Genomic exploration of durable wheat rust resistance by integrating data from multiple worldwide populations

Abstract

Global wheat (Triticum aestivum L.) production faces significant challenges due to the destructive nature of leaf (Puccinia triticina; leaf rust [Lr]), stem (Puccinia graminis; stem rust [Sr]), and stripe (Puccinia striiformis; stripe rust [Yr]) rust diseases. Despite ongoing efforts to develop resistant varieties, these diseases remain a persistent challenge due to their highly evolving nature. Overcoming these challenges requires the identification and deployment of genetically diverse resistance genes in future cultivars. This study explored durable resistance against rust diseases by integrating data from five global populations. The populations exhibit diverse origins and were phenotypically evaluated in 16, 13, and 19 global field experiments, with total phenotypic observations of 12,694, 10,725, and 16,281 for Lr, Sr, and Yr, respectively. Field experiments showed moderate heritability of 0.43, 0.62, and 0.41 for Lr, Sr, and Yr, respectively. Genetic correlations were moderate among experiments for the same disease (0.34–0.59), but low among the three diseases (<0.21). The meta‐genome‐wide association studies (metaGWAS) analysis identified 19 quantitative trait loci (QTLs) associated with the resistance to Lr, 17 with the resistance to Sr, and five with the resistance to Yr. Six QTLs controlling resistance to more than one rust disease were also identified. Additionally, the study unveiled 13 potentially new QTLs (five for Lr and Yr each and three for Yr), contributing valuable insights into the genetic basis of wheat rust resistance. The integration of diverse populations and environments through metaGWAS enhanced the detection of stable QTL. This research provides breeders with additional resistance loci to combat rust pathogens.

Core Ideas

• Integrated data from five global populations to study wheat rust resistance.

• Identified several quantitative trait loci (QTLs), of which six control resistance to more than one rust disease.

• MetaGWAS enhanced detection of stable QTL across diverse environments.

• Provides breeders with new resistance loci to combat wheat rust pathogens.

Plain Language Summary

Wheat rust diseases, including leaf, stem, and stripe rust, pose significant threats to global wheat production. This study aimed to identify durable resistance genes against these diseases by analyzing data from five global wheat populations. Researchers discovered 19 genes for leaf rust, 17 for stem rust, and five for stripe rust, with six genes showing resistance to more than one rust disease. These findings provide valuable genetic information to help breeders develop wheat varieties that are more resistant to rust diseases, ensuring better wheat production worldwide.

Abbreviations

BLUE

best linear unbiased estimator

GWAS

genome‐wide association study

LD

linkage disequilibrium

Lr

leaf rust

PCA

principal component analysis

QTL

quantitative trait locus

REML

restricted maximum likelihood

SNP

single‐nucleotide polymorphism

Sr

stem rust

Yr

stripe rust

1. INTRODUCTION

The occurrence of rust diseases can severely limit wheat yield. Leaf rust (Lr) (caused by Puccinia triticina), stem rust (Sr) (P. graminis), and stripe rust (Yr) (P. striiformis) are considered to be the most destructive fungal diseases of wheat globally (R. P. Singh et al., ²⁰¹⁵). These pathogens can lead to 100% yield loss in extreme cases, especially when infections occur early in development on highly susceptible cultivars (Afzal et al., 2007; Olivera Firpo et al., ²⁰¹⁵; R. P. Singh et al., ²⁰¹¹). In response to these challenges, wheat breeders use various genetic strategies in coordination with integrated disease management to mitigate the impact of these diseases, thus sustaining regional and global wheat production (Wellings, 2011). These strategies include the use of fungicides, crop rotation, and the development of resistant wheat varieties using traditional and molecular techniques (Khan et al., 2013). Nevertheless, rust diseases continue to pose a significant threat to global wheat production due to their ability to mutate and evolve to overcome resistance (McIntosh et al., 1995). Therefore, continuous research and monitoring are essential to stay ahead of these threats to ensure food security (Chen et al., 2002).

Resistance to rust diseases is primarily conferred by resistance genes that can be broadly classified into two categories: major and minor resistance genes (R. P. Singh et al., ²⁰¹⁶). Major resistance genes are often race‐specific but condition high level of resistance, whereas minor resistance genes impart partial resistance individually (Kolmer et al., 2009). However, combinations of major and minor genes confer higher levels of resistance (Silva et al., 2015). Unlike major genes, minor genes are less likely to break down; thus the resistance is durable, despite some reports of the emergence of pathotypes virulent against adult genes (Park & McIntosh, 1994; Sørensen et al., ²⁰¹⁴). The identification and deployment of both major and minor resistance genes is a key focus of current wheat breeding programs.

