Characterization of molecular diversity and genome-wide association study of stripe rust resistance at the adult plant stage in Northern Chinese wheat landraces

Background Stripe rust is a serious fungal disease of wheat (Triticum aestivum L.) caused by Puccinia striiformis f. sp. tritici (Pst), which results in yield reduction and decreased grain quality. Breeding for genetic resistance to stripe rust is the most cost-effective method to control the disease. In the present study, a genome-wide association study (GWAS) was conducted to identify markers linked to stripe rust resistance genes (or loci) in 93 Northern Chinese wheat landraces, using Diversity Arrays Technology (DArT) and simple sequence repeat (SSR) molecular marker technology based on phenotypic data from two field locations over two growing seasons in China. Results Seventeen accessions were verified to display stable and high levels of adult plant resistance (APR) to stripe rust via multi-environment field assessments. Significant correlations among environments and high heritability were observed for stripe rust infection type (IT) and disease severity (DS). Using mixed linear models (MLM) for the GWAS, a total of 32 significantly associated loci (P < 0.001) were detected. In combination with the linkage disequilibrium (LD) decay distance (6.4 cM), 25 quantitative trait loci (QTL) were identified. Based on the integrated map of previously reported genes and QTL, six QTL located on chromosomes 4A, 6A and 7D were mapped far from resistance regions identified previously, and represent potentially novel stripe rust resistance loci at the adult plant stage. Conclusions The present findings demonstrated that identification of genes or loci linked to significant markers in wheat by GWAS is feasible. Seventeen elite accessions conferred with stable and high resistance to stripe rust, and six putative newly detected APR loci were identified among the 93 Northern Chinese wheat landraces. The results illustrate the potential for acceleration of molecular breeding of wheat, and also provide novel sources of stripe rust resistance with potential utility in the breeding of improved wheat cultivars. Electronic supplementary material The online version of this article (10.1186/s12863-019-0736-x) contains supplementary material, which is available to authorized users.


Background
Stripe rust caused by Pst is also known as yellow rust because of the spore color during its asexual infection cycle on wheat [1]. Stripe rust is a serious disease of wheat worldwide that mainly damages leaf tissues. Stripe rust significantly reduces wheat yield by at least 10%, and up to 100% under severe infections [2]. The stripe rust fungus has diversified into a large number of races possessing different combinations of virulence genes. These races have the capability of circumventing the host resistance genes, and in combination with their capacity for long-distance dispersal, subsequently creating the potential for destructive epidemics in susceptible varieties under favorable conditions [3,4]. In China, the most severe epidemics of wheat stripe rust occurred in 1950, 1964, 1990, and 2002, and caused substantial yield losses of wheat, which were estimated at 6.00, 3.20, 2.65 and 1.40 million metric tons, respectively. In 2017, a stripe rust epidemic affected 1.65 million hectares in 12 provinces [4,5]. Application of fungicides is widely used in the control of stripe rust, however, this practice adds considerable cost to wheat production. In contrast, growing resistant cultivars is considered to be the most effective, environment-, and consumer-friendly means to manage stripe rust [2,[6][7][8].
Resistance to stripe rust can be classified into two types, on the basis of the growth stage of resistance expression: seedling/all stage resistance (ASR) and APR [2,9,10]. The ASR is effective at seedling and adult plant stages and is usually race specific and qualitatively inherited, but it can be overcome by new races of the pathogen. In contrast, APR is only effective at adult plant stages when warm weather is prevalent, and is usually non-race specific and quantitatively inherited, and more likely to be durable [11]. Previously, 80 Yr genes for stripe rust resistance have been identified and formally named [12,13], however, the majority of these resistance genes are ineffective against new Pst races [14][15][16]. Therefore, identification of novel sources of resistance for deployment in breeding programs is a matter of urgency.
Wheat landraces are traditional varieties that were selected by farmers in the field for preferable agronomic traits, but concurrently were also indirectly selected for disease resistance [17]. As the included resources in the primary gene pool, wheat landraces harbor many novel and stable resistance genes that can be utilized for the improvement of modern high-yielding cultivars [18]. The landraces carry homologous chromosomes that readily recombine with those of hexaploid wheat [19]. Wheat landraces are regarded as untapped genetic resources with potentially useful genetic diversity in view of their limited use in modern breeding programs. The utilization of wheat landraces as a valuable source of disease resistance has been demonstrated previously [20]. Numerous QTL for stripe rust resistance have been identified in recent decades [21]. The usual method for identification of QTL is traditional QTL mapping, also known as linkage mapping. The technique is applied to identify underlying genetic variations that co-segregate with a trait of interest using a bi-parental mapping population [22]. However, QTL mapping is fundamentally limited to the comparatively low allelic diversity of the two parents used for a cross and low recombination events which impair the mapping resolution [23]. Alternatively, GWAS has been used successfully in mapping QTL in different species, such as rice [24], barley [25], maize [26], soybean [27], cotton [28], oat [29], secale [30], eggplant [31], tomato [32], perennial ryegrass [33], chickpea [34], grape [35], sugarcane [36] and Brassica napus [37]. In addition, GWAS has been applied to study diverse traits in wheat, such as rust resistance [3], abiotic stress [38], yield-related traits [39] and agronomic traits [40]. Thus, GWAS is proven to be an appropriate approach for identification of novel genetic loci.
In this study, we evaluated 93 wheat landraces grown in the Northern Chinese wheat-growing zone (I-Northern Winter Wheat Zone and VII-Northern Spring Wheat Zone) [41] for resistance to Pst. The accessions were evaluated at the adult plant stage using a mixture of Pst races prevalent in China over 2 years in two field locations. We identified 32 high-confidence associations and further compared their chromosomal locations with previously mapped Pst resistance genes and QTL on the integrated map. The identified loci are suitable for marker-assisted selection (MAS) and further genetic dissection.

