A total of 197 spring durum wheat accessions that are maintained by the Debre Zeit Agricultural Research Center (DZARC) and Sinana Agricultural Research Center (SARC) as single seed descent (SSD) progenies were used to assemble the association mapping panel. The panel was genotyped using the 90 K Illumina iSelect wheat SNP array (Wang et al., 2014) and 182 accessions with missing marker data frequencies less than 10% were used in GWAS analyses (Supplemental Data Sheet 1). These accessions are cultivated in different wheat-producing regions of Ethiopia, including Wello, Akaki, Bichena, Gojam, Gonder, Tigray, Shoa and Bale, as well as 11 lines that are now cultivated in the United States. The collection includes 160 landraces and 22 cultivars released by DZARC or SARC.

Four Pst races (PSTv-14, PSTv-37, PSTv-40, and PSTv-51) (Wan and Chen, 2014) collected from bread wheat in the United States and two races (PSTv-106 and PSTv-110) collected from bread and durum wheat in Ethiopia (Table 1, Wan et al., 2016) were chosen to conduct single-race seedling test experiments. Three seeds for each of the 197 accessions were planted in 96-well trays containing soil mix (6 L peat moss, 2 L perlite, 3 L sand, 3 L potting soil mix, 4 L vermiculite, 100 g Osmocote, and 2 L water). ‘AvS’ was included in each tray to serve as a susceptible check. To confirm the race identity, a core set of Pst differentials containing 18 Yr-single gene lines (Wan and Chen, 2014) was included in each single-race test. Seedlings were grown under controlled greenhouse conditions (a diurnal temperature cycle of 20–24°C, a nocturnal cycle of 15–18°C and 16 h photoperiod). When the second leaves were fully expanded, the plants were inoculated with a single race and incubated in a 10°C dark dew chamber with 100% relative humidity for 24 h. Then, seedlings were transferred to a rust-free growth chamber at a diurnal temperature cycle gradually changing from 4°C at 2:00 am to 20°C at 2:00 pm and 16 h light and 8 h dark and grown for 18–20 days before recording the infection type (IT) data based on the 0–9 scale (Line and Qayoum, 1992). ITs of three seedlings for each accession were recorded as one score if reactions were the same. If different reactions were present, accessions were re-tested against the Pst races. Accessions with IT from 0 to 3 were considered as resistant, IT from 4 to 6 as intermediate and IT from 7 to 9 as susceptible. Resistant accessions were re-tested to verify their resistance to the corresponding Pst races.

To evaluate the reactions to Pst challenge at adult-plant stages, the 197 accessions and checks were planted in single rows without replicates in six environments of the US Pacific Northwest under natural infection of Pst. The predominant races in the local environments are presented in Supplementary Table S1. These environments included Whitlow farm located in eastern Washington State (WA), USA (46.721275° N, 117.150170° W) in 2014 (WHT14); Spillman Farm located in eastern WA (47.467413° N, 122.931436° W) in 2014 (SPM14) and 2015 (SPM15); Mount Vernon located in western WA (48.421215° N, 122.334047° W) in 2014 (MTV14) and 2015 (MTV15), and Central Ferry located in central WA (46.644795° N, 117.781377° W) in 2015 (CLF15). Spring wheat variety ‘Avocet Susceptible’ (AvS), known to be susceptible to stripe rust at both the seedling and adult-plant stages, was used as susceptible check and planted every 20 rows. Another susceptible variety, ‘Lemhi,’ was planted around each trial as borders to ensure the uniformity of Pst inoculum. For each accession, approximately 150 seeds were planted in rows of 0.5 m long and 0.3 wide. Two disease phenotypes, IT and disease severity (SEV, rating as a percentage of diseased areas covering flag leaves with a 0–100% scale) (Line and Qayoum, 1992), were recorded when the flag leaves of susceptible check ‘AvS’ were rated with IT from 7 to 9 and SEV from 70 to 100%. Since plant height and heading date might play a role in the stripe rust resistance phenotypes during adult-plant stages, these two traits were also scored in the field during the growing seasons of 2014 and 2015.

Best linear unbiased predictors (BLUPs) for IT and SEV of each line across six environments were calculated using PROC MIXED statement in SAS v.9.3 (SAS Institute, Inc., Cary, NC, USA).

