Genetic diversity and population structure of a core collection of Mediterranean durum wheat accessions using DArTseq markers

Abstract

Background

Durum wheat was domesticated in the Fertile Crescent and later spread to the Mediterranean, where human and natural selection led to the establishment of local landraces. Since the 1970s, elite durum wheat varieties largely replaced these landraces. However, landraces continue to play an important role in broadening genetic diversity and contributing valuable traits to modern wheat, particularly in the Mediterranean region, which accounts for 45% of global durum wheat production. This study examined the genetic diversity and population structure of 258 durum wheat genotypes, comprising 125 landraces, 54 breeding lines, 43 cultivars, 34 accessions of uncertain improvement status, and two near-isogenic lines. These genotypes originate primarily from various countries in the Mediterranean basin, including Algeria (39 genotypes), Egypt (3), France (45), Israel (2), Italy (54), Jordan (2), Morocco (2), Portugal (51), Spain (50), Tunisia (6), and Turkey (4).

Results

The analysis was conducted using 8,015 single nucleotide polymorphism (SNP) markers and 28,200 SilicoDArT markers. Homoeologous A- and B-genome chromosomes exhibited distinct linkage disequilibrium (LD) decay patterns, confirming genetic differences between the two genomes for both SNP and SilicoDArT markers. SNP analysis identified two subpopulations among the tested accessions: Subpopulation 1 contained 34.11% of the accessions, Subpopulation 2 contained 39.53%, and 26.36% were classified as admixture. Similarly, SilicoDArT markers revealed two subpopulations, with 53.10% of accessions in Subpopulation 1, 27.91% in Subpopulation 2, and 18.99% classified as admixture. For SNP markers, Analysis of Molecular Variance (AMOVA) showed that genetic variation was 12.77% between populations, 74.86% among samples within populations, and 12.77% within samples. Discriminant Analysis of Principal Components (DAPC) identified nine groups. AMOVA based on DAPC population assignments indicated that genetic variation was 25.95% between populations, 61.30% among samples within populations, and 12.76% within samples. DAPC was most effective in distinguishing populations based on either their level of improvement or their geographical origin.

Conclusions

This study highlighted the extent of genetic exchange in durum wheat among Mediterranean countries and further supported two primary dispersal pathways: the northern route (north and east of the Mediterranean basin) and the southern route, including North Africa and the Iberian Peninsula. The analyses demonstrated the efficiency of DArTseq markers in differentiating between genotypes and provided a more comprehensive view of available genetic resources, which can be leveraged in breeding programs to introgress beneficial alleles.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-025-12232-2.

Background

Durum wheat (Triticum turgidum, 2n = 4x = 28, AABB) is an economically important crop and the tenth most cultivated cereal worldwide, covering 18 million hectares and accounting for 10% of the global wheat cultivation area, with an estimated production of 40 million tons [1–4]. The Mediterranean Rim and North America (Canada and the United States) are the primary producers, together contributing approximately 60% of global durum wheat production [1, 5, 6].

Domesticated around 12,000 years ago in the Fertile Crescent, durum wheat spread westward to the Mediterranean Rim and the Iberian Peninsula [4, 5]. The evolution of domesticated emmer wheat (Triticum turgidum ssp. dicoccoides) through natural and human selection led to the establishment of local landraces, which exhibit high genetic diversity and resilience to both abiotic and biotic stresses. During its spread across the Mediterranean basin, distinct adaptation strategies emerged, resulting in landraces with diverse phenological traits suited to different agroecological zones [3, 4, 7].

Owing to their heterogeneity and high phenotypic and genetic diversity, landraces were extensively used in breeding programs across the Mediterranean region and at international research institutes such as the International Maize and Wheat Improvement Center (CIMMYT) and the International Center for Agricultural Research in the Dry Areas (ICARDA) [5, 6, 8]. These breeding efforts led to the development of high-yielding cultivars with enhanced grain quality and resistance to several abiotic and biotic stresses [8]. Additionally, durum wheat production increased significantly due to improved agronomic practices [4, 9]. While these advancements helped mitigate the challenges of rapid population growth following the Green Revolution, they also contributed to the decline of landrace cultivation, as farmers increasingly adopted modern cultivars. This shift reduced the availability of genetically diverse landraces, leading to the widespread cultivation of genetically uniform, high-performing varieties and, consequently, to genetic diversity erosion. The reliance on a limited number of ancestral lines in breeding further narrowed the genetic background of modern cultivars. As a result, the global distribution of a restricted set of cultivars likely facilitated the emergence of new biotic stresses capable of overcoming host genetic resistance [4, 8].

