The 340 forage oat germplasm resources utilized in this experiment were preserved by the lab of Sun Yan from the College of Grassland Science and Technology at China Agricultural University (Beijing). Among these germplasm materials, 282 were sourced from the National Crop Germplasm Resource Center of China (Beijing), 28 from the Qinghai Academy of Animal Husbandry and Veterinary Sciences (Qinghai Province), and 15 from the Seed Resource Medium-term Repository of the Chinese Academy of Agricultural Sciences (Inner Mongolia Autonomous Region); 10 were purchased from Beijing Zhengdao Ecological Technology Co., Ltd.; 3 were purchased from Beijing Crowgrass Technology Development Center; and 2 were purchased from BaiLv Group (Jiangsu Province). All germplasm materials were planted in 2022 and 2023, respectively, in the Changping Experimental Base in Beijing (116.23°E, 40.17°N) and the Pingluo Experimental Base in Ningxia Hui Autonomous Region (106.59°E, 38.95°N).

The experiment adopted a drilling planting method with a row spacing of 30 cm, and the sowing density was 6,400,000 seeds per hectare. Irrigation was applied once after sowing, and urea was top-dressed at the jointing stage at a rate of 30 g/m². Phenotypic data such as plant height (from the surface of the soil to the tip of the plants), panicle length (from the base to the top of the panicle), stem diameter (in the middle of stem between the first and second node near the root), node number (the total number of nodes from the root to the top of the plant), length (from the bottom to the tip of the leaf), and width (in the direction perpendicular to the middle of the leaf area, the widest part of the leaf was measured) of the flag leaf and the second-to-last leaf were measured in this study. These eight phenotypic traits were selected because they can basically depict the overall phenotypic characteristics of oat germplasm. Plant height reflects the maximum height that oat plants can reach by the milk-ripe stage, panicle length approximately indicates the size of the plant’s panicle, leaf length and width describe the morphological characteristics of the plant’s leaves, and stem diameter and number of stem nodes reflect the characteristics of the plant’s stem.

When the forage oats reached the milk-ripe stage, yield-related agricultural traits were measured. Five plants from the middle of each row were selected to measure plant height, panicle length, stem diameter, node number, and the length and width of the flag leaf and second-to-last leaf.

Years of multisite phenotype data were converted into best linear unbiased prediction (BLUP) data, with BLUP calculations performed using in-house R scripts. The mathematical model for the BLUP transformation is as follows:

Where Y j k h is the observation of individual ith in the hth year at the kth location, m is the mean value, l k is the effector at the kth location, g j is the effector of the jth genotype,   y h is the effector in the hth year, g l j k is the interact effector between genotype and location, and g y j h is the interact effector between genotype and year. e j k h is residuals. The line BLUPs will be used for subsequent GWAS.

A representative individual was selected from the field, and leaf samples were taken back to the laboratory for genomic DNA extraction using the TIANGen DNA Extraction Kit (Yuntai Biotechnology Co., Ltd, Beijing). After extraction, the quality and concentration of the DNA were assessed using a NanoDrop 2000 ultraviolet-visible spectrophotometer (Thermo Fisher Scientific, USA). Once the DNA samples passed quality control, the SLAF genome sequencing process was carried out. First, the qualified genomic DNA samples were digested with the restriction enzyme HaeIII. The resulting enzyme-digested fragments (SLAF tags) were treated with 3′-end A addition, ligation of dual-index sequencing adapters, PCR amplification, purification, pooling, and size selection of DNA fragments with lengths of 414–444 bp by gel electrophoresis, which were defined as SLAF tags. After library quality inspection, sequencing was performed with the help of the NovaSeq 6000 sequencing platform at Beijing Biomarker Technologies Corporation. Each sample had approximately 10 Gb of sequencing data, with a Q30 sequencing quality of 85%. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Chen et al., 2021) in the National Genomics Data Center (Memberspartners, 2021), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA019491) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