Adult-stage responses to stripe rust and estimation of heritability
In the field, we recorded the stripe rust response of the 93 wheat landraces grown in four environments at Mianyang and Chongzhou in 2016 and 2017 (Additional file 1). The landraces displayed diverse adult plant disease responses to a mixture of races prevalent in China. Under each of the four environments and the best linear unbiased estimates (BLUE) for all environments (BLUE_ALL), the phenotypic performance of the 93 landraces varied from 1 to the maximum of 6 in IT, and from 0 to 100% in DS (Fig. 1 The remaining 76 accessions showed higher IT and DS values in one environment, while showing lower values of IT and DS in another environment. Variation in the prevalent pathogen races in the different trials and interactions between genotype and environment may lead to differences in the numbers of resistant accessions across environments. Despite these differences, we observed high correlations coefficients (0.478-0.958, P < 0.001) between IT and DS values recorded in the different environments (Additional file 2). These strong correlations were mainly attributed to the similar Pst populations that we inoculated in the two planting areas. Broad-sense heritability (H 2 ) for both DS and IT were high across environments, with values of 0.80 and 0.85, respectively (Table 1). Fig. 1 Violin plots illustrating the density distribution of stripe rust response in four environments and BLUE_ALL. The IT data for 2016M, 2016C, 2017M and 2017C were converted from 0, 0; 1, 2, 3 and 4 to 1, 2, 3, 4, 5 and 6 scale, respectively, to allow comparison across all data sets. The white dot displays the median, the top and bottom of the thick black vertical bars represent first and third quartiles, respectively, and the green fill shows DS and IT estimates (n = 93). The two graphs were drawn using the omicshare online tool violin2 (http://www.omicshare.com/tools/Home/Soft/violin2)

