Genomic strategies to facilitate breeding for increased β-Glucan content in oat (Avena sativa L.)

Abstract

Background

Hexaploid oat (Avena sativa L.) is a commercially important cereal crop due to its soluble dietary fiber β-glucan, a hemicellulose known to prevent cardio-vascular diseases. To maximize health benefits associated with the consumption of oat-based food products, breeding efforts have aimed at increasing the β-glucan content in oat groats. However, progress has been limited. To accelerate oat breeding efforts, we leveraged existing breeding datasets (1,230 breeding lines from South Dakota State University oat breeding program grown in multiple environments between 2015 and 2022) to conduct a genome-wide association study (GWAS) to increase our understanding of the genetic control of beta-glucan content in oats and to compare strategies to implement genomic selection (GS) to increase genetic gain for β-glucan content in oat.

Results

Large variation for β-glucan content was observed with values ranging between 3.02 and 7.24%. An independent GWAS was performed for each breeding panel in each environment and identified 22 loci distributed over fourteen oat chromosomes significantly associated with β-glucan content. Comparison based on physical position showed that 12 out of 22 loci coincided with previously identified β-glucan QTLs, and three loci are in the vicinity of cellulose synthesis genes, Cellulose synthase-like (Csl). To perform a GWAS analysis across all breeding datasets, the β-glucan content of each breeding line was predicted for each of the 26 environments. The overall GWAS identified 73 loci, of which 15 coincided with loci identified for individual environments and 37 coincided with previously reported β-glucan QTLs not identified when performing the GWAS in single years. In addition, 21 novel loci were identified that were not reported in the previous studies. The proposed approach increased our ability to detect significantly associated markers. The comparison of multiple GS scenarios indicated that using a specific set of markers as a fixed effect in GS models did not increase the prediction accuracy. However, the use of multi-environment data in the training population resulted in an increase in prediction accuracy (0.61–0.72) as compared to single-year (0.28–0.48) data. The use of USDA-SoyWheOatBar-3 K genotyping array data resulted in a similar level of prediction accuracy as did genotyping-by-sequencing data.

Conclusion

This study identified and confirmed the location of multiple loci associated with β-glucan content. The proposed genomic strategies significantly increase both our ability to detect significant markers in GWAS and the accuracy of genomic predictions. The findings of this study can be useful to accelerate the genetic improvement of β-glucan content and other traits.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-024-11174-5.

Background

Hexaploid oat (Avena sativa L.; 2n = 6x = 42, AACCDD) is an important cereal crop for both food and animal feed, with an annual worldwide grain production of 22.6 million tons (FAOSTAT 2021) [1]. In the last twenty years, oat has gained popularity among consumers and culinary experts due to their numerous health benefits and unique nutritional profile [2]. Among the nutritious components of oat, is a unique water-soluble fiber, beta-glucan (β-glucan), which is present in the endosperm and aleurone layer of oat groats and constitutes up to 70% by weight of the endosperm cell walls [2, 3]. Barley (Hordeum vulgare L.) and oat are considered to have higher β-glucan content than wheat (Triticum aestivum L.) and rye (Secale cereale L.). The total β-glucan content by kernel weight ranges from 2.5 to 11.3% for barley, from 2.2 to 7.8% for oat, from 1.2 to 2.9% for rye, and from 0.2 to 1.4% for wheat [4–6]. Although some barley varieties have been reported with higher β-glucan content than oat [7], barley use for human consumption is limited; oat-based food products are more common and constitute the main source of β-glucan intake.

Regular consumption of oat β-glucan can lower blood cholesterol, reduce the risk of coronary heart disease [8], improve the immune response [9], reduce the risk of obesity, and prevent cancer [10, 11]. This has led various agencies around the globe to approve health claims for β-glucan. Among them, in 1997, the Food and Drug Administration (FDA) approved the claim that a daily amount of 3 g of oat β-glucan can help to reduce blood cholesterol and promote cardiovascular health [12]. Increasing the β-glucan content in oat could help consumers to achieve these dietary goals [13].

The effect of genotype, environmental conditions, and agronomic practices on β-glucan content has been reported in various studies [14–23]. Genotype plays a significant role on groat β-glucan content [16, 22, 24]. Some report minor effects of genotype × environment (GE) interactions [25], but others have indicated that GE interactions strongly influence the β-glucan content [26]. The broad-sense heritability estimates for β-glucan content are moderate to high, with values ranging from 0.27 to 0.45 [26], from 0.44 to 0.63 [27], from 0.52 to 0.73 [28], from 0.43 to 0.63 [29], and from 0.80 to 0.85 [30] depending upon the genetic architecture of populations under study and the environments in which they were evaluated. Environmental conditions such as temperature, precipitation, and fertilizers also affect β-glucan content in oat. Previous studies have found that β-glucan content was higher in warm, dry climates and lower in cool/wet environments [17, 31–33]. Furthermore, an increase in nitrogen fertilization resulted in higher levels of β-glucan [14, 17, 34].

Breeding efforts to increase β-glucan content in oat have been facilitated by the development of NIR calibrations that can quickly and accurately measure β-glucan content on whole groats [35]. This considerably increased throughput as compared to traditional methods that required measurement of ground groat flour to estimate β-glucan content. Efforts would further be facilitated by the integration of genomic strategies for germplasm improvement in this crop. QTLs for β-glucan have been identified in several studies using bi-parental populations [27, 36, 37]. GWAS has also been carried out to further elucidate the trait architecture [29, 38–41]. An association analysis using a worldwide diversity panel revealed three DArT markers significantly associated with β-glucan content, and one of the markers sequence showed homology with rice’s (Oryza sativa L.) CslF (cellulose synthase-like) gene family [38]. The detection of a low number of associations in this study might be due to insufficient marker density [38]. A GWAS analysis for β-glucan content was also conducted for 446 North American elite oat breeding lines with phenotypic data collected from historical and balanced two-year trials (Ames, IA) [29]. A total of 24 and 37 significant marker associations for β-glucan content were detected for the historical and balanced data sets, respectively [29], with most of the markers identified corresponding to genomic regions previously found in QTL analysis in bi-parental populations [26, 36]. The use of historically unbalanced data however led to higher genotype × environment interaction and resulted in multiple loci with low R² values [29].

More recently, a GWAS was performed using three panels of elite accessions (Spring, Winter, and World Diversity) that were genotyped using the Oat 6 K Custom Infinium iSelect BeadChip [42], and through a meta-analysis identified 58 markers significantly associated with β-glucan content [39]. In contrast to the above studies that used North American oat germplasm, a GWAS for β-glucan content conducted using a Federal University of Rio Grande do Sul (UFRGS) oat panel consisting of 413 genotypes evaluated under subtropical conditions led to the identification of seven quantitative trait loci (QTL) associated with β-glucan content with three of those genomic regions being conserved between oat and barley [40].

Collaborative work between Pepsico and Corteva Agriscience released the full genome sequence of hexaploid oat line OT3098 which can be accessed at https://wheat.pw.usda.gov/jb/?data=/ggds/oat-ot3098v2-pepsico [43]. The oat reference genome was used to map QTL affecting agronomic and grain quality traits such as β-glucan content using five different bi-parental recombinant inbred line populations and three QTL linked with β-glucan content were identified on chromosomes 6A (2 QTL) and 7D (1 QTL) [41]. These QTL were co-localized with the QTL found in populations ‘Kanota’ x ‘Ogle’ and Kanota x ‘Marion’ as well as in the CORE and UFRGS diversity panels and with three β-glucan candidate genes (CslF11_chr6A_298 Mb, CslF9_chr6A_416 Mb, and CslF6_chr7A_399 Mb) [41]. They also found a strong association among QTL and candidate genes affecting heading date and quality traits including β-glucan content [41]. Recently, an inventory of genes and QTL identified for various traits [44], including β-glucan content, was created by assigning genes/QTL to the recent OT3098 v2 hexaploid oat physical map found on the GrainGene database [43]. This inventory facilitates the comparison of identified QTL/genes among studies.

Previous GWAS studies have reported many genomic regions associated with β-glucan content, suggesting that marker-assisted selection may not be the most efficient approach to improve β-glucan content. On the other hand, genomic selection (GS) has shown promise for improving β-glucan content in oat [45]. GS considers the effect of both minor and major genetic factors [46, 47]. GS consists of using a statistical model to predict the performance of a specific set of lines (testing) built upon a training population for which both phenotypic and genotypic data are available [48]. Multiple strategies have been proposed to increase prediction accuracy, including using molecular markers as a fixed effect and using multi-environment data. When QTL/markers that influenced phenotypic value by 10% or more were considered as a fixed effect in GS models, prediction accuracy increased by 7.70–33.4% for flowering time, 7.0–20.5% for plant height, and 12.8–62.8% for yield in rice and by 10% for heading date, 8–14% for resistance to powdery mildew, and 9–17% for plant height in wheat depending upon the training size and environment [49, 50]. In addition, integrating multi-environment information into the GS model has been suggested to overcome the genotype × environment interaction and increase prediction accuracy in maize (Zea mays L.), wheat, and potato (Solanum tuberosum L.) [51]. These strategies can be explored in oat to improve the prediction accuracy for β-glucan content.