The sequencing data were filtered with Trimmomatic software (Bolger et al., 2014), and sequencing adapters, primer sequences, and low-quality reads were removed. Subsequently, the BWA-MEM (v0.7.10-r789) software (Li, 2013) was employed to align the filtered data to the reference genome. The SAMtools (v1.9) software (Li et al., 2009) was then employed to convert SAM files generated from a previous alignment into sorted BAM files. The Picard (v1.94) software (Toolkit, 2019) was employed to mark duplicate sequences, and the GATK (v3.8) software (Van der Auwera et al., 2013) was utilized to correct indel interference with SNP results. SNPs were detected by the SAMtools (v1.9) (Li et al., 2009) software. The default parameters were used for all software mentioned above. Finally, VCFtools (v0.1.15) (Danecek et al., 2011) was employed for filtering, with criteria of less than 10% missing data, a minor allele frequency (MAF) of at least 0.05, and a minimum sequencing depth of 5.

The population structure analysis of forage oats was performed using the Admixture software, with the optimal number of subpopulations (K) determined using cross-validation (CV error) assessment. The clustering type with the smallest CV error was identified as the optimal population structure. The population structure visualization was created using the R package pophelper (Francis, 2017). Principal component analysis (PCA) was performed using GAPIT3, and the PCA results were visualized using the R package ggplot2 (Wickham, 2011).

GWAS was performed using the BLINK model in the GAPIT3 software (Huang et al., 2019; Wang and Zhang, 2021), with the first 20 principal components (PC1–20) included as covariates to control for population structure. We also used other models to perform the association analysis, such as the GLM, MLM, SUPER, CMLM, and MLMM models, but the Q-Q plot showed that none of these models performed better than the BLINK model in this study ( Supplementary Figures S1 , S2 ). A Bonferroni correction was applied to set the significance threshold for the association analysis at 8.30⁻⁸ (0.05/602572). The results were visualized using Manhattan and Q-Q plots generated by the GAPIT3 software.

Firstly, the TBtools software (Chen et al., 2023) was used to acquire the positional information of all genes within a 200-kb range upstream and downstream of the SNP loci from the genome annotation file. Subsequently, based on this positional information, the genome sequences of all target genes were retrieved from the publicly available oats genome sequence on the NCBI website. Finally, the BLASTP was employed to obtain annotation information for these sequences.

A core collection was sampled from the whole collection using the sampleCore function from the Core Hunter 3 package (Thachuk et al., 2009). The genotypic and phenotypic data, including the traits SLL, FLL, SLW, FLW, PH, PL, SD, and NN, were used as input data together. The size of the core collection was set as 10% and 15% of the whole collection, respectively. Moreover, we employed the evaluateCore function in the Core Hunter 3 package to evaluate the proportion of alleles observed in the full genotypic dataset that is retained in the selected core collection. Furthermore, a PCA analysis was conducted for our collection to illustrate the distribution of the chosen samples within the proposed core collection.

A total of 340 accessions were evaluated under field conditions, and the eight traits were screened for assessment. The distributions of the eight traits were approximately normal in all assessed environments ( Figure 1 ). The correlation analysis of the eight traits of test materials in different years and different locations was carried out. As shown in Figure 1 , the phenotypic correlations of seven traits of the test germplasm materials except stem diameter were significantly correlated in different years and different locations (P < 0.05). There was no significant correlation between the stem diameter in Beijing in 2022 and that in Ningxia in 2022.

In order to detect the phenotypic differences of individuals of the same germplasm material under different environmental conditions, ANOVA was performed on the phenotypic data obtained under four environments. As shown in Supplementary Table 1 , there were extremely significant differences in phenotype between different environmental conditions (P < 0.0001). Tests of genotype × location and genotype × year interaction on original traits showed significant genotype × location and genotype × year interaction ( Supplementary Table 1 ), respectively, for the eight traits. Although there was a significant genotype × year interaction and a significant genotype × location interaction, data were analyzed jointly because most of the observations were significantly correlated across four environments ( Figure 1 ), and this enables the identification of stable alleles across environments.