Population structure and genetic diversity
The optimal number of subpopulation in the 93 wheat landraces panel was determined to be two based on calculation with the STRUCTURE software using a Bayesian clustering model [42] and subsequent application of STRUCTURE HARVESTER (http://taylor0.biology.ucla. edu/structureHarvester/) [43] (Fig. 2a, b). Subpopulation 1 contained 66 accessions, whereas subpopulation 2 contained 27 accessions (Additional file 1). Similarly, construction of a distance-based neighbor-joining tree resulted in a dendrogram in which clustering of the accessions was consistent with the STRUCTURE analysis (Fig. 2c).
The 78 accessions from I-zone were divided into two groups, sixty-four accessions were classified in subpopulation 1 and accounted for 97% of all accessions in subpopulation 1, whereas 14 accessions were classified in subpopulation 2 and accounted for 52% of all accessions in subpopulation 2. Among the 15 accessions from VII-zone included in the study, 13 accessions were grouped in subpopulation 2, accounting for 48% of all accessions in  subpopulation 2. Two accessions from VII-zone were grouped in subpopulation 1, comprising 3% of the accessions in subpopulation 1.
In a summary, winter wheat accessions comprised the dominant proportion of subpopulation 1, whereas subpopulation 2 consisted of winter wheat and spring wheat accessions in similar proportions (Additional file 1).
After filtering, 7107 polymorphic DArT markers and 120 SSR markers with 792 polymorphic allele variations were retained. Among the 7899 polymorphic markers, 2577 were located on the A genome chromosomes, 3655 in the B genome, and 1667 in the D genome. The maximum number (831) of polymorphic markers was observed for chromosome 3B and the minimum number (102) for chromosome 4D (Table 2). Genetic diversity was analyzed using the 7899 markers. Gene diversity and polymorphism information content (PIC) of the whole genome ranged from 0.1017 to 0.5000 and from 0.0966 to 0.3750, with averages of 0.3109 and 0.2540, respectively. Minor allele frequencies (MAF) attained a maximum of 0.3750, with an average of 0.2222 (Table 2, Additional file 3). These analyses revealed a highly significant difference between the two subpopulations with regard to gene diversity and PIC values. Both diversity indices were significantly higher in subpopulation 2 in all of the 21 wheat chromosomes (Additional file 4).

Linkage disequilibrium
Among the 7899 polymorphic markers, the map position of 5486 markers was known on the wheat consensus map version 4.0 (Additional file 3). These mapped markers (genome A = 1845 markers, B = 2871 markers, and D = 770 markers) were used to estimate LD values (Additional file 3). Scatter plots of LD values, represented as squared allele-frequency coefficients, between intra-chromosomal markers against the genetic distance are shown in Fig. 3. The fitted model suggested that LD decayed to r 2 < 0.3 at 12.7 cM (Fig. 3b), 1.8 cM (Fig. 3c), and 4.4 cM (Fig. 3d) in the A, B, and D genomes, respectively. The LD decayed to the critical r 2 value (0.30) for the entire genome at about 6.4 cM (Fig. 3a), which was used to determine the confidence interval for declaring distinct QTL. Thus, for markers that were significantly associated with stripe rust and located on the same chromosome, marker were considered to represent the same locus only if the genetic distance between the markers was less than 6.4 cM or the r 2 value between the markers was greater than 0.3.

GWAS of stripe rust resistance at the adult plant stage
Using the data for 7899 polymorphic SSR and DArT markers that had a missing data frequency less than 0.05 and MAF higher than 0.05, a GWAS was performed on IT and DS. The exploratory threshold for definition of marker-trait associations (MTAs) as significant was taken to be P < 0.001 (−log 10 (P) > 3.00) [38]. To obtain reliable MTAs, phenotypic data collected from the four environments and BLUE_ALL were used. Using quantile-quantile (Q-Q) plots, we compared a general linear model (GLM) and MLM, both of which use the population structure or population structure and kinship as parameters for model calculation. The MLM method using a kinship matrix was the most efficient ( Fig. 4). Association analyses between the two resistance traits (IT and DS) and the polymorphic markers showed that there were 32 significantly associated SSR and DArT markers (P < 0.001), among which 13 markers were significantly associated with IT and 19 markers were significantly associated with DS. Using the data recorded from the four environments and BLUE_ALL, 28 and 5 significantly associated markers were detected, respectively. Among the 28 markers, the majority (23) were detected at Mianyang in both years, and half (14) were detected in 2016 at the two locations (  (Table  3). Considering the LD decay distance observed in this study, significant markers within 6.4 cM were combined as a QTL, hence a total of 25 QTL regions were assigned based on IT and DS (Table 3). Although we developed the integrated map based on the linkage maps reported previously, a portion of the markers used in the present study could not be mapped because of the lack of sufficient common markers in the previous maps.

