Genetic analysis of resistance to stripe rust in durum wheat (Triticum turgidum L. var. durum)

Stripe rust, caused by the fungal pathogen Puccinia striiformis Westend. f. sp. tritici Eriks, is an important disease of bread wheat (Triticum aestivum L.) worldwide and there is an indication that it may also become a serious disease of durum wheat (T. turgidum L. var. durum). Therefore, we investigated the genetic architecture underlying resistance to stripe rust in adapted durum wheat germplasm. Wheat infection assays were conducted under controlled conditions in Canada and under field conditions in Mexico. Disease assessments were performed on a population of 155 doubled haploid (DH) lines derived from the cross of Kofa (susceptible) and W9262-260D3 (moderately resistant) and on a breeding panel that consisted of 92 diverse cultivars and breeding lines. Both populations were genotyped using the 90K single-nucleotide polymorphism (SNP) iSelect assay. In the DH population, QTL for stripe rust resistance were identified on chromosome 7B (LOD 6.87–11.47) and chromosome 5B (LOD 3.88–9.17). The QTL for stripe rust resistance on chromosome 7B was supported in the breeding panel. Both QTL were anchored to the genome sequence of wild emmer wheat, which identified gene candidates involved in disease resistance. Exome capture sequencing identified variation in the candidate genes between Kofa and W9262-260D3. These genetic insights will be useful in durum breeding to enhance resistance to stripe rust.


Introduction
Durum wheat (Triticum turgidum L. var. durum) is an important food crop in regions with relatively dry climates. Worldwide, 21 countries produce durum wheat across an average area of approximately 18 million hectares, with an annual production of approximately 35 million tonnes (Mt) [1]. Canada is the second largest producer of durum wheat in the world, with already widely used in bread wheat, should be the focus in durum wheat breeding. Thus, efforts are needed to identify new sources of resistance for durum wheat, including genes that can be combined or pyramided for durable resistance.
Combining multiple resistance genes into breeding lines can be difficult if based exclusively on phenotypic selection, especially when all-stage resistance genes are involved. As an alternative to phenotypic selection alone, molecular markers that are tightly linked to resistance genes are used for stacking resistance genes [21]. Although numerous studies have investigated and developed genetic markers for stripe rust resistance, research has primarily focused on hexaploid wheat and much less is known about the genetic basis of stripe rust resistance in durum wheat or from other tetraploid sources. Among the 78 officially named resistance genes to date [22], roughly seven were detected in durum wheat [22][23][24].
The aim of this study was to identify additional stripe rust resistance genes in modern durum wheat germplasm that could be used in breeding programs to develop more durable resistance. The 90K SNP iSelect assay made it possible to genotype a large number of accessions and simultaneously investigate the genetic architecture of stripe rust resistance in durum wheat. We performed linkage analysis on a bi-parental doubled haploid (DH) mapping population from the cross of Kofa (susceptible) and W9262-260D3 (moderately resistant), which was evaluated for seedling resistance against two Pst isolates. QTL from the DH population were also tested in a breeding panel consisting of 92 diverse durum cultivars and breeding lines.

Populations
Two populations were used in the present study; the first was a bi-parental population consisting of 155 DH lines, derived from the cross between the susceptible parent Kofa and the moderately resistant parent W9262-260D3 (Kyle Ã 2/Biodur) [25]. The second population consisted of a breeding panel of 92 elite cultivars and breeding lines collected from 13 countries, representative of the major durum wheat breeding programs of the world (S1 Table). In addition, three stripe rust susceptible checks (Avocet, Brigade and DT749) and two resistant checks (DT546 and Lillian) were included in the panel.

