A multi-reference parent nested-association mapping population to dissect the genetics of quantitative traits in durum wheat

Durum wheat (Triticum durum Desf.) breeding programs face many challenges surrounding the development of stable varieties with high quality and yield. Therefore, researchers and breeders are focused on deciphering the genetic architecture of biotic and abiotic traits with the aim of pyramiding desirable traits. These efforts require access to diverse genetic resources, including wild relatives, germplasm collections and mapping populations. Advances in accelerated generation technologies have enabled the rapid development of mapping populations with significant genetic diversity. Here, we describe the development of a durum Nested Association Mapping (dNAM) population, which represents a valuable genetic resource for mapping the effects of different alleles on trait performance. We created this population to understand the quantitative nature of drought-adaptive traits in durum wheat. We developed 920 F6 lines in only 18 months using speed breeding technology, including the F4 generation in the field. Large variation in above- and below-ground traits was observed, which could be harnessed using genetic mapping and breeding approaches. We genotyped the population using 13,393 DArTseq markers. Quality control resulted in 6,785 high-quality polymorphic markers used for structure analysis, linkage disequilibrium decay, and marker-trait association analyses. To demonstrate the effectiveness of dNAM as a resource for elucidating the genetic control of quantitative traits, we took a genome-wide mapping approach using the FarmCPU method for plant height and days to flowering. These results highlight the power of using dNAM as a tool to dissect the genetics of durum wheat traits, supporting the development of varieties with improved adaptation and yield.

enabled the rapid development of mapping populations with significant genetic diversity. Here, we describe the development of a durum Nested Association Mapping (dNAM) population, which represents a valuable genetic resource for mapping the effects of different alleles on trait performance. We created this population to understand the quantitative nature of drought-adaptive traits in durum wheat. We developed 920 F 6 lines in only 18 months using speed breeding technology, including the F 4 generation in the field. Large variation in above-and below-ground traits was observed, which could be harnessed using genetic mapping and breeding approaches. We genotyped the population using 13,393 DArTseq markers. Quality control resulted in 6,785 high-quality polymorphic markers used for structure analysis, linkage disequilibrium decay, and marker-trait association analyses. To demonstrate the effectiveness of dNAM as a resource for elucidating the genetic control of quantitative traits, we took a genome-wide mapping approach using the FarmCPU method for plant height and days to flowering. These results highlight the power of using dNAM as a tool to dissect the genetics of durum wheat traits, supporting the development of varieties with improved adaptation and yield.

Introduction
Durum wheat (Triticum durum Desf.) contains two sub-genomes, A and B (2n = 4x = 28), and the high protein content makes it an ideal ingredient for pasta. Durum wheat is a major staple food in Mediterranean regions, including Italy, Spain, Morocco, Algeria, Tunisia, Turkey, the Middle East and Ethiopia, and it is widely produced as a cash crop in North and South America and India. Recent evidence indicates that the origins of all cereals including durum wheat trace back to the Fertile Crescent in southeastern Turkey and northern Syria (Lev-Yadun et al. 2000). Durum wheat is one of the earliest cultivated cereals, resulting in a genetic pool rich in phenotypic and genotypic variation; such variation is essential for breeding (Mengistu et al. 2016). The demand for durum wheat has increased dramatically in recent years due to the global pandemic, the recent war in Ukraine and durum wheat's high sensitivity to changes in climate, including severe drought and heat in the key durum production regions of western Canada and north Africa (Glasser et al. 2022). The challenge in meeting this demand is complicated by climate variability, with unprecedented high temperatures and the increased frequency and duration of drought in Europe already negatively affecting crop production (Hochman et al. 2017). To overcome this challenge, researchers seek to better understand the genetic architecture of biotic and abiotic traits and to empower breeding programs to assemble these traits for the development of resilient durum wheat varieties.
