,042 SNP markers for 244 lines in the panel subset representing 9,838 alleles

Chrom	Number of markers	Physical position (bp)- first and last marker across chrom*	Number of HB	Number of markers per HB	Number of alleles
1H	1721	49,952–516,332,560	325	2–12	1097
2H	2611	58,711–665,302,457	490	2–13	1751
3H	2336	114,611–620,927,757	441	2–13	1487
4H	1540	195,516–609,903,777	297	2–11	1012
5H	2838	250,030–587,779,056	547	2–14	1920
6H	1972	77,448–561,748,098	392	2–10	1306
7H	2024	121,803–632,166,840	373	2–9	1265

*Physical positions of the first and the last SNP on an individual chromosome

Chrom = Chromosome number

In the combined environments, GWAS based on haplotypes revealed 12 significant haplotype-trait associations (HTAs) on chromosomes 2H (11) and 4H (1) for DON in hull (2), height (2), DON in rachis (3), and heading date (5) (Fig. S5). Table S4 shows the physical distance covered by each haplotype block, the specific haplotypes observed within each block and their effects, as well as the percentage phenotypic variance of each block. Across the haplotype blocks, there were at least two markers with a coverage of 0.17–361 Kb. The haplotype “BBBBAAAABB” at chr2_hap92 had the lowest mean for DON in hull, DON in rachis, and heading date, and was significantly different (p < 0.05) from others except for BBAAAABBAA in DON in hull and DON in rachis (Fig. 5). This suggests that this haplotype may be beneficial in improving naked barley for these traits. The phenotypic variation explained by the haplotype blocks varied between 8.17 and 10.53% for DON in hull, 4.42 and 11.30% for DON in rachis, 6.44 and 28.57% for heading date, and 9.96 and 28.24% for height (Table S4).

Fig. 5.

The distribution of haplotypes at block chr2_hap92 associated with DON in hull, DON in rachis, and heading date

All the haplotype blocks declared significant for heading date contained at least two markers that were previously detected by single-SNP GWAS (Table S4). However, 22 MTAs uncovered for DON in hull, DON in rachis, height, hull DON proportion, hull mass, and hull-kernel DON ratio were binned into at least one of the blocks associated with the heading date (Table S4). Among them, markers JHI-Hv50k-2016–73487 and JHI-Hv50k-2016–73617 associated with DON in hull in single-SNP GWAS were binned into chr2_hap91 and chr2_hap93, respectively, both of which were associated with the heading date. Nonetheless, the haplotype blocks chr2_hap90 and chr2_hap92, which were associated with DON in the hull, were located 2.2 Kb and 5.4 Kb away from chr2_hap91 and chr2_hap93, respectively. Overall, an additional seven genomic regions were uncovered through haplotype analysis for DON in hull (2), DON in rachis (3), and height (2), but not through the single-marker analysis. In contrast, single-SNP GWAS identified one QTL region for DON in hull, two for DON in rachis, and three for height, which were not detected by haplotype-based GWAS. Two haplotype blocks (chr2_hap92 and chr2_hap93) were detected for heading date in the combined environments, Crookston and Saint Paul. The same haplotype at each of the blocks had the lowest mean heading date in the two locations, thus suggesting the stability of these associations (Table S5). The comparison of the phenotypic variation explained by the most significant haplotype block and marker revealed differences of − 16.67% for DON in rachis, − 0.97% for DON in hull, 7.38% for heading date, and 25.91% for height (Fig. 6).

Fig. 6.

The R² values of the most significant marker (yellow) and haplotype block (blue) for DON in hull, DON in rachis, heading date, and height. Only traits with significant associations for both GWAS approaches are shown (color figure online)