Most of the major durum wheat producers are also the largest consumers, with Algeria, Italy, and Tunisia among the top countries [10]. Durum wheat is deeply embedded in the dietary traditions of the Mediterranean region and remains a staple food, with key end-use products including pasta, semolina, couscous, and bulgur [6]. The Mediterranean Rim is characterized by hot dry summers and mild wet winters, making October the typical sowing period and June the harvest season. Despite its adaptability to high temperatures and semi-arid environments, durum wheat production remains highly vulnerable to climate change [8, 11]. Average yield varies depending on genotype, geographical location, and climatic conditions. Additionally, production is influenced by soil fertility, agronomic practices, and both abiotic and biotic stresses.

In 2024, global durum wheat consumption surpassed production, which decreased by 10% compared to the previous cropping season [12]. In the Mediterranean region, durum wheat production has gradually declined because of its lower profitability as opposed to bread wheat [4]. Durum wheat production is highly influenced by climate conditions, with rising temperatures and fluctuating precipitation leading to lower yields and grain quality. Consequently, its genetic potential has been reduced, a trend expected to worsen in the future [1, 2, 13]. For instance, in 2024, an extended period of drought in Spain and France and harvest-season rainfall in Italy resulted in a 7% reduction in durum wheat production in the European Union compared to the previous year [14].

Understanding genotype responses to environmental changes is crucial for assessing climate change impacts on yield. For breeders, stable performance across diverse environments is essential for developing resilient varieties with durable resistance to biotic and abiotic stresses while maintaining high grain quality. Achieving food security amid population growth, water scarcity, and emerging pests requires integrating novel traits into high-yielding cultivars.

Landraces and wild relatives offer a valuable genetic reservoir for breeding programs, facilitating the introgression of beneficial alleles for multi-environment adaptation, high yield, and improved quality [3]. These resources have contributed to gene discoveries and trait enhancement in modern cultivars [15]. However, many landraces remain underutilized due to limited knowledge of their genetic diversity, agronomic performance, and/or resistance traits. To fully harness these resources, understanding genetic diversity, population structure, and marker associations for climate resilience is essential, especially given durum wheat’s sensitivity to environmental conditions. While breeding programs have traditionally prioritized yield, multi-trait selection offers a balanced approach, integrating yield with resistance to biotic and abiotic stresses [1, 8]. Expanding the genetic base of cultivars enhances stability and productivity.

Recent breakthroughs in biotechnology, statistics, genomics, and phenomics are playing a crucial role in addressing current and future challenges in plant breeding. The rapid expansion of genomic tools has enabled a more detailed characterization of the molecular genetic diversity of durum wheat and facilitated the identification of chromosomal regions associated with valuable traits. Advancements in next-generation sequencing (NGS) technologies have made sequencing large genomes, such as durum wheat, more cost-effective. Several high-throughput platforms, including single nucleotide polymorphism (SNP) arrays and genotyping-by-sequencing (GBS) methods like DArTseq (developed by Diversity Arrays Technology Pty Ltd., Canberra, Australia), are now widely available for wheat research. For years, assembling the large and complex genome of wheat posed significant challenges, and a complete reference genome for the 10.45-gigabase (Gb) durum wheat cultivar ‘Svevo’ was only recently achieved [16]. The availability of this high-quality reference genome has facilitated research in disease resistance, abiotic stress tolerance, grain quality, comparative genomics, functional genomics, and synthetic biology, enabling more effective breeding strategies and improved wheat varieties.

Understanding the genetic diversity and evolutionary history of durum wheat is essential for its conservation and improvement in breeding programs. This study aimed to assess the genetic diversity, population structure, and linkage disequilibrium of a core collection of durum wheat accessions, including landraces, breeding lines, and cultivars from the Mediterranean Rim, using two types of molecular markers: SNPs and SilicoDArTs.