Collection of Pst races
Two isolates (W009 and W015) and a field collection (FC) of stripe rust were provided by the Cereal & Flax Pathology Laboratory at the Crop Development Centre, University of Saskatchewan. W009 was isolated in Richardson, SK (50˚24'N, 104˚29'W) in 2011 and W015 was isolated in Lethbridge, AB (49˚43'N, 112˚48' W) in 2010 [13]. These isolates were propagated to obtain inoculum by infecting susceptible wheat plants (cv. 'Avocet S') with urediniospores that were obtained from a single pustule. Brar and Kutcher [13] reported that W009 and W015 are genetically uniform isolates with different avirulence / virulence formulae. Isolate W009 is race C-PST-2, which is the second most common race among 59 isolates collected from 2005 to 2013 in western Canada. Isolate W015 is race C-PST-30, and was determined previously to have the widest spectrum of virulence among western Canadian isolates when evaluated on the Avocet differential set [13].
The FC for the seedling assays was collected from naturally infected susceptible spring wheat lines in Lethbridge, Alberta in 2011, and the composition is unknown. Stripe rust urediniospores were collected from multiple infected wheat leaves, combined, and stored in a freezer at -80˚C. The FC was also propagated on the susceptible cultivar 'Avocet S' to obtain additional inoculum.

Stripe rust resistance evaluation
Evaluation of seedling stripe rust resistance was performed in an environmentally controlled growth chamber with a diurnal temperature cycle of 18˚C in darkness and 22˚C in light, with a photoperiod of 8 h of darkness and 16 h of light. Seedlings of both populations were inoculated at the two-leaf stage (approximately 10 days after planting). Inoculations were made using urediniospores suspended in mineral oil (Bayol1, Esso Canada, Toronto, ON) at a concentration of 0.01 g of urediniospores per 900 μl mineral oil. Inoculated seedlings were left to dry and transferred to a high humidity growth chamber (10˚C) and kept in darkness for 24 h. Seedlings were then moved to a growth chamber with a temperature of 10˚C in darkness and 15˚C in light, with a photoperiod of 8 h of darkness and 16 h of light. The DH population was inoculated with spores of W009 and W015 as independent experiments, while the breeding panel was inoculated with W009, W015, and FC, as independent experiments. The DH population and the breeding panel were arranged in an alpha-lattice design with three replications and four seedlings per replication (S2 Table). Ten to eighteen days post-inoculation, the second leaf was evaluated for disease based on the 0 (resistant) to 9 (susceptible) scale rating of infection type (IT) [12,18]. The IT was scored twice approximately two days apart, and scores were analyzed independently.
The DH population was also evaluated for stripe rust resistance in replicated field trials in Mexico in 2014. Similarly, the breeding panel was evaluated in Mexico over two consecutive years, 2013 and 2014. The experiments were conducted in Toluca (19˚17'N, 99˚39'W, 2,680 m above sea level, sandy clay loam soil). A randomized complete block design (RCBD) with three replications was employed in each field test (S3 Table). The experimental lines were planted in single row plots (1.5 m) with 0.5 m space between rows, with approximately 30 plants per plot. Susceptible spreader rows were planted around the experimental plots to facilitate the spread of the disease. Both spreader rows and experimental plots were artificially inoculated with a mixture of stripe rust races with the widest virulence spectrum in Mexico. Disease severity (DS) in the field trials was evaluated for the flag leaf based on the modified Cobb scale, on a per plot basis [26]. The first disease evaluation in 2014 was performed when the flag leaves of susceptible checks reached 40% severity. For both 2013 and 2014, subsequent ratings were performed every five days with the final rating performed when the susceptible checks reached 100% severity, at approximately Zadoks growth stage (GS) 55 [27]. The area under the disease progress curve (AUDPC) for the field data collected in Mexico was used to obtain an estimate of disease accumulation on each plot. The AUDPC was calculated using the following formula; where, n is the total number of ratings, Y i is the stripe rust severity for the i th rating and T i is the day of the i th rating [28].
The disease evaluation data was analyzed using PROC MIXED in SAS V9.3 with durum accessions as a fixed effect, while replications (Rep), blocks and interacting factors were considered random. The least square means (LSMeans) for the stripe rust disease ratings were calculated using LSMEANS in SAS V9.3 for both growth chamber and field experiments [29]. The deviations of observed and expected frequencies of resistant individuals in the DH population were tested using the Pearson's chi-squared test.