The RR-BLUP model was used to conduct genomic prediction for the traits, and the model’s performance was assessed using repeated five-fold cross-validation. The mean predictive ability varied from 0.38 for DON in hull to 0.72 for height (Fig. 7). Also, we compared the effectiveness of two markers identified by single-SNP GWAS (Fig. S6a, b), both present in haplotype block “chr2_hap92” along with haplotypes (Fig. S6c) and predicted values from the RR-BLUP model (Fig. S6d) for selecting early heading genotypes within the subset panel. The means of genotypes homozygous for alleles “B” and “A” at markers JHI_Hv50k_2016_73422 and BK_15, respectively, were lowest for heading date and were significantly different from those carrying the two copies of the other allele. About 12% of the subset panel had a haplotype combination of BBBBAAAABB at chr2_hap92, with their heading date ranging from 40 to 51 days, and their mean heading date was significantly different from the values of other groups. Interestingly, all the genotypes with haplotype BBBBAAAABB were homozygous for the desirable alleles at markers JHI_Hv50k_2016_73422 and BK_15. However, selecting solely based on these loci would include a few genotypes with later heading dates. The mean heading date of the genotypes with the desirable haplotype (44.40 days ± 2.75) and the mean predictions for the same genotypes using genome-wide markers (47.84 days ± 2.89) were quite similar; however, the latter has a higher standard deviation. This suggests that the haplotype block may be in strong LD with the causal variant.

Fig. 7.

Predictive abilities for ten traits evaluated in the F_DIV panel using the RR-BLUP model

Discussion

Utilizing naked barley to manage deoxynivalenol contamination

Naked barley has considerable promise to deliver benefits to farmers and end-users for food, feed, and malting (Meints and Hayes 2019). A major obstacle to realizing these benefits is the risk associated with the accumulation of the mycotoxin DON in the harvested grain. Fusarium Head Blight is a very challenging disease to manage, particularly in organic systems, because of restrictions on the use of some fungicides that are capable of reducing DON concentration in grains by about 35–60% (Cowger et al. 2023). Previous work examining a limited number of barley genotypes showed that naked barley can have up to 59% less DON compared to covered barley, and the physical removal of the hull from covered barley through pearling can reduce DON by 86% (Clear et al. 1997). Although the latter was more effective in reducing the toxin content, it can lead to a significant loss of grain mass (House et al. 2003). Thus, using naked barley cultivars would be a more effective way to eliminate DON that has accumulated in the hull.

To confirm the importance of the hull in accumulating DON, we examined a diverse panel of 244 naked barley genotypes and directly measured the concentration of DON in the hull and naked kernel. Up to 32% of the total DON in the grain of naked barley was found in the hull, indicating that utilizing naked barley cultivars can provide substantial benefit in managing DON contamination. We also found moderate heritabilities and substantial variation for the DON in the hull and the proportion of DON in the hull versus the naked kernel. This suggests that it may be possible to breed naked barley to reduce the amount of DON that ends up in the harvested product.

Association of heading date and disease is different in naked barley compared to covered barley

Numerous genetic studies in covered barley have observed a correlation between heading date and DON concentration (Choo et al. 2004; Dahleen et al. 2003; Mamo & Steffenson 2015; Mesfin et al. 2003; Sallam et al. 2024). In most cases, the correlation is negative with later heading and taller plant height associated with lower disease severity and DON concentration. This is presumably due to delayed exposure to inoculum in late heading plants and reduced inoculum exposure in taller plants farther from the ground (Huang et al. 2018; Sallam et al. 2024). In our study with naked barley, we observed a positive correlation between heading date and DON concentration in the hull, but not with DON concentration in the kernel. Also, the heading date QTL we identified on chromosome 2H overlaps with a QTL for DON concentration in the hull, but we did not identify a QTL in that region for DON concentration in the kernel. This suggests that later infections result in greater localization of F. graminearum in the hull and that deeper penetration to the kernel is inhibited by the shorter window of infection time.