Analysis of variance (ANOVA) was computed using PROC MIXED COVTEST statement in SAS v.9.3 treating genotypes, environments and genotype by environment interactions as random factors. The broad-sense heritability (H²) were calculated using the formula:

where σ_G² = variance component of genotypes, σ_E² = variance component of environments, σ_G∗_E² = variance of environment and genotype interactions, σ_e² = variance of residuals, n = number of environments.

Pearson’s correlation coefficients of IT and SEV among six environments were analyzed using PROC CORR statement in SAS v.9.3.

The tissue samples of each accession were collected from 25 7-day-old seedlings that were the same source of seeds used in the phenotypic evaluations. The genomic DNA was extracted using the DNeasy 96 Plant Kit (Qiagen GmbH, Hilden, Germany). SNP markers used in this study were generated using the Illumina^® iSelect 90 K wheat SNP assay (Wang et al., 2014). Marker genotypes were called with the GenomeStudio v2011.1 software package (Illumina, San Diego, CA, USA). Before exporting genotype data from GenomeStudio, calls showing residual heterozygousity were entered as missing values. Individuals with >15% missing SNP calls and markers with >10% missing and <10% minor allele frequency (MAF) were eliminated. The chromosome positions of SNPs were based on a high-density consensus map of tetraploid wheat generated by Maccaferri et al. (2015a). Markers with unknown chromosome positions were removed. After filtering the genotypic data, 182 accessions and 7,867 polymorphic markers were retained for the following analyses. Among 7,867 SNPs, 3,464 and 4,403 SNPs were located in A and B genomes, respectively.

Tightly-linked or diagnostic markers for Yr5, Yr15, Yr36, and Yr30/Sr2 were genotyped on the current panel, including SNP-derived Kompetitive allele-specific PCR (KASP) maker-IWA6121 and IWA4096 (Naruoka et al., 2016) for Yr5, barc8 for Yr15 (Yaniv et al., 2015), uhw89 for Yr36 (Distelfeld et al., 2006) and SNP-derived KASP marker wMAS000005¹ for Yr30/Sr2. The primer information of these diagnostic markers is provided in Supplementary Table S2.

To analyze population structure, Bayesian model-based clustering and distance-based hierarchical clustering methods were performed in STRUCTURE v.2.3.4 (Pritchard et al., 2000; Falush et al., 2003) and JMP^® Genomics v6.0 (SAS Institute, Inc., Cary, NC, USA 2012), respectively. For population structure analysis, 1,086 tag-SNPs were selected by a linkage disequilibrium (LD)-based method implemented in HAPLOWVIEW v4.2 (Barrett et al., 2005) with a setting of r² = 0.5. Five independent iterations were run for each subpopulation (settings from 1 to 10) under an admixture model of population structure with correlated allele frequencies. For each iteration, a 50,000 length burn-in period and 100,000 Markov Chain Monte Carlo (MCMC) replications after burn-in were applied. STRUCTURE outputs were collated by a web-based tool STRUCTURE HARVEST (Earl, 2012) and the optimum numbers of subpopulations were chosen based on an ad hoc Δk statistic method (Evanno et al., 2005). The output of STRUCTURE HARVEST was imported into the program Clumpp v1.1.2 (Jakobsson and Rosenberg, 2007), which generates the population structure matrix (Q) based on multiple independent replicates of STRUCTURE runs. Q bar graphs were depicted by the program Distruct v1.1 (Rosenberg, 2004). To examine the genetic divergence among subpopulations, fixation index (F_ST) was measured among each pair of subpopulations by JMP^® Genomics v6.0.

The pairwise kinship matrix (K) among all the 182 accessions was estimated using the identity-by-state (IBS) setting incorporated in JMP^® Genomics v6.0. After setting r²= 1.0 of tagger function implemented in HAPLOWVIEW v4.2, a subset of 3,022 unique tag-SNPs were chosen to calculate the K-matrix, which was then incorporated in the mixed linear model (MLM) marker-trait association (MTA) tests. To confirm the population stratification results made by Bayesian model-based clustering, hierarchical cluster analysis was conducted in JMP^® Genomics 6.0 using the Fast Ward algorithm method.