Influence of favorable alleles number on response to Pst
A total of 32 significantly associated markers with reactions to Pst were identified. The number of favorable alleles among the 93 landraces ranged from 5 to 22. After ranking the accessions in increasing order based on the number of favorable alleles, comparison of the bottom 5% with a mean number of 7 favorable alleles to the top 5% (mean number 21.2 of favorable alleles) revealed that the former accessions showed significantly higher mean values of IT (4.73) and DS (50.93%), whereas the latter accessions showed lower mean values of IT (2.57) and DS (13.48%). The accessions that harbored relatively few of the identified resistance-associated favorable alleles showed a comparatively high disease index. Similar to a previous report by Maccaferri et al. [20], resistance to Pst was enhanced continuously with increase in the number of favorable alleles (Fig. 6), which revealed the additive effect of accumulation of alleles. The favorable alleles of 990,726, 1,270,827, Xgwm169-4 and 995,958 were present in all 17 stable-resistance landraces, whereas the favorable alleles of Xgwm169-9, Xcfd95-2, 5,010,940 and 1,074,322 were only harbored in some of the 17 accessions with lower frequencies (6-18%), which revealed that Pst resistance in the landraces was polygenic (Additional file 5).

Discussion
Phenotypic variability and molecular diversity of the northern Chinese wheat landraces germplasm The utilization of wheat landraces has gained increased interest in recent years. An enhanced understanding of     (Table 3 and Additional file 6) the extent of genetic diversity and the genetic basis of responses to stripe rust in wheat may improve the effectiveness of exploration of genetic resources and enrich breeding for durable stripe rust resistance in wheat [60]. In the current study, 93 Northern Chinese wheat landraces were surveyed under four environments and challenged with a mixture of Pst isolates. The data revealed that the Northern Chinese landraces, in particular the 27 accessions grouped in subpopulation 2, possess abundant variation for resistance to stripe rust. The 17 accessions showed stable stripe rust resistance across four artificial inoculation environments, which indicated that these accessions carried genes effective against the Pst isolates prevalent in the study years and might be useful as parental breeding lines for improvement of stripe rust resistance in wheat. The remaining 76 accessions displayed high IT and DS in one environment, and low IT and DS in another environment (Fig. 1, Additional file 1). Variation in the prevalent pathogenic races in different trials and genotype by environment interactions may have led to difference in the number of resistant accessions across environments. Such findings have been reported in many previous studies [61,62]. Genetic diversity is the probability of two randomly chosen alleles from the population being different; PIC estimates the detection power and informativeness of the molecular markers [63,64]. The genome-wide average gene diversity values were 0.25 and 0.34 for subpopulations 1 and 2, respectively, and the PIC values were 0.20 and 0.27 for the respective subpopulations (Additional file 3). The values of both parameters were similar to those reported in previous studies [3], but also higher in subpopulation 2 than in subpopulation 1 in the present study. These results revealed the potential utility of the landraces for GWAS.