Methods

Plant materials

A total of 258 spring tetraploid wheat accessions obtained from the United States Department of Agriculture (USDA) National Small Grains Collection (https://www.ars.usda.gov/pacific-west-area/aberdeen-id/small-grains-and-potato-germplasm-research/docs/national-small-grains-collection/), Aberdeen, ID, were investigated for their genetic diversity and population. The collection included 125 landraces, 54 breeding lines, 43 cultivars, 34 accessions of uncertain improvement status, and two near-isogenic lines (Supplementary Table S1). The accessions were collected from eleven different countries in the Mediterranean basin, including Algeria (39 accessions), Egypt (3), France (45), Israel (2), Italy (54), Jordan (2), Morocco (2), Portugal (51), Spain (50), Tunisia (6), and Turkey (4) (Supplementary Table S1).

All 258 accessions were multiplied and checked for uniformity during the 2016–2017 and 2017–2018 cropping seasons at the CRP Wheat Septoria Precision Phenotyping Platform, Kodia Experimental Station (36°32′51.89 N, 9°0′40.73 E), National Institute of Field Crops, Bou Salem, Tunisia, as described by Laribi et al. [17]. Seeds from a single plant were selected for each accession, kept distinct from the original population, and used for seed multiplication in the following season, as well as for DNA extraction and sequencing.

DNA extraction and genotyping

Five seeds from each selected plant were grown at 20 °C for 2 weeks under growth cabinet conditions (16 h light/8 h dark) at the University of Alberta, Edmonton, Canada. Leaves were then collected, flash-frozen in liquid nitrogen, and stored at −80 °C. Samples were lyophilized for 24 h using a freeze dryer (Labconco, Kansas, USA) and sent for DNA extraction and genotyping at the Genetic Analysis and Service for Agriculture (SAGA) Laboratory in El Batán, Mexico. Genomic DNA was extracted using the cetyltrimethylammonium bromide (CTAB) method in 96-well plates [18]. The quality and quantity of the DNA were assessed with a NanoDrop 8000 spectrophotometer (V 2.1.0; Thermo Fisher Scientific, Waltham, MA, United States), and genotyping was conducted using DArTseq™ technology (Diversity Arrays Technology Pty Ltd.), as described by [17]. This technology generated two types of markers: co-dominant (SNP) and dominant markers representing the presence or absence of restriction fragments of specific sequences (SilicoDArT), which were exported using DArTsoft14. All markers were mapped to the reference durum wheat genome, Svevo RefSeq Rel. 1.0, available online at https://wheat.pw.usda.gov/GG3/node/759. The physical positions of the markers on the reference genomes (Svevo, Chinese spring and consensus map) and their sequence reads were provided by SAGA.

Marker filtering

For SNP markers, accessions missing > 20% of genotype data and those with a minor allele frequency (MAF) > 5% were removed through filtering. SilicoDArT markers were filtered based on several criteria: first, monomorphic markers and those with a polymorphic information content (PIC) < 0.05 were excluded. Subsequently, markers with a read depth ≤ 5, reproducibility ≤ 0.96, and a call rate ≤ 0.88 were also eliminated [19].

Genetic diversity indices

Gene diversity (GD), PIC, and MAF were determined using PowerMarker 3.25 for the retained SNPs after filtering [20]. The PIC for SilicoDArT markers was provided by SAGA and recalculated using the “DArTR” package in R [19] for markers retained after filtering. The density of SNP and SilicoDArT markers was plotted within a 10 Mb window size across genomes A and B, as well as scaffolds (ChrUn) using the R package “CM-plot” [21]. Observed heterozygosity (Ho), expected heterozygosity (He), and total heterozygosity (Ht) were calculated using the gl.basic.stats function in the “DArTR” package in R [19].

The Analysis of Molecular Variance (AMOVA) was conducted for the subpopulations determined by STRUCTURE and Discriminant Analysis of Principal Components (DAPC) to estimate genetic differentiation, using the “hierfstat” and “adegenet” packages [22, 23].

Linkage disequilibrium analysis

Linkage disequilibrium (LD) for both filtered SNP and SilicoDArT markers was assessed using TASSEL v.5.2.94, with an LD window size of 50 bp for each chromosome [24]. The critical pairwise marker correlation value r², which represents the critical threshold below which the LD is attributed to true genetic linkage, was estimated as the 95th percentile of the distribution of square root transformed inter-chromosomal LD values (r²) of 2,023 SNP and 16,211 unlinked SilicoDArT markers [25]. A false discovery rate (FDR) of 0.01 was established using the Benjamini and Hochberg [26] method to account for multiple testing and reduce potential bias. The LD curve was calculated and presented as a locally-weighted polynomial regression (LOESS) [26] in R using the “ggplot2” package [27]. The intersection points between the LD curve and the r² threshold determined the LD decay value, which is defined as the genetic distance (Mb) at which the average maximum value of r² drops to half for each chromosome. The LD decay pattern analysis was performed according to the Hill and Weir function at the half decay value (r² = 0.228 for SNPs and 0.270 for SilicoDArTs, p < 0.01) [28, 29] in R software.