A total of 5,918,177 population SNP markers were identified in this study. SNPs were filtered using the criteria of minor allele frequency (MAF) ≥0.05, missing rate <10, and minimum sequencing depth >20. There were 602,572 high-quality markers obtained. The individual heterozygosity of most samples is less than 0.1 ( Figure 2A ). The distribution of minor allele frequency showed a decrease in the number of markers as the MAF value increased, with the highest number of markers falling within the MAF range of 0.05~0.1 ( Figure 2B ). Marker distributions are displayed as a heatmap on 21 chromosomes ( Figure 2C ). The average SNP density was 17,652.86 bp/SNP.

A total of 602,572 high-quality SNPs were used for the population structure analysis, and we conducted PCA to assess the genetic diversity of forage oat germplasm resources. The results revealed that the first three principal components, PC1, PC2, and PC3, explained 6.81%, 4.12%, and 3.85% of the genetic variation, respectively, and a total of 14.78% of the genetic variation ( Figure 3 ). The principal component analysis revealed minimal variation among the germplasm samples, implying a subtle population structure. This finding mirrors previous research on globally diverse oat germplasm, where similarly weak population structures were observed (Newell et al., 2011).

The Admixture software was employed to assess population structure by calculating the CV error across a range of subpopulation numbers (K) from 2 to 10. The results indicated that the minimum CV error occurred at K = 9, suggesting an optimal subpopulation composition ( Figure 4 ). The population structure diagram, which showed changes from two to nine subpopulations, illustrated increasing mixture between populations with higher K values ( Figure 5 ). According to the clustering results, apart from N22, N23, N27, N31, N41, and AN254, more than 20 naked oat materials were assigned to the same group.

In this study, phenotypic data of forage oats were collected from four different environments: Changping (Beijing) in 2022 and 2023 and Pingluo (Ningxia Hui Autonomous Region) in 2022 and 2023. These phenotypic values were then converted into BLUP values, which were used for the GWAS analysis. The results of the GWAS analysis are shown in Figure 6 , revealing a total of six SNP loci significantly associated with node number (NN) ( Figure 6C ) and three SNP loci significantly associated with flag leaf length ( Figure 6D ). Two SNP loci (1D_201215487 and 1D_461680555) were located on chromosome 1D, one SNP locus (2A_297564389) on chromosome 2A, one SNP locus (4C_270343673) on chromosome 4C, one SNP locus (5C_33721066) on chromosome 5C, and one SNP locus (7D_302568573) on chromosome 7D. Among these, the locus 7D_302568573 exhibited the lowest P-value and the highest correlation with NN ( Table 1 ).

Three SNP loci significantly associated with flag leaf length were located on chromosomes 5A (5A_9808941), 6A (6A_276363495), and 7A (7A_90499576) ( Figure 6D ). Among these, locus 6A_276363495 exhibited the lowest P-value and the highest correlation with flag leaf length ( Table 1 ). Additionally, GWAS analysis was performed on six other phenotypes: plant height, spike length, second-to-last leaf length and width, flag leaf width, and stem diameter. However, no significant SNP loci associated with these traits were identified. We screened SNP loci whose P <10⁻⁵, and then we got 8, 18, 9, 10, 9 and 8 SNPs associated with FLW, PH, PL, SD, SLL, and SLW, respectively. The detailed location of the SNP in chromosomes and the P-value are provided in Supplementary Table S3 .

Phenotypic differences among various genotypes were indicated in boxplots ( Figure 7 ). Notably, several SNP markers, such as eight out of the nine SNP loci associated with stem node number and flag leaf length (1D_201215487, 2A_297564389, 4C_270343673, 5C_33721066, 7D_302568573, 5A_9808941, 6A_276363495, 7A_9049576), exhibited a lack of heterozygous genotypes in the test population. These observations indicated a relatively high level of homozygosity within the tested population. The sequences corresponding to the 50-bp regions upstream and downstream of the SNP markers are provided in Table 2 .