DNA extraction, genotyping, and construction of the genetic map for the DH population
Genomic DNA was extracted from leaves of one-week-old seedlings using the cetyltrimethylammonium bromide (CTAB) protocol [30]. Standard gel electrophoresis, using a 1.5% (w/v) agarose gel with known size standards, was used to evaluate the quality and integrity of the DNA samples. DNA was quantified by the PicoGreen fluorometric assay [31]. Each sample was genotyped using 500 ng of DNA in 10μl as input for the Wheat 90K iSelect SNP assay [32]. Genotypic data was analyzed by the genotyping module of Illumina GenomeStudio data analysis software GSGT V1.9.4 (Illumina, San Diego, CA). For the DH population, in addition to the 10,641 SNP markers from the 90K SNP Infinium iSelect assay [32], 109 SSR and 125 DArT markers [25] were analyzed. The polymorphism information content (PIC) of both SNP and SSR markers were determined by PowerMarker V3.25 [33]. A genetic map of the DH population was constructed using both MSTMap [34] and MapDisto [35], as described previously [36]. For the breeding panel, conservative filtering (S4 Table) was applied to filter out noninformative SNPs. There were 244 simple sequence repeat (SSR) markers that were also used in the analysis of the breeding population [37].

QTL analysis
Composite interval mapping (CIM) was performed by Qgene V4.3.10 [38], using the genetic map of the DH population. We used a stepwise cofactor selection method, where the 'maximum number of cofactors', 'F-to-add', and 'F-to-drop' thresholds were set as 'auto'. One thousand permutation tests were performed to estimate the critical logarithm of the odd (LOD) threshold. Significant QTL were illustrated with diagonally hatched bars using MapChart V2.2 [39]. The additive and epistatic effects of QTL were investigated using SAS V9.3.

Confirmation of QTL from the DH population in the breeding panel
Using data from the breeding panel, the linkage disequilibrium (LD) between pair-wise markers was calculated among SNPs with known genetic positions according to the SNP-based consensus map of durum wheat [40]. Using TASSEL V3.0, LD was measured using the squared correlation between loci (r 2 ) and was plotted against genetic distance between adjacent markers. The LD decay against genetic distance was simulated in a nonlinear regression model [41]. The critical r 2 value referred to the 95% quantile of r 2 values between unlinked SNP markers (i.e. markers that were localized to different chromosomes).
The phylogenetic tree was constructed using allele frequency-based distances between accessions. Population structure within the breeding panel was estimated using principal component analysis (PCA) using TASSEL V3.0 [42], and Bayesian clustering analysis was performed using STRUCTURE V2.3.4 and Structure Harvester [43] to determine the coancestry coefficient (Q matrix). The Bayesian clustering analysis was applied to SSR markers with high PIC identified previously [37]. The Q matrix was used as a covariate in the general linear model (GLM) or mixed linear model (MLM). The kinship matrix (k matrix) was estimated using genetic data in TASSEL V3.0, and incorporated as a random effect in the MLM.
Association analyses were performed using three different models in TASSEL V3.0; these included a naïve model (GLM without any correction for population structure), GLM (with Q matrix as a correction for population structure) and MLM (with Q and P matrix as corrections for population structure) [44]. The model that best corrected for bias in population structure and kinship was chosen based on minimizing systematic inflation or deflation of P-values using the Quantile-Quantile (Q-Q) plot. The P-value was adjusted using a positive false discovery rate (pFDR) method. When the MLM was employed, an 'FDR Q-value' was estimated using the R package 'fdrtool' and Q 0.05 was used as threshold to determine significant associations [45,46].