Map-based comparison of significant stripe rust resistance loci with previously published Yr genes
To identify novel genes for effective resistance to Pst in China, we performed field evaluations at two locations in Sichuan with entirely different environments where stripe rust is endemic. The strong correlations among environments were also reflected in high heritability for IT (0.85) and DS (0.80) (Table 1), which supported the conclusion that the tested landraces were suitable for identification of significant associations in GWAS analyses. Accompanied with high heritability values, 32 significantly associated markers were detected at the exploratory threshold of P < 0.001 in this study (Table 3). Among the 32 significant markers, 28 were detected in the four environments and four in BLUE_ALL. However, the markers associated with each individual environment did not overlap with the BLUE_ALL associated markers. The results may reflect the small size of the study panel [65][66][67][68][69]. Considering the LD decay distance (6.4 cM) in the present study and marker position in the integrated map, a total of 16 QTL were detected (Table 3, Additional file 6).
The QTL QYr.sicau-5A was identified for IT on chromosome 5A ( Table 3). The Yr genes and QTL reported on chromosome 5A may be located on the long arm, such as Yr48, Yr34 and QYr.caas-5AL, as shown on the integrated map of chromosome 5A (Fig. 5, Additional file 6). QYr.sicau-5A was associated with the SSR marker Xwmc410, which is a flanking marker for QYr.caas-5AL [52]. Thus, it is likely that the two QTL were at the same locus.
QYr.sicau-6B was detected with the DArT markers 3,025,054 and 1,074,322 for DS at positions 31.18 and 26.85cM on chromosome 6B (Table 3). The majority of Yr genes and QTL detected on chromosome 6B are located on the short arm, such as Yr35, Yr36, Yr78 and QTL as displayed on the integrated map for 6B (Fig. 5, Additional file 6). 3,025,054 overlapped with Yr78 [56] and 1,074,322 overlapped with QYrst.wgp-6BS.2 on the consensus map [57]. Yr78 was designated as synonymous with QYr.wgp-6BS.1 which is close to the centromere of the 6BS chromosome but different from QYrst.wgp-6BS.2 close to the telomere of chromosome 6B short arm [56,57]. In the present study, by comparing the positions of the four QTL on the integrated map, QYr.sicau-6B was speculated to be synonymous with Yr78 and QYr.wgp-6BS.1.
The QTL QYr.sicau-7D.1 and QYr.sicau-7D.2 were detected for DS at position 65.83 and 94.32-97.09 cM on chromosome 7D (Table 3). As shown in the integrated map (Fig. 5), the majority of reported QTL on chromosome 7D, as well as Yr18, Yr33 and YrYL were distributed on the short arm, except Yr33, which was located on the long arm (Additional file 6) and linked with the SSR markers Xgwm437 and Xgwm11 [70]. The closely linked markers 995,958, 1,095,389 and 993,762 of the two QTL QYr.sicau-7D.1 and QYr.sicau-7D.2 were all located far from the reported loci. Therefore, it was speculated that they represent newly detected QTL for stripe rust.
The power of QTL detection by GWAS depends on sample size, number of markers, high LD and trait heritability [20,62,68]. The number of landraces (93) used in the current study was larger than that used for genome-wide association mapping of resistance to pre-harvest sprouting (80 Chinese wheat founder parents) [65], rust resistance mechanisms (33 orchardgrass accessions) [67], phenotypic traits (81 Canadian western spring wheat cultivars) [68], late maturity α-amylase activity (91 synthetic hexaploid wheat accessions) [69], and comparable to the 93 bread wheat accessions used for mapping agronomic traits [66]. However, the population size of the present study was smaller than that of several previous studies, which would lower the power of QTL detection. The landraces were highly diverse, as evidenced from the high rate of whole genome LD decay (6.4 cM, r 2 = 0.30, Fig. 3a) with the 7899 DArT and SSR markers, and the average gene diversity and PIC of the whole genome were 0.3109 and 0.2540, respectively ( Table 2). On the other hand, the 93 landraces represented the majority of the landraces collected by the Chinese Academy of Agricultural Sciences (CAAS) from the Northern Wheat-growing Zone in China. The landraces exhibited substantial and significant phenotypic variation in response to stripe rust (IT, 1.50-6.00; DS, 0-91%; Table 1, Fig. 1) among the four environments. The findings from the present research not only furnish valuable and practical information for acceleration of molecular breeding, but also provide novel sources of rust resistance for ongoing wheat improvement.