Recent advances in sequencing technology have driven down costs, enabling researchers to access a wealth of resources and tools to better dissect the architecture of complex traits, such as drought adaptation. A key example is the sequencing of the reference genome of the modern durum wheat variety Svevo (Maccaferri et al. 2019). However, to effectively harness the potential benefits of these new genomic tools, rich allelic diversity is required. Yet, constraints associated with the genetic diversity and resolution of available durum wheat mapping populations could limit the potential benefits of deploying these new genomic tools (Kidane et al. 2019). Diversity panels are typically deployed for marker-trait association studies due to a high level of historical recombination. Unfortunately, such panels lack the statistical power that classical pedigree-based mapping provides (Mackay et al. 2009). Novel structured mapping populations such as Multi-parent Advanced Generation Inter-Cross (MAGIC; Mackay et al. 2014;Milner et al. 2016) and Nested Association Mapping (NAM) can provide a high degree of genetic diversity in addition to a dramatic increase in allelic recombination frequencies (Ladejobi et al. 2016).
A NAM structure is based on crossing a panel of donor lines with a number of reference lines (often varieties or elite lines) and then producing recombinant inbred lines (RILs) from each donor and reference line cross (Guo et al. 2013). The design of such structure introduces chromosomal diversity through recombination in the form of 'mosaic' segments from each donor line sharing the same reference background. The NAM structure combines the advantages of well-described bi-parental and association mapping approaches while minimizing the disadvantages of each approach (Yu et al. 2008). The advantages of NAM population include (but are not limited to) high allelic richness, high statistical power, a low number of markers required for whole-genome scanning and high mapping resolution (Buckler et al. 2009). NAM was first applied by Yu et al. (2006) to study the effects of many quantitative trait loci (QTLs) controlling flowering time in maize. Since then, NAM populations have been developed for many strategic crops, including sorghum (Mace et al. 2013;Bouchet et al. 2017;Perumal et al. 2021), wheat (Richard et al. 2015;Chidzanga et al. 2021), barley (Ziems et al. 2017), rice (Fragoso et al. 2017;Kitony et al. 2021), soybean (Guo et al. 2013;Maranna et al. 2019), and Ethiopian durum wheat (Kidane et al. 2019). The Ethiopian durum NAM population (EtNAM) combines 49 varieties developed by Ethiopian farmers with one internationally representative modern variety to bridge the gap between farmer-developed and modern varieties while generating rich allelic diversity for biotic traits (Kidane et al. 2019). Although EtNAM is an invaluable resource, it was created to explore the genetics of disease and adaptation to Ethiopian production regions. This created the need for a more globally representative NAM population targeting Mediterranean and similar water-limited growing environments.
Here, we describe the development of a durum NAM (dNAM) population that combines the genetics of ICARDA elite donor lines carrying desirable drought-adaptive traits with two Australian varieties. The population segregated for plant height (PH), days to flowering (DTF), and several droughtadaptive traits, including above-and below-ground water use traits, which can be explored using mapping approaches. We genotyped the entire dNAM population using 13,395 DArTseq markers on the DArTseq genotyping-by-sequencing platform, allowing the characterisation of population diversity, population structure and linkage disequilibrium (LD). To highlight the usefulness of dNAM, we performed a genome-wide association study (GWAS) for two commonly phenotyped agronomic traits, PH and DTF, and detected several marker-trait associations. The novel genomic resource developed in this study will be highly valuable for dissecting complex traits, especially those that contribute to improved drought adaptation, which should facilitate the breeding of climate-resilient durum wheat varieties.