Identifying rust resistance genes, especially those of minor effect, is a first step toward establishing durable resistance against wheat rust diseases by accumulating diverse resistance alleles (Park, 2007). However, these genes are often difficult to detect due to their small effects (Evangelou & Ioannidis, 2013) and limited phenotypic and genotypic wheat rust data available on populations of adequate size from a single source. In this context, large global populations that are integrated from multiple sources provide the scale required (Hayes et al., 2009). Large populations increase the power to detect minor genes, which are often overlooked in smaller populations (Hou et al., 2021). However, while creating such large populations can be challenging due to logistical and resource constraints, they can be collected from different data holders (Jighly et al., 2022). MetaGWAS (meta‐genome‐wide association studies) is a powerful tool that allows the integration of summary statistics inferred from different populations to create a large and a diverse population (Joukhadar & Daetwyler, 2022). This approach increases the statistical power needed to detect minor genes, thus allowing investigation of complex genetic architectures (Bolormaa et al., 2014). The combination of rust resistance gene detection, the use of large global populations, and the application of metaGWAS represents a promising strategy for achieving durable resistance to rust diseases and ensuring the sustainability of global wheat production.

The present study collected phenotypic data on the three rust diseases from five different global populations genotyped with the Illumina iSelect 90k single‐nucleotide polymorphism (SNP) bead chip array. The populations originated from diverse origins and were phenotyped in geographically different areas of the world. Summary statistics (SNP effects and their standard errors) were calculated for each experiment independently and then integrated for each disease (Lr, Sr, and yellow rust) across all experiments, and then metaGWAS used to calculate a global p‐value for each SNP.

2. MATERIALS AND METHODS

2.1. Collected data

Five populations genotyped with Illumina iSelect 90k SNP bead chip array and scored for resistance to different rust diseases across diverse global locations were studied (Table 1). The five populations were selected due to their diverse origins and genetic backgrounds, which is crucial for identifying novel and durable rust resistance genes and for enabling comprehensive genetic analysis. Furthermore, their selection was based on their use of the same genotyping platform and public availability or generation by the authors, facilitating integrated analysis. The first population (AVR) was previously published in Joukhadar et al. (2020) and comprised 2300 worldwide wheat accessions, including 615 obtained from the Australian Grains Genebank, 552 landraces from the Watkins collection, 792 from the USDA National Small Grains Laboratory, and 341 BC₁F_4‐6 synthetic derivatives developed by crossing the Australian wheat varieties Annuello, Correll, or Yitpi to different primary synthetic lines. The population was phenotyped in 2014 and 2015 at five Australian locations. Lr and Sr were phenotyped at Cobbitty, NSW, and Horsham, VIC, and Carnarvon, WA, while Yr was phenotyped in the first two environments plus Manjimup, WA, and Wagga Wagga, NSW.

TABLE 1.

Number of trials and populations screened and analyzed for reaction to three rust diseases: Leaf (Lr), stem (Sr), and stripe (Yr).

Disease	Population	Trials	NoRecords	h 2	SD
Lr	AVR	6	5606	0.37	0.04
	SydU	5	4847	0.78	0.03
	TCAP	3	1832	0.19	0.06
	Kaz	2	409	0.35	0.09
	Total/mean	16	12,694	0.43	0.05
Sr	AVR	6	5397	0.43	0.04
	SydU	5	4919	0.80	0.03
	Kaz	2	409	0.45	0.09
	Total/mean	13	10,725	0.62	0.05
Yr	AVR	8	9521	0.43	0.04
	SydU	5	4852	0.71	0.04
	TCAP	3	1259	0.14	0.06
	Kaz	2	409	0.24	0.08
	China	5 a	240	0.57	0.08
	Total/mean	19	16,281	0.41	0.05

^a Only best linear unbiased estimator (BLUE) values over the five trials were available in Jia et al. (2020).

Abbreviations: NoRecords, the cumulated size of phenotyped individuals across all trials; h2, average single‐nucleotide polymorphism (SNP)‐based heritability across trial; SD, standard deviation.