Based on previous studies on linkage disequilibrium (LD) in forage oats, the LD distance in the forage oat genome is approximately 200 kb (Newell et al., 2011). Thus, we extracted 200 kb interval information upstream and downstream of the associated SNP loci, and then candidate genes were identified. Within the above interval, several genes linked to six SNP loci associated with stem node number were identified, and the annotation information of the gene closest to the associated SNP locus is shown in Table 1 .

The marker 7D_302568573, associated with NN, was localized to the 7D chromosome at position 302568573 bp. The closest gene named 7D_gene23644 was detected at 90.1 kb upstream of the associated marker. This gene showed similarity with the hypothetical protein QYE76_059829 of Lolium multiflorum which was validated by BLASTP.

The 1D_201215487 marker was localized to chromosome 1D of Avena sativa at position 201,215,487 bp. A potential candidate gene named 1D_gene11508 was detected at 28.2 kb downstream from this marker. According to BLASTP comparison, 1D_gene11508 shared 66.67% similarity with a zinc finger MYM-type 1-like protein of Triticum dicoccoides.

The associated marker 1D_461680555 was also localized on chromosome 1D at position 461,680,555 bp. One potential candidate gene was identified 14.3 kb upstream of this marker. This gene, 1D_gene55679, is likely to encode the ervatamin-B-like gene since it shares 73.96% similarity with the Lolium rigidum ervatamin-B-like gene.

The marker 2A_297564389 was localized to the 2A chromosome at position 297,564,389 bp. A potential candidate gene named 2A_gene19938 was detected at 10.7 kb upstream from this marker. This gene showed similarity with the Pimeloyl-ACP methyl ester carboxylesterase of L. rigidum which was validated by BLASTP.

The 5C_33721066 marker was localized to chromosome 5C of A. sativa at position 33,721,066 bp. A potential candidate gene named 5C_gene3312 was detected at 21.9 kb upstream from this marker. According to BLASTP comparison, 5C_gene3312 shared 88.1% similarity with the ACT domain-containing protein ACR4-like protein of L. rigidum.

The marker 4C_270343673 was localized to the 4C chromosome at position 270,343,673 bp. The closest gene named 4C_gene32033 was detected at 9.7 kb downstream of the associated marker. This gene showed similarity with the hypothetical protein EJB05_32782 of Eragrostis curvula which was validated by BLASTP.

The candidate genes closest to the three SNP loci associated with flag leaf length within the given interval were 6A_gene22964, 7A_gene17036, and 5A_gene1085, which were annotated as putative aquaporin PIP2-2, triacylglycerol lipase OBL1-like, and scarecrow-like protein 21 ( Table 1 ) by BLASTP comparison, respectively.

The marker 6A_276363495, associated with FLL, was localized to the 6A chromosome at position 276,363,495 bp. The closest gene named 6A_gene22964 was detected at 8.1 kb upstream of the associated marker. This gene showed similarity with the putative aquaporin PIP2-2 of Triticum urartu which was validated by BLASTP.

The 7A_90499576 marker was localized to chromosome 7A of A. sativa at position 90,499,576 bp. A potential candidate gene named 7A_gene17036 was detected at 52.5 kb downstream from this marker. According to BLASTP comparison, 7A_gene17036 shared 81.12% similarity with a triacylglycerol lipase OBL1-like protein of Lolium perenne.

The associated marker 5A_9808941 was localized on chromosome 5A at position 9808941 bp. One potential candidate gene was identified 102.7 kb downstream of this marker. This gene, 5A_gene1085, is likely to encode scarecrow-like protein 21 since it shares 89.76% similarity with L. rigidum scarecrow-like protein 21.