Physical anchoring of QTL and exome sequencing
The genome sequence for wild emmer wheat (WEW) was used to determine the physical interval of the QTL. Markers were compared to the genome by GMAP v2014-12-29 [47] and filtered for matches with 95% sequence identity and 80% coverage. Genes within the intervals were extracted from the available gene annotations of WEW. For exome sequencing, DNA from Kofa and W9262-260D3 were enriched for coding regions using the wheat exome capture array according to the procedures outlined previously [48]. High-throughput sequencing was performed on the Illumina HiSeq2500 platform with 2 x 100 bp PE chemistry. Raw sequence reads were deposited in the NCBI Sequence Read Archive (Accession: SRP154228). Reads were processed in Trimmomatic v0.32 [49] and aligned to the genome of WEW using Novoalign v3.02.05. Duplicate read mappings and improper read pairs were removed using samtools v1.3.1 [50] and picard-tools [51]. Sequence variations were determined using freebayes v1.0.2-16-gd466de [52], with ploidy set to 4. Predictions on the effect of the variants on gene function were determined within high confidence gene models by SnpEff [53].

Genetic mapping for stripe rust resistance in the DH population
The two parents of the DH population had clear differences in their phenotypic response to stripe rust infection. In the seedling stage, the 1 st and 2 nd leaves of the moderately resistant parent W9262-260D3 had few uredia that were surrounded by chlorotic/necrotic areas, which were staggered on the leaf surface (IT 3-4). In contrast, both leaves of the susceptible parent Kofa were covered with many uredia, which showed no necrosis or chlorosis (IT 6-7). There was no observable difference in the number of uredia between isolates W009 and W015. The IT frequency distribution approximated a bimodal distribution in the DH population, with a range of 1-8 for isolate W009 and 2-8 for isolate W015 (Fig 1A and 1B). The frequency distribution of resistance to W009 had two peaks, with 17.1% of individuals with IT ranging from 2 to 3, and 71.9% of individuals with a more susceptible IT range of 6 to 8 ( Fig 1A). The frequency distribution of resistance to W015 also had two peaks, with 20.0% of individuals with IT ratings that ranged from 3 to 4, and 61.3% of individuals with IT ratings that ranged from 5 to 7 ( Fig 1B). Transgressive segregation for seedling resistance was also observed for the DH population. When inoculated with W009, 18.7% of the DH population had a higher level of resistance than W9262-260D3, while 23.2% were more susceptible than Kofa; similarly; for W015, 12.3% of the DH population were more resistant than W9262-260D3, while 35.5% were more susceptible than Kofa.
The distribution of stripe rust DS assessed in field experiments in Mexico in 2014 is shown in Fig 1C. The moderately resistant parent W9262-260D3 had some uredia on the surface of the flag leaf (DS of 10-20%), while susceptible parent Kofa had uredia on half of the flag leaf area (DS of 50%). DS of the DH population ranged from 0 to 90%; however, DS ranged from 10% to 50% for 87.7% of the population. Similar to the seedling reactions, the DH population also showed transgressive segregation for adult plant resistance to stripe rust (Fig 1). For adult plant resistance in Mexico in 2014, 5.8% of the DH population were more resistant than W9262-260D3, whereas 6.5% were more susceptible than Kofa.
In order to identify QTL associated with the resistance in the DH population, we first constructed a genetic map using genotypic data from the 90K iSelect SNP array. The genetic map spanned 2,639.7 cM with an average interval size of 0.64 markers/cM. Gaps larger than 10 cM were found on chromosomes 1A, 1B, 3A, 3B, 4A, 5A and 6A (S5 Table). Two QTL for seedling resistance to isolates W009 and W015 were then identified by CIM (permutation P < 0.05), QYr.usw-5B and QYr.usw-7B (Fig 2; Table 1; S1 Appendix). The moderately resistant parent W9262-260D3 carried the resistant alleles for both QYr.usw-5B and QYr.usw-7B. The QTL positions also coincided perfectly for both fungal isolates. QYr.usw-7B had the largest effect and was also identified in the field reaction to the Mexican races in Toluca in 2014 (Fig 2, Fig 3C and Table 1). QYr.usw-7B was flanked by markers BS00003929 and BS00075300_51 with an interval of 2.9 cM; the peak marker, BS00075300_51, was located at 222.5 cM on chromosome 7B. QYr. usw-5B was identified in seedling experiments, but was not detected in field trials against the Mexican races in 2014. QYr.usw-5B was flanked by markers RAC875_c38873_1118 and wsnp_Ku_c4427_8029592 with an interval of 1.3 cM; the peak marker, wsnp_Ku_c4427_8029592, was located at 225.3 cM on chromosome 5B. Based on seedling stripe rust resistance to W009 and W015, there was a significant epistatic interaction between QYr.usw-5B and QYr.usw-7B (P < 0.01). The two-QTL interaction explained 12.7% of seedling resistance to W009 and 17.1% for W015 (Fig 3A and 3B).