The second population (USyd) was previously used in Joukhadar et al. (2021) and comprised 2468 lines phenotyped for the three rust diseases at Cobbitty, NSW, Australia, for five seasons between 2016 and 2020. The population included 1961 diverse lines developed at the Plant Breeding Institute, 457 from the International Maize and Wheat Improvement Center (CIMMYT) and the International Center for Agricultural Research in the Dry Areas, 36 Australian checks, and 14 miscellaneous lines, including parents of different bi‐parental mapping populations.

A collection of another 216 cultivars (Kaz) and breeding materials (Table 1) was also used in the present study, including 91 cultivars from Kazakhstan and Russia, 38 cultivars from Europe, and 86 cultivars and lines from Kazakhstan, Russia, United States, Canada, Mexico, Germany, and Australia (Genievskaya et al., 2020). The population was phenotyped for three rust diseases in 2018 and 2019 at the Research Institute of Biological Safety Problems fields in Gvardeisky, South Kazakhstan.

A population (China) of 229 cultivars collected from geographically distant locations in China, representing the currently cultivated wheat in China, plus 10 elite lines obtained from CIMMYT as well as one Australian cultivar, were also screened (Jia et al., 2020). The lines were evaluated for Yr resistance in Pixian and Xindu in Sichuan province in China in 2015. These were also evaluated in Wuhan in Hubei province, China, in 2013, 2016, and 2018. However, only the best linear unbiased estimator (BLUE) values over the five locations were available in the Supporting Information Materials of Jia et al. (2020).

Finally, a population (TCAP) was downloaded from https://wheat.triticeaetoolbox.org/, which comprised 642 individuals (Table 1), of which a subset was phenotyped for Lr in St. Paul, MN, in 2011 and 2012, as well as Crookston, MN, in 2012. Another subset was phenotyped for Yr in 2014 at UC Davis, under both irrigated and drought conditions.

The final cumulated dataset comprised 12,694 phenotypic records for Lr, 10,725 for Sr, and 16,281 for Yr, representing a total of 16, 13, and 19 experiments for the three diseases, respectively (Table 1).

Core Ideas

• Integrated data from five global populations to study wheat rust resistance.

• Identified several quantitative trait loci (QTLs), of which six control resistance to more than one rust disease.

• MetaGWAS enhanced detection of stable QTL across diverse environments.

• Provides breeders with new resistance loci to combat wheat rust pathogens.

2.2. Principal component analysis (PCA), SNP‐based heritability, and genetic correlation

The PCA was calculated using PLINK V2 software (Purcell et al., 2007). The SNP‐based heritability of each disease resistance in each experiment, which is equivalent to the narrow‐sense heritability (h ²), was estimated using the restricted maximum likelihood model, implemented in the MTG2 software (Lee & Van der Werf, 2016). The computed heritability values were then averaged over all trials. Additionally, the bivariate model of MTG2 was employed to calculate the genetic correlations between each pair of traits across all field experiments. These values were averaged across experiments to obtain a comprehensive estimate of the genetic correlations within and among different disease resistances.

2.3. Linkage disequilibrium (LD) decay

We evaluated the LD decay by calculating the r ² value between each pair of SNPs located on the same chromosome using PLINK2 software (Chang et al., 2015). These r ² values were plot against the physical distance between the corresponding pair of SNPs. The critical r ² value, or background LD, beyond which physical linkage is more likely to cause the LD, was calculated considering the 95th percentile of all r ² values calculated between SNPs located on different chromosomes (inter‐chromosomal LD).

2.4. MetaGWAS

The initial phase of MetaGWAS analysis involves performing a univariate GWAS analysis for each phenotypic trait (rust resistance) in every environment. The BLUEs were examined using the mixed linear model, implemented in GEMMA software with the default parameters (Zhou & Stephens, 2012, 2014):

$$y=\mathrm{\mu }+X\mathrm{\beta }+\mathrm{\alpha }+\mathrm{\epsilon }$$

Here, y represents the BLUEs for all individuals, μ is the intercept, X represents the SNP genotypes, β is the substitution effect of the SNP, ε is the residual, and α is the random effects derived from the following multivariate normal distribution $\mathrm{\alpha }\sim \mathrm{MVN}(0,\mathrm{\lambda }{\mathrm{\tau }}^{-1}G)$; where λ is the ratio between variant components, ${\mathrm{\tau }}^{-1}$ is the variance of the errors, and G is the genomic relatedness matrix calculated as per Yang et al. (2010) using the same software.