To develop a core collection, 10% and 15% of the lines were sampled from the whole collection, respectively. According to the allele coverage (CV) analysis, CV values of 10% and 15% core collection were 0.9975 and 0.9990, respectively ( Supplementary Figure S3 ). We proposed the 15% subset as the core collection since this subset covered more alleles of the oat germplasm, and the size of this core collection is more suitable for use in breeding. The lines of the core collection are listed in Supplementary Table S4 . PCA distribution of the samples selected in the core collection compared to the whole collection is shown in Supplementary Figure S4 . The core collection members were evenly distributed throughout the entire collection. This core collection can then be directly used for forage breeding or improved through breeding procedures.

The large amount of variation and phenotypic diversity in these eight descriptive traits observed among panels indicated abundant genetic diversity among the genotypes. Flag leaf is one of the most functional leaves. It is important to genetically analyze the morphological characteristics of functional leaves, especially flag leaf in oat improvement. The CV and H of oat germplasm resources showed a higher variation in FLW ( Supplementary Table S1 ). This suggested that genetic diversity is higher in FLW than in other traits. For the same trait, the H values of our germplasm resources were similar in all four environments ( Supplementary Table S1 ). This implied that our phenotype data are consistent under the four different environments. In general, the genetic diversity of these agronomic traits in 340 genotypes was shown to be abundant and cover extensive genetic variation. The results indicated that phenotypic variation of the eight traits was mainly controlled by genetic factors and is suitable for trait–marker association studies.

Previous studies have shown that the phenotypic variations in forage oats were greatly influenced by their growth environment (Huang et al., 2020). The results of this study are consistent with previous findings. There were significant variations among different germplasm materials ( Supplementary Table 1 ). Further analysis revealed significant phenotypic correlations among seven traits, excluding stem diameter, across different years and locations (P < 0.05) with R ² ranging from 0.14 to 0.59 ( Figure 1 ). In 2022, no significant correlation was observed between stem diameter of materials in Beijing and Pingluo. This may be attributed to the considerable variation in stem diameter between different tillers of experimental plants, leading to substantial measurement errors. Consequently, the measured values failed to reflect the stem diameter of forage oats in this environment accurately, resulting in poor correlation between different locations.

Furthermore, this study employed strip sowing. Compared to the phenotype values obtained from spot seeding, the differences in phenotype between germplasm materials under strip sowing conditions may be reduced, thereby increasing the difficulty of identifying phenotype and genotype-associated loci. However, the phenotype values of plants under strip sowing conditions more accurately reflect their true performance under field conditions, so the identified SNP loci may possess higher practical application value.

The population structure analysis was conducted on 602,572 high-quality SNPs obtained by SLAF-seq technology. The PCA results indicated that the first three principal components explained a total of 14.78% of the genetic variation. The forage oat population structure analyzed in this study was relatively weak, consistent with the findings of Huang et al. (2020) and Zimmer et al. (2020). According to the PCA and Admixture analysis results, most germplasm did not form distinct structural populations. However, the distribution of some germplasm in G1, G2, G3, and G9 on the PCA plot differed significantly from the majority of the germplasm.

With the help of the Admixture software, population structure information was assessed and the CV error was estimated for the number of subpopulations (K) ranging from 2 to 10. The results indicated that the minimum CV error occurred when the K = 9 ( Figure 4 ). The group structure diagram of subgroups changing from 2 to 9 revealed that as the number of subgroups increased, the mixing between groups gradually became more apparent ( Figure 5 ). Based on the clustering results, the majority (over 20 samples) of naked oats and a few hulled oat germplasms were grouped together into G9. The SNP information obtained in this study suggests that naked oats and hulled oats were not segregated into two distinct populations. This could be attributed to the possibility that some germplasm resources utilized in this study may have resulted from the hybridization of hulled oats and naked oats.

The germplasm resources tested in this study were collected from several sources. Considering the results of population structure analysis may be associated with the source of lines, we checked out the distribution of lines from the same source in the nine groups ( Supplementary Table S5 ). The data showed that the lines from the same source were randomly distributed in the nine groups.