JMP^® Genomics v6.0 was used to calculate squared Pearson’s correlation coefficient (r²) of LD among pair-wise SNP markers and fit the LD decay curve by a smoothing spline regression line (λ = 10,000) at the genome level. The pair-wise LD was evaluated for all of the 7,867 polymorphic SNP markers. The population-specific critical value of r², which is an evidence of genetic linkage, was determined by taking the 95th percentile of r² distribution of unlinked SNP pairs following the method described in Breseghello and Sorrells (2006). The intersection of the fitted curve with this r² threshold was considered as the average estimated LD extent in the genome. Also, the mean r² and the percentage of significant marker pairs (P < 0.01) for each chromosome and genome were calculated to compare the degree of LD among chromosomes and genomes.

The intersection of LD decay curve and the standard critical r² = 0.30 (Maccaferri et al., 2015b; Bulli et al., 2016) was estimated at 2.25 cM. Therefore, the window size of quantitative trait loci (QTL) determined in the Ethiopian durum wheat panel was a genetic distance of ±2.25 cM from the peak of the significant associations.

A total of 7,867 high-quality SNP markers and 182 accessions were used in this association analysis. R package GAPIT (Lipka et al., 2012) software was applied to analyze the associations between SNP markers and stripe rust reaction data. Strong population structure in this panel was revealed by the model-based Bayesian clustering and the IBS-based hierarchical clustering. To reduce the spurious associations caused by population structure, MLM with Q and K as covariates (Yu et al., 2005) was applied for all GWAS analyses. Loci with marker-wise significance of P < 0.005 were reported for the seedling resistance loci. To determine the confidence interval for the potential QTL identified in this study, multiple SNPs significantly associated to the disease response that were co-locating together and/or adjacent within intervals inferior to the general LD-decay distance of ±2.25 cM from the association peak were considered as tagging a single QTL region. For each QTL region, the SNP with the P-value showing the strongest association was considered as the QTL-representative marker (tag-SNP).

To identify race-specific seedling resistance loci, GWAS was performed on each race response data set separately. To detect field resistance loci, two steps of GWAS were conducted for each environment and BLUPs data of IT and SEV across all environments. The first GWAS used all 182 accessions and the second GWAS used 127 accessions whose IT of BLUPs were greater than three. The purpose of the second GWAS was to detect loci of small effects that may be hidden by the larger effect loci of seedling or field resistance. Loci that were significant (P < 0.005) in at least two environments and BLUPs data were reported for the field resistance. Since plant height and heading date may interact with stripe rust reaction at the adult-plant stage, GWAS was also conducted on these traits. Common significant loci (P < 0.005) with field resistance and these two traits were eliminated from the reports of field resistance.

An integrated map of hexaploid wheat, consisting of published stripe rust resistance genes and QTL as well as various types of molecular markers (Rosewarne et al., 2013), was generated by Maccaferri et al. (2015b). Since the map positions of the resistance QTL identified in this durum wheat panel were derived from a tetraploid wheat consensus map (Maccaferri et al., 2015a), the first step used was to align the map positions of the identified QTL in the integrated map to those in the consensus tetraploid map, and a second step was conducted to compare the positions of the identified QTL with those of previously published Yr genes and QTL in the integrated map (Maccaferri et al., 2015b; Bulli et al., 2016). The alignment results are presented in Supplementary Table S6.

Single Pst race evaluation during the seedling stage was conducted using four races that were commonly detected in the United States in recent years and two races that predominated in Ethiopia in 2013–2014. The distributions of reaction to the six Pst races are summarized in Figure 1A and ITs of 182 accessions to individual races are presented in Supplemental Data Sheet 1. No evident differences were observed in the frequency of resistance and susceptibility to races collected from the United States and Ethiopia. About 36, 30, 31, and 29% of accessions were resistant (IT = 0 to 3) to US races PSTv-14, PSTv-37, PSTv-40 and PSTv-51, and the similar percentages of accessions were resistant to PSTv-106 and PSTv-110 from Ethiopia, with 31 and 32%, respectively. Higher percentages of accessions were susceptible (IT = 7–9) to US races PSTv-37 (56%), PSTv-40 (45%) and Ethiopian race PSTv-106 (46%) than to the US races PSTv-14 (23%), PSTv-51 (30%) and Ethiopian race PSTv-110 (37%). Among the 182 accessions, six were resistant and four were susceptible to all six races tested.

Highly significant differences were observed among accessions and accessions by environments (P < 0.0001) in ANOVA analyses for stripe rust response in the field tests, but no significant difference was detected among environments. Heritabilities of IT and SEV across six environments were high, at 0.87 for IT and 0.86 for SEV. Significant (P < 0.0001) correlations of IT and SEV were detected in the six environments, with correlation coefficient values ranging from 0.42 to 0.88, with an average of 0.59 (Table 2).