Population structure analysis

Population structure analysis was conducted using STRUCTURE with both SNP and SilicoDArT markers. The analysis included 258 durum wheat genotypes and utilized the admixture model implemented in STRUCTURE v2.3.4 [30]. The number of clusters tested ranged from 1 to 10, with 10 iterations and no prior information regarding genotype assignments. Each run had a burn-in length of 100,000 iterations and MCMC run length of 100,000 permutations. The ΔK method was employed to determine the optimal number of clusters, utilizing the online tool Structure Harvester [31, 32]. A membership probability threshold of 0.70 was applied to assign genotypes to different subpopulations; genotypes not meeting this threshold were considered admixed.

The discriminant analysis of principal components (DAPC) was employed to identify the number of clusters using the R packages “adegenet” and “popr” [22] with both SNP and SilicoDArT markers. The Bayesian information criterion (BIC) was utilized to determine the optimal number of clusters, with the number of subpopulations set from 1 to 15. The K value corresponding to the lowest BIC, after which the BIC value increases, was considered the optimal number of clusters. Subsequently, discriminant analysis was performed using the groups identified through k-mean clustering.

Cluster analysis

The genetic diversity among the 258 durum wheat accessions was assessed using the Neighbor-Joining (NJ) method implemented in TASSEL 5.2.94 [24] for both SNPs and SilicoDArTs. Kinship clustering for these markers was performed using the “GAPIT” package in R [33].

Results

Marker distribution and diversity

A total of 48,846 SNPs and 80,178 SilicoDArT markers were used to screen the 258 genotypes. Among the SNPs, 16,366 and 17,528 markers were located on the A- and B-genomes, respectively, while 14,952 were on scaffolds. After filtering, 8,015 SNP markers were retained for further analysis, with 2,854 on the A-genome, 3,138 on the B-genome, and 2,023 unlinked markers without chromosomal positions (Table 1). For the SilicoDArT markers, 16,591 and 18,774 were positioned on the A- and B-genomes, respectively, while 44,813 were on scaffolds. After filtering, a total of 28,200 markers were retained, comprising 5,328 on the A-genome, 6,661 on the B-genome, and 16,211 on scaffolds (Table 1).

Table 1.

Distribution and diversity indices of 8,015 single nucleotide polymorphism (SNP) markers and 28,200 silicoDArT markers in a set of 258 durum wheat accessions

Chromosome	SNP Markers (Before Filtering)	SNP Markers (After Filtering)	Polymorphic Markers	Gene Diversity	PICa	SilicoDArT Markers (Before Filtering)	SilicoDArT Markers (After Filtering)	PICa
1A	1863	340	328	0.293	0.240	1870	567	0.269
2A	2529	495	474	0.282	0.232	2726	867	0.271
3A	2389	380	368	0.296	0.242	2411	666	0.278
4A	1956	264	259	0.278	0.229	2199	743	0.268
5A	2466	445	427	0.310	0.253	2263	672	0.270
6A	1930	282	267	0.270	0.224	2036	730	0.270
7A	3233	648	621	0.264	0.221	3086	1083	0.258
Subtotal/mean	16,366	2854	2744	0.285	0.235	16,591	5328	0.269
1B	2440	437	415	0.281	0.232	2601	912	0.286
2B	3003	552	540	0.316	0.258	3299	1209	0.297
3B	2796	515	494	0.274	0.227	2981	1119	0.286
4B	1507	278	271	0.291	0.239	1718	483	0.267
5B	2526	426	411	0.292	0.240	2682	861	0.282
6B	2496	438	425	0.289	0.238	2605	965	0.296
7B	2760	492	473	0.287	0.237	2888	1112	0.294
Subtotal/mean	17,528	3138	3029	0.290	0.239	18,774	6661	0.287
1	4303	777	743	0.287	0.236	4471	1479	0.277
2	5532	1047	1014	0.299	0.245	6025	2076	0.284
3	5185	895	862	0.285	0.235	5392	1785	0.282
4	3463	542	530	0.284	0.234	3917	1226	0.268
5	4992	871	838	0.301	0.246	4945	1533	0.276
6	4426	720	692	0.280	0.231	4641	1695	0.283
7	5993	1140	1094	0.275	0.229	5974	2195	0.276
Subtotal/mean	33,894	5992	5773	0.287	0.237	35,365	11,989	0.278
Unassigned
Subtotal/mean	14,952	2023	1990	0.273	0.227	44,813	16,211	0.255
Total A + B + Unassigned	48,846	8015	7763	0.280	0.232	80,178	28,200	0.266