In traditional crop breeding research, the oat phenotype was heavily influenced by environmental factors and is typically controlled by multiple genes. Due to the scarcity of phenotype markers, molecular marker technology has become a promising tool for breeders.

In this study, based on phenotypic data and SNP information at the whole-genome level of 340 forage oat genotypes, association analysis was carried out to identify the genomic regions significantly associated with the target traits using the BLINK model. The BLINK model is a machine learning approach used in GWASs to analyze large-scale genetic data. BLINK incorporates linear mixed models to account for population structure and relatedness among individuals offering advantages in removing false-positive sites. According to the Q-Q results, the distribution of observed P-values was closest to the distribution of expected P-values ( Figure 6 ), indicating that the results of GWAS analysis performed using the BLINK model in this study were reliable. To reduce the interference of false positives, we choose 8.30⁻⁸ as the common threshold for this study. Finally, we identified six SNP loci significantly associated with NN and three SNP loci significantly associated with FLL by GWAS analysis. Molecular marker primers could be developed based on the flanking sequences of these SNP markers, which could be used directly for molecular marker-assisted breeding.

Six candidate genes, hypothetical protein QYE76_059829, zinc finger MYM-type protein 1-like isoform X1, ervatamin-B-like, Pimeloyl-ACP methyl ester carboxylesterase, ACT domain-containing protein ACR4-like, and hypothetical protein EJB05_32782, were identified within a 200-kb range upstream and downstream of the SNP loci significantly associated with NN.

The gene associated with the locus 5C_33721066 is the ACT domain-containing protein ACR4-like. This gene is located 21.9 kb upstream of 5C_33721066 ( Table 1 ). ACR4 is closely related to the number of tissue cell layers in Arabidopsis sepals and ovules (Gifford et al., 2003). In the root system, ACR4 is involved in controlling the division of cambium cells (De Smet et al., 2008). The function of the zinc finger MYM-type protein 1-like isoform X1 gene near the locus 1D_201215487 in plants has not been reported. In animal cells, zinc finger MYM-type protein 1 is associated with the metastasis of gastric cancer, and this protein exerts its transcriptional repression function by directly physically binding to the target gene DNA sequence (Yue et al., 2019). The ZMYM2 gene, which belongs to the same family, is involved in controlling the cell cycle and is essential for G1/S transition in cells (Cibis et al., 2020). Based on the above evidence, it is speculated that oats may regulate stem node number through cell division-related genes such as ACR4-like and MYM-type proteins.

The genes associated with the 1D_461680555 and 2A_297564389 loci are ervatamin-B-like and Pimeloyl-ACP methyl ester carboxylesterase, respectively. Ervatamin-B-like is a member of the papain-like cysteine protease gene (PLCPs) found in the latex of the tropical plant Ervatamia coronaria. PLCPs are classified into clan CA because of the structural similarity of their conserved catalytic residues to papain (Rawlings et al., 2018). PLCPs are generated as inactive precursors featuring an autoinhibitory prodomain that prevents unwanted protein degradation (Coulombe et al., 1996). The prodomain can block the access of substrate to the active site and is also involved in protein folding and subcellular targeting (Taylor et al., 1995; Santamarıa et al., 1998). Cysteine proteases are an important class of proteolytic enzymes involved in the programmed cell death (PCD) process of the anther tapetum during the development of many plant organs (Song et al., 2016). Recently, ervatamin-B-like protein is identified as associated with the morphogenesis of walnut (Juglans regia L.) stamen development (Li et al., 2024) and photosynthetic gene expression (Alomrani et al., 2021). UCH1/2, a PLCP gene found in Arabidopsis, significantly impacts stem architecture and greatly increases the number of leaves when overexpressed (Yang et al., 2007). Ervatamin B-like may play a crucial role in regulating stem node number in oats, since grasses have an equal number of leaves and stem nodes. PLCPs also play roles in maintaining the yield of Arabidopsis under low nitrogen (LN) conditions (James et al., 2018) and enhancing resistance to salt and drought stress of sweet potato (Ipomoea batatas L.) when overexpressing (Chen et al., 2010).