A MetaGWAS analysis was conducted for each disease resistance trial across populations to identify stable QTLs for each disease. The model developed by Bolormaa et al. (2014) was utilized, which computes a global p‐value for each SNP considering its effects and standard error that was previously calculated in each univariate GWAS analysis. The method calculates a chi‐squared value for each SNP across all analyzed trials analyzed with the univariate model with the following equation:

$${\mathrm{\chi }}_i^2=t_i^{\prime }{V}^{-1}{t}_i$$

where $t_i$ and $t_i^{\prime }$ are the signed t‐values for the SNP i and its transpose, respectively; $V^{-1}$ is the inverse of the genetic correlation matrix among the t‐values for all univariate analyses. The t‐values can be calculated as follows:

$$t=\frac{b}{\mathrm{se}\left(b\right)}$$

Here, b represents the effect of allelic substitution for the SNP, as determined in each univariate analysis, and se(b) is its corresponding standard error. Associations between SNPs and disease resistance were grouped into distinct QTLs if neighboring loci on the chromosome exhibited high LD, with r ²  > 0.5. The Bonferroni correction was applied to estimate the significant threshold, accounting for multiple testing.

3. RESULTS

The distribution of the five populations across the first two principal components is illustrated in Figure 1. Collectively, these two principal components account for 19.2% of the total variation in the SNP data. Notably, the AVR population emerges as the most diverse, exhibiting a range of accessions that overlap with those of other populations. This diversity is attributed to the inclusion of landraces and synthetic wheat derivatives within the AVR population. Conversely, the remaining populations, primarily composed of cultivars and breeding materials, are observed to cluster together. Among these, the USyd population stands out, displaying the broadest distribution along the two principal components. This extensive spread is attributed to the presence of numerous lines descended from landraces and wild ancestors within the population. The analysis reveals distinct patterns in population distribution, underscoring the unique characteristics and diversity within each group. The critical r ² value for the LD decay analysis was equal to 0.121. Figure S1 shows the LD decay plot, which started to decay below the critical value at around 3 Mb.

FIGURE 1.

Principal component analysis (PCA) showing the distribution of the lines collected from five different populations.

Table 1 provides a detailed account of the average SNP‐based heritability (h ²) for each rust disease across the experiments. The average number of phenotypic records per experiment was 793, 825, and 857 for Lr, Sr, and Yr, respectively. The heritability values varied across populations for the three diseases, with the highest h ² values across the three diseases recorded for USyd (Table 1). For Lr, the highest value was observed in the USyd population (h ²  = 0.78) and the lowest in the TCAP population (h ²  = 0.19), resulting in an overall mean heritability of 0.43 across 16 experiments. The heritability values for AVR population (0.37) and Kaz population (0.35) were almost equal to the mean value. For Sr, the highest h ² was observed in the USyd population (h ²  = 0.80) and lowest in the AVR population (h ²  = 0.43), leading to a mean heritability of 0.62 across the 13 experiments, which was higher than that observed for Lr. There was, however, a slight difference between the Kaz (0.45) and AVR (0.43) populations for heritability. For Yr, the heritability was highest in the USyd population (h ²  = 0.71) and lowest in the TCAP population (h ²  = 0.14), culminating in an overall average heritability of 0.41 over 19 experiments. The heritability value of Kaz population was closer to the TCAP population, whereas heritability values for AVR (0.43) and China (0.57) populations were above average.

We conducted an in‐depth exploration of genetic correlations across all populations and experiments for three rust diseases (Table 2). The mean genetic correlation across the 16 Lr experiments was 0.52, indicating a moderate level of correlation. Similarly, the 13 Sr experiments had a mean genetic correlation of 0.34, while Yr had a higher mean genetic correlation of 0.59. In terms of inter‐disease correlations, the mean genetic correlation between Lr and Sr, and Lr and Yr remained relatively low, at 0.21 and 0.20, respectively. Notably, Sr and Yr had the lowest mean genetic correlation at 0.02. The average standard deviation of these estimations was 0.11, which signifies a moderate level of genetic correlation variability across different diseases and populations. This comprehensive analysis highlights the nuanced relationships within and between rust diseases.

TABLE 2.