Conclusion
Breeding for stripe rust resistance in modern wheat cultivars continues to be impeded by the narrow genetic basis of resistance in elite genetic backgrounds. Hence, wheat landraces, as an excellent genetic resource for bread wheat improvement, have attracted the attention of wheat researchers in recent years. The present study reports the presence of valuable genetic variation for multiple Pst races in the field among Northern Chinese wheat landraces. The landraces that showed stable resistance at the adult plant stage could be crossed with cultivars that exhibit desirable agronomic traits to enhance stripe rust resistance, particularly those accessions that harbor improved resistance alleles at novel loci. High-density, whole-genome DArT-seq markers revealed a high degree of genetic diversity and relatively rapid LD decay, which indicated that these 93 wheat landraces are suitable for GWAS to directly identify markers closely linked to the causal loci. In the GWAS analyses, 32 loci significantly associated with Pst resistance in the field were detected. Six QTL (QYr.sicau-4A.1, QYr.sicau-6A.1, QYr.sicau-6A.2, QYr.sicau-6A.3, QYr.sicau-7D.1 and QYr.sicau-7D.2) were mapped to chromosomal regions in which no stripe rust resistance genes have been reported previously. This finding indicates that the landraces possess useful alleles currently underexploited in modern breeding germplasm, and that the landraces might carry novel resistance genes to stripe rust. However, allelism tests are required to confirm which of the identified QTL represent novel resistance genes and which represent alleles of previously mapped genes. The present results reveal the presence of novel Pst resistance loci in Northern Chinese Wheat Zone landraces that could be pyramided into common wheat cultivars by MAS, and provide closely linked markers to accelerate their validation and deployment in wheat breeding programs.

Plant materials
A total of 93 wheat landraces from the Northern Wheat-growing Zone in China were obtained from the CAAS. Accessions from eight provinces in the Northern Winter Wheat Zone (I) and Northern Spring Wheat Zone (VII) [41] in China were represented, including accessions from Inner Mongolia (12) In all field trials, accessions in the stripe rust nurseries were evaluated as non-replicated three rows. Rows were 1.5 m long with 0.3 m spacing between rows and the susceptible check ' Avocet S' , was planted every 20 rows. 'SY95-71' , a highly susceptible wheat line, was planted as spreader rows bordering the nurseries to ensure production of sufficient inoculum to provide uniform stripe rust infection. At the fourth leaf stage, all seedlings of the spreaders and susceptible checks were artificially inoculated with a mixture of races prevalent in China, which included the officially named Chinese Pst races CYR31, CYR32, CYR33 and CYR34, and a series of pathotypes, for example, Guinong 22-14, Shui 4, and Shui 5, which were provided by the Plant Protection Institute of the Gansu Academy of Agricultural Sciences, Gansu, China. CYR34 shows the broadest spectrum of virulence in China, and it is avirulent to Yr5, Yr8, Yr15, Yr24, Yr32, and YrTr1, but is virulent to Yr1, Yr2, Yr3, Yr4, Yr10, Yr25, Yr26, Yr44, and Yr76, which are widely utilized in Chinese wheat cultivars [4,71].
Stripe rust resistance was evaluated three times when disease severity on the flag leaves of the susceptible checks attained 60-100%. The stripe rust IT was estimated using a 0 to 4 scale (0, 0;, 1, 2, 3, 4) as described previously by Bariana and McIntosh [72] when the susceptible checks showed abundant sporulation. The scale values 0, 0;, 1, 2, 3 and 4 was converted to 1, 2, 3, 4, 5 and 6, respectively, prior to statistical analysis. Plants of IT 1-4 and of 5-6 were considered to be resistant and susceptible, respectively. Stripe rust DS was recorded weekly as the percentage leaf area with disease symptoms, from when the disease severity on the flag leaves of the susceptible checks attained 80-90% until after the peak severity. The DS thresholds of 0-20(%), 21-40(%) and 41-60(%) represented high, moderate and low levels of APR, respectively, whereas 61-80(%) and 81-100(%) indicated moderate and high susceptibility to stripe rust, respectively.