Selection of durum NAM parental lines
The dNAM population was generated by crossing ICARDA elite donor lines with two Australian durum varieties, Jandaroi and DBA Aurora, which served as reference varieties (Fig. 1a). A description of the parental lines, including pedigree information, is provided in Table 1. Selection of genotypes from ICARDA's germplasm pool in Morocco was based on three decades of detailed characterisation of grain yield, agronomic traits, and biotic and abiotic traits in Mediterranean water-limited environments ( Table 2). The elite durum donor lines that were imported into Australia include Fastoz2, Fastoz3, Fastoz6, Fastoz7, Fastoz8, Fastoz10, Outrob4, and Fadda98. Furthermore, the ICARDA lines have been used as parents in breeding programs targeting marginal rainfall Family 6 = Jandaroi × Fastoz8; Family 7 = Jandaroi × Fastoz10; Family 8 = Jandaroi × Fastoz6; Family 9 = Jandaroi × Fastoz2; and Family 10 = Jandaroi × Outrob4. b Crossing and most single seed descent (SSD) generations were produced under speed breeding conditions (brown), and one generation was produced in the field (green). (Color figur online) regions in West Asia and North Africa. The two Australian durum commercial reference varieties, DBA Aurora and Jandaroi, are widely grown and are preferred varieties of Australian farmers. This is largely due to their high yield potential and protein content and their grain qualities, which are preferred by the industry.

Development of the durum NAM population
The dNAM population was generated in the speed breeding (SB) facility at The University of Queensland, St Lucia, Brisbane, QLD, Australia, with one generation progressed in the field (Fig. 1b). SB allows six plant generations of spring durum wheat to be grown in a single year Alahmad et al. 2022). The parental lines were initially grown under SB conditions, and at flowering stage, eight ICARDA donor lines were crossed with the two Australian reference varieties. The crossing scheme, which is shown in Fig. 1a, resulted in the production of 10 families. The first five families shared DBA Aurora as the common parent, and the remaining five families shared Jandaroi as the common parent. Two donor lines (Outrob4 and Fastoz8) were crossed to both reference varieties. The number of recombinant inbred lines (RILs) per family, their cross-IDs and the traits carried by the donor lines are listed in Table 2. Following crossing, F 1 seeds were planted under SB conditions and 500 F 2 seeds per family were harvested. The F 2 and F 3 segregating populations were re-sown and progressed under SB using the single seed descent (SSD) method (Fig. 1b). The F 4 plants were grown in single 2-m-long rows under field conditions at The University of Queensland, Gatton Research Station (27°32′34″S; 152°19′59″E)  to generate 50-100 g of bulk F 4 -derived seed. Plant height (PH) and days to flowering (DTF) were recorded on F 4 -derived rows grown in the field. Ninety-two lines per family were subsequently selected based on agronomic appearance and a broad representation of genetic diversity within families. A single spike per field row was selected, harvested and progressed in the SB system for two more generations to generate F 6 homozygous lines. Phenotypic data used in the association analysis included the arithmetic mean values for the two PH measurements/genotype/row and the single value/genotype for DTF. An individual was considered flowering when 50% of the spikes on the main stem reached anthesis in the row of plants. The F 4 generation, which was evaluated in the field at Gatton Research Station, displayed large variation for PH and flowering time in both reference variety backgrounds (Fig. 2a).
Genotyping method and marker quality control Leaf tissue samples for the 920 introgression lines and the 10 parental lines were sampled from F 6 generation plants grown under SB conditions. Leaf samples were freeze-dried, and DNA was extracted from the samples using Diversity Arrays Technology (DArT) protocols. The 10 families were then genotyped using the DArTseq platform. Genotyping resulted in 13,395 DArTseq markers, which were ordered according to their genetic positions in the wheat consensus map (version 4.0) provided by DArT, Canberra, Australia. High-quality markers were filtered using the R package "SelectionTools" (http:// cran-r. uni-giess en. de/ ~user/). Quality control and filtering thresholds were applied. For example, markers with a minimum gene diversity frequency of ≤ 10% and genotypes with ≤ 20% missing marker information were removed and the remaining markers were imputed for missing marker calls using Beagle v4.1 (Browning and Browning, 2016). A total of 6,785 high-quality DArTseq markers and 926 individuals were used in subsequent analyses.
Population structure and linkage disequilibrium decay All population structure analyses were conducted using the R package "SelectionTools".
To investigate population structure in the dNAM panel, genetic distance was calculated with pairwise Roger's distances between all genotypes using the 6,785 high-quality markers. Based on the resulting distance matrix, a principal coordinate analysisbased k-means clustering approach was performed, assuming k = 15 subgroups as a starting point. To determine the best number of clusters, the R package "NbClust version 3" was used. This package relies on 30 inbuilt indices as well as multiple factors, including distance measures, the number of clusters and the number of indices used in the package (Charrad et al. 2014). The appropriate number of clusters was confirmed based on the highest number of indices, for which the first three principal components were considered. Based on the crossing scheme, the dNAM populations were divided into two main hierarchical clusters representing the genetic backgrounds of the two Australian reference varieties (DBA Aurora and Jandaroi). To visualise the results, a genetic distance matrix was constructed by plotting principal coordinates (Fig. 3), and a heat map of genetic relatedness and NJ trees were constructed based on the clustering results ( Fig. 4a, b).
Linkage disequilibrium (LD) decay analysis was performed for the entire dNAM population, as well as analysis of decay within each of the 10 dNAM families. Pearson correlation coefficients (r 2 ) for pairs of adjacent markers along the durum genome were calculated for 6,785 markers to determine the level of decay due to recombination. Locally estimated scatterplot smoothing (LOESS) curves were constructed to display the level of LD decay within each family and across the entire dNAM population, with the genetic distance shown in megabase pairs. Marker pairs with a genetic distance exceeding 50 Mbp were considered unlinked and eliminated from the analysis (Roncallo et al. 2021).
Marker-trait associations for plant height and days to flowering Marker-trait associations (MTAs) between individual polymorphic markers and each phenotype, PH, and DTF were identified using the 6,785 quality-controlled DArTseq markers. GWAS was conducted in GAPIT Version 3 (Wang et al. 2021) using the Fixed and Random model Circulating Probability Unification (FarmCPU) model. This model, which was selected due to its high statistical power in detecting significant MTAs, enables efficient computation, eliminates confounding factors, and is significantly superior in avoiding false positives compared to mixed linear model approaches. FarmCPU implements two separate models that run and update iteratively. The fixed model handles false positive  (3) is displayed as shapes, and the entire dNAM population is clustered by family and coloured according to the donor parents associations by testing one marker at a time and considering the rest of the associated markers as covariates. The random model uses the associated markers to obtain the kinship, which is included as a covariate in the model to account for the relationship between the tested lines in the model. This eliminates the risk of overfitting while controlling for false positives (Liu et al. 2016 nucleotide, Q is the population structure using PCA, K is the kinship matrix and e is the residual error. Three principal components were used as covariates to control for population structure in the model. Model fitness was evaluated using quantile-quantile plots between expected and observed − log10(p) values. The significance p-value threshold for the marker-trait association was set to (− log10(p) ≥ 3), corresponding to a significance level of p ≤ 0.001. Fig. 4 Calculation of the genetic relatedness of 926 F 6 dNAM lines using 6,785 DArTseq high-quality markers. a Heatmap displaying two major clusters representing the Australian reference parents Jandaroi (blue box) and DBA Aurora (red box). Lines with high levels of genetic similarity clustered together within each reference parent, representing 10 sub-dNAM populations (dark brown). The dendrogram at the top is colourcoded according to the dNAM families, and the dendrogram on the left corresponds to the three distinct clusters displayed based on the calculated k-means values. b Neighbour-Joining unrooted dendrogram analysis discriminated individuals across and within genetic backgrounds. Individuals sharing Jandaroi as a reference clustered into five families (in blue), as did individuals sharing DBA Aurora as a reference (in red). (Color figure online)

Plant height
Following crossing and the generation of F 1 seeds, we observed segregation for several traits during SSD under SB conditions from F 1 through to F 4 , which was expected due to the occurrence of multiple recombination events during self-pollination. The F 4 :F 5 dNAM lines grown under field conditions displayed a high degree of variation for PH among the lines within each family, across families with a common reference parent, and across families which did not share the reference parent. For instance, PH ranged from 49-106 cm for families derived from DBA Aurora as a common reference variety and 49-96.5 cm for families derived from Jandaroi. The highest degree of within-family variation was observed among Family 4 lines derived from the cross between DBA Aurora (PH = 68 cm) and Fadda98 (PH = 79 cm). Plant height for this family ranged from 55.5-106 cm, suggesting transgressive segregation for the PH trait, with several lines showing taller or shorter PH than the respective parents (Fig. 2b). The lowest degree of segregation was observed in Family 9, as the individuals derived from the cross between Jandaroi (56 cm) and Fastoz2 (67.3 cm) displayed PH values ranging from 55.5-75.5 cm, with several plants slightly taller than the donor parent (Fig. 2b).

Days to flowering
We obtained phenotypic data for DTF by subtracting the flowering date from the sowing date for the F 4 :F 5 generation in the field. A reasonable degree of variation in DTF was observed among the dNAM lines across all families, ranging from 79-106 days. In families derived from DBA Aurora (DTF = 93 days), DTF ranged from 80-106 days, and in families derived from Jandaroi (DTF = 88 days), DTF ranged from 79-94 days. Family 3, derived from a cross between DBA Aurora and Fastoz8 (DTF = 85 days), and Family 4, derived from a cross between DBA Aurora and Fadda98 (DTF = 92 days), displayed high degrees of transgressive segregation, with DTF values of 80-98 days and 81-106 days, respectively. The least within-family variation for DTF was observed in Families 1, 6, and 9, with DTF values of 85-95, 80-91, and 79-90 days, respectively. The across-family differences for DTF (i.e., least variation and most variation for DTF) for the entire dNAM population was 15 days.
The overall correlation between PH and DTF among individuals of the entire dNAM under field conditions was r = 0.2 (p < 0.01). However, the correlation between the two traits varied within each family. For instance, Families 2, 3, 4, 5, and 7 displayed no association or a very weak association, with Pearson's correlation coefficients ranging from r = 0.01 to 0.16. A significant negative correlation between PH and DTF was observed for individuals in Family 1 (r = − 0.2; p < 0.01). Finally, significant positive correlations were observed for both traits between individuals within Families 10 and 8 (r = 0.27; p < 0.01 and r = 0.44; p < 0.001, respectively; Supplementary Table 1).

Genetic relatedness and linkage decay
We performed quality control on the 13,393 DArTseq markers and retained 6,785 high-quality polymorphic markers. Thereafter, markers were physically mapped to both sub-genomes of Svevo durum wheat (Maccaferri et al. 2019). Of these DArTseq markers, 3,171 were distributed in the A genome and 3,708 were distributed in the B genome, with the number of markers per chromosome ranging from 225 (1A) to 803 (7A) for sub-genome A and from 295 (4B) to 708 (7B) for sub-genome B. We used these markers for subsequent genetic analyses, including calculating genetic distance using the pairwise Roger's distance approach. We determined that additive genetic variance was explained by the first 10 principal components (PC1 = 31.8%, PC2 = 9.6%, PC3 = 7.8%, and PC4 to PC10 ranging from 6.0-1.49%). The large variation explained by the first three PCs accounted for 49.2% of the genetic variability.
The optimum number of clusters for the 926 genotypes was three, as determined by k-means clustering based on seven indices (Fig. 3). The first cluster consisted of individuals that shared Jandaroi and DBA Aurora as reference parents, as well as individuals with the same ICARDA elite donor parents (Outrob4 and Fadda98; n = 183 individuals). The second cluster included individuals that shared Jandaroi as a reference parent (n = 282 individuals), and the third cluster included individuals that shared DBA Aurora as a reference parent (n = 462 individuals). We calculated the genetic relationships between individuals across dNAM, which allowed us to discriminate between individuals from different genetic backgrounds in the form of a heatmap (Fig. 4a) and a neighbour-joining unrooted dendrogram (Fig. 4b). This resulted in the clustering of five families in the background of each reference variety. The LD of the entire dNAM population quickly decayed and reached a threshold of r 2 = 0.2 at a distance of ~ 3 Mbp and r 2 = 0.1 at a distance of ~ 11 Mbp. Variation in LD decay was evident among dNAM families and reached a threshold of r 2 = 0.2 at ~ 8 Mbp (Family 1) to ~ 13 Mbp (Family 2) (Fig. 5).

GWAS for plant height and anthesis time for the durum NAM lines
Based on quantile-quantile plots between expected and observed − log10(p) values with a cut-off threshold (− log10(p) ≥ 3), we identified 35 significant MTAs for PH and 20 for DTF (Fig. 6). The identified MTAs corresponded to 22 and 13 unique QTLs for PH and DTF, respectively (Supplementary Table 2).
These QTLs were distributed across the durum genome, with QTLs for PH distributed on 11 chromosomes and QTLs for DTF distributed on 10 chromosomes. Notably, for PH, six QTLs detected in this GWAS were also identified in previous studies, including the RhtB1 gene (Mazucotelli et al. 2020;Wang et al. 2020;Zhang et al. 2021). In particular, marker '2,371,505' had the most significant association (− log10(p) = 30.9) with PH and was aligned with RhtB1(Mazucotelli et al. 2020). However, GWAS for PH using the dNAM populations revealed 16 novel QTLs underpinning variation in PH across dNAM lines. Similarly, six QTLs detected for DTF in this study were also identified in previous studies, including the Ppd-A1, TaGI, VRN1, and VRN3 genes (Shi et al. 2019;Gupta et al. 2020;Semagn et al. 2021;Mangini et al. 2021). Interestingly, marker '982,956' had the most significant association (− log10(p) = 25.8) with DTF and was precisely aligned with Ppd-A1 (Mangini et al. 2021). However, six novel QTLs for DTF were also detected in durum wheat.

Discussion
Here, we rapidly developed dNAM, the first durum wheat NAM population, via SB. We developed this population to dissect the genetics of target traits and enhance adaptation to Mediterranean environments.

Fig. 5
Comparison of linkage disequilibrium (LD) decay of the entire dNAM population and the 10 families. LD decay is shown as the intersection of the LOESS curves at a threshold of r 2 = 0.1, 0.2. Each LOESS curve represents one family (coloured using the same key). The LD decay of the entire dNAM is referred to as ALL We genotyped this population using 13,393 DArTseq markers. dNAM exhibits significant trait variation and diversity, with huge variation in abiotic stress traits for durum wheat. We observed a positive correlation between PH and DTF, which was also reported by Ruan et al. (2020). This observation indicates that partially shared genetic regions underpin both traits, which we detected on chromosomes 1B, 7B, 2A, 4A, and 5A; this conclusion was also described in previous reports (Ruan et al. 2020; Supplementary Table 2). The structure and statistical power of this population make it ideal for mapping the underlying genetic architecture of important traits. This notion was confirmed by our identification of previously well-described QTLs and genes for PH and DTF, as well as the discovery of novel QTLs for both traits.
We provided insights into the rapid development of the population, the genetic makeup of dNAM, its population structure, LD decay and the relatedness among individuals. To demonstrate the power of dNAM to detect significant MTAs, we mapped two important agronomic traits using FarmCPU. The genetic architecture of countless numbers of traits could be explored using this technique. Such analysis would enable the efficient identification and discovery of novel QTLs and the rapid introgression of these genes into adapted materials using trait pyramiding backcrossing schemes, thereby enhancing the productivity and end-use grain quality of modern durum varieties. The dNAM population represents a valuable resource for durum wheat breeders and geneticists. Durum NAM population development and diversity Genetic variation allows plant breeders to increase production more sustainably with crops that have enough plasticity to withstand climate change and evolving pathogens. Plant breeding requires ongoing cycles of selection and the progression of the best performing lines to increase production and help meet the demands of several billion people for highquality food (Tester and Langridge, 2010). A powerful pre-breeding approach is to develop new sources of genetic variation such as NAM populations, segregating for several adaptive traits which could be mapped using traditional mapping approaches for genetically less complex traits or meta QTL analyses for traits with complex genetic architecture (Ariagada et al. 2022). The concept of creating variation using a nested crossing structure has been used for many winter cereals (Richard et al. 2015;Ziems et al. 2017;Kitony et al. 2021;Kidane et al. 2019), pulses (Maranna et al. 2019), and oilseed crops (Hu et al. 2018) to investigate the genetic effects of complex traits that are desired by breeders. However, developing such useful populations is challenging due to the time required to complete multiple generation cycles to create inbred lines (Alahmad et al. 2022). Consequently, in the current study, we used SB to develop the first durum wheat NAM population. SB permits rapid generation cycling using a prolonged photoperiod and controlled temperatures. The significantly reduced time required for a single generation allows RILs to be developed in a reasonably short period of time. For instance, six generations of durum wheat can be produced in a single year using protocols detailed previously Watson et al. 2018).
In the current study, a significant level of variation was explained by the first two PCs, likely due to the genetic background shared by the two reference parents. The remaining PCs explain the variability across different families in the background of each reference variety, DBA Aurora and Jandaroi. The population structure and genetic relatedness between dNAM lines were as expected, with clear population stratification. The genomic relationship matrix perfectly matched the crossing scheme, which involved five families within each reference parent, as shown in Fig. 4. Indeed, previous NAM populations developed for durum and bread wheat showed a clear structure and hierarchical relationship matching the designed crossing schemes (Kidane et al. 2019;Chidzanga et al. 2021). Based on genetic linkage information, we calculated the genome-wide LD decay across the 10 dNAM families. The LD of the dNAM population quickly decayed and reached a threshold of r 2 = 0.2 at an ~ 3 Mbp distance compared to the average LD decay distance of approximately 7.4 Mbp across the EtNAM (Kidane et al. 2019). Analysing global LD decay provides insight into the genetic diversity of a population due to the recombination of LD blocks for key drought-adaptive and developmental traits, thereby assisting plant breeders to apply selection within each population (Voss-Fels et al. 2015).

Durum NAM is a valuable genetic resource
The value of using dNAM populations lies in the significant range of genetic variation generated in a sufficiently large population, which is likely due to the reshuffling of the donor genes in the reference variety backgrounds. This allows for the identification of gene variants with the greatest potential for improving complex traits such as yield and quality. NAM populations are especially useful when paired with high-throughput phenotyping and screening techniques (Perumal et al. 2021). For instance, dNAM provides the opportunity to map breeders' most targeted quality traits including protein content, vigorousness and semolina which could be aligned with previously mapped QTL in the recently reported consensus map for quality traits in durum wheat . Additionally, we previously identified a QTL for seminal root angle (qSRA-6A) with major effects under controlled conditions (Alahmad et al. 2019) using the high-throughput 'clear pot' method, which is suitable for root screening small grains such as winter cereals (Richard et al. 2015;Robinson et al. 2016;Alahmad et al. 2018). We also successfully screened another subset of the dNAM population for crown rot disease under field conditions, resulting in the identification of QTLs underpinning tolerance to crown rot (qCR-6B), staygreen traits (qSG-integral-6B, qSG-0.1-6B), and PH (qPH-6B), as reported by Alahmad et al. (2020).