Pimeloyl-ACP methyl ester carboxylesterase is a coenzyme involved in carboxylation reactions and is a key gene in the biotin biosynthesis pathway (Lin, 2012). Biotin is an essential vitamin for plants, which can enhance the resistance to carbonate stress (Wang et al., 2020) and modulate primary root growth in Arabidopsis (Gibbs et al., 2021). Biotin also can enhance the biomass in algal culture (Varaprasad et al., 2021). The strongest correlation was observed between the 7D_302568573 locus and NN, with numerous transposon genes located approximately 200 kb upstream and downstream of this locus. The only non-transposon gene identified is located 90.1 kb upstream of this locus, which is a putative gene predicted by bioinformatics. Its true existence and function remain unknown.

Leaf is the main photosynthesis organ, and the agronomic traits of leaves are also very important for forage oat due to the close relationships with grass yield. In this research, we identified three SNP loci significantly associated with FLL ( Table 1 ). The genes closest to these three SNP loci are putative aquaporin PIP2-2, triacylglycerol lipase OBL1-like, and scarecrow-like protein 21. Scarecrow-like protein 21 (SCL21) is an important transcription factor in plants belonging to the GRAS family. The GRAS family is named after three representative members: GAI (Gibberellin Insensitive), RGA (Repressor of ga1-3), and SCARECROW (Sc) (Pysh et al., 1999). These proteins are closely related to the phytochrome A signaling pathway in Arabidopsis and regulate the elongation of the embryonic axis (Torres-Galea et al., 2013). The function of SCL21 in leaves has not been reported, but several studies have revealed the functions of SCL21 on plant morphogenesis. SCL21 acts redundantly with the related PHYTOCHROME A SIGNAL TRANSDUCTION 1 (PAT1) and SCARECROW-LIKE 5 (SCL5) to activate the expression of the DNA-BINDING ONE FINGER 3.4 (DOF3.4) transcription factor gene, whose expression is in an ETHYLENE RESPONSE FACTOR 115 (ERF115)-dependent manner (Bisht et al., 2023). ERF115 plays a predominant role in the activation of regenerative cell divisions (Heyman et al., 2013). The overexpression of SCL21 results in a notable decrease in the count of meristematic cortex cells. Co-overexpression of ERF115 with SCL21 resulted in a disorganized shoot and a strong inhibition of root growth (Bisht et al., 2023). According to these lines of evidence, it can be assumed that the function of SCL21 on leaf development may be inducible and requires collaboration with other factors to achieve its effects.

PIP2;2 (Plasma Membrane Intrinsic Protein 2;2) may be a potential regulator of FLL traits. It is a subtype of intrinsic proteins located in the plasma membrane, belonging to the PIP2 subfamily. Aquaporin PIP2;2 is associated with traits such as fiber elongation in cotton (Li et al., 2013), root water absorption, drought tolerance in rice (Oryza sativa L.), and salt tolerance in barley (Hordeum vulgare L.) (Bai et al., 2021; Patel and Mishra, 2021; Sharipova et al., 2022). PIP2;2 is involved in regulating water flow in plant cells, which is crucial for water management and osmotic pressure regulation in plants (Javot et al., 2003). In maize, the expression level of PIP2;2 in the elongation zone is higher than that in the mature zone, which strongly indicates that PIP2;2 is involved in leaf elongation (Maistriaux et al., 2024). Furthermore, overexpression of PIP2;2 was confirmed to enhance the tolerance of O. sativa to mild salt stress, and overexpression of PIPs in transgenic tobacco improves plant height and leaf dry weight (Aharon et al., 2003). On the contrary, knockout of either PIP2;1 or PIP2;2 results in the leaves of Physcomitrella patens being more bent and twisted. In legume species, cell division and expansion are both sensitive to water stress and are both probably involved in the net effect of water stress on leaf size (Gindel, 1968; Bunce, 1977). All these lines of evidence suggest that OsPIP genes may play important roles in regulating water homeostasis and leaf size. Therefore, PIP2;2 may regulate the development of flag leaf by maintaining cell turgor pressure, promoting plant growth, and responding to environmental changes.