Disease reaction and testing of QYr.usw-5B and QYr.usw-7B in the breeding panel
Phenotypic variation was observed in the infection assays for the breeding panel. For the seedling assays, the IT ranged from highly resistant (IT < 2) to highly susceptible (IT > 7) (S6 Genetic analysis of resistance to stripe rust in durum wheat (Triticum turgidum L. var. durum) Table). The IT LSmeans for the disease assessment across the entire breeding panel ranged from 1.4 to 9.1 (S6 Table). The correlation among the infection assays using FC, W009 and W015 was high, with Pearson's correlation coefficients larger than 0.77 (S7 Table). Under field  Table). DS and AUDPC were highly correlated between the two years (r 2 ! 0.90). The adult plant resistance observed in 2013 and 2014 was moderately correlated with seedling resistance, with correlation coefficients ranging from 0.58 to 0.71. Genetic analysis of resistance to stripe rust in durum wheat (Triticum turgidum L. var. durum) The 92 accessions were clustered into three subpopulations (SP1, SP2 and SP3) based on geographic distribution, phylogenetic relationship, and Bayesian clustering (S1 Fig) [54]. The first subpopulation, SP1, was comprised of 33 varieties from diverse origins, SP2 of 18 Italian accessions, and SP3 of 41 varieties from North America (Fig 4A). The LSmeans (from seedling and field experiments) across all cultivars within a subpopulation then were compared. The LSmeans for stripe rust resistance of SP3 (North American accessions) were significantly (P < 0.05) more susceptible compared to SP1 and SP2 (Fig 4A and 4B). The accessions in SP1 and SP2 had a more balanced distribution for resistance and susceptibility against stripe rust, and the average DS was not significantly different between SP1 and SP2. Similar trends were observed in the seedling and adult plant resistance.
To facilitate QTL testing, the breeding population was genotyped with the 90K iSelect SNP array (S9 Table; S2 Fig; S1 Appendix). The major locus QYr.usw-7B from the DH population was identified within a 3.0 cM region near the telomere of chromosome 7B in the breeding panel. Fifteen SNPs and one SSR marker were associated with stripe rust in some or all of the datasets (S10 Table). The proportion of phenotypic variance explained by the QTL when inoculated with W009, W015, and the FC ranged from 17.1% to 32.3% for seedling resistance, while for field resistance, it ranged from 11.6% to 22.6%. The QYr.usw-5B QTL identified in the DH population was not found in the breeding panel. The alleles at the marker wsnp_Ku_ c4427_8029592 were nearly fixed in most lines in this panel. Ninety-five percent of the lines studied carried the same allele as W9262-260D3 (resistant genotype from the DH population) at marker wsnp_Ku_c4427_8029592 (peak marker of QYr.usw-5B) and five carried the same genotype as Kofa (susceptible genotype from the DH population). Lines carrying the susceptible genotype were all highly susceptible to stripe rust at all stages. Average disease scores placed these five individuals within the top 25% of the most susceptible lines; Mexa (IT = 6.7), Kofa (IT = 6.9), Westbred (IT = 7.2), Ocotillo (IT = 7.3) and Pathfinder (IT = 7.6). Mexa, Kofa, Westbred and Ocotillo also carried the susceptible genotype (same as Kofa) at the QYr.usw-7B locus. The only exception was Pathfinder, which had the resistant genotype (same as W9262-260D3) at QYr.usw-7B, but was still among the most severely affected by stripe rust among these five genotypes.

Identification of gene candidates
Gene candidates for QYr.usw-5B and QYr.usw-7B were identified by mapping the markers for the QTL to the available genome sequence of WEW. QYr.usw-5B mapped to 3.63 Mbp interval ranging from 681.19 Mbp (RAC875_c38873_1118) to 684.82 Mbp (wsnp_Ku_c4427_8029592) on chromosome 5B. This interval contains 67 genes, of which 36 have high confidence annotations ( Table 2). Two of these genes, TRIDC5BG077260and TRIDC5BG077630,are annotated as leucine-rich repeat receptor-like protein kinases, which have a role in pathogen recognition and disease resistance. Similarly, QYr.usw-7B mapped to a very small 267 kbp interval ranging from 719.49 Mbp (BS00075300_51) to 719.75 Mbp (BS00003929_51) on chromosome 7B. This interval contains seven genes, of which only four have high confidence annotations ( Table 2). Two of the genes within QYr.usw-7B, TRIDC7BG070820and TRIDC7BG070830,are annotated as disease resistance proteins with nucleotide-binding site leucine-rich repeat domains.
Exome sequencing identified 642 sequence variants, SNPs and small insertions or deletions (InDels), between Kofa and W9262-260D3 in QYr.usw-5B and QYr.usw-7B (S2 Appendix). While most of the variants identified were in intergenic regions and introns, which may have a role in regulating gene expression, 78 were within the coding sequences. Fifteen genes within QYr.usw-5B had sequence variation in their coding sequence, including InDels in the genes TRIDC5BG077150(IAA-amino acid hydrolase ILR1-like 4) and TRIDC5BG077600(unknown function) that are predicted to cause a shift in the reading frame and have a major effect on gene function. Variation was also detected within three genes from QYr.usw-7B, including predicted disease resistance genes TRIDC7BG070820and TRIDC7BG070830,which had four missense mutations each (S2 Appendix). All four missense variants for TRIDC7BG070820 occurred within InterPro predicted leucine-rich repeat or nucleotide binding domains. Two out of four missense variants for TRIDC7BG070830occurred in InterPro predicted leucinerich repeat or nucleotide binding domains, while the remaining two were part of the leucinerich repeat unintegrated signature (S2 Appendix).

Discussion
Stripe rust, a destructive disease of wheat occurring worldwide, can cause complete yield loss in extreme cases. In recent years, regional epidemics have occurred in North America (particularly the Pacific North-West), East Asia, South Asia, Australia, and East Africa [3]. In Canada, stripe rust has appeared more frequently in regions east of the Rocky Mountains, with epidemics occurring in southern Alberta and Saskatchewan in 2006, 2010 and 2011 [6]. Given the increased incidence of stripe rust and the limited number of effective disease resistance genes characterized in durum wheat, additional effort is required to identify new resistance genes that can be combined to mitigate the breakdown of currently available resistance [3,19,20]. In this study, we identified QTL on chromosomes 5B and 7B that confer resistance to stripe rust in durum wheat. The QTL on chromosome 5B was detected only in the DH population at the seedling stage. A second QTL, on chromosome 7B, was identified through QTL mapping of both seedling evaluations under controlled conditions and adult plant field trials in the DH population, and was also identified in global breeding lines.

Genomic regions associated with stripe rust resistance on chromosome 7B
The effect of QYr.usw-7B on chromosome 7B in the DH population was confirmed through the discovery of a QTL at the same position in our breeding panel (S10 Table). The experimental evidence presented here suggests that this QTL contains a broad-spectrum resistance gene that is effective against multiple races of stripe rust, and is effective at all stages of plant development. QYr.usw-7B maps to the region on chromosome 7B known to harbor Yr67 [50,55] and YrZH84 [56]. According to previous studies, 78 officially named resistance genes (Yr1 to Yr78) and many temporarily designated genes (www.ars.usda.gov and www.shigen.nig.ac.jp) have been identified. Of the resistance genes currently known, Yr2, Yr6, Yr39, Yr52, Yr59, YrZH84, Yr67 and YrMY37 are located on chromosome 7B. The genes Yr39, Yr52 and Yr59 are adult plant resistance genes, which are not effective at the seedling stage and are therefore unlikely candidates for QYr.usw-7B [57]. Although Yr2, Yr6, Yr67, YrZH84 and YrMY37 could qualify as candidates because of their effectiveness at all growth stages, the differential lines carrying Yr2 and Yr6 were not effective against W009 and W015 [13]; therefore, QYr.usw-7B is functionally distinct from Yr2 and Yr6. YrMY37 was also ruled out as a candidate corresponding to the QTL we investigated because it is located on chromosome 7B close to the centromere [58], while QYr.usw-7B is near the distal end of chromosome 7B. Yr67 (previously named YrC591) [50,55] and YrZH84 [56] are also all-stage dominant resistance genes mapped to the telomeric region of chromosome 7B. Lines carrying Yr67 and YrZH84 had different reactions to a panel of Pst isolates, indicating they are in fact distinct resistance genes [59]. Both genes share one significant flanking SSR marker with QYr.usw-7B, namely, Xcfa2040-7B. The relationship among QYr.usw-7B, Yr67 and YrZH84 needs further study, with allelism testing to estimate the genetic distance among them. However, both Yr67 and YrZH84 are outside of the physical interval for QYr.usw-7B, indicating that QYr.usw-7B may be a different gene (S3 Fig). The physical interval for QYr.usw-7B is very narrow and contains two genes annotated as pathogen receptors, each of which contained four missense mutations between Kofa and W9262-260D3 (Table 2, S2 Appendix). Additional research is required to validate the role of these gene candidates and the SNPs we identified, including their potential involvement in stripe rust resistance. Furthermore, linkage between QYr. usw-7B and other genes on chromosome 7B, including genes for yellow pigment and resistance to leaf rust, should also be investigated for breeding purposes [60].

Genomic regions associated with stripe rust resistance on chromosome 5B
The QTL on chromosome 5B, QYr.usw-5B, was found to confer resistance to Pst isolates W009 and W015 in the Kofa/W9262 DH population. The QTL was flanked by RAC875_c38873_1118 and wsnp_Ku_c4427_8029592, with the peak situated nearest to the latter marker. The QTL explained 25.1% and 11.0% of phenotypic variance in seedling resistance to W009 and W015, respectively. QYr.usw-5B showed strong epistatic interaction with QYr.usw-7B in the DH mapping population but was not identified in the breeding panel. Upon closer inspection, it was found that the susceptible allele at marker wsnp_Ku_c4427_8029592, the peak marker for QYr. usw-5B, was nearly absent in the breeding panel, which may have limited our ability to detect the QTL. The five lines that carried the susceptible allele all had severe disease ratings in all trials. The reduced representation of this allele in such a diverse set of lines suggests that breeding programs have selected for the resistant allele at this locus. In previous studies, seedling resistance genes Yr19, Yr47, and Yr74 were also found to be located on chromosome 5B. Yr47 was identified as a seedling resistance gene located on chromosome 5BS [61], whereas QYr.usw-5B is located on 5BL. Recently, Yr74 was identified as an all stage resistance gene in the Australian hexaploid wheat Avocet R (AvR)-AUS 90660; however, there is no information regarding this gene in durum wheat and the assocciated DArT-Seq markers are unavailable for comparative analysis to QYr.usw-5B [62]. Yr19 is a dominant resistance gene located on chromosome 5B discovered in an F 2 population from crosses of disomic aneuploidy lines of Chinese Spring [63], but no marker information is available for Yr19. The relationship between QYr.usw-5B, Yr74, and Yr19 is unknown at this time and warrants further investigation. Comparison of QTL across populations can be facilitated by using a common set of markers or a common reference genome sequence. Our study provides a physical location for QYr.usw-5B in an available tetraploid reference genome and lists gene candidates from the interval as well as sequence variation within genes and intergenic regions that may be involved in stripe rust resistance. However, the sequence variation identified in this study using exome capture sequencing does not fully capture all genes and their variants; therefore, additional variants within genes and intergenic regions likely exist between Kofa and W9262-260D3 for both QYr. usw-5B and QYr.usw-7B that could be identified by a more comprehensive sequence strategy. Additional research is also required to determine if the variation within these gene candidates are involved in stripe rust resistance.

Interaction between rust resistance QTL on chromosomes 5B and 7B
The full expression of resistance in the DH population requires both QYr.usw-5B and QYr.usw-7B (Table 1). The statistical analysis indicated that the epistatic interaction between QYr.usw-5B and QYr.usw-7B was significant (P < 0.01), and genes are most effective when stacked. Epistatic interactions that involve rust resistance genes have been documented in other studies. For example, an epistatic interaction was identified between a QTL for stripe rust resistance on chromosome 2AS and QTL on chromosome 6AL, which resulted in some inbred lines that had resistance that was equal to or greater than the resistant parent [64]. Similarly, an epistatic interaction was identified between QTL that conferred resistance to stem rust resistance in a DH population derived from the Canadian wheat cultivars AC Cadillac and Carberry [65]. Epistatic interactions were also observed between multiple pairs of QTL in other studies, including QTL on chromosomes 4B and 5B, 4B and 7B, 5B and 6D, 6D and 3B, and 6D and 7B. In research conducted by Yu et al. [66] on stem rust resistance of CIMMYT spring wheat, multiple significant pairwise QTL interactions were detected. The Sr2 locus on chromosome 3BS and the wPt1859 locus on 7DL interacted with other loci on the same chromosome and with markers on chromosome 6B. Interactions also involved loci on chromosomes 1B, 4A and 2B. Yu et al. [67] also conducted research on stem rust resistance of winter wheat to Ug99, and showed that multiple loci were involved in the QTL interaction, including loci on chromosomes 3BS, 6BS, 2BS and 7DS. This suggests that complex genetic control for adult plant resistance to stem rust isolate Ug99 exists in winter wheat. Kumar et al. [68] studied leaf rust and stripe rust resistance in the International Triticeae Mapping Initiative (ITMI) population and identified eight QTL interactions for each trait, as well as and epistatic-QTL interaction between loci on chromosomes 1D and 3B. Together, epistatic interactions between QTL relating to resistance to rust pathogens is common, and the interaction we identified between QTL on chromosomes 7B and 5B will be actively incorporated into breeding programs to enhance resistance to stripe rust in Canadian germplasm.

Germplasm from the breeding panel can be used in future crosses to enhance resistance to stripe rust
Genetic exchange among breeding programs is critical to widen the genetic base of any program in general, and to enhance stripe rust resistance in particular.

Conclusions
We used genetic mapping to detect two QTL (QYr.usw-5B and QYr.usw-7B) that conferred stripe rust resistance in a DH population, one of which (QYr.usw-7B) was also identified in a diverse breeding panel. The physical locations of the QTL were determined in the WEW genome and candidate resistance genes and genetic variations within the interval were identified. Within the DH population, a strong epistatic interaction was observed between QYr.usw-5B and QYr.usw-7B. Combining the major QTL from this study with other effective resistance genes via marker assisted selection could be applied in future breeding programs to develop durable resistance to stripe rust. The 90K wheat SNP assay greatly facilitated the identification of the QTL in the DH population and provided a standard set of SNP markers that could be used to pyramid the QTL we investigated with those identified from other studies.