In the current study, we performed a GWAS to identify genomic regions controlling two phenological traits, PH and DTF. To highlight the power of mapping dNAM populations to dissect the genetic basis of complex traits, we used the FarmCPU model to detect significant QTLs associated with PH and DTF. We discovered 13 and 22 QTLs that were significantly associated with DTF and PH, respectively. The GWAS revealed the complex genetic architecture and quantitative nature of DTF and PH, as the detected QTLs were distributed across the durum wheat genome. Interestingly, the QTLs identified using dNAM for PH on chromosomes 1B, 2B, 4A, 5B, 6A, 6B, 7A, and 7B were also reported in previous studies (Milner et al. 2016). Further, the most significant QTL associated with DTF in this study is aligned with the recently reported flowering genes ppd-1B, Vrn1, and Vrn2, which were previously mapped to chromosomes 2A, 2B, and 5A, respectively (Gupta et al. 2020). In addition to the previously reported QTLs for PH and DTF, six novel QTLs controlling DTF and six for PH were detected using the dNAM population developed in our study. Notably, dNAM has also been successfully used to map key drought-adaptive traits such as canopy dynamics and root system architecture (Alahmad et al. 2019(Alahmad et al. , 2020. Perhaps the QTLs reported for above-and below-ground traits using the dNAM populations developed in this study could be evaluated in multiple environments and used as markers for assisted breeding and introgression into elite germplasm. This would provide insights into the genetic architecture of a trait and GxE interactions, empowering plant breeders with knowledge about how to strategically select for these traits in durum breeding programs.

Future applications and material availability
The dNAM developed in this study is a structured population based on 10 donor ICARDA elite lines that were crossed with two Australian reference parents. The progenies were forwarded in a series of six generations of selfing until homozygosity. In this population, the entire genomes of the donor lines were re-shuffled in the form of chromosomal blocks into the background of the shared reference parent. While mapping traits by analysing the whole dNAM population would be desirable, it would be more cost effective to analyse the 10 parental lines for a trait of interest. Based on the results, sub-populations derived from donors carrying the desirable trait could then be evaluated, enabling a more targeted mapping approach (Alahmad et al. 2019(Alahmad et al. , 2020. This approach would be particularly suitable for time-consuming assays or field experiments that are expensive and require large numbers of genotypes. The evaluated subset for the trait of interest could subsequently serve as a useful resource and be used as a training population for other breeding tools such as genomic selection. This would enable the fast-tracking and pyramiding of traits (including quantitative traits such as grain quality) into future varieties using genomewide marker information (Hayes et al. 2009).
While we described the development of 920 dNAM lines, the population could be rapidly expanded by performing additional crosses to the reference parents via speed breeding, allowing additional genetic variation to be harnessed. The dNAM population could also be expanded by targeting traits of interest such as a disease. Individual donor lines could be selected from gene banks using the Focused Identification of Germplasm Strategy (FIGS) method (Endresen et al. 2012). We would like to use the dNAM population to establish collaborations with researchers and pre-breeders targeting adaptive traits and seeking to mine unique genetic diversity that combines Australian and ICARDA genetic pools. Seeds from the 920 dNAM lines and their parental lines, as well as the GBS DArTseq marker data, can be obtained upon request from the corresponding authors. Please direct requests to The University of Queensland, St Lucia, QLD, Australia or ICARDA, Rabat, Morocco using a Standard Material Transfer Agreement (SMTA).
Author's contributions SA, LH, and FMB conceived and designed the scheme for dNAM population development; SA and LH performed the crosses and developed the dNAM population. SA collected phenotypic data. LH and FMB genotyped the dNAM population. ED provided advice about structure and diversity analyses; SA, YK, ED, DJ, and HR drafted the original manuscript; all authors revised the manuscript. All authors read and approved the final manuscript. Data availability Data and materials are available upon request.