^aPIC Polymorphic information content

SNP markers were distributed across all chromosomes, with the highest number on chromosome 7A (648) and the lowest on chromosome 6A (282). For SilicoDArT markers, the highest count was on chromosome 2B (1,209), while the lowest was on chromosome 4B (483) (Table 1; Fig. 1). Coverage for SNPs ranged from 542 on chromosomes 4A and 4B to 1,140 on chromosomes 7A and 7B (Table 1; Fig. 1). Similarly, SilicoDArT markers had coverage ranging from 1,226 on chromosomes 4A and 4B to 2,195 on chromosomes 7A and 7B (Table 1; Fig. 1). The distribution and density of SNP and SilicoDArT markers across durum wheat chromosomes and scaffolds are illustrated in Fig. 2. Both types of markers were unevenly distributed among the 14 chromosomes (Table 1; Fig. 2), with SilicoDArT markers showing a more equitable distribution and better overall coverage than the SNPs (Table 1; Fig. 2). Coverage within ten megabase (Mb) pairs confirmed that SNPs exhibited lower density compared to SilicoDArTs (Fig. 2).

Fig. 1.

Distribution of SNP and SilicoDArT markers across all chromosomes in the durum wheat ‘Svevo’ reference genome

Fig. 2.

(A) SNP density plot showing the number of SNPs within a 10 Mb window size along the A- and B-genomes, including scaffolds (ChrUn). (B) SilicoDArT density plot displaying the number of SilicoDArT markers within a 10 Mb window size across the A- and B-genomes and scaffolds (ChrUn). The horizontal axis represents chromosome length (Mb), with different colors indicating marker density

Marker polymorphism

The mean PIC for SNPs and SilicoDArTs were 0.280 and 0.266, respectively (Table 1). For SNPs, the genetic diversity in the A-genome, B-genome, and scaffolds were similar, with Nei’s gene diversity and PIC values of 0.285 and 0.235 for the A-genome, 0.290 and 0.239 for the B-genome, and 0.273 and 0.227 for the scaffolds (Table 1). Among SNP markers, the highest genetic diversity in the A-genome was observed on chromosome 5A (Nei’s = 0.310; PIC = 0.253), while the lowest was on chromosome 7A (Nei’s = 0.264; PIC = 0.221) (Table 1). In the B-genome, chromosome 2B exhibited the highest genetic diversity (Nei’s = 0.316; PIC = 0.258), whereas chromosome 3B had the lowest (Nei’s = 0.274; PIC = 0.227) (Table 1). Nei’s genetic diversity and PIC values were generally consistent across homoeologous chromosome pairs, ranging from 0.275 to 0.301 and from 0.229 to 0.246, respectively (Table 1). For SilicoDArT markers, PIC values were highest in the B-genome (0.287), followed by the A-genome (0.269) and scaffolds (0.255) (Table 1). In the A-genome, chromosome 3A had the highest PIC value (0.278), while chromosome 7A had the lowest (0.258). In the B-genome, the highest PIC value was observed on chromosome 2B (0.297), and the lowest on chromosome 4B (0.267) (Table 1). Across homoeologous chromosome pairs, PIC values ranged from 0.268 to 0.28 (Table 1).

The observed heterozygosity (Ho) was 0.34 for SNPs and 0.68 for SilicoDArT markers, while the expected heterozygosity (He) was 0.22 for SNPs and 0.40 for SilicoDArTs. Total heterozygosity (Ht) was 0.23 for SNPs and 0.41 for SilicoDArTs. These results highlight the high genetic diversity of the 258 durum wheat accessions.

Linkage disequilibrium

The homoeologous chromosomes of the A- and B-genomes showed different LD decay patterns for both SNP and SilicoDArT markers, confirming genetic differences between both genomes (Fig. 3). For SNPs, the LD halving distance exhibited a slower decay in the A-genome for chromosomes 1, 5, and 7 (Fig. 3A), whereas in the B-genome, chromosomes 1, 4, and 5 showed a slower decay (Fig. 3B). The LD decay patterns of chromosomes 5 A and 5B were highly similar (Figs. 3A-B).