While previous studies have been critical in increasing our understanding of the genetic control of β-glucan content in oat, further work is needed to fully integrate those findings into breeding schemes. As part of breeding programs’ activities, valuable data is collected that can further confirm previous findings and lead to enhanced breeding strategies. In comparison to the use of diverse panels for conducting GWAS, advanced breeding material offers the advantage of directly applying findings to ongoing breeding programs. In the present study, existing breeding data were leveraged to improve our understanding of the genetic control of β-glucan content and to evaluate strategies to increase β-glucan content in oat. The utilization of an existing data set encompassing multiple years of breeding activities in multiple environments is expected to improve the power, precision, and accuracy of GWAS and enable the detection of variants with smaller effect. The objectives of the study were to (1) map the genomic regions associated with β-glucan content among experimental breeding lines of a North American oat breeding program, (2) compare the significant genomic regions associated with β-glucan content in the present study with those previously reported in the literature, (3) provide new molecular markers to be used in oat breeding programs to improve β-glucan content in oat, and (4) evaluate GS strategies to predict β-glucan content.

Materials and methods

Plant material and field experiments

The GWAS panel consisted of six sets of advanced breeding lines (~ 148–233 lines) evaluated in the preliminary yield trials (PYT) of the South Dakota State University oat breeding program from 2015 to 2022. In total, 1,230 oat breeding lines were considered in this study. Each year, experimental material was planted at three to five locations in South Dakota. The testing sites were located in Brookings (BRK) (44° 18’ N, 96°47” W), South Shore (SSH) (45° 6’ N, 96° 55’ W), Volga (VLG) (44° 19’ N, 96° 55’ W), Beresford (BRF) (43° 4’ N, 96°46’ W), Pierre (PIE) (44° 22’N, 100°21’ W), and Winner (WIN) (43° 22’ N, 99°51’ W) (Table 1). Phenotypes were collected in 26 different environments. The combination of location and year was considered as an individual environment and abbreviations were denoted in Table 1. In each environment, the experimental design followed an augmented design with at least four checks replicated nine times. The plot size was 1.52 × 1.83 m² at SSH and VLG and 1.52 × 3.65 m² at BRF, BRK, PIE, and WIN. The experiments were managed following regional standard cultural practices and recommendations.

Table 1.

List of years, locations within each year, set of lines, and SNPs used for GWAS analysis for β-glucan content

Year	Location	Environmenta	Number of lines	Number of markers
2015	BRF, SSH, VLG, WIN	BRF_15, SSH_15, VLG_15, WIN_15	221	8,299
2016	SSH, VLG, WIN	SSH_16, VLG_16, WIN_16	148	11,714
2017	BRF, SSH, VLG, WIN	BRF_17, SSH_17, VLG_17, WIN_17	225	11,356
2020	BRK, BRF, SSH, VLG, WIN	BRK_20, BRF_20, SSH_20, VLG_20, WIN_20	224	9,301
2021	BRK, BRF, SSH, VLG, PIE	BRK_21, BRF_21, SSH_21, VLG_21, PIE_21	230	8,625
2022	BRK, BRF, SSH, VLG, PIE	BRK_22, BRF_22, SSH_22, VLG_22, PIE_22	233	10,902
Across all years	-	ACRb	1,230	9,341

^aCoded as year and location combination, where BRK, Brookings; BRF, Beresford; SSH, South Shore; VLG, Volga; PIE, Pierre; WIN, Winner and 15, 16, 17, 20,21, and 22 for the years 2015, 2016, 2017, 2020, 2021, and 2022, respectively

^bLines from all the years together

Phenotypic data and statistical analysis

The GWAS panel was phenotyped for β-glucan content (as dry-weight%) using Near Infrared (NIR) spectroscopy on whole oat groats using a calibration developed at SDSU [35]. Prior to 2017, β-glucan content was measured on the oat flour instead of the whole oat groat. The phenotypic data for the GWAS panel in multi-location, multi-year trials were curated before analysis. Outliers in the data were identified based on the interquartile range method. In this method, data points having values less than Q1 – (1.5 * IQR) and more than Q3 + (1.5 * IQR) were removed from further analysis where Q1 was the first quartile, Q3 was the third quartile, and IQR was the interquartile range. Phenotypic data from the years 2018 and 2019 were excluded from the analysis due to the poor quality of genotyping data. A linear mixed model was applied to calculate the best linear unbiased estimator (BLUE) value for each line within each environment and across all the environments for a specific year using lme4 r-package [52] as per the equations below and further used in the analysis:

$${\text{Y}}_{\text{ijk}}=\mathrm{\mu }+{\text{Gen}}_{\text{i}}+{\text{Row}}_{\text{j}}+{\text{Column}}_{\text{k}}+{\text{e}}_{\text{ijk}}$$ 1

$${\text{Y}}_{\text{ijk}}=\mathrm{\mu }+{\text{Gen}}_{\text{i}}+{\text{Loc}}_{\text{j}}+{\left(\text{Gen},,\times ,,\text{Loc}\right)}_{\text{ij}}+{\text{e}}_{\text{ij}}$$ 2

where ${\text{Y}}_{\text{ijk}}$ is the groat β-glucan content (expressed in %) of an individual line, $\mathrm{\mu }$ is the mean effect, ${\text{Gen}}_{\text{i}}$ represents the effect of genotypes, ${\text{Row}}_{\text{j}}$ is the effect of the row, ${\text{Column}}_{\text{k}}$ is the effect of the column, ${\text{Loc}}_{\text{j}}$ is the effect of location, ${\left(\text{Gen},,\times ,,\text{Loc}\right)}_{\text{ij}}$ is the interaction effect of genotype and location, and ${\text{e}}_{\text{ijk}}$ or ${\text{e}}_{\text{ij}}$ is the error effect. All the effects were considered as random effects except the genotype effect in BLUE value calculations. The broad sense heritability (H²) for β-glucan in each environment and across environments was calculated from the variance estimates from the linear mixed model. Descriptive statistics and Pearson correlations among locations within each year for β-glucan content were determined in R [53].

Genotyping

The entire GWAS panel was genotyped using the Genotyping-by-Sequencing (GBS) method at the USDA-ARS (U.S. Department of Agriculture-Agricultural Research Service) Genotyping Lab, Fargo, ND. The PYT set from 2015 to 2016 were genotyped as part of the Public Oat Genotyping Initiative led by Dr. Jean-Luc Jannink. A subset of this panel was also genotyped with the USDA-SoyWheOatBar-3 K Illumina iSelect array (Jason Fiedler, personal communication). For DNA extraction, each line seed was grown in a standard potting mix of small pots in a greenhouse. Young leaf tissue was collected into a 96-well plate having silica sand to keep the sample dry. Genomic DNA was extracted from ground dried tissue by first lysing with 10% sodium dodecyl sulfate at 65°C, precipitating proteins with 6 M ammonium acetate, precipitating DNA with cold 70% isopropanol, and washing with 70% ethanol. GBS protocol was previously described [54]. Libraries were prepared using double restriction enzymes PstI and MspI, ligated to unique barcode adapters. Paired-end sequencing of the GBS libraries was performed on an Illumina HiSeq 2000 system with 150 bp paired ends. All the Illumina paired end reads were cleaned with bbduk [55] using the following parameters ktrim = r; K = 23; mink = 11 and hdist = 1). Following the cleanup, sequencing reads were aligned to the PepsiCo oat reference genome OT3098V2 [43] using Bowtie2 [56]. Sample read alignment files were then sorted, indexed using SAMtools [57]. Raw single nucleotide variants and indels were called using SAMtools and bcftools v1.14 [58] and were filtered with a minimum mapping quality score of 30 and a minimum reads depth coverage of 4.

Initially, genotypic data consisted of 38,070 SNP markers. Genotypic data for the set of lines present in each year was extracted and curated individually. For quality control, SNP markers with minor allele frequency (MAF) of less than 5%, more than 5% heterozygosity, and unassigned to any oat chromosomes were filtered out. Missing SNP data points were imputed using an LD-kNNi method in TASSEL 5.0 software [59]. Genotypes that had more than 60% of marker data missing and marker heterozygosity of 0.15 were also excluded. The filtered set of 8,299 SNPs and 221 lines in 2015, 11,714 SNPs and 148 lines in 2016, 11,356 SNPs and 225 lines in 2017, 9,301 SNPs and 224 lines in 2020, 8,625 SNPs and 230 lines in 2021, 10,902 SNPs and 233 lines in 2022, and 9,341 SNPs and 1,230 lines for all years combined were used for association analysis (Table 1). The Illumina assay for the USDA-SoyWheOatBar-3 K array was generated using the manufacturer’s protocol. After filtering markers with the same metrics as for GBS data, 2,290 SNPs markers were available for further use in this study.