Infection type (Figure 1B) and SEV (Figure 1C) of this durum population displayed a broad and continuous distribution in the field. Relatively high frequencies of resistant accessions were observed in MTV14, SPM14, WHT14, MTV15 and CLF15, with 48, 47, 45, 56, and 61% of accessions classified as resistant, respectively. Across the six environments, 29, 61, and 10% of accessions showed resistant, intermediate resistant, and susceptible responses, respectively. The mean IT of this panel across six environments ranged from 3.3 to 5.5 and mean SEV ranged from 14.6 to 32.5%. Considering all phenotypic data, three accessions, ‘WC-2no.100,’ ‘WC-4no.78’ and ‘2000/01population37-30BDIno.63,’ were resistant at both seedling and adult-plant stages.

The optimal number of subgroups in the durum wheat panel was determined to be three based on the Δk method of model-based Bayesian clustering analysis using STRUCTURE (Evanno et al., 2005). Subpopulation one, two and three contained 71, 86, and 25 accessions, respectively. A similar clustering result was obtained using the distance-based Fast Ward clustering method (Figure 2) with a high correlation (95.8%) between these two approaches. Among 22 cultivars included in this panel, 21 were grouped into subpopulation three, which accounts for 84% of the subpopulation three accessions. One cultivar was grouped into subpopulation one. Landraces comprised 100% of accessions in subpopulation two.

The groups of geographic origins admixed into the marker-based clusters. Accessions in subpopulation one were sampled from Bale (59%), DZARC (15%), Wello (11%), Gojam (6%), Tigray (4%), Shoa (3%), and Bichena (1%). Landrace accessions in subpopulation two comprised accessions from Bale (28%), Wello (28%), Akaki (23%), USA (13%), Bichena (7%), and DZARC (1%). In subpopulation three, cultivars released by DZARC and SARC comprised 56 and 28% of accessions, and landraces collected from Bale, Akaki, and Gojam comprised 8, 4, and 4% of accessions, respectively. It was noteworthy that some sampling regions were aggregated into specific marker-based clusters. All accessions originating from the Shoa and Tigray regions were grouped into subpopulation one. The accessions originating from the USA and SARC were all grouped into subpopulation two and three, respectively.

The maker-based familial relatedness was estimated by IBS and a heat map of the kinship matrix is presented in Figure 2B. Observing the heat map, subpopulation one and two had a relatively high level of genetic similarity compared to the other pair-wise comparison subgroups. This result was confirmed by a lower F_ST value between subpopulation one and two (Supplementary Table S3).

Linkage disequilibrium r² values among SNP marker pairs were estimated for all chromosomes. For the whole durum wheat genome, 54% of the pair-wise LD comparisons were significant at P < 0.01, with the mean r² of 0.116. A genome had 56% significant marker pairs and an average r² of 0.122, and B genome had 54% significant marker pairs and an average r² of 0.108. Chromosomes 1A (63%) and 5A (61%) had the highest percentage of significant marker pairs of LD r², whereas chromosomes 6B (50%) and 7A (46%) showed the lowest.

In order to investigate the extent of LD in each genome, LD decay rates were evaluated at individual genome level (Figure 3). Considering genome-wide LD decay in this Ethiopian durum wheat population, LD decayed below r² = 0.3 at 2.25 cM. The threshold of the 95th percentile of the r² between unlinked SNP pairs was 0.15. The interaction between the LD decay curve and threshold (r² = 0.15) was 5.75 cM. The initial LD r² was 0.53 and decreased to its 50% of value at about 2.75 cM. For A genome, the LD decay curve crossed the r² = 0.15 at 5.76 cM. The LD r² decreased to its 50% of initial value (0.55) at about 3.30 cM. For B genome, the LD r² decay to 0.15 at about 6.20 cM, and LD r² declined to half of its initial value (0.49) at about 3.25 cM.