Fig. 3.

Linkage disequilibrium (LD) decay pattern according to the Hill and Weir function for: (A) A-genome SNP markers, (B) B-genome SNP markers, (C) A-genome SilicoDArT markers, and (D) B-genome SilicoDArT markers

The LD half decay was estimated for all 14 chromosomes based on an r² threshold of 0.1, following the method of Hill and Weir [34]. It ranged from 2.78 Mb for chromosome 6A to 6.42 Mb for chromosome 1B (Table 2; Fig. 3A-B). The average LD half decay was 4.56 Mb for the A-genome and 4.83 Mb for the B-genome, with an overall average of 4.69 Mb for both genomes combined (Table 2). Chromosome 7A had the highest critical r² value (0.29), followed by chromosomes 3A, 1A, and 2A, while chromosomes 4A, 6A, 3B, and 7B had the lowest critical r² values (Table 2).

Table 2.

SilicoDArT and SNP markers and extent of intra-chromosomal linkage disequilibrium in 258 durum wheat accessions from the mediterranean basin

Marker	Chromosome	Marker Pairs	No. and Rate of markers in Significant LDa	Mean r2of All Marker Pairs	Critical r2	% r2 < 0.1	LD decay (Mb)b	LD Half Decay (r2 = 0.1) (Mb)	LD (r2 = 1)
SNP	1A	14,416	4118 (28.57)	0.078	0.27	29.72	8.40	5.38	1.75
	2A	22,754	6714 (29.51)	0.085	0.26	32.38	7.15	4.19	4.94
	3A	17,953	4457 (24.83)	0.080	0.27	38.32	7.13	4.67	4.51
	4A	11,790	2484 (21.07)	0.074	0.17	40.46	6.20	3.43	2.46
	5A	20,397	5154 (25.27)	0.074	0.24	41.95	7.50	5.72	4.58
	6A	12,895	2680 (20.78)	0.072	0.18	38.73	4.36	2.78	2.50
	7A	29,635	8446 (28.50)	0.074	0.29	24.25	9.50	5.78	3.81
	1B	20,043	5339 (26.64)	0.079	0.23	32.55	10.18	6.42	1.93
	2B	26,114	7035 (26.94)	0.081	0.22	37.50	7.46	5.02	3.04
	3B	22,893	4852 (21.19)	0.076	0.18	37.41	4.68	2.84	2.25
	4B	12,524	3141 (25.08)	0.074	0.23	35.82	8.63	5.61	1.75
	5B	19,283	4543 (23.56)	0.073	0.24	36.03	8.78	5.52	2.60
	6B	19,277	4583 (23.77)	0.072	0.23	36.11	8.45	4.83	1.88
	7B	22,131	4572 (20.66)	0.073	0.18	40.18	5.67	3.55	2.14
	A-genome	129,840	34,053(26.23)	0.077	0.24	35.12	7.71	4.56	3.79
	B-genome	142,265	34,065 (23.94)	0.075	0.22	36.51	7.61	4.83	2.30
	A + B	272,105	68,118 (25.03)	0.076	0.23	35.82	7.50	4.69	3.04
SilicoDArT	1A	27,075	5308 (19.60)	0.058	0.21	38.00	0.46	2.37	0.72
	2A	43,350	11,199 (25.83)	0.085	0.24	35.28	0.48	2.45	2.56
	3A	33,300	6444 (19.35)	0.059	0.21	39.23	0.47	2.49	1.23
	4A	37,150	7702 (20.73)	0.052	0.16	42.09	0.22	2.45	1.34
	5A	33,600	6406 (19.07)	0.053	0.18	45.57	0.52	2.36	1.48
	6A	36,500	9055 (24.81)	0.066	0.22	35.48	0.20	1.29	0.95
	7A	54,150	11,209(20.70)	0.055	0.15	38.45	0.20	1.22	1.06
	1B	45,600	11,932(26.17)	0.069	0.21	36.75	0.44	1.43	0.91
	2B	60,450	16,561(27.40)	0.077	0.25	37.26	0.46	2.82	1.63
	3B	55,950	11,673(20.86)	0.054	0.18	41.22	0.24	2.09	1.12
	4B	24,150	4169 (17.26)	0.043	0.09	45.67	0.29	1.12	1.20
	5B	43,050	8780 (20.39)	0.052	0.17	42.16	0.37	1.30	1.30
	6B	48,250	11,815(24.48)	0.063	0.19	39.16	0.24	1.08	0.78
	7B	55,600	12,974(23.33)	0.059	0.20	39.69	0.30	1.73	0.81
	A-genome	265,125	57,323(21.62)	0.061	0.20	39.16	0.36	2.09	1.52
	B-genome	333,050	77,904(23.39)	0.059	0.18	40.27	0.34	1.65	1.04
	A + B	598,175	135,227(22.61)	0.060	0.19	39.72	0.35	1.87	1.24