Population structure and linkage disequilibrium

To determine the underlying population structure in the GWAS panel, the principal component analysis (PCA) approach was used, with the first five PCAs being considered to account for sub-structure. Linkage disequilibrium (LD) analysis was performed for the whole genome as well as for sub-genomes of oat by calculating squared correlation coefficients (r²) for the pairwise marker comparisons using TASSEL v5.0 software [59]. To visualize the extent of LD, r² was plotted against the map distance (base pairs), and the LOESS curve was fitted. This nonlinear regression curve was used to estimate the LD decay over a genetic distance where the average r² dropped to half its maximum value [60].

BLUE values calculation for across years GWAS analysis

Most breeding lines were evaluated in only one year. The number of lines common from year to year ranged from 2 to 15. This can lead to bias in calculating BLUE values across all years. Therefore, the β-glucan content values for all lines present across years (i.e., 1230 lines) were predicted (genomic estimated breeding value (GEBV)) using a univariate RRBLUP GS model (Eq. 3). The RRBLUP model was implemented using mixed.solve function of the RRBLUP R package which facilitates the partitioning of random and fixed effects [61] and is represented as:

$$\text{Y}=\mathrm{\mu }+\text{Zu}+\text{e}$$ 3

where $\text{Y}$ is the vector of BLUEs for the phenotypic trait, $\mathrm{\mu }$ is the overall mean, $\text{Z}$ is the N × M matrix of markers, $\text{u}$ is a vector of normally distributed random marker effects with constant variance $\text{u}\sim \text{N}(0,\text{I}{\mathrm{\sigma }}_{\mu }^2)$ and e is the residual error variance distributed as $\text{e}\sim \text{N}(0,\text{I}{\mathrm{\sigma }}_e^2)$. Data collected in a specific environment was used as a training set to predict the GEBV value of all other lines within that environment. This was done separately for all 26 environments. In this strategy, each PYT set was predicted in all 26 environments, which resulted in a balanced data set (GEBVs) for all 1230 lines. This balanced dataset was used to calculate BLUEs across all the years using a linear mixed model:

$${\mathrm{Y}}_{\mathrm{ij}}=\mu +{\mathrm{Gen}}_{\mathrm{i}}+{\mathrm{Env}}_1+{(\mathrm{Gen}\times \mathrm{Env})}_{\mathrm{ij}}+{\mathrm{e}}_{\mathrm{ij}}(\mathrm{across}\mathrm{years})$$ 4

where Y_ij is the predicted groat β-glucan content (expressed in %) of line I in environment j, μ is the mean effect, Gen_i represents the effect of genotypes, Env_j is the effect of environment, (Gen × Env)_ij is the interaction effect of genotype and environment, and e_ij is the error effect. All the effects were considered as random effects except the genotype effect.

Association analysis

GWAS was conducted for each environment, across environments for a specific year, and across years using respective SNP data and β-glucan content BLUE values. The BLUE values for each environment and across environments for a specific year were calculated using Eqs. 1 and 2, respectively, whereas across years BLUE values were predicted using Eq. 3. GWAS was performed using a generalized linear model (GLM) and mixed linear model (MLM) in TASSEL v5.0, and with fixed and random model Circulating Probability Unification (FarmCPU) and Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway (BLINK) methods in R environment with the Genomic Association and Prediction Integrated Tool (GAPIT) [62]. Population structure, either represented by Q-values or PCAs, and kinship coefficient matrix (K) were included as covariates in the GWAS analysis. The kinship coefficient matrix (K) for the GWAS panel was calculated using the Identify-by-Descent (IBD) method in TASSEL. The FarmCPU and BLINK models reduce false positive associations without compromising true positive associations [63, 64]. The population structure and kinship matrix calculated using TASSEL software were included as covariates in MLM, FarmCPU, and BLINK models. All models (GLM, MLM, FarmCPU, and BLINK) were compared based on quantile-quantile (Q-Q) plots to select the best-fitted model for GWAS. Based on the fitness of the model and the power to detect significant associations, the BLINK model performed better and was selected for the association analysis reported here. The Bonferroni-corrected threshold was used to declare a significant association. The estimated allelic effects for each significant marker were obtained directly from the BLINK model. Based on the LD decay value, significant SNPs within 28 Mb were clustered into one locus. Significant SNPs detected in at least two environments were considered a stable locus.

Co-localization with previous studies and candidate gene identification

Significant SNPs for β-glucan content identified in the present study were compared with the β-glucan QTL reported in previous biparental and GWAS studies, and their position was confirmed based on the updated version of Oat Sang v1 and PepsiCo OT3098 v2 hexaploid reference genome [43, 44, 65]. Genomic regions underlying significant β-glucan SNPs were searched for candidate genes with function directly related to β-glucan synthesis i.e., CslF genes using GrainGenes oat genome browser [43]. The genes were searched in the vicinity of each significant SNP with a flanking window of ± 10 Mb upstream and downstream estimated based on the LD-decay value.

Genomic selection

The prediction accuracy (PA) of different genomic prediction scenarios for β-glucan content was compared. The validation strategy consisted of forward predictions in which models were built using previous year(s) PYT data to predict the most recent PYT (2022). Four GS scenarios were evaluated: (i) GS model with no markers as a fixed effect using PYT data from 2015 to 2021 as the training set (TS) to predict PYT 2022 (GSM1); (ii) GS model with a specific set of markers as a fixed effect (selected based on GWAS results) using PYT data from 2015 to 2021 as TS to predict PYT 2022 (GSM2); (iii) GS model with no markers as a fixed effect using data from PYT 2021 as TS to predict PYT 2022 (GSM3); (iv) GS model with a specific set of markers as a fixed effect using data from PYT 2021 as TS to predict PYT 2022 (GSM4). To compare the effect of genotyping platforms on prediction accuracy, the four GS scenarios were evaluated using two types of genotyping data: GBS (9,341 SNPs) and 3 K (2,290 SNPs). Based on GWAS results, four markers were used as a fixed effect in the above scenarios were: (i) Schr1D_372963087 on 1D, Schr2C_67182758 on 2 C, Schr_2D_142397650 on 2D, and Schr_7D_329527800 on 7D from GBS data, and (ii) GMI_ES02_c16996_412 on 1D, PepOat_178 on 2 C, PepOat_245 on 2D, and GMI_ES22_c18772_417 on 7D from 3 K array data. If markers at the same position were not available within the 3 K marker set, a 3 K marker in close physical proximity was identified to reproduce/compare the results. The above GS scenarios were ran using the kin.blup function of the rrBLUP package in R [61]. This linear mixed model partitioned the effects into random and fixed as per Eq. 5:

$$\text{Y}=\text{X}\mathrm{\beta }+\text{Zu}+\text{e}$$ 5

where $\text{Y}$ is the vector of phenotypic observations (predicted GEBV values from Eq. 3 for GS scenario (i and ii) and BLUE values from Eq. 2 for GS scenario (iii and iv)), $\text{X}$ full-rank design matrix for the fixed effects ($\mathrm{\beta }$), $\text{Z}$ is the design matrix for the random effects ($\text{u}$) with $\text{u}\sim \text{N}(\text{G}{\sigma }_{\mu }^2$), where $\text{G}$ is a genomic relationship matrix calculated using all the SNP markers and $\text{e}$ is for the residuals ($\text{e}\sim \text{N}(0,\text{I}{\sigma }_e^2$) with $\text{I}$ being the identity matrix [61].

Results

Phenotypic analysis

Large phenotypic variation was observed in the overall dataset for β-glucan content with a minimum value of 3.02% (VLG_21) to a maximum value of 7.24% (BRK_17) and with the environmental means ranging from 4.21% (BRF_20) to 5.82% (BRF_15) (Table 2). Overall, there is sufficient phenotypic diversity in the GWAS panel to explore the genetic basis of β-glucan content (Fig. 1). Broad-sense heritability (H²) estimates for β-glucan content ranged from 0.30 (SSH_16) to 0.91 (BRF_22). We observed lower heritability in the years 2015, 2016, and 2017 as compared to 2020, 2021, and 2022. Many factors can affect H² estimates, including genetic variance and error variance. The variation in heritability could be due to the use of more diverse materials as parents of the PYT after 2016. The higher heritability values in later years could also suggest that the new NIRS calibration may be more accurate at estimating β-glucan content. The overall moderate to high heritability of β-glucan content indicates the robustness of the data and the strong genetic control for β-glucan content in our breeding populations. There was a significant (p < 0.001) positive moderate to high correlation among environments in each year for β-glucan content, ranging from 0.27 to 0.85 (Supplementary Table 1). Low correlation among environments was observed in the years prior to 2017.

Table 2.