The higher concentration of DON in hull compared to the naked kernel may be attributed to differences in the trichome density, mycelial colonies, and vascular bundle colonization of the two tissues. Firstly, Imboden et al. (2017) found that trichomes, which are sites for fungal penetration and invasion, are abundant on the surface of the palea compared to the caryopsis, thus creating an opportunity for pathogen colonization in the hull. Secondly, in an experiment on the distribution of mycelia colonies on florets a few days post-infection, Lewandowski et al. (2006) found more colonies on the palea, followed by the lemma, while only a few were found on the developing kernel. Furthermore, they observed that the palea had more colonies on the exterior, particularly near the base of the furrow, than on the interior surfaces. As the mycelium grows out of the furrow and spreads laterally into the crevice between the lemma and palea, it increases colonization of the inner surfaces of the hull, which is further enhanced by the presence of trichomes. Finally, a secondary colonization of the vascular bundles in the hull by the fungus beyond the initial infection site could have contributed to the findings (Lewandowski et al. 2006; Imboden et al. 2017). Considering these insights along with Pearson’s correlation value (r = 0.68, p < 2.2 × 10⁻¹⁶) between DON in the hull and DON in the kernel observed in this study, further work should confirm the importance of mycelial load and trichome density for DON accumulation. If this is established, breeding strategies for developing genotypes with reduced mycotoxin in the kernel and hull can then be optimized by incorporating the use of high-throughput imaging for precise phenotyping of trichome density.

Influence of rachis structure on Type II resistance

More than 60% of our naked barley lines had higher DON concentrations in the rachis than in other tissues. A similar finding was also reported by Tucker et al. (2019) in covered barley grains harvested, particularly at the soft-dough stage. This observation was surprising, as barley has been reported to possess superior Type II resistance, which restricts the subsequent spread of infection (Langevin et al. 2004; Hao et al. 2020). This happens by inhibiting the entry of the fungus hyphae at the rachis node and rachilla, irrespective of whether the mycotoxin is present or absent (Jansen et al. 2005). In wheat, on the other hand, the production of trichothecene mycotoxins by F. graminearum contributes to its virulence, allowing the pathogen to spread from one infected floret to adjacent ones by hyphal growth through the rachis internodes (Desjardins and Proctor 2007). However, in the absence of trichothecene production, the fungus is prevented from moving into the rachis because of the strong cell wall developed in the rachis node (Jansen et al. 2005). Previous research on barley suggested that infection of trichomes on the rachis may provide a route for F. graminearum to spread throughout the spike, thereby overcoming Type II resistance (Jansen et al. 2005). In our study, we only measured the final concentration of DON at harvest; therefore, we do not know if toxin translocation from nearby infected spikelets occurred. Additional research is needed to understand better the role of Fusarium in the colonization and spread of the rachis, as well as the apparent differences between wheat and barley with respect to Type II resistance.

Breeding for improved naked barley using markers, haplotypes, and genomic selection

The use of haplotypes in GWAS is a complementary approach to single-SNP association mapping in many crops (Contreras-Soto et al. 2017; Maldonado et al. 2019; Helal et al. 2021). We found that most of the MTAs and HTAs detected were harbored on chromosome 2H. This further confirms the chromosome's pivotal role in contributing to the variation observed for FHB disease, DON accumulation, and agro-morphological traits (Pillen et al. 2003; Huang et al. 2018; Sallam et al. 2024). Within the QTL region (24.13–27.99 Mb) associated with heading date, three stable MTAs were identified: BK_15 and JHI-Hv50k-2016–73422, both binned into haplotype block “chr2_hap92”, and SCRI_RS_143250. BK_15 and SCRI_RS_143250 were previously reported by Huang et al. (2018), whereas JHI-Hv50k-2016–73422 by Sharma et al. (2025) for the same trait. Interestingly, the authors found that both BK_15 and JHI-Hv50k-2016–73422 are located within the Ppd-H1 gene on chromosome 2H. This gene is known to regulate barley flowering time through photoperiod response, which significantly affects heading date (Turner et al. 2005). The stable MTAs and HTAs for heading date identified across Saint Paul and Crookston can be valuable tools in naked barley breeding programs to utilize favorable alleles for early heading.