LD Linkage disequilibrium, Mb megabase, r² the critical pairwise marker correlation value, SNP single nucleotide polymorphism

^a The number and percentage of marker pairs in significant Linkage Disequilibrium (LD) (p < 0.001)

^b LD decay was estimated from the projection of the intersection between the fitted curve of the data points and the 95th percentile of unlinked r² threshold line onto the physical distance axis

The proportion of marker pairs in significant LD (p < 0.001) was 26.23% and 23.94% for the A- and B-genomes, respectively, with an overall proportion of 22.61% and a mean r² value of 0.076. This proportion varied among sub-genomes, ranging from 20.66% for chromosome 7B to 29.51% for chromosome 2A. Across all markers, 26.40% and 7.21% of significant marker pairs had r² values above 0.2 and 0.7, respectively. Chromosome 7B had the lowest proportion of significant marker pairs in LD (20.66%), while chromosome 2A had the highest (29.51%). The critical r² values were 0.24 for the A-genome, 0.22 for the B-genome, and 0.23 for the whole genome (Table 2).

The extent of LD decay across the 14 chromosomes ranged from 4.36 to 10.18 Mb, with an average of 7.71 Mb for the A-genome, 7.61 for the B-genome, and 7.50 Mb for the whole genome (Table 2). The LD decay range for individual chromosomes was 4.36–9.50 Mb in the A-genome and 4.68–10.18 Mb in the B-genome (Table 2). The percentage of significant marker pairs in complete intra-chromosomal LD (r² = 1) was 3.79% for the A-genome, 2.30% for the B-genome, and 3.04% for the whole genome. Chromosomes 2A, 2B, 5A, and 7A contained the highest number of marker pairs in complete LD, while 1A, 4A, and 6A had the lowest (Table 2).

For SilicoDArTs, the LD halving distance showed a faster decay in the A-genome for chromosomes 6 and 7 (Fig. 3C), whereas in the B-genome, chromosomes 2 and 3 exhibited a slower decay (Fig. 3D). The LD half decay at r² = 0.1 ranged from 1.08 Mb for chromosome 6B to 2.82 Mb for chromosome 2B (Table 2; Fig. 3C-D). The average LD half decay was 2.09 Mb for the A-genome and 1.65 Mb for the B-genome, with an overall average of 1.87 Mb across both genomes (Table 2).

Chromosome 2B had the highest critical r² value (0.25), followed by chromosomes 2A and 6A, while chromosomes 4B, 7A, and 4A had the lowest critical r² values (Table 2). The proportion of marker pairs in significant LD (p < 0.001) was 21.62% for the A-genome and 23.39% for the B-genome, with an overall proportion of 25.03% and a mean r² value of 0.060. This proportion varied across sub-genomes, ranging from 17.26% for chromosome 4B to 27.40% for chromosome 2B. Among significant marker pairs, 7.55% had r² values above 0.2, while 1.72% had r² values above 0.7. Chromosome 4B had the lowest proportion of significant marker pairs in LD (17.26%), while chromosome 2B had the highest (27.40%). The critical r² values were 0.20 for the A genome, 0.18 for the B genome, and 0.19 for the whole genome (Table 2).

The extent of LD decay across the 14 chromosomes ranged from 0.20 to 0.52 Mb, with an average of 0.36 Mb for the A-genome, 0.34 for the B-genome, and 0.35 Mb for the whole genome (Table 2). The LD decay range was 0.20–0.52 Mb for A-genome chromosomes and 0.24–0.46 Mb for B-genome chromosomes (Table 2). The percentage of significant marker pairs in complete intra-chromosomal LD (r² = 1) was 1.52% for the A-genome, 1.04% for the B-genome, and 1.24% for the whole genome. Chromosomes 2A and 2B contained the highest number of marker pairs in complete LD, while chromosomes 1A, 4B, and 3A had the lowest (Table 2). The frequency of r² values in each chromosome is shown in Supplementary Figure S1.