Mean values for genetic correlations across all populations and trials for three rust diseases: leaf (Lr), stem (Sr), and stripe (Yr) rusts. The average standard deviation of the estimation was equal to 0.11.

Disease	Lr	Sr	Yr
Lr	0.52	0.21	0.20
Sr	0.21	0.34	0.02
Yr	0.20	0.02	0.59

The metaGWAS analysis, integrating diverse field experiments across five distinct wheat populations, has exposed the genetic architecture underlying resistance to Lr, Sr and Yr. The −log10(p) values for Lr, Sr, and Yr signify the statistical significance of each association, with a stringent threshold set at >5.8 for significance. Of the 35 QTLs identified in this analysis, 19 (from chromosomes 1B, 1D, 2A, 2B, 3A, 3B, 3D, 4B, 5B, 7A, and 7D) were associated with resistance to Lr, 17 (from chromosomes 1B, 2A, 2B, 2D, 3A, 3B, 3D, 4D, 6A, and 6B) to SR, and five (from chromosomes 2D, 3B, 3D, and 5D) to Yr (Table 3). Five QTLs conferred resistance to both Lr and Sr, whereas only one QTL was associated with Lr and Yr resistance.

TABLE 3.

Quantitative trait locus (QTL) detected for the three rust diseases leaf rust (Lr), stem rust (Sr), and stripe rust (Yr).

ID	SNP	Chr	GenPos	PhyPos	Lr	Sr	Yr	QTL
1	Excalibur_c64975_90	1B	11.0	94,166,445	7.0	1.2	0.0	Novel
2	BS00078431_51	1B	70.8	467,200,434	9.1	1.2	0.8	Kaz (Kankwatsa et al., 2017)
3	BS00069668_51	1B	–	618,543,821	0.0	6.3	0.0	Kankwatsa et al. (2017)
4	BS00063511_51	1D	167.1	485,708,706	7.7	0.6	0.2	Kaz
5	BobWhite_c14476_80	2A	102.0	86,901,274	10.7	1.3	1.2	Kaz
6	wsnp_Ex_c2887_5330787	2A	99.3	110,465,085	0.0	7.0	0.0	Kankwatsa et al. (2017); PM (N. Liu et al., 2017)
7	Excalibur_c51281_802	2A	–	534,631,831	0.0	8.5	0.0	Novel
8	BobWhite_c31854_189	2B	16.9	293,046,517	6.9	0.4	0.3	Novel
9	Excalibur_rep_c106124_239	2B	93.5	159,207,418	5.9	0.5	0.0	Bokore et al. (2023)
10	Excalibur_c20376_615	2B	–	108,232,773	15.0	3.5	0.1	Kaz
11	RAC875_c1578_1198	2B	110.5	762,956,082	0.1	9.1	0.0	AVR
12	Excalibur_c5592_178	2D	9.5	–	0.7	0.4	18.8	Novel
13	BobWhite_c40561_305	2D	80.4	574,173,698	0.0	8.3	0.0	SydU
14	RAC875_c20989_348	2D	–	10,334,965	0.1	0.0	6.0	Novel
15	D_contig77683_350	2D	–	458,162,961	0.0	7.8	0.0	Novel
16	Ra_c8717_520	3A	98.4	–	0.6	7.5	0.1	Novel
17	Tdurum_contig42150_3417	3A	108.9	645,995,656	8.2	3.5	0.0	Sr (Kankwatsa et al., 2017)
18	Kukri_s111374_55	3A	173.2	724,646,829	12.2	6.2	0.4	FHB (Mwaniki, 2017)
19	Excalibur_c8386_1009	3B	14.1	7,188,324	7.5	0.1	12.5	AVR
20	BobWhite_c4514_298	3B	–	40,238,983	0.1	7.3	0.6	AVR
21	RAC875_rep_c107110_137	3B	67.5	466,614,415	12.4	7.4	0.0	Novel
22	Excalibur_rep_c103408_632	3B	132.1	804,670,139	12.7	6.1	0.0	FHB (Mwaniki, 2017)
23	CAP11_c7974_175	3D	4.0	–	0.2	0.4	6.0	Novel
24	IAAV4004	3D	143.0	–	17.4	6.0	0.0	AVR, SydU
25	Excalibur_c27349_166	4B	78.0	–	6.2	1.8	0.6	Kaz
26	Tdurum_contig56887_322	4B	109.1	660,676,250	6.8	0.0	0.5	SydU
27	JD_c3554_64	4D	–	118,623,219	13.9	6.4	0.0	Novel
28	wsnp_Ra_c17506_26393195	5B	117.8	598,505,470	6.2	0.0	0.9	Novel
29	BS00000929_51	5D	147.2	479,879,340	0.1	0.1	8.2	Cr (Rahman, 2018)
30	Excalibur_c50953_96	6A	43.1	–	0.1	5.8	0.1	Novel
31	BobWhite_s67148_292	6A	99.0	582,253,768	0.1	5.8	0.0	Novel
32	BobWhite_rep_c49333_223	6B	–	7,146,212	1.1	8.0	0.1	FHB (Mwaniki, 2017)
33	BS00041633_51	6B	–	688,607,609	0.2	8.2	0.1	SydU
34	BobWhite_c24063_231	7A	127.7	232,746,015	7.0	1.3	0.5	Kaz
35	Ku_c67089_214	7D	95.7	–	6.4	1.8	0.0	Yr (J. Liu et al., 2015)

Note: Values in the columns Lr, Sr, and Yr represents −log₁₀(p) with significant threshold > 5.8.

Abbreviations: Chr, chromosome; Cr, crown rot; FHB, Fusarium head blight; GenPos, genetic position (cM); PhyPos, physical position (bp); PM, powdery mildew.

The metaGWAS analysis identified six QTLs with associations between two rust diseases (Table 3). These shared QTLs suggest a potential genetic basis for common resistance mechanisms against multiple rust diseases. Interestingly, these QTLs exhibited an exceptionally high level of significance for at least one of the two diseases, with a −log₁₀(p) value exceeding 12. Five of the six QTLs were located on homoeologous group 3 chromosomes (3A, 3B, and 3D), whereas one was placed in chromosome 4D (Table 3). Of the homoeologous group 3 chromosomes, three were located on chromosome 3B (at 7.2, 466.6, and 804.7 mb), and one each QTL was from chromosomes 3A (at 724.6 mb) and 3D (at 612.9 mb). One QTL on chromosome 3B was located on the short arm at 7.2 mb and associated with both Lr and Yr. The remaining five QTLs were all associated with Sr and Lr. Specifically, one was located on the long arm of chromosome 3A at 724.6 mb, and two QTLs were found on chromosome 3B at 466.6 mb and 804.7 mb. One QTL was located on the long arm of chromosome 3D at 612.9 mb, while the final QTL was positioned on the short arm of chromosome 4D at 118.6 mb.

4. DISCUSSION

Various factors play a crucial role in influencing the power of GWAS analysis. These factors include population size, the magnitude of the QTL effect, its frequency within the population, and LD between the genotyped variant and the QTL (Purcell et al., 2003; Sham et al., ²⁰⁰⁰). Consequently, large population size emerges as pivotal for the identification of QTL with low frequency or marginal effects on the trait, facilitating the detection of statistically significant variations in the population studied (Spencer et al., 2009). In the pursuit of robust GWAS outcomes, the importance of large population size cannot be overstated.

Thirteen QTLs identified in this study appear to be novel and were not previously reported (Table 3; the “QTL” column). The remaining QTLs were either reported in the original research aggregated in the current study or documented by other studies investigating rust resistance across different germplasm and fungal diseases. Specifically, six Lr QTLs situated on chromosomes 1B, 2D, 2A, 2B, 4B, and 7A were previously reported by Genievskaya et al. (2020) in their investigation of resistance to three rust diseases in the Kaz population. Among these, the 1B QTL was also reported by Kankwatsa et al. (2017). Two Sr QTLs located on chromosomes 2B and 3B were reported in the AVR population by Joukhadar et al. (2020), who also identified the Lr/Sr QTL on chromosome 3D and the Lr/Yr QTL on chromosome 3B. A study of the USyd population reported two Sr QTLs on chromosomes 2D and 6B, an Lr QTL on chromosome 4B, and the Lr/Sr QTL on 3D. Additionally, Kankwatsa et al. (2017) reported two Sr QTLs that overlapped with the Sr QTL on chromosomes 1B and 2A detected in the present study, while Bokore et al. (2023) detected an Lr QTL that shares the same location as our Lr QTL on chromosome 2B (Table 3).

Several QTLs exhibited associations with other fungal diseases distinct from those explicitly identified as significantly associated in this study, thereby demonstrating a potential pleiotropic effect. The Sr QTL on chromosome 2A was previously reported by N. Liu et al. (2017) to be associated with powdery mildew. Mwaniki (2017) detected several QTLs associated with Fusarium head blight, of which three overlapped with the Lr QTL detected on chromosome 6B and two Lr/Sr QTLs located on chromosomes 3A and 3B in this study. Rahman (2018) also reported a crown rot QTL at the same position as the Yr QTL on chromosome 5D in this analysis, while J. Liu et al. (2015) detected a Yr QTL at the same position as the Lr QTL on chromosome 7D from this investigation. Similarly, Kankwatsa et al. (2017) detected an Sr QTL at the same location as the Lr QTL on chromosome 3A. Although this QTL exhibited a −log₁₀(p) of 3.5 for Sr in the current study, it fell slightly short of the stringent significance threshold. Further exploration of the underlying genes and allelic interactions within these QTLs regions promises to provide new insights into the genetic architecture of wheat rust resistance, and it will also allow refinement of more than one QTL on the same chromosome based on the physical locations. Such physical map controversies do exist and need to be verified based on pedigrees of contributing genotypes.

Previous rust resistance research also exemplified the principle of using large populations, often comprising several thousand individuals, in simple GWAS. For example, F. Liu et al. (2020) assessed 133 genotypes and their 1574 hybrids for Lr resistance, while Gao et al. (2017) characterized Sr resistance in a global collection of 2152 accessions. Joukhadar et al. (2020) utilized a multivariate linear mixed model approach (Zhou & Stephens, 2014) to combine field experiments for the AVR population used in the current study. Consequently, the cumulative size of the populations analyzed in the current study far surpassed that of any previously documented GWAS study on wheat rust diseases and led to detection of 13 potentially new QTLs, five for Sr and Lr each and three for Yr (Table 3). This substantial population size is imperative for enhancing the sensitivity and reliability of the GWAS analysis, particularly when dealing with traits characterized by low frequency or subtle effects within the population. The detection of only five QTLs for Yr is surprising. This can possibly be attributed to poor performance of Yr nurseries and/or assessment of Yr infection too late in the season combined with unsuitable weather conditions at certain locations (H.S. Bariana, personal observation). These facts would have led to removal of higher number of genotypes from the final analysis.

The concept of metaGWAS, originally introduced in human genetics to address challenges associated with the impracticality of raw data sharing (Winkler et al., 2014), has evolved into a powerful analytical tool with wide‐ranging applications. Both theoretical and empirical demonstrations consistently highlight the enhanced power of metaGWAS analyses compared to standard GWAS on individual datasets, equating this power to an analysis of the complete raw data (Evangelou & Ioannidis, 2013; Lin & Zeng, ²⁰¹⁰; Panagiotou et al., ²⁰¹³). Even in scenarios where raw data sharing is feasible, metaGWAS offers an additional advantage in crop breeding, particularly when dealing with unbalanced multi‐trait or multi‐environmental datasets (Jighly et al., 2022; Joukhadar et al., ²⁰²⁰). Unbalanced datasets are common in breeding programs, where a portion of the population with lower performance is systematically replaced each year with new individuals from new crosses (Battenfield et al., 2018). Summary statistics for SNPs from multiple studies can be integrated through metaGWAS regardless of the population used, thus allowing integration of different population structures and trait architectures (Bolormaa et al., 2014). Despite its potential, there has been a limited application of metaGWAS to enhance the statistical power of GWAS analyses in wheat. Joukhadar et al. (2020) applied metaGWAS to 11 agronomic and quality traits phenotyped under optimal and heat‐stressed conditions, with a maximum of 13,959 accumulated phenotypic data points for grain yield. Battenfeld et al. (2018) applied metaGWAS to evaluate wheat milling and baking quality on 4095 samples across various environments, while S. Singh et al. (2021) used metaGWAS, to integrate three Yr experiments resulting in 2785 phenotypic records. The current study is a new benchmark in data aggregation and provides a basis for unraveling the intricate genetic mechanisms underpinning rust resistance in wheat.

MetaGWAS enables the integration of different populations phenotyped under different environmental conditions, thereby enhancing the detection of more stable QTL across environments with increased power (Joukhadar et al., 2020). While it may result in loss of environment‐specific QTL compared to individual GWAS analyses, the distinctive strength of metaGWAS is the identification of QTLs with stable effects across wider ranges of environments and rust disease pathotypes (Joukhadar & Daetwyler, 2022). This stability is vital to the development of cultivars that adapt across diverse geographical regions (Singh et al., 2021). Unlike individual GWAS analyses, metaGWAS captures QTL that transcend specific environments, providing a basis for designing breeding strategies to improve broad adaptation (Evangelou & Ioannidis, 2013). We acknowledge that the low heritability of a trait (e.g., TCAP data reported in the present study) can influence the power to detect genetic regions in individual studies. However, the metaGWAS approach followed here addresses this, as it accounts for differences in reliability and consistency across experiments, giving more weight to robust trials (Bolormaa et al., 2014). This integrated approach significantly enhances our ability to identify stable QTL, overcoming the limitations of single studies that might show lower heritability.

While the metaGWAS approach offers significant advantages in identifying robust and stable QTLs across diverse environments, we acknowledge certain limitations. The integration of large‐scale data, by prioritizing broadly consistent genetic effects, might lead to the diminished detection of environment‐specific QTLs that are highly relevant only in particular conditions (Li et al., 2024). Therefore, for breeding programs targeting specific environments, a dedicated focus on these environment‐specific QTLs would be crucial alongside these QTLs detected using metaGWAS. Furthermore, the genetic associations identified in this study, while statistically significant, require further functional validation through methods such as gene expression variation, gene editing, transgenic, or single cell sequencing approaches to confirm their role in conferring rust resistance and to translate these findings effectively into breeding programs (Cano‐Gamez & Trynka, 2020).

While our metaGWAS approach successfully identified 13 potentially novel QTLs and several loci conferring resistance to multiple rust diseases, it is important to consider the limitations in fine‐scale gene resolution. LD decay analysis in our wheat populations revealed extensive LD blocks of around 3 Mb, which is expected in self‐pollinated crops like wheat (Jighly et al., 2018). Given the high gene density within these large genomic regions, precisely narrowing down individual candidate genes for each QTL is challenging. This broad LD also limits our ability to definitively distinguish whether multi‐disease resistance loci are attributable to pleiotropic genes or multiple tightly linked resistance genes at the current resolution. Future high‐resolution mapping and functional studies will be essential to dissect these complex genomic regions and identify the underlying causal genes.

5. CONCLUSION

In conclusion, this study demonstrated that how the integration of data from five populations using a metaGWAS approach has identified additional genetic variation for resistance to rust diseases that was not feasible with standard GWAS based on small populations. The study highlights the effectiveness of metaGWAS in detecting stable QTL. The potential pleiotropic effects and associations with resistance to other fungal diseases identified in this study provides a basis for further research into understanding whether resistance to different pathogens is linked in coupling or repulsion phases. These insights provide a foundation for targeted and effective breeding strategies, providing promising solutions to mitigate the impact of rust diseases on global wheat production. Ultimately, these findings provide a robust foundation for informing marker‐assisted selection and genomic selection pipelines, enabling the integration of these newly identified and validated resistance loci into pre‐breeding efforts and pyramiding strategies to develop more resilient wheat cultivars.

AUTHOR CONTRIBUTIONS

Reem Joukhadar: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; resources; software; validation; visualization; writing—original draft. Richard M. Trethowan: Conceptualization; data curation; funding acquisition; project administration; resources; supervision; writing—review and editing. Urmil Bansal: Investigation; methodology; writing—review and editing. Rebecca Thistlethwaite: Data curation; investigation; methodology; writing—review and editing. Josquin Tibbits: Writing—review and editing. Harbans Bariana: Investigation; methodology; writing—review and editing. Matthew J. Hayden: Project administration; resources; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Figure S1. Linkage disequilibrium decay of the lines collected from five different populations

Supplementary Materials

Joukhadar, R. , Trethowan, R. M. , Bansal, U. , Thistlethwaite, R. , Tibbits, J. , Bariana, H. , & Hayden, M. J. (2025). Genomic exploration of durable wheat rust resistance by integrating data from multiple worldwide populations. The Plant Genome, 18, e70093. 10.1002/tpg2.70093

DATA AVAILABILITY STATEMENT

All data used in the present study was previously published in the original papers that used them. AVR data can be shared upon request subject to an institutional Material Transfer Agreement.