Descriptive statistics and broad sense heritability (H²) of β-glucan content in GWAS panel evaluated at different locations in multiple years

Year	Location	Environmenta	µ (%)	Min – Max (%)	Std	H2
2015	SSH	SSH_15	5.66	4.63–6.69	0.48	0.46
	BRF	BRF_15	5.82	4.70–6.97	0.38	0.40
	VLG	VLG_15	5.69	4.60–6.73	0.40	0.59
	WIN	WIN_15	5.31	4.23–6.39	0.40	0.49
	Across	ACR_15	5.62	4.77–6.31	0.30	0.29
2016	SSH	SSH_16	5.54	3.84–6.72	0.53	0.30
	VLG	VLG_16	5.31	3.91–6.52	0.48	0.72
	WIN	WIN_16	5.37	4.21–6.84	0.50	0.50
	Across	ACR_16	5.41	4.09–6.39	0.38	0.28
2017	SSH	SSH_17	5.01	3.65–6.48	0.57	0.54
	BRF	BRF_17	5.60	3.82–7.24	0.66	0.70
	VLG	VLG_17	5.04	3.42–6.68	0.61	0.69
	WIN	WIN_17	5.00	3.44–6.89	0.61	0.54
	Across	ACR_17	5.16	3.86–6.63	0.50	0.47
2020	BRK	BRK_20	4.42	3.45–5.34	0.41	0.88
	SSH	SSH_20	4.56	3.41–5.70	0.42	0.79
	BRF	BRF_20	4.21	3.21–5.69	0.45	0.85
	VLG	VLG_20	4.53	3.27–5.57	0.44	0.84
	WIN	WIN_20	4.48	3.27–5.84	0.45	0.85
	Across	ACR_20	4.44	3.50–5.50	0.39	0.71
2021	BRK	BRK_21	4.27	3.50–5.84	0.33	0.76
	SSH	SSH_21	4.94	4.12–5.96	0.35	0.60
	BRF	BRF_21	4.31	3.35–5.37	0.32	0.76
	VLG	VLG_21	4.30	3.02–5.63	0.38	0.36
	PIE	PIE_21	4.22	3.19–5.30	0.38	0.78
	Across	ACR_21	4.40	3.67–5.57	0.29	0.34
2022	BRK	BRK_22	4.74	3.60–6.46	0.40	0.85
	SSH	SSH_22	4.97	3.93–6.32	0.45	0.89
	BRF	BRF_22	4.92	3.67–6.50	0.46	0.91
	VLG	VLG_22	4.95	3.83–7.03	0.45	0.83
	PIE	PIE_22	4.70	3.62–6.47	0.45	0.65
	Across	ACR_22	4.86	3.80–6.32	0.39	0.66

µ, Min, Max, and Std are the mean, minimum, maximum, and standard deviation of β-glucan content (%) in a specific environment

^aCoded as year and location combination, where BRK Brookings, BRF Beresford, SSH South Shore, VLG Volga, PIE Pierre, WIN Winner, ACR Across locations for the specific year and 15, 16, 17, 20,21, and 22 for the years 2015, 2016, 2017, 2020, 2021, and 2022, respectively

Fig. 1.

Box plots showing the distribution of β-glucan content in GWAS panels evaluated at different locations in multiple years. Cross (×) represents the mean β-glucan content in each respective location

Population structure and linkage disequilibrium

Principal component analysis (PCA) was used to infer the population structure of the GWAS panels within each year and for all the lines together. The first two PCs account for 13.42 to 20.68% of the genotypic variation present in the GWAS panels. This along with no discrete clustering patterns indicates weak population sub-structure (Supplementary Fig. 1). Linkage Disequilibrium (LD) decay was calculated as the squared correlation coefficient (r²) among the marker pairs. The LD for the whole genome decayed at 28 Mb (Fig. 2). Among the three genomes of oat, the LD decay was ~ 13 Mb for genome A, ~ 142 Mb for genome C, and ~ 4 Mb for genome D (Supplementary Fig. 2). We found sub-genome C had a slow LD decay rate as compared to A and D sub-genomes.

Fig. 2.

Genome-wide linkage disequilibrium (LD) decay and the decline of LD-r² between SNP marker pairs are presented as a function of physical distance (in Mb)

Association analysis

We performed GWAS for each environment individually, across the environments within each year, and across all years. GWAS models such as GLM, MLM, FarmCPU, and BLINK were compared based on a Q-Q plot to select the best-fitted model to conduct GWAS. The BLINK model performed better than others based on the Q-Q plot with fewer false associations and was selected for the association analysis moving forward (Fig. 3). Across all analyses for individual environments and across locations within a specific year, a total of 38 significant marker trait associations (MTAs) for β-glucan content were positioned on fourteen oat chromosomes (1A, 1D, 2A, 2C, 2D, 3C, 4C, 5A, 5D, 6A, 6C, 6D, 7C, and 7D) (Table 3). The phenotypic variation explained by significant SNPs ranged from 1.78 to 57.54% (Table 3). The position of the significant SNPs identified on oat chromosomes for the different environments is presented in Fig. 4. Based on the whole genome LD decay value, significant SNP markers flanking a region of 28 Mb were clustered as a single locus, which reduced the number from 38 to 22 loci. Out of these 22 loci, 12 were reported previously, and 10 are new loci. Among the new loci, 4 of them (Locus 12_IND on chromosome (chr) 4C, Locus 17_IND on chr 6D, Locus 18_IND on chr 7C, and Locus 19_IND on chr 7C) were detected in more than one environment (Table 3). Those loci contribute to a significant proportion of variance for β-glucan content in the breeding material (Table 3).

Fig. 3.

A quantile-quantile plot of observed -log10(p) vs. expected -log10(p) for GWAS models comparison (GLM: Generalized linear model; MLM: Mixed linear model; FarmCPU: Fixed and random model Circulating Probability Unification; BLINK: Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway)

Table 3.

Significant SNP markers linked with β-glucan content in the GWAS panel identified by association analysis for individual environments and across environments for specific years and its co-localization with β-glucan QTLs in previous studies and reported CslF genes/QTLs

Locusa	Envb	SNPc	Chrd	Position (bp)e	-log10(P)f	Allelic Effectg	R2(%)h	Fav Allelei	QTL Referencej	Reported QTL/ CslF genek
1IND.	BRF_22	Schr1A_460912744	1A	460912744	8.29	−0.15	8.09	AA/TT
2 IND.	WIN_15	Schr1A_530450138	1A	530450138	11.89	0.17	8.84	CC/TT	39	QTL QBG.CORE-BG_18.1
	ACR_15	Schr1A_530450138	1A	530450138	5.94	0.07	1.83	CC/TT
3l IND.	SSH_21	Schr1D_372963087	1D	372963087	6.47	−0.15	19.77	CC/GG	39	QTL QBG.CORE-BG_1.2
	BRK_21	Schr1D_394700343	1D	394700343	6.30	0.10	15.41	AA/GG
4 IND.	VLG_20	Schr2A_435494783	2A	435494783	6.29	0.13	7.35	GG/TT	43
5l IND.	ACR_20	Schr2C_67182758	2C	67182758	5.55	0.11	8.12	AA/GG	43	QTL QBG.CORE-BG_13.2; CslF3/4/8_chr2C_87 Mb
6 IND.	WIN_17	Schr2D_105868877	2D	105868877	5.47	0.16	5.18	CC/TT
7l IND.	WIN_15	Schr2D_143950575	2D	143950575	6.02	−0.12	7.84	AA/GG	39	QBG.CORE-BG_8.3
8 IND.	SSH_22	Schr3C_57082338	3C	57082338	6.59	−0.16	12.59	AA/GG
9 IND.	SSH_22	Schr3C_494512246	3C	494512246	6.13	−0.14	6.08	CC/TT
10l IND.	BRF_17	Schr4C_18577723	4C	18577723	6.59	0.20	11.10	AA/GG	39,41	QTL QBG.CORE-BG_11.1, Qbgl.UFRGS.L17N-Mrg11.1; CslF9_chr4C_24Mb;
11l IND.	ACR_20	Schr4C_161363513	4C	161363513	5.80	0.10	13.72	AA/GG	41	Qbgl.UFRGS.E18N-Mrg11.6
	BRK_20	Schr4C_161363513	4C	161363513	5.46	0.10	4.77	AA/GG
	SSH_22	Schr4C_171146697	4C	171146697	5.91	−0.10	1.78	AA/TT
12l IND.	BRK_22	Schr4C_706039374	4C	706039374	10.08	0.31	57.54	AA/GG
	BRF_22	Schr4C_706039374	4C	706039374	7.92	0.30	39.95	AA/GG
13l IND.	VLG_15	Schr5A_436924985	5A	436924985	7.22	0.29	48.31	GG/TT	41	QTL QBG.CORE-BG_24.4
	ACR_15	Schr5A_436924985	5A	436924985	7.27	0.19	11.12	GG/TT
	SSH_22	Schr5A_450770290	5A	450770290	6.46	0.21	13.45	GG/TT
14l IND.	PIE_22	Schr5D_397064548	5D	397064548	6.30	−0.13	4.72	CC/TT	39	QBG.CORE-BG_6.4
15l IND.	BRK_20	Schr6A_367992170	6A	367992170	6.68	−0.12	5.40	AA/GG
16 IND.	ACR_15	Schr6C_586192808	6C	586192808	6.48	0.08	3.76	AA/TT	28	QBG.ISU-BG_36
17l IND.	SSH_17	Schr6D_283928205	6D	283928205	6.24	0.24	33.41	AA/GG	-	CslF9/11_Chr6D_254 Mb
	ACR_17	Schr6D_283928205	6D	283928205	6.04	0.18	19.81	AA/GG
18l IND.	VLG_21	Schr7C_9754827	7C	9754827	6.27	−0.15	18.25	CC/TT
	ACR_15	Schr7C_20935319	7C	20935319	8.52	−0.15	6.39	AA/GG
19l IND.	PIE_22	Schr7C_79586276	7C	79586276	5.59	0.24	12.75	AA/GG
	WIN_17	Schr7C_94029283	7C	94029283	5.47	0.17	4.45	AA/GG
	BRF_17	Schr7C_103345888	7C	103345888	6.06	−0.20	11.16	AA/GG
	ACR_17	Schr7C_103345888	7C	103345888	5.57	−0.15	13.39	AA/GG
20 IND.	PIE_22	Schr7D_216827089	7D	216827089	7.71	−0.17	10.70	CC/TT
21 IND.	ACR_15	Schr7D_428906183	7D	428906183	5.37	−0.11	4.09	CC/TT	39,41	Qbgl.UFRGS.L17N-Mrg02.2
	BRK_20	Schr7D_454257773	7D	454257773	11.25	−0.16	7.17	GG/TT
22l IND.	BRF_20	Schr7D_466875467	7D	466875467	7.26	0.18	35.15	CC/TT	39,41	QBG.CORE-BG_2.2b, Qbgl.UFRGS.L17E-Mrg02.1
	VLG_20	Schr7D_467361253	7D	467361253	5.41	0.15	9.08	AA/GG
	WIN_20	Schr7D_467361253	7D	467361253	6.17	0.16	21.77	AA/GG
	WIN_17	Schr7D_518690781	7D	518690781	5.51	0.22	7.87	CC/TT

^aGenomic loci associated with β-glucan content and SNP present within a confidence interval of 28 Mb clustered in individual loci. Subscript IND. denotes locus detected either in a single environment or across environments in a single year

^bCoded as year and location combination, where BRK Brookings, BRF Beresford, SSH South Shore, VLG Volga, PIE Pierre WIN Winner, ACR Across locations for the specific year and 15, 16, 17, 20, 21, and 22 for the years 2015, 2016, 2017, 2020, 2021, and 2022, respectively

^cSignificant markers associated with β-glucan content

^dChromosome on which significant SNP is present

^ePhysical position of associated SNP in base pair (bp) based on A. sativa reference genome version2.0

^f-log10 of the P-value of the significant SNP market

^gAllelic effect of the significant marker. The allelic effect is a difference in mean β-glucan content between genotypes with major and minor alleles

^hPercentage of phenotypic variance explained by the significant SNP marker (R²)

ⁱMajor/minor allele of significant SNP. The favorable allele (increases the β-glucan content) for that specific marker is bolded

^jPrevious studies where the identified locus also reported

^kPreviously reported QTL/CslF gene present in the confidence interval of locus identified in the present study

^lLocus also identified in the across all-years analysis

Fig. 4.

Distribution over oat chromosomes of significant SNP associated with β-glucan content in the GWAS panel identified in individual environments/year (IND, orange highlighted) and in multi-years GWAS analysis (MULTI, blue highlighted). Rectangle over the locus showed the loci identified in both single and multi-year GWAS analysis. The star on the left side of the chromosome represents the QTLs associated with β-glucan content found at the same or near positions in previous studies [27, 29, 36, 39–41, 44, 66, 67]

Using genomic predicted values for each genotype in each environment, we also performed an association analysis across all six years (2015, 2016, 2017, 2020, 2021, and 2022) and found a total of 73 loci distributed over all oat chromosomes except 2A (Table 4). Loci identified in individual environments/years are differentiated from loci identified in the multi-year GWAS by adding subscripts “IND” and “MULTI”, respectively, along with the locus number for clarity (Tables 3 and 4). A total of 13 loci were detected by both the single-year analyses and the multi-year analysis (Locus 3_IND on chr 1D; 5_IND on chr 2C; 7_IND on chr 2D; 10_IND, 11_IND, and 12_IND on chr 4C; 13_IND on chr 5A; 14_IND on chr 5D; 15_IND on chr 6A; 17_IND on chr 6D; 18_IND and 19_IND on chr 7C; 22_IND on chr 7D) with 9 of those loci being previously reported (Fig. 5). The four new loci (Locus 12_IND on chr 4C, Locus 17_IND on chr 6D, Locus 18_IND on chr 7C, and Locus 19_IND on chr 7C) identified in more than one single environment GWAS were also detected in the multi-year analysis. Colocalization of loci identified in the individual environments and in multi-year analysis increases the confidence of genomic regions being truly associated with β-glucan content. Out of the 73 loci detected in the multi-year analysis, 37 loci were detected in the multi-year analysis and in previous studies but not in single environment analyses. The remaining 21 loci detected in the multi-year analysis were not reported previously in the literature and could be novel genomic regions associated with β-glucan content.

Table 4.

GWAS results for β-glucan content when the analysis was conducted across all the years (2015 to 2022) and its co-localization with β-glucan QTLs in previous studies and reported CslF genes/QTLs

Locusa	SNPb	Chrc	Position (bp)d	-log10(P)e	Coincide Individual env. locusf	QTL Referenceg	Reported CslF geneh
1MULTI	Schr1A_384797276	1A	384797276	29.64		36,39,44
2MULTI	Schr1A_493308933	1A	493308933	8.15		39,44
3MULTI	Schr1C_20530564	1C	20530564	8.45
	Schr1C_37214405		37214405	10.81
4MULTI	Schr1C_76291638	1C	76291638	7.97
5MULTI	Schr1D_70591432	1D	70591432	12.77		39
6MULTI	Schr1D_254437491	1D	254437491	8.36
7MULTI	Schr1D_333161751	1D	333161751	11.05		39,44
8MULTI	Schr1D_367551307	1D	367551307	11.79	3	39
	Schr1D_372963087		372963087	10.32
	Schr1D_377173026		377173026	17.10
	Schr1D_382578378		382578378	6.86
9MULTI	Schr1D_420289976	1D	420289976	12.54	3	41
	Schr1D_448548603		448548603	6.70
10MULTI	Schr1D_481203247	1D	481203247	6.31
11MULTI	Schr2C_67182758	2C	67182758	15.84	5	39
12MULTI	Schr2C_346291638	2C	346291638	10.82		39
13MULTI	Schr2C_367576622	2C	367576622	22.46		39
14MULTI	Schr2C_476138324	2C	476138324	28.58		39
15MULTI	Schr2C_547058747	2C	547058747	6.92		39
16MULTI	Schr2C_555114495	2C	555114495	7.47		39
17MULTI	Schr2D_45694651	2D	45694651	34.26
	Schr2D_56178195		56178195	8.06		27,44
	Schr2D_59094977		59094977	9.53
18MULTI	Schr2D_142397650	2D	142397650	25.28	7	39
19MULTI	Schr2D_181009434	2D	181009434	6.89		39
	Schr2D_189050926		189050926	7.90
20MULTI	Schr3A_4984378	3A	4984378	17.23
21MULTI	Schr3A_194568800	3A	194568800	11.87		29,39,44
22MULTI	Schr3A_377182544	3A	377182544	8.60		39,44,66,67
23MULTI	Schr3A_421912601	3A	421912601	8.25		39
24MULTI	Schr3C_294355	3C	294355	24.66
25MULTI	Schr3C_374039611	3C	374039611	16.83
26MULTI	Schr3C_443733273	3C	443733273	19.38
27MULTI	Schr3C_611975641	3C	611975641	8.20		39
28MULTI	Schr3D_19577733	3D	19577733	9.58
29MULTI	Schr3D_451241845	3D	451241845	7.92		39,41
30MULTI	Schr4A_203506415	4A	203506415	26.39
	Schr4A_229092078		229092078	6.60
31MULTI	Schr4A_280002673	4A	280002673	7.40		29,44
32MULTI	Schr4A_435265955	4A	435265955	8.40
33MULTI	Schr4C_33900744	4C	33900744	12.65	10	40	CslF9_chr4C_24 Mb
	Schr4C_52140370		52140370	16.19
34MULTI	Schr4C_83002071	4C	83002071	22.23		39,44
35MULTI	Schr4C_127448389	4C	127448389	9.60		29,40,44
36MULTI	Schr4C_172148894	4C	172148894	17.81	11	40
37MULTI	Schr4C_688571895	4C	688571895	20.71	12
38MULTI	Schr4D_261944205	4D	261944205	8.22
39MULTI	Schr4D_347891017	4D	347891017	18.23
40MULTI	Schr4D_393489739	4D	393489739	9.11		27,41,44
	Schr4D_415401302		415401302	7.18
	Schr4D_426900840		426900840	8.35
	Schr4D_432568653		432568653	8.22
	Schr4D_433400944		433400944	5.53
	Schr4D_435485375		435485375	7.34
	Schr4D_443871497		443871497	6.49
	Schr4D_444182728		444182728	7.16
	Schr4D_444372059		444372059	5.54
	Schr4D_444434309		444434309	5.61
	Schr4D_444526825		444526825	5.58
	Schr4D_444526867		444526867	5.72
	Schr4D_444535579		444535579	5.61
41MULTI	Schr5A_3987284	5A	3987284	11.94		40
42MULTI	Schr5A_94477736	5A	94477736	6.15
43MULTI	Schr5A_335671941	5A	335671941	8.25		39
	Schr5A_402281419		402281419	12.01
44MULTI	Schr5A_436463393	5A	436463393	8.27	13	39
	Schr5A_436924985		436924985	7.05
	Schr5A_437216254		437216254	7.33
	Schr5A_437216264		437216264	7.71
	Schr5A_437345826		437345826	5.34
	Schr5A_437345885		437345885	65.00
	Schr5A_437441308		437441308	8.05
	Schr5A_438690717		438690717	8.17
	Schr5A_439200480		439200480	6.95
	Schr5A_439200517		439200517	6.43
	Schr5A_439200520		439200520	6.43
	Schr5A_439200524		439200524	6.43
	Schr5A_442135864		442135864	7.36
	Schr5A_442408039		442408039	7.88
	Schr5A_442408104		442408104	8.22
	Schr5A_442408107		442408107	8.22
	Schr5A_442408113		442408113	8.22
	Schr5A_444061480		444061480	6.83
	Schr5A_444088899		444088899	7.54
	Schr5A_444088914		444088914	7.54
	Schr5A_444624291		444624291	6.67
	Schr5A_444624294		444624294	6.67
	Schr5A_444624355		444624355	6.67
	Schr5A_444624357		444624357	6.20
	Schr5A_444624380		444624380	6.23
	Schr5A_444624403		444624403	6.20
	Schr5A_444928281		444928281	6.58
	Schr5A_445419515		445419515	8.86
	Schr5A_450770290		450770290	5.51
	Schr5A_455456754		455456754	5.44
	Schr5A_455456789		455456789	5.44
	Schr5A_455456796		455456796	5.44
	Schr5A_455456804		455456804	5.44
	Schr5A_455456809		455456809	5.44
	Schr5A_455456825		455456825	5.44
	Schr5A_455456831		455456831	5.44
	Schr5A_455456837		455456837	5.44
	Schr5A_455456839		455456839	5.44
	Schr5A_455456841		455456841	5.44
	Schr5A_455456860		455456860	5.44
	Schr5A_455456862		455456862	5.44
	Schr5A_459620200		459,620,200	7.10
	Schr5A_479430988		479430988	22.38
45MULTI	Schr5C_20094258	5C	20094258	18.62		29,39,44
	Schr5C_22624452		22624452	33.19
46MULTI	Schr5C_152476214	5C	152476214	20.96		39,44
47MULTI	Schr5C_372915413	5C	372915413	22.83		39,44
	Schr5C_374966630		374966630	5.98
48MULTI	Schr5C_531636904	5C	531636904	5.92		39,41
	Schr5C_535293220		535293220	8.27
	Schr5C_545123612		545123612	7.01
	Schr5C_568292149		568292149	6.73
	Schr5C_596791583		596791583	8.00
49MULTI	Schr5D_410021112	5D	410021112	21.67	14	39,41
50MULTI	Schr5D_442808910	5D	442808910	9.70		39
51MULTI	Schr5D_475690512	5D	475690512	16.58		39
52MULTI	Schr6A_234858768	6A	234858768	29.37
53MULTI	Schr6A_267195489	6A	267195489	12.63
	Schr6A_292728342		292728342	7.66
54MULTI	Schr6A_382776022	6A	382776022	12.72	15	39,41
	Schr6A_398040302		398,040,302	23.98			CslF9_chr6A_416 Mb
	Schr6A_414716719		414716719	18.30
55MULTI	Schr6C_24190624	6C	24190624	8.51		39,44
	Schr6C_45173202		45173202	13.56
56MULTI	Schr6C_65706004	6C	65706004	10.56		39
	Schr6C_90813862		90813862	16.79
57MULTI	Schr6C_118611070	6C	118611070	7.16		41
58MULTI	Schr6C_256378545	6C	256378545	11.49		39
59MULTI	Schr6C_482355608	6C	482355608	7.40		29,39,44,67
60MULTI	Schr6D_200889441	6D	200889441	5.72
61MULTI	Schr6D_274125462	6D	274125462	12.16	17		CslF9/11_chr6D_254 Mb
62MULTI	Schr7A_51412334	7A	51412334	8.92
63MULTI	Schr7A_487098818	7A	487098818	7.11
64MULTI	Schr7C_17256941	7C	17256941	7.59	18
65MULTI	Schr7C_45870666	7C	45870666	36.31	18
	Schr7C_70734181		70734181	22.33
66MULTI	Schr7C_102880155	7C	102880155	7.60	19
67MULTI	Schr7C_186204467	7C	186204467	7.54
68MULTI	Schr7C_521748351	7C	521748351	8.07
69MULTI	Schr7D_7926065	7D	7926065	11.10		29,36,39,40,44
70MULTI	Schr7D_65830000	7D	65830000	11.55		29,36,39,40,44
71MULTI	Schr7D_181512834	7D	181512834	35.75		29,36,39,40,44
72MULTI	Schr7D_329527800	7D	329527800	5.35		29,36,39,40	CslF6_chr7D_301 Mb
73MULTI	Schr7D_418639913	7D	418639913	15.84	22	40,39,29,36,44
	Schr7D_434959047		434959047	11.78
	Schr7D_454257791		454257791	88.45
	Schr7D_466875464		466875464	41.87

^aGenomic loci associated with β-glucan content and SNP present within a confidence interval of 28 Mb clustered in individual loci. Subscript MULTI denotes locus detected across all year GWAS analysis

^bSignificant markers associated with β-glucan content

^cChromosome on which significant SNP is present

^dPhysical position of associated SNP in base pair (bp) based on A sativa reference genome version2.0

^e-log10 of the p-value of the significant SNP marker

^fCorresponding locus identified when analysis conducted for a single environment and single year

^gPrevious studies in which the identified locus was also reported

^hPreviously reported QTL/CslF gene near/present in the confidence interval of locus identified in the present study

Fig. 5.

Venn diagram representing the number of loci detected for β-glucan content in individual environment/year analyses, multi-year GWAS analysis, and previously reported [44]

Co-localization with previous studies and candidate gene identification

The significant genomic regions identified in the present study were compared with previous QTL studies. We searched these regions with ± 10 Mb confidence interval for previously identified β-glucan QTLs as well as for candidate genes that have direct or indirect functions related to β-glucan synthesis. We found that Locus 2_IND (at 530 Mb on 1A), Locus 3_IND (at 372–394 Mb on 1D), Locus 4_IND (at 435 Mb on 2A), Locus 5_IND (at 67 Mb on 2C), Locus 7_IND (at 143 Mb on 2D), Locus 10_IND (at 185 Mb on 4C), Locus 11_IND (at 161–171 Mb on 4C), Locus 13_IND (at 436 [27, 42–44]– 450 Mb on 5A), Locus 14_IND (at 397 Mb on 5D), Locus 16_IND (at 586 Mb on 6C), Locus 21_IND (at 425–454 Mb on 7D), and Locus 22_IND (at 466–518 Mb on 7D) coincided with the genomic regions previously reported [29, 39–41] for β-glucan content (Table 3; Figs. 4 and 5). Candidate gene analysis found that Locus 5_IND (at 67 Mb on 2C), Locus 10_IND (at 18 Mb on 4C), and Locus 17_IND (at 283 Mb on 6D) are located near Cellulose synthase-like CslF genes, CslF3/4/8_chrom2C_87 Mb, CslF9_chrom4C_24 Mb, and CslF9/11_chr6D_254 Mb, respectively [43] (Table 3).

GWAS across all years (multi-year analysis) resulted in the identification of 73 loci, among which 37 were not detected in the single environment/year analyses but were previously reported for their association with β-glucan content (Fig. 5). The list of loci co-localized with previous studies is provided in Table 4 [26, 28, 36, 39, 41, 43, 44, 66, 67]. Three loci, i.e., Locus 33_MULTI (at 33–52 Mb on 4C), Locus 54_MULTI (at 382–414 Mb on 6A), and Locus 72_MULTI (at 329 Mb on 7D), are identified in the vicinity of CslF genes CslF9_chr4C_24Mb, CslF9_chr6A_416 Mb, and CslF6_chr7D_301 Mb, respectively. These three loci were also reported to be associated with β-glucan content in previous studies [39–41].

Breeding for higher β-glucan content in oat

Pyramiding of favorable alleles of significant markers can improve the trait performance. Four loci were selected: Schr1D_372963087 (Locus 3_IND) which was detected in multiple environments, was reported in previous studies, and explained a large proportion of phenotypic variance; Schr2C_67182758 (Locus 5_IND) which was detected in multiple environments, was reported in previous studies, and is located in the vicinity of a β-glucan synthesis genes (CslF); Schr2D_142397650 (Locus 7_IND) which was detected in multiple environments and was reported in previous studies, and Schr7D_329527800 (Locus 72_MULTI, across year analysis) which was also detected in multiple environments, reported in previous studies, and located in the vicinity of a β-glucan synthesis genes (CslF). The 1,230 lines used in the GWAS were scanned for favorable and unfavorable alleles at those 4 loci. Favorable alleles for these four markers resulted in an increase in β-glucan content which was dependent on the number of favorable alleles at those loci (Fig. 6). The average predicted value of β-glucan content ranged from 4.86% (zero favorable allele) to 5.22% (four favorable alleles). These markers could be exploited to develop Kompetitive allele-specific PCR (KASP) markers to screen oat germplasm in the early breeding cycle.

Fig. 6.

Box plots showing the β-glucan content values for each stack of favorable alleles in the GWAS panel. The legend represents the number of favorable alleles stacked, with zero meaning no favorable allele and four meaning all four favorable alleles

The results from this GWAS study confirm that many loci are involved in controlling β-glucan content in oat. Therefore, GS is a valid approach to increase β-glucan content in oat. Using the same breeding data as used for the GWAS, we compared several factors and their effect on genomic prediction accuracy (PA) for β-glucan content. We investigated the source genotyping platform used (GBS vs. 3 K), if important markers should be considered as a fixed effect in GS models, and the inclusion of multiple years of data in the training set. The accuracy of each scenario varies depending on the set of lines used as training sets (TS). In scenarios GSM1_GBS and GSM2_GBS, we used data from PYT 2015 to 2021 as TS to predict lines from the 2022 PYT. In scenarios GSM3_GBS and GSM4_GBS, however, only data from the 2021 PYT was used as TS to predict PYT 2022 lines. A significant improvement in PA was observed when multiple years of PYT data (2015–2021) were used to build the model (0.61 to 0.72) as compared to when only one year of data (2021) was used as a training set (0.38 to 0.48) to predict the genotypes evaluated in 2022 (Table 5). Using markers as a fixed effect on the other hand did not result in drastic improvement in PA (no marker as fixed effect: 0.38 to 0.71 versus marker as fixed effect: 0.39 to 0.70). Models built using data from the Illumina 3 K chip platform were based on a reduced number of markers (2290 SNPs) compared to the GBS dataset (9,341 SNPs), but the reduction in markers did not significantly impact the accuracy of the prediction models (Table 5).

Table 5.

Accuracies of β-glucan content predictions for four different genomic prediction scenarios (GSM)

Scenario	BRK_22	SSH_22	BRF_22	VLG_22	PIE_22	ACR_22
GSM1_GBS	0.62	0.62	0.66	0.61	0.63	0.71
GSM2_GBS	0.64	0.63	0.66	0.63	0.63	0.72
GSM3_GBS	0.38	0.39	0.47	0.38	0.48	0.48
GSM4_GBS	0.39	0.41	0.46	0.40	0.48	0.48
GSM1_3K	0.62	0.62	0.66	0.61	0.63	0.71
GSM2_3K	0.63	0.61	0.65	0.62	0.62	0.71
GSM3_3K	0.38	0.38	0.44	0.34	0.44	0.45
GSM4_3K	0.32	0.32	0.36	0.28	0.36	0.37

BRK Brookings, BRF Beresford, SSH South Shore, VLG Volga, PIE Pierre, WIN Winner, and 22 for the year 2022

GSM1-GSM4 are four different GS scenarios, GBS: genotyping by sequencing data, 3 K: SNP array data. GSM1: GS model with no markers as a fixed effect using PYT data from 2015 to 2021 as the training set (TS) to predict PYT 2022; GSM2: GS model with four markers as a fixed effect (located at loci 3 IND ; 5 IND ; 7 IND ; and 72 MULTI ) using PYT data from 2015 to 2021 as TS to predict PYT 2022; GSM3: GS model with no markers as a fixed effect using data from PYT 2021 as TS to predict PYT 2022; GSM4: GS model with a specific set of markers as a fixed effect using data from PYT 2021 as TS to predict PYT 2022

Discussion

Although oat production has decreased worldwide, it is still highly valued for its positive health benefits due to the presence of water-soluble β-glucan. A better understanding of the genetic architecture of β-glucan content will help breeders increase β-glucan content in oat groats and ultimately enhance the nutritional quality of oat-based food products. An association analysis for β-glucan content was conducted using a panel of oat breeding populations from a North American oat breeding program. Broad sense heritability estimates for β-glucan content were moderate to high in most environments (except SSH_16 and VLG_21) for the breeding populations considered, indicating a relatively strong effect of the genotype relative to the environment (Table 2). Many factors can affect H² estimates, including genetic variance and error variance. The variation in heritability could be due to the use of more diverse materials as parents of the PYT after 2016. The higher heritability values in later years could also suggest that the new NIRS calibration may be more accurate at estimating β-glucan content. The overall moderate to high heritability of β-glucan content indicates the robustness of the data and confirms the strong genetic control for β-glucan content in our breeding populations. Similar levels of heritability for β-glucan content was reported previously [25, 29]. Even though the breeding germplasm used in this study was derived from a limited number of progenitors (parents), there was sufficient phenotypic variability and high heritability in the present panel to capture reliable associations including variations controlled by rare alleles [68, 69]. Although phenotypic variation may not be as large as a diverse population of non-related lines, the associated genes regions found here will have less problems with non-desirable linkage drag and should be more easily deployed.

There was no defined population structure in the GWAS panels as the breeding lines were derived from parents developed primarily from South Dakota State University oat breeding programs. The sum of genotypic variations explained by the first two principal components (PCs) for individual years ranges from 13.42 to 20.68% (Supplemental Fig. 1), which is less compared to other crops such as barley that was ~ 90% [70]. The low level of population structure in the GWAS panel reduces the chance of false associations. The extent of LD and rate of LD decay across the whole genome affect the power and resolution of association mapping. Previous studies reported different ranges (1.4–4.1 cM) of whole genome LD decay, which is difficult to convert in terms of physical distance (base pairs) [42, 71–74]. However, these studies found that the rate of LD decay is slower in North American oat germplasm; this is consistent with the 28 Mb in the present study. In contrast, a whole genome LD decay of 2.29 Mb was reported for a diverse panel of worldwide oat accessions [75]. Differences in LD decay among studies might be due to the inclusion of different landraces/accessions in the GWAS panel and differences in selection practices [76]. In predominantly self-pollinating species such as oat and barley, it is expected that LD decay occurs over longer map distances, and this influences the power and resolution of GWAS. The extent of LD decay for each genome varies and we found that sub-genome C had higher LD (142 Mb) as compared to A (13 Mb) and D (4 Mb) genomes. Genome C also had more marker coverage with less gaps in comparison to the other two genomes (Supplementary Fig. 3).

Both false positives and true negatives occur in GWAS. These errors can be corrected by using a statistical model that controls false associations, and BLINK outperformed other models in this regard (Fig. 4), as found in previous studies [64, 77–79]. Stringent conditions such as Bonferroni-corrected threshold value of -log₁₀(P) ≥ 5.2 were used to identify positive and significant marker-trait associations. In this study, the association analysis for β-glucan content was conducted using six breeding panels and was performed for a total of 26 environments. Using multiple GWAS panels and evaluating lines in multiple environments can enhance the power and accuracy of GWAS to capture significant associations. This approach differs from previous studies that either evaluated the same population in multiple environments or different populations in fewer environments [27, 29, 36, 39, 40]. Overall, our study confirmed the quantitative inheritance of β-glucan content [36, 41] and resulted in the identification of a high number of genomic regions associated with the trait, providing further insight into the genetic architecture of β-glucan content in oat.

Conducting GWAS individually for each of the 26 environments and across locations within a specific year, we identified 22 genomic loci associated with β-glucan content on 14 different chromosomes, and 12 loci that were congruent with previously identified β-glucan QTLs (Table 4) based upon their physical position on the oat consensus map. The use of fewer genotypes (~ 230) and fewer environments in single-year GWAS analyses led to a much smaller number of loci being detected as compared to the multi-year analysis, and only few loci co-localized with previously detected QTLs. However, single location/year GWAS analyses led to the identification of 10 loci that have not been previously reported and four of those loci were detected in more than one environment when analyzing each environment separately, suggesting that they are true associations. Furthermore, a search for candidate genes in the vicinity of those loci revealed the presence of CslF9/11 (positioned at 245 Mb) in proximity to the loci 17_IND/61_MULTI located on 6D.

Simultaneously using the information from all PYTs from 2015 to 2022 to conduct GWAS, we identified 73 loci associated with β-glucan content, of which 15 co-localized with the genomic regions identified in individual environments and 47 were reported in previous studies (Fig. 5). Direct comparison of QTL identified in the present study with previous linkage mapping and GWAS studies is facilitated by the recent QTL inventory [44], as the previously reported QTLs are assigned to chromosomes (in base pairs) according to the OT3098 oat genome version 2 [43]. Co-localization of QTLs among individual environments, across all-year GWAS analysis, and with previously reported QTL and candidate genes for β-glucan synthesis increase our confidence in their true association with β-glucan content. To the best of our knowledge, we are the first to report a GWAS conducted using genomic predicted values computed for many environments to increase the power of detection. Genomic prediction was used to achieve the evaluation of 1,230 breeding lines in 26 environments. While this approach may violate some assumptions typically required for GWAS, it enabled the detection of a higher number of loci compared to single environment analyses and confirmed the importance of 47 loci (detected in the multi-environment GWAS and reported previously in the literature) in controlling β-glucan content that would not have been otherwise detected when performing single location/year analyses. New associations were also discovered. Among the 25 new associations, four were also detected in the single environment analyses suggesting that they are valid associations. Further evaluation is needed to confirm the validity of the remaining 21 new associations. But overall, the multi-environment analysis was more powerful in detecting associations than the GWAS conducted for single environment/single year. This study highlights the importance of performing GWAS in multiple environments to obtain robust results.

Several loci merit consideration for marker-assisted breeding. Locus 3_IND (Schr1D_372963087- Schr1D_ 394700343)/8_MULTI and 9_MULTI explained from 15 to 19% of the PV and coincided with markers GMI_ES14_c3227_73 and avgbs_109934 (named as QTL QBG.CORE-BG_1.2) identified previously [39]. Locus 5_IND (Schr2C_67182758)/11_MULTI and Locus 7_IND (Schr2D_143950575)/18_MULTI fall in the confidence interval of QTL QBG.CORE-BG_13.2 and QBG.CORE-BG_8.3, respectively, which were identified in the Oat CORE panel [39, 43]. Locus 10_IND/33_MULTI and Locus 11_IND/36_MULTI explained 11–13.72% of the PV and are co-localized with Qbgl.UFRGS.L17N-Mrg11.1 and QTL Qbgl.UFRGS.E18N-Mrg11.6 in the UFRGS Oat Panel [41]. Finally, Locus 13_IND (Locus 44_MULTI) and Locus 22_IND (Locus 73_MULTI) explained more than 10% of the PV and are mapped in the genomic position of QTL QBG.CORE-BG_24.4 and QBG.CORE-BG_2.2b/Qbgl.UFRGS.L17E-Mrg02.1, respectively, which were detected in the Oat CORE Panel [39, 41].

Four Loci, Locus 5_IND/11_MULTI, Locus 10_IND/33_MULTI, Locus 17_IND/61_MULTI, and Locus 72_MULTI are present near CslF genes CslF3/4/6_chr2C_87 Mb, CslF9_chr4C_24 Mb, CslF9/11_chr6D_254 Mb, and CslF6_chr7D_301 Mb genes, respectively [41]. Cellulose synthase-like (Csl) genes play a major role in cellulose synthesis and are associated with β-glucan production in small grains, with CslF likely being of major importance [80–82]. In the Poaceae, gene families belonging to cellulose synthase (CeS) and cellulose synthase-like genes (Csl) are involved in the biosynthesis of β-glucan in the endosperm and include CesA, CslA, CslB, CslC, CslD, CslE, CslF, CslG, and CslH [80, 83]. Out of these, CslF and CslH glucosyltransferase families are only present in grasses/monocotyledons such as oat, rice, wheat, and barley, and CslF family genes are considered potential candidate genes controlling β-glucan synthesis [80, 81, 84]. In wheat, down-regulation of CslF6 resulted in a 30–53% decrease in β-glucan content, while overexpression of this gene led to an increase in β-glucan content [81, 82]. In barley, bgl (β-glucan-less mutant) trait was mapped on chromosome 7H, where candidate gene HvCslF6 is located [85]. In oat, the level of expression of AsCslF6_C relative to its homeologs was associated with β-glucan content [39]. Surprisingly, no QTL was previously reported for β-glucan content on chromosome 7C where AsCslF6_C is located [41, 44]. In this study, five genomic regions were reported on chromosome 7C, however, none coincided with the location of AsCslF6_C, the closest associated SNP is located approximately 100 MB from AsCslF6_C.

Despite the important role played by Csl and CeS genes in controlling β-glucan content, the detection of many loci associated with β-glucan content in this study and in previous GWAS studies suggest that even traits such as β-glucan content that have relatively simple biological control can be affected by variation at many loci across the genome. Interestingly, in SDSU breeding lines, the five loci that were detected near CsFl genes were not among those that had the strongest effect on the PV. Of interest are the presence of loci on 1D (3_IND), 5A (13_IND), and 7D (22_IND) which explained a larger proportion of the phenotypic variation (19–48%) than the loci located near CsFl genes (5–11%). QTLs were previously detected in the region of those three loci located on 1D, 5A and 7D reinforcing the importance of those three QTLs in controlling β-glucan content in oat. Furthermore, results indicate that the genomic regions associated with β-glucan content found in this study are common with the ones detected in germplasm from different parts of the globe, such as the Oat CORE Panel and the UFRGS Oat Panel. We used advanced breeding lines to capture marker-trait association instead of using a diverse GWAS panel, yet we were able to capture many MTAs that were reported in QTLs studies using diverse panels. Our observation demonstrates the presence of enough variation for β-glucan content in SDSU breeding population to increase β-glucan content in future oat varieties. We expect that this is also the case in other US oat breeding programs as germplasm exchange is common among programs.

Because the trait is controlled by many loci across the genome, GS is a promising strategy to increase β-glucan content. Leveraging the data collected in the PYTs over years as part of our breeding activities (2015–2021) increased prediction accuracy by 28% when predicting the 2022 PYT as compared to building a model using the data collected for the PYT from the previous year only (2021), even though data from five locations in 2021 were used to build the model. There are two likely reasons for these results. First, our strategy of leveraging breeding data over multiple years allows us to increase the genetic diversity used to build the model by using a much larger set of genotypes (997 genotypes versus 230 genotypes). Second, the multi-year strategy built a model based on marker effects from 26 environments. Despite β-glucan content having high heritability, the effect of each marker on the phenotypic value depends on the environment (as observed when performing the GWAS in single locations, i.e. markers detected and their effect on phenotypic value varied based on the location in each year). It is therefore not surprising that a model that captures each marker effect over a larger number of environments would be a more robust model. On the other hand, the genotyping platforms did not affect prediction accuracy. The lower number of markers on the USDA 3 K array as compared to GBS did not reduce prediction accuracy; the cost effectiveness of the 3 K array makes it a promising genotyping platform for implementing GS in oat breeding programs.

Conclusion

In summary, a new approach was proposed to perform GWAS and build genomic prediction models. The approach consisted of leveraging data collected as part of breeding activities to increase our ability to detect loci associated with specific traits and to increase the robustness of genomic prediction models. The study demonstrated that performing GWAS multiple environments increase the number of significant associations that are detected and is therefore necessary to obtain robust association results. Similarly, building GS models based on large multi-environment trials data leads to more robust models, even for traits with high heritability such as beta-glucan content. In addition to providing valuable insights into the genetic basis of β-glucan content in oat, the findings of this study will facilitate the implementation of GS to accelerate the genetic progress for increasing β-glucan content in oat and will be of value for improving other traits in oat and other crops.

Abbreviations

Csl

Cellulose synthase-like (Csl)

GBS

Genotyping by Sequencing (GBS)

GEBV

Genomic estimated breeding value

GS

Genomic selection

GWAS

Genome-wide association study

H²

Broad sense heritability

KASP

Kompetitive allele-specific PCR

LD

Linkage disequilibrium

MAF

Minor allele frequency

MAS

Marker-assisted selection

NIRS

Near-infrared spectroscopy

PA

Prediction accuracy

PCA

Principal component analysis

QTL

Quantitative trait loci

RRBLUP

Ridge regression best linear unbiased prediction

SNP

Single nucleotide polymorphism

Authors’ contributions

SKB performed the statistical analyses and wrote the first draft. GO assisted with data analyses. JDF and RSD helped with genotyping data processing and imputation. JLJ supported the genotyping in 2015 and assisted with the genotyping in 2016 and 2017. MC oversaw the project, secured funding, and participated in writing the manuscript. All authors were involved with editing the manuscript.

Funding

This study was supported by the South Dakota Agricultural Experiment Station, South Dakota State University Agronomy Horticulture and Plant Science Department, the National Institute of Food and Agriculture under hatch proposal SD00H529-14, the South Dakota Crop Improvement Association, Grain Millers, and General Mills.

Data availability

The genotyping data is available on T3/Oat (triticeaetoolbox.org) under the genotyping projects named SDSU_2015_GBS, SDSU_2016_GBS, SDSU_2017_GBS, SDSU_2020_GBS, SDSU_2021_GBS, SDSU_2022_GBS, and 3K_SDSU_PYT2015_2022. Small seed samples from the breeding lines used in this study are available for research purposes upon request (melanie.caffe@sdstate.edu).

Declarations

Ethics approval and consent to participate

All methods were performed in accordance with the relevant guidelines and regulations of institutional, national, and international guidelines and legislation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Supplementary Information

Supplementary Material 1.