All the haplotype blocks associated with heading date contained two or more markers detected for the same trait using a single-SNP approach. This suggests that the use of haplotype blocks in GWAS has the potential to capture cis‐interactions among multiple SNP variants in a region with strong LD (Liu et al. 2008). Interestingly, individual markers within the blocks associated with other traits were not significant in the single-SNP analysis. However, they were located close to the significant SNPs identified by individual SNP tests, suggesting that haplotype blocks could have tagged a local genomic region harboring causal variants that may not be fully captured by single-SNP analysis. All the markers and haplotype blocks with the lowest p-values captured more than 10% of the genetic variation in the subset panel, except for marker JHI-Hv50k-2016–73689 associated with height. The PVEs significantly increased from 21.19% in single-SNP GWAS to 28.57% in haplotype-based GWAS for heading date and from 2.33% to 28.24% for height, but the reverse was observed for DON in hull and DON in rachis. These contrasting patterns imply that the best resolution for the GWAS approaches may be trait-dependent, thereby reflecting differences in the underlying genetic architecture. A similar finding was reported by Lu et al. (2010), where haplotype-based analysis, compared to single-SNP-based analysis, significantly improved the mapping efficiency of the anthesis-silking interval in maize, as revealed by the increase of PVE from 5.4 to 23.3%. In another study on covered spring barley, the use of haplotype information resulted in higher cumulative PVE for days to maturity and yield, while a contrasting trend was observed for heading time and specific weight (Abed and Belzile 2019). Also, a simulation study using German winter barley cultivars revealed that the haplotype-based approach had a higher likelihood of detecting association than the single-marker approach (Gawenda et al. 2015). However, they reported that a higher false discovery rate for SNP detection was observed for the haplotype-based approach because population structure was not accounted for in the model. It may be desirable to utilize both approaches in genetic mapping studies to maximize the opportunity to identify and characterize important loci controlling traits. These selection tools must be technically and biologically validated in independent populations to assess their prediction metrics before their routine use in a breeding program (Ige et al. 2021).

In addition to the two approaches used for GWAS to identify loci for marker-assisted selection, we also explored the use of genomic selection. This provides a more comprehensive approach for predicting the overall genetic merit of individuals in a population, often utilizing single-SNP and haplotype information as part of the analysis. The accuracy of the RR-BLUP models utilizing 15,042 SNPs ranged from 38% for DON in hull to 72% for height, demonstrating a moderate to high level of precision. The values showed that genomic prediction models can accurately predict line performance, even though the accuracy can be improved by leveraging on multi-trait model that incorporates phenotypic information from correlated traits (Massman et al. 2023; Rutkoski et al. 2012). The inclusion of test weight data in predicting threshability in the full Naked Barley DIV panel increased predictive ability by 8% (Massman et al. 2023). Also, several authors have reported the effectiveness of using haplotypes over single SNPs in increasing the predictive accuracy of many traits (Hess et al. 2017; Sallam et al. 2020; Difabachew et al. 2023). Thus, incorporating haplotype metrics into genomic prediction models may be considered for further study.

Conclusion

The loss of 32% of the total DON in the grain to the threshed and discarded hull makes naked barley an effective option in managing DON contamination. Given the great challenge in improving FHB resistance in barley, developing naked barley cultivars that meet the needs for food, feed, and malting could significantly reduce the risk of DON contamination. Exploiting the variation for the proportion of DON in the hull through breeding could improve the effectiveness of naked barley to complement current efforts to improve disease resistance. As breeders, farmers, and end-users consider the merits of naked barley for these various end-uses, this additional advantage should be weighed carefully, particularly in regions with a higher risk of FHB disease.

Supplementary information

Below is the link to the electronic supplementary material.

Author contribution statement

Conceptualization was done by Kevin Smith and Brigid Meints; Methodology was executed by John Hawkins, Kevin Smith, Adenike Damilola Ige, and Yanhong Dong; Formal analysis and investigation were performed by Adenike Damilola Ige; Writing—original draft preparation was done by Adenike Damilola Ige, John Hawkins, and Kevin Smith; Writing—review and editing was done by Adenike Damilola Ige, John Hawkins, Yanhong Dong, Brigid Meints, and Kevin Smith; Funding acquisition was carried out by Brigid Meints and Kevin Smith; Supervision was done by Kevin Smith.

Funding

This work was supported by funding from USDA-NIFA-OREI (Grant numbers: 2020–51300-32179 and 2023–51300-40964).

Data availability

The datasets generated and analyzed during this study, and the R script used to convert haplotype alleles to numbers, are available in the GitHub repository [https://github.com/UMN-BarleyOatSilphium/OREI-Project].

Declarations

Competing interest

The authors declare they have no financial or non-financial interests.