A total of 154 SNP markers distributed on all 14 chromosomes were identified to be significantly (P < 0.005) associated with seedling resistance for one or more Pst races (Supplementary Table S4) and these 154 SNP markers were grouped into 68 loci based on the confidence interval defined as the general LD-decay distance of ±2.25 cM from the association peak. Among the 68 loci, six loci showed significant associations with at least two Pst races (Table 3). Only YrEDWL-1BS.2, located on the short arm of chromosome 1B, was associated with both US races PSTv-14, PSTv-37 and Ethiopian race PSTv-106. The other five loci, YrEDWL-1AS, YrEDWL-1BS.1, YrEDWL-3AS, YrEDWL-4BL and YrEDWL-5BL, were exclusively associated with resistance to US races.

Compared to the other five seedling resistance loci, YrEDWL-1BS.2 exhibited a higher level of contribution for IT with phenotypic variances (R²) ranging from 8.6 to 11.7%. Resistance locus YrEDWL-1BS.2 contained a block of seven SNP markers (IWB36298, IWB31756, IWB10480, IWB38291, IWB74352, IWB74353, and IWB669) in strong LD with an average LD r² of 0.98. On the short arm of chromosome 1A, YrEDWL-1AS showed significant associations with US races PSTv-14 and PSTv-37 and explained from 4.5 to 4.6% of the phenotypic variance. YrEDWL-1BS.1 on the short arm of chromosome 1B, YrEDWL-3AS on the short arm of chromosome 3A and YrEDWL-4BL on the long arm of chromosome 4B were associated with US races PSTv-14 and PSTv-51, and accounted for 3.9–4.8%, 3.8–5.1%, and 4.0–5.1% of the phenotypic variances, respectively. YrEDWL-5BL, mapped to the distal part of the long arm of chromosome 5B. It was associated with three US races PSTv-14, PSTv-37, and PSTv-51 with R² values ranging from 3.4 to 4.3%.

Sixty-two loci were associated with response to single races at the seedling stage (Supplementary Table S4), including 13 loci with PSTv-14, 15 loci with PSTv-37, 15 loci with PSTv-40, seven loci with PSTv-51, seven loci with PSTv-106 and five loci with PSTv-110.

Two-step sequential GWAS analyses identified 24 loci that were significantly (P < 0.005) associated with field resistance in at least two environments as well as BLUP data across six environments at the adult stages. Six loci, YrEDWL-1AS, YrEDWL-1BS.1, YrEDWL-1BS.2, YrEDWL-3AS, YrEDWL-4BL and YrEDWL-5BL, also showed significant associations with IT at the seedling stage and most likely they represent valuable all-stages resistance genes. Consequently, the six loci were not considered in the pool of loci associated with putative adult plant field-resistance only. An ANOVA model with population structure (Q) as a covariate was used to re-test the significance of these 18 loci. Twelve loci passed the adjusted P < 0.05 of ANOVA test, including seven loci, QYrEDWL-1AL, QYrEDWL-1B.1, QYrEDWL-1BL.2, QYrEDWL-2BS, QYrEDWL-4AL, QYrEDWL-5AL.1 and QYrEDWL-5AL.2, that were detected in the first GWAS and five loci, QYrEDWL.par-1AS, QYrEDWL.par-2BS, QYrEDWL.par-3BL, QYrEDWL.par-4B.1 and QYrEDWL.par-4BL.2, that were detected in the second GWAS (Table 4). The R² explained by the individual loci ranged from 1.5 to 7.6% for BLUP-IT and from 1.5 to 10.2% for BLUP-SEV across all environments. The cumulative R² explained by the 12 loci were 46.3% for BLUP-IT and 46.5% for BLUP-SEV, after subtracting variances due to population structure (Supplementary Table S5).

Profiling tightly linked or diagnostic markers for Yr5, Yr15, Yr36, and Yr30/Sr2 on the Ethiopian durum wheat panel, none of these genes were detected in any of 182 accessions. In addition, MTA analyses in this study did not identify any loci that overlapped with the chromosome regions of Yr5, Yr15, Yr36, and Yr30/Sr2. GWAS of plant height and heading date were also conducted with the aim to check the possible impacts of these two traits on the stripe rust response at the adult stage. No common significant SNPs were detected between these two traits and field resistance.

Three distinct subgroups within this Ethiopian durum wheat association mapping population were revealed by both the model-based Bayesian clustering and the IBS-based hierarchical clustering approach. The number of genetic-based subgroups was less than the number of cultivated regions where the accessions were collected. A high level of admixture of geographic origins/cultivation regions was observed within each of the genetic-based subpopulations, indicating that an extensive exchange of germplasm between families and villages happened in Ethiopia. However, some collection regions have strong enrichment in specific genetic-based subpopulations. A similar detection was also reported in another GWAS panel of Ethiopian durum wheat by Mengistu et al. (2016), who speculated that the reasons for this enrichment might be the preferences of local farmers toward some specific wheat landraces/pedigrees, forced by local climate conditions and/or socioeconomic elements related to quality/end use.