Triacylglycerol lipase OBL1 is an enzyme widely present in plants, primarily involved in the metabolism of fatty acids and glycerol. OBL1 participates in lipid-mediated signaling pathways (Testerink and Munnik, 2005). By regulating lipid metabolism and the flux of acyl groups, OBL1 influences various physiological processes such as stress response, growth, and development (Müller and Ischebeck, 2018). For instance, under adverse conditions, OBL1 may help plants adapt to environmental changes by modulating the levels of lipid signaling molecules (Rottet et al., 2015).

In this study, we only identified candidate genes based on their proximity to the nearest SNP markers. Interestingly, we found that there is a FAR-RED ELONGATED HYPOCOTYL 3 (FHY3) gene located 51.3 kb upstream of OBL1. FHY3 was also selected as a candidate gene in a GWAS study of leaf length in Arabidopsis thaliana (Zhang et al., 2023). Therefore, there may be some limitations in the criteria used to screen candidate genes. We only determine candidate genes based on their proximity to SNP markers, which may lead to the omission of genes that are functionally related to SNPs but located further away. Furthermore, gene expression and regulation are complex processes involving interactions between multiple genes, transcription factors, and regulatory elements. Therefore, even if a gene is located closest to an SNP marker, it may not necessarily be the key gene influencing the trait. To accurately identify the key genes affecting a trait, it is necessary to conduct a comprehensive analysis incorporating more biological information and experimental evidence.

Apart from flag leaf length and node number, no significant SNPs were identified for the other six phenotypes when applying the Bonferroni correction threshold set at 8.30⁻⁸. This result may be attributed to several factors, including the large and complex oat genome and the insufficient number of identified SNP loci. After all, SLAF-seq does not yield all SNP locus like genome resequencing technology. On the flip side, environmental variation or trait complexity may also be the reasons because no significant SNP loci were found for other traits. The phenotype of oat plants is influenced by various factors such as genotype, water condition, temperature, and others. For example, oat plants grow taller when there is sufficient rainfall. When the temperature is too high, some genotypes of oats will bloom earlier. Future research will employ advanced sequencing methods to acquire a greater abundance of SNP loci. Subsequent GWAS analyses on existing phenotypes will aim to uncover SNP information associated with additional traits. Additionally, to overcome the limitations of GWAS in the study of complex traits controlled by multiple genes, genomic prediction methods may be a better choice for future breeding researchers. Next, we will continue to analyze the vegetative traits of forage oat by combining GWAS and genome prediction.

SNP markers identified in the study will serve as ideal tools for MAS programs. By being tightly linked to target traits, such as FLL and NN, these markers allow breeders to indirectly select plants with desirable traits through marker detection, rather than relying solely on phenotypic observations. This indirect selection method improves the efficiency of the breeding process and accelerates breeding, as breeders can quickly identify and select plants with the desired genetic characteristics early in the growth cycle of oats.

Due to the vast size and heterogeneity of global germplasm collections, it often becomes challenging to evaluate and utilize them comprehensively. A core collection addresses this issue by providing a manageable sample that represents the genetic variability of a crop species and its relatives. To effectively use and improve our collection through breeding procedures, we assembled a core collection based on genotypic and phenotypic data using Core Hunter 3 software. This 15% core collection not only captured the representation of the total diversity of the entire collection but also had a manageable number of accessions for trait evaluations and oat breeding utilization. Since our germplasm resource population is not large enough, we will continue to collect oat germplasm resources. With the expansion of the germplasm resource bank, the core germplasm population also needs to be continuously updated and expanded. Nevertheless, the construction of this core collection will greatly promote our oat breeding process.