Genotyping
Genomic DNA was extracted from young leaf tissue of 2-week-old seedlings using modified cetyltrimethyl ammonium bromide method as essentially described by Saghai Maroof et al. [73]. The DNA concentration was determined and diluted to a working solution of 50-100 ng/μL. The collection of 93 wheat landraces was genotyped using the DArT-seq (Diversity Arrays Technology, Canberra, ACT, Australia) genotyping-by-sequencing (GBS) platform. A total of 89,284 probes from the DArT-seq (DArT and DArT_GBS) were used for genotyping. The accessions were also screened with 450 SSR markers, which were obtained from GrainGenes database (http://wheat.pw.usda.gov) and those reported by Peng et al. [74], Suenaga et al. [75] and Li et al. [76]. PCR amplification of SSR markers was performed with the following thermal cycling conditions: initial denaturation at 94°C for 5 min, followed by 35 cycles of denaturation at 94°C for 40 s, annealing at 50-65°C for 30 s depending on the primers, and extension at 72°C for 1 min with final elongation of 7 min. After amplification, PCR products in a reaction volume of 3 μL were resolved by electrophoresis in 6% denaturing polyacrylamide gel and revealed by silver staining in according with the method of Bassam et al. [77]. The GWAS marker data were filtered based on the following criteria. Monomorphic markers and markers with maximum missing values of 5% were discarded and only those with MAF ≥ 0.05 were used for further analyses.

Phenotypic data analysis
An analysis of variance (ANOVA) was used to test for additive variance between genotypes, environments, and the interaction between genotypes and environments using SAS V8.0 (SAS Institute, Cary, NC, USA). Broad-sense heritability (H 2 ) was calculated using the ANOVA model to estimate the variance components on an accession mean basis. BLUE values were obtained across locations and years when considering genotypes as a fixed effect in the model using QTL IciMapping. BLUE values were also used to perform GWAS with the four environments resistance phenotype. The correlation between different environments was calculated with Spearman's correlation coefficient.
Genome-wide association study for stripe rust resistance The association of the two marker sets (DArT and SSR markers) and stripe rust disease phenotype based on the adult stage in the field evaluation was carried out using a unified mixed linear model as implemented in TASSEL 3.0 software [78] (http://www.maizegenetics.net). Power-Marker V3.25 was used to estimate the genetic diversity of the DArT and SSR data [79]. The population structure of the 93 wheat landraces was assessed using the Bayesian clustering algorithm conducted with STRUCTURE V2.3.4 with a burn-in period at 50,000 iterations and a run of 100,000 replications of Markov Chain Monte Carlo (MCMC) after the burn in [80][81][82]. Manhattan plots were generated using the "Manhattan" function in the "qqman" package [83] in R × 64 3.4.3 (R Core Team, 2014). The Q-Q plots were used to assess the fit of the model.

Comparison of significant resistance loci with previously reported Yr genes and QTL
For comparison with previous studies, we generated an integrated map of the stripe rust resistance genes and QTL reported previously (referred to herein as the integrated map), including 80 officially named Yr genes, 67 temporarily named Yr genes and 327 previously mapped QTL of different marker types [21,84]. The map positions of the Yr gene or QTL in the integrated map were based on the 'Synthetic' × 'Opata' DH GBS map [85], the 9 K SNP consensus map [86], the sequential projection of the 90 K SNP consensus map [87], the tetraploid consensus map [88], the Diversity Array Technology consensus map V4.0 (http://www.diversityarrays.com), the 2004 SSR consensus map [89] and the 'Synthetic' × ' Opata' ITMI BARC SSR map [90]. The DArT and SSR markers significantly associated with stripe rust resistance in this study were mapped based on this integrated map using BioMercator V4.2.