Population structure explained 15.9 and 13.7% of phenotypic variances for IT and SEV across six environments (Supplementary Table S5), illustrating that the population structure of the current population had some associations with the stripe rust resistance in the field. The IT and SEV of subgroup three was significantly different from subgroup one and two (Supplementary Table S3). Subgroup three, comprised of 84% modern cultivars, exhibited a higher level of disease resistance to stripe rust races tested in this study than the other two subpopulations (Figure 2D). The high level of resistance is most likely attributed to the enrichment of favorable resistance alleles of QYrEDWL-1B.1, QYrEDWL-2BS, and QYrEDWL-5AL.1 in subgroup three (Table 5).

Co-linearity was found for the map positions of all the significant SNP markers in the current panel between the tetraploid wheat consensus map (Maccaferri et al., 2015a) and hexaploid integrated map (Maccaferri et al., 2015b) (Supplementary Table S6), indicating feasibility of using the integrated map for aligning significant loci of tetraploid wheat in this study. However, alignment results should be cautiously utilized in future research due to the inherent bias in the integrated and consensus maps.

Among six seedling resistance loci, YrEDWL-1BS.2 tagged by IWB36298 was positioned into the confidence interval of Yr64 on chromosome 1BS. Identified from Ethiopian durum wheat accession PI 331260, Yr64 was resistant to all tested races including PSTv-14, PSTv-37, and PSTv-40 (Cheng et al., 2014). YrEDWL-1BS.2 in this study was associated with responses to PSTv-14, PSTv-37, and PSTv-106 but not PSTv-40. The different resistance spectrum indicated that seedling resistance locus/alleles might not be the same as Yr64 as described in Cheng et al. (2014). Four SNPs (IWB36298, IWB10480, IWB38291, and IWB669) at this locus also showed significant associations with resistance to PSTv-14 and PSTv-37 at the seedling stage in a global collection of elite durum wheat population (Liu et al., 2017), indicating that this seedling resistance locus may be widely distributed among the durum wheat germplasm. YrEDWL-1BS.2 was significantly associated with resistance to races PSTv-14, PSTv-37 and PSTv-106, and 16% of accessions in the Ethiopian durum population carry the resistance-associated SNP marker alleles.

The objective of GWAS analyses for field resistance in this study is to identify loci consistently associated with APR. Hence, the MTAs that were consistently detected in at least two environments plus BLUP data, but not associated with seedling resistance, were further considered. A total of 12 field resistance loci were detected in the current population, and six of them co-located with previously reported APR QTL (Supplementary Table S6).

On chromosome 1B, QYrEDWL-1B.1, significantly associated with stripe rust resistance in three environments, was mapped within confidence intervals of QYrdr.wgp-1B.1L and QYr.cim-1BS. Both QTL were identified in hexaploid bread wheat ‘Druchamp’ and ‘Pastor,’ and are commonly present in bread and durum wheat backgrounds (Rosewarne et al., 2012; Hou et al., 2015).

Many APR QTL have been reported on the short arm of chromosome 2B (Boukhatem et al., 2002; Carter et al., 2009; Lan et al., 2010; Lowe et al., 2011; Chen et al., 2012; Mcintosh et al., 2014). Strong associations were detected in the Ethiopian durum wheat panel in the distal part of the short arm of chromosome 2B. A minor QTL for stripe rust resistance identified in the current study, QYrEDWL.par-2BS, resides in the chromosomal regions of QYr.inra-2BS. QYr.inra-2BS was detected in hexaploid bread wheat cultivar ‘Renan,’ and was stably detected for more than 4 years and expressed during earlier adult stages compared to other QTL characterized in ‘Renan’ (Dedryver et al., 2009).

On the long arm of chromosome 4A, represented by tag-SNP IWB2634, QYrEDWL-4AL was significantly associated with stripe rust resistance in two environments. In the same region, a recent study using the GWAS approach in a global winter wheat collection identified a stripe rust APR locus, QYr.wsu-4A.4 tagged by IWA4651 (Bulli et al., 2016). However, IWA4651 was neither significantly associated with any phenotype in the current population, nor in LD with IWB2634, indicating that QYrEDWL-4AL identified in the current study is different from QYr.wsu-4A.4.