Population structure analysis

The optimal number of K was determined using the method of Evanno et al. [31]. Accessions were assigned to subpopulations based on a membership probability threshold of ≥ 0.70 for both SNP and SilicoDArT markers. SNP markers indicated the presence of two subpopulations among the tested accessions. Subpopulation 1 contained 88 accessions (34.11%), while Subpopulation 2 comprised 102 accessions (39.53%). The remaining 68 accessions (26.36%) were classified as admixture (Fig. 4A, Supplementary Table S1). Similarly, SilicoDArT markers identified two subpopulations. Subpopulation 1 contained 137 accessions (53.10%), Subpopulation 2 contained 72 accessions (27.91%), and 49 accessions (18.99%) were classified as admixture (Fig. 4B, Supplementary Table S1).

Fig. 4.

Bayesian clustering of 258 durum wheat accessions from 11 Mediterranean countries based on: (A) 8,015 SNP markers sorted by Q, (B) 28,200 SilicoDArT markers sorted by Q, (C) assignment of SNP subpopulations in accordance with SilicoDArT subpopulations, and (D) assignment of SilicoDArT subpopulations in accordance with SNP subpopulations

Based on SNP marker assignments, the admixture group included accessions from all countries except Turkey. Subpopulation 1 primarily consisted of accessions from Portugal (45) and Spain (20), with the remaining accessions originating from other countries, excluding Israel and Jordan (Supplementary Table S2). Subpopulation 2 was mainly composed of accessions from Algeria (18), France (27), Italy (38), and Spain (14), while the remaining accessions came from various countries, excluding Morocco and Portugal. Most landraces (60) were assigned to Subpopulation 1, whereas 40 breeding lines and 29 cultivars were classified under Subpopulation 2. Accessions with uncertain improvement status were relatively evenly distributed among subpopulations: 14 in Subpopulation 1, 9 in Subpopulation 2, and 11 in the admixture group (Supplementary Table S2).

Based on SilicoDArT marker assignments, the admixture group primarily included accessions from France (24) and Algeria (8), while accessions from other countries ranged from 0 to 5 (Supplementary Table S2). Subpopulation 1 mainly consisted of accessions from Portugal (50) and Spain (35), with the remaining accessions originating from other countries, excluding Jordan. Subpopulation 2 included accessions only from Algeria (14), France (10), Italy (36), Spain (11), and Turkey (1). Most landraces (94) were assigned to Subpopulation 1, while 28 breeding lines and 20 cultivars were classified under Subpopulation 2 (Supplementary Table S1). Additionally, 22 accessions of uncertain improvement status were assigned to Subpopulation 1 (Supplementary Table S2).

Comparison of population structure assignment based on SNPs and SilicoDArTs

Among the 68 accessions classified as admixture based on SNP markers, 23.53% remained in the admixture category, while 76.47% were reassigned to Subpopulation 1 according to STRUCTURE results for SilicoDArT markers (Fig. 4C, Supplementary Table S1). Of the 88 accessions assigned to Subpopulation 1 using SNP markers, 3.41% were classified as admixture, and 96.59% remained in Subpopulation 1 based on SilicoDArT marker analysis. Similarly, among 102 accessions assigned to Subpopulation 2 using SNP markers, 29.41% were classified as admixture, while 70.59% remained in Subpopulation 2 according to STRUCTURE results for SilicoDArT markers (Fig. 4C, Supplementary Table S1).

Among the 49 accessions classified as admixture based on SilicoDArT markers, 36.65%, remained in the admixture category, while 2.94% and 44.12% were reassigned to Subpopulation 1 and Subpopulation 2, respectively, according to STRUCTURE results for SNP markers (Fig. 4D, Supplementary Table S1). Of the 137 accessions assigned to Subpopulation 1 using SilicoDArT markers, 37.96% were classified as admixture, while 62.04% remained in Subpopulation 1 based on SNP marker analysis. All 72 accessions assigned to Subpopulation 2 using SilicoDArT markers were consistently classified as Subpopulation 2 according to STRUCTURE results for SNP markers (Fig. 4D, Supplementary Table S1).

Discriminant Analysis of Principal Components (DAPC)

DAPC analysis was conducted using both SNP and SilