Genome-wide association analysis of stripe rust resistance in modern Chinese wheat

Background Stripe rust (yellow rust) is a significant disease for bread wheat (Triticum aestivum L.) worldwide. A genome-wide association study was conducted on 240 Chinese wheat cultivars and elite lines genotyped with the wheat 90 K single nucleotide polymorphism (SNP) arrays to decipher the genetic architecture of stripe rust resistance in Chinese germplasm. Results Stripe rust resistance was evaluated at the adult plant stage in Pixian and Xindu in Sichuan province in the 2015–2016 cropping season, and in Wuhan in Hubei province in the 2013–2014, 2016–2017 and 2018–2019 cropping seasons. Twelve stable loci for stripe rust resistance were identified by GWAS using TASSEL and GAPIT software. These loci were distributed on chromosomes 1B, 1D, 2A, 2B, 3A, 3B, 4B (3), 4D, 6D, and 7B and explained 3.6 to 10.3% of the phenotypic variation. Six of the loci corresponded with previously reported genes/QTLs, including Sr2/Yr30/Lr27, while the other six (QYr.hbaas-1BS, QYr.hbaas-2BL, QYr.hbaas-3AL, QYr.hbaas-4BL.3, QYr.hbaas-4DL, and QYr.hbaas-6DS) are probably novel. The results suggest high genetic diversity for stripe rust resistance in this population. The resistance alleles of QYr.hbaas-2AS, QYr.hbaas-3BS, QYr.hbaas-4DL, and QYr.hbaas-7BL were rare in the present panel, indicating their potential use in breeding for stripe rust resistance in China. Eleven penta-primer amplification refractory mutation system (PARMS) markers were developed from SNPs significantly associated with seven mapped QTLs. Twenty-seven genes were predicted for mapped QTLs. Six of them were considered as candidates for their high relative expression levels post-inoculation. Conclusion The resistant germplasm, mapped QTLs, and PARMS markers developed in this study are resources for enhancing stripe rust resistance in wheat breeding. Supplementary information Supplementary information accompanies this paper at 10.1186/s12870-020-02693-w.


Background
Stripe rust (yellow rust), incited by obligate biotroph fungus Puccinia striiformis Westend. f. sp. tritici (Pst), is a significant disease of bread wheat (Triticum aestivum L.) worldwide. An outbreak will fast destroy green leaves and, in turn, dramatically reduce photosynthesis, resulting in stunted and weakened plants, reduced grain numbers per spike, shriveled grains, and lower grain weights. Grain yield losses of 100% can occur in fields sown to susceptible cultivars [1]. Annual grain yield losses caused by this disease are currently estimated to be 5.47 million tons [2].
With the co-evolution of host and pathogen, epidemics of stripe rust follow a boom-and-bust cycle. Extensive use of single resistance genes usually leads to the emergence of new virulent Pst races, followed by subsequent disease epidemics. The breakdown of resistance genes Yr1 in Bima 1 and Yr9 in 1BL.1RS wheat cultivars caused huge economic losses in China's wheat production [6]. Continuous gene discovery is urgently needed to enrich the resistance gene diversity to slow the boomand-bust cycle. For the longer-term, quantitative resistances controlled by minor QTLs are encouraged to be pyramided to achieve broad-spectrum and durable resistance to stripe rust.
As a powerful tool for QTL mining, genome-wide association studies (GWAS) have been widely used for gene discovery of multiple traits of wheat, including stripe rust resistance in a worldwide collection of spring wheat landraces [14], Ethiopian wheat [15], International Maize and Wheat Improvement Center (CIMMYT) elite wheat [16], European winter wheat [17], northern Chinese wheat landraces [18], and Sichuan wheat [19].
In this study, we undertook a GWAS on stripe rust resistance in 240 wheat accessions to: 1) study the phenotypic variance of stripe rust response and 2) detect the genetic loci underlying the stripe rust resistance. The results should help to understand the genetic basis of stripe rust resistance in Chinese modern wheat cultivars and facilitate the improvement of stripe rust resistance through marker-assisted selection (MAS).

Phenotypic variation
The best linear unbiased predictor (BLUP) values of stripe rust maximum disease severity (MDS) across five environments ranged from 8.8 to 89.1%, with an average of 49.8% (Additional file 1; Additional file 2), presenting a wide range of disease responses across the population. The levels of stripe rust symptoms varied across environments. The mean stripe rust MDS of the GWAS population were 43.6, 34.9, 71.2, 58.1 and 41.0% in Wuhan 2014, Wuhan 2017, Wuhan 2019, Xindu 2016, and Pixian 2016, respectively. The highest MDS was observed in Wuhan 2019 and disease severity in Wuhan 2017 was lighter than other environments. This might be attributed to not only the different weather conditions during inoculation and disease development but also the amount of pathogen in the natural environments. The variation of symptom levels might cause differences in the frequency distributions of the stripe rust MDS. Analysis of variance (ANOVA) revealed highly significant differences among genotypes, environments, and genotype × environment interactions (Table 1). Broad-sense heritability (H 2 ) for MDS was estimated to be 0.91, indicating less environmental variance than genotypic variance; thus, this panel was suitable for further genetic analysis. High Pearson's correlation coefficients (0.62-0.79) of stripe rust MDS were observed across five environments (Fig. 1). The highest correlation coefficient was observed between Pixian 2016 and Xindu 2016, and the lowest was between Wuhan 2017 and Xindu 2016.
Cultivars from Gansu had the highest stripe rust resistance with an average MDS of 20.0%, followed by Shaanxi and Sichuan at 35.4 and 37.0%, respectively. The cultivars from Henan (55.1%), Shandong (57.5%), and Jiangsu (59.5%) tended to be more susceptible (Fig. 2). CIMMYT lines presented good resistance to stripe rust with an average MDS of 28.6%.
Marker coverage, genetic diversity, and population structure After removing SNPs (single nucleotide polymorphism) with minor allele frequency (MAF) < 5% and missing rate > 20%, 14,578 SNPs were used for subsequent analyses. Of these, 5778 (39.6%), 6588 (45.2%), and 2212 (15.2%) were from the A, B and D genomes. The A (1.2 SNPs/Mb) and B (1.3 SNPs/Mb) genomes had higher marker density than the D (0.5 SNP/Mb) genome. Similarly, the A and B genomes had higher genetic diversity and polymorphism information content (PIC) values than the D genome (Additional file 3). A weak kinship was observed among cultivars in the association population [20]. The population was grouped into three subpopulations. The result was consistent with the geographic origin and pedigree. Most of the samples from Zone I (Northern Winter Wheat Zone), II (Yellow and Huai River Valleys Facultative Wheat Zone), VIII (Northwestern Spring Wheat Zone), and CIMMYT were grouped in sub-population I (Additional file 4), most of the Zone IV (Southwestern Autumn-Sown Spring Wheat Zone) cultivars in sub-population II, and the Maker-trait associations (MTAs) and geographical distribution of favorable alleles Twelve stable loci for stripe rust resistance were identified ( Table 2). These loci were distributed on chromosomes 1B, 1D The favorable alleles of QYr.hbaas-2AS, QYr.hbaas-3BS, QYr.hbaas-4DL, and QYr.hbaas-7BL were rare in the present panel, with frequencies ranging from 0.05 to 0.43. The frequencies of the other seven QTLs were above 0.60. Relatively higher favorable allele frequencies (0.44) of QYr.hbaas-2AS and QYr.hbaas-4DL were observed in Gansu cultivars, compared with 0-0.19 in cultivars from other provinces. The favorable allele of QYr.hbaas-3BS was rare in Chinese wheat (0-0.14) compared with lines from CIMMYT (0.45). In contrast, QYr.hbaas-4BL.1 was widely adapted in Chinese wheat but rare in CIMMYT lines (Fig. 5).
Fifty-one and 192 wheat accessions from the GWAS panel were speculated to contain Yr62 and Yr64, respectively based on linked SSR markers analysis (Additional file 1). Six and 52 wheat accessions were tested to contain YrSP and Yr7, respectively using gene-specific markers. None of the wheat samples contains Yr5 and Yr15 indicating a lot of room for stripe rust resistance improvement in China.

Relationship between stripe rust MDS and the number of favorable alleles
To illustrate the pyramiding effects of favorable alleles in different QTLs, we examined the number of favorable alleles of 12 mapped loci in each accession. The number of favorable alleles ranged from 1 to 12 (Additional file 1). Linear regression (r 2 = 0.87) showed the dependence of disease severity on the number of favorable alleles (Fig. 6). Accessions with more favorable alleles, such as Lantian 15 (11 favorable alleles), Lantian 26 (11), Lantian 21 (10), Lantian 12 (10), and Zhongmai 12 (10) exhibited strong stripe rust resistance.

PCR-based markers for mapped loci
A set of 11 penta-primer amplification refractory mutation system (PARMS) markers were successfully developed to detect the presence of stripe rust resistance QTLs (Additional file 8). Primers for these 11 markers are given in Additional file 9, and the protocols are described in Additional file 10. Developed markers were validated using the 240 GWAS accessions; the results produced very low frequencies of inconsistency (3.7-5.9%) with chip data.

Phenotypic variation and genetic diversity of stripe rust resistance
The levels of resistance to stripe rust in different provinces could be assessed with the mean stripe rust MDS. Samples from Gansu, Shaanxi, and Sichuan had the highest levels of resistance to stripe rust in China. This result is in accordance with the severe occurrence of stripe rust and substantial resistance-breeding efforts in these three provinces.
Additionally, CIMMYT germplasm showed high level of resistance to stripe rust. This is partly due to the long history of breeding for durable resistance to stripe rust at CIMMYT. The pathologists and wheat breeders from CIMMYT have focused on partial resistance for more than 30 years. This resistance source has been successfully used in the spring wheat region in China, such as Sichuan province [19]. The different resistance backgrounds of Chinese wheat in the CIMMYT germplasm suggest its usefulness in wheat breeding in China.
According to the favorable alleles frequency analysis, resistant stocks from Gansu, Shaanxi, Sichuan and CIMMYT carry different resistance loci. The high resistance level and diversity of resistance genes make them valuable sources of stripe rust resistance in breeding. This diversity also allows wheat breeders to pyramid various QTLs to achieve high level, broad-spectrum, and even durable resistance to stripe rust.

Novelty of the mapped QTLs
QYr.hbaas-1BS was mapped at the distal region of 1BS (9.7 Mb). This chromosome arm is rich with stripe rust resistance genes/QTLs, including Yr64 [21] and Yr15 [12]. Most of the previously mapped loci are far from QYr.hbaas-1BS based on physical position (Additional file 12). The genotype of linked marker for Yr64 is inconsistent with QYr.hbaas-1BS suggesting they are not the same loci. QYr.hbaas-1BS is different from Yr15 either as no sample in the GWAS panel was detected to contain this gene based on    QYr.hbaas-2AS mapped in the present study is also located in this region.
On 2BL, Yr5/YrSP, Yr7 [34,35], and many other loci were mapped or isolated (Additional file 12). Among them, a QTL associated with wsnp_JD_c744_1111659 is the closest to QYr.hbaas-2BL, with a distance of 126.7 Mb. The distinct locations between QYr.hbaas-2BL and other reported loci on 2BL indicated that QYr.hbaas-2BL is a new QTL for stripe rust resistance. The genotypes of gene-specific markers for Yr5, YrSP, and Yr7 were distinct from QYr.hbaas-2BL further indicating their difference.
QYr On 4DL, a QTL linked to Xwmc399 (484.7 Mb) in oligoculm wheat [44], was close to QYr.hbaas-4DL. Most of the varieties used in China come from CIMMYT, while the oligoculm variety is from Israel; it is unlikely that oligoculm is used in China, so this QTL may not be the QTL associated in this study. Although the two QTLs were mapped to similar regions, they appear to have different origins. Thus, QYr.hbaas-4DL is probably new.

Application of resistant germplasm and MTAs in breeding for stripe rust resistance
The GWAS population represents a large proportion of the diversity of recently cultivated wheat in China. Fiftyfour cultivars from the population have achieved a peak Grain yield and end-use quality are the main objectives for most breeding programs, however, resistance or tolerance to abiotic and biotic stresses is also a major concern for grain yield stabilization. As a minor QTL alone does not provide adequate resistance, pyramiding minor loci is necessary to achieve an acceptable level of resistance. For the long term, resistance QTLs are worth being gradually pyramided into founder parents for further improvement in stripe rust prevalent regions. PARMS markers developed in the present study will expedite this procedure with advantages of high throughput, low cost, and stability. Furthermore, the DNA template for PARM S can be prepared with alkaline lysis, which is convenient and time-saving [51]. In the stripe rust prevalent area, minor QTLs can also be pyramided through phenotypic selection which was verified to be feasible at CIMMYT [52]. Further, well-characterized resistance genes Yr5 and Yr15 effective against prevalent Chinese Pst races have not been incorporated in Chinese wheat cultivars. Combining these genes with minor QTLs will be another solution for stripe rust resistance breeding.
When selecting the resistance alleles of stripe rust QTLs through MAS, a large part of the respective chromosome will be fixed as identical by descent (IBD) which means the surrounding genes from the donor will be introduced along with the target QTLs. From this point of view, the resistant donors should be carefully selected not only for the target trait but also the agronomic performance to minimize linkage drag. Compared with introduced germplasm, using widely planted cultivars in the corresponding region may lower the risks of linkage drag. To avoid narrowing down the genetic diversity of stripe rust resistance, introducing resistant cultivars from other agro-ecological regions can enrich the resistance gene pool. When using these unadapted donors, 1-2 backcross with large populations will help getting rid of undesired traits.
Breeding for quantitative resistance is a challenge due to its complex inheritance. Genomic selection can facilitate parental selection, potentially accurately predict the phenotypes, and increase the genetic gain. Mean prediction accuracies reached 0.34 to 0.71 for stripe rust APR using different genomic prediction models [53] indicating its effectiveness in wheat breeding for stripe rust resistance. Markers for validated QTLs and wellcharacterized genes can used as fixed effects in genomic selection models to increases the prediction accuracy.

Plant materials
The wheat association panel comprises 240 geographically diverse cultivars and elite lines (Additional file 1; Fig. 2 [54]. Field studies were conducted in accordance with local practices. The stripe rust response was scored as MDS when the severity of the susceptible cultivar Mingxian 169 reached the maximum level. The MDS under five environments and the BLUP across environments (hereafter referred to as BLUP) were used for subsequent analysis.

Genotyping
The 240 wheat accessions were genotyped with an Illumina 90 K SNP array [56]. Calling and filtering of SNPs, kinship, and population structure analysis are described in our previous study [20]. Physical positions of SNPs were referred to the Chinese Spring reference genome sequences RefSeq v1.0 (http://www.wheatgenome.org). Genetic diversity [57], PIC, and MAF were computed by PowerMarker v3.25 [58].
Linked SSR markers Xgwm251 [43] and Xgdm33 [59] were used to test Yr62 and Yr64, respectively in the GWAS panel. The PCR products were detected by 12% polyacrylamide gel electrophoresis. Gene-specific markers Y15K1 [12] for Yr15 and Yr5-Insertion for Yr5 [11] were used to test corresponding genes. The PCR product was detected by 1.0% agarose gel electrophoresis. Linked KASP marker Sr2_ger9 3p was used to test Yr30 [60]. Gene-specific KASP markers Yr7-A, Yr7-D for Yr7 and YrSP for YrSP [11] were used to test corresponding genes.

Genome-wide association analysis
GWAS was conducted using the MLM (PCA + K) in TASS EL v5.2.53 (http://www.maizegenetics.net/tassel) and GAPIT (http://zzlab.net/GAPIT). Due to the higher extent of linkage disequilibrium (LD) in wheat and the complex genetic architecture of stripe rust resistance, markers with an adjustedlog 10 (P) ≥ 3.0 were regarded as significant for the trait. This threshold was also used in previous GWAS on stripe rust in bread wheat [17][18][19]61]. Loci significant in at least two environments were reported in the results and considered stable QTLs. 'CMplot' package (https://github.com/YinLiLin/R-CMplot) was used to draw the Manhattan plots and quantile-quantile (Q-Q) plots with R. Alleles leading to lower MDS were referred to as favorable, whereas those leading to higher MDS were unfavorable. The frequencies of favorable alleles and their allelic effects were calculated based on the representative SNPs associated with the resistance loci.

Development of PARMS markers for significant SNPs
To facilitate the use of mapped QTLs, significantly associated SNPs were developed as PARMS markers [51]. Primers were designed using PolyMarker (http://polymarker.tgac.ac.uk). Fluorescence signals were screened with PHERAstar Plus (BMG LABTECH) and analyzed with KlusterCaller (LGC Genomics).

Prediction of candidate genes
Annotated high confidence genes with a distance no more than 1 Mb from the representative SNPs of mapped QTLs were examined to consider the candidacy. The annotations of high confidence genes were referred to IWGSC RefSeq Annotation v1.0 (www.wheatgenome.org). Expressions of disease-related genes were analyzed in the public wheat expression database Triticeae Multi-omics Center (http://202.194.139.32) using the study performed by Zhang et al. (2014) [62]. Genes with a TPM > 0.5 were reported as the candidate genes for mapped QTLs. The highest expressions of these genes post-inoculation were also compared to those before inoculation to further confirm their candidacy.