QYrEDWL.par-4B.1, spanning the centromeric region of chromosome 4B, has been mapped in the same region of previously reported QYr-4B in French durum wheat ‘Sachem’ (Singh et al., 2013) and QYr.ufs-4B in South African spring wheat ‘Palmiet’ (Agenbag et al., 2012). QYr-4B, tagged by wPt-0872, was a pleiotropic QTL conferring effective resistance to both stripe rust and stem rust in the field. It is difficult to clarify the relationship between QYr-4B and QYrEDWL.par-4B.1 since durum wheat ‘Sachem’ was not included in the present Ethiopian durum wheat panel. Targeted comparisons of high-density mapped SNP profiles from the relevant genotypes ‘Sachem,’ ‘Palmiet’ and selected accessions of the Ethiopian tetraploid panel would help in clarifying the identity by descent relationships at the haplotype level. Additionally, in a worldwide GWAS panel of 232 elite durum wheat lines, IWB53059 was significantly associated to field resistance and fell into the mapping intervals of QYr-4B and QYr.ufs-4B, 4.8 cM away from QYrEDWL.par-4B.1 (unpublished data). On the distal long arm region of chromosome 4B, QYrEDWL.par-4BL.2, tagged by IWB74594 in this study, resides in the same region where another QTL for stripe rust resistance (QYr-4BL) has been identified in Israeli wheat ‘Oligoculm’ (Suenaga et al., 2003). QYr-4BL was consistently detected to have strong associations with leaf rust and stripe rust resistance across different environments and its effect on stripe rust resistance increased as plants advanced into adult stages.

Finally, QYrEDWL-5AL.1 identified in the current study was located within the genomic region of QYr.cim-5AL on chromosome 5AL. QYr.cim-5AL was a minor APR QTL derived from spring wheat ‘Avocet-YrA’ and effective in trials conducted in both Mexico and China (Lan et al., 2014). Whether QYrEDWL-5AL.1 and QYr.cim-5AL are the same APR locus needs further investigation.

Five loci, YrEDWL-1AS, YrEDWL-1BS.1, YrEDWL-3AS, YrEDWL-4BL, and YrEDWL-5BL on chromosome arms 1AS, 1BS, 3AS, 4BL, and 5BL were significantly associated with seedling resistance, but do not overlap with known seedling resistance genes, and therefore should be considered as newly discovered seedling resistance loci. Of particular interest, on the short arm of chromosome 3A, a stem rust seedling resistance locus, identified in an elite durum wheat panel using GWAS (Letta et al., 2014), may be linked to YrEDWL-3AS based on the consensus map distance of 5.0 cM between these loci. The seedling resistance to stem rust was tagged by wPt-1923 and significantly associated with seedling resistance to stem rust races TTTTF, TTKSK (Ug99) and JRCQC (Letta et al., 2014).

Six loci associated with resistance to Pst in field screening nurseries, QYrEDWL.par-1AS, QYrEDWL-1AL, QYrEDWL-1BL.2, QYrEDWL-2BS, QYrEDWL.par-3BL and QYrEDWL-5AL.2, are located at chromosome positions distinct from previously reported stripe rust resistance QTL. QYrEDWL-1BL.2 co-located with SSR marker barc81, which has been associated with seedling resistance to stem rust races TTTTF and TTKSK (Ug99) in a GWAS population of elite durum wheat (Letta et al., 2014). Further research to validate these loci and determine if these stem and stripe rust resistance genes are tightly linked, and their linkage phase, is warranted.

Collectively, GWAS based on this Ethiopian durum wheat population resulted in the detection of five newly documented seedling resistance loci and six newly documented field resistance loci. These newly documented loci represent 83.3 and 50.0% of total identified seedling and field resistance loci characterized in this GWAS. The historical isolation of Ethiopian durum wheat germplasm from the wheat currently under cultivation in other countries (Tsegaye and Berg, 2007; Mengistu et al., 2016), combined with regular pressure from wheat rusts in this epidemic hot-spot area of East Africa (Dawit et al., 2009; Wan et al., 2016) appears to have resulted in selection for unique diversity of stripe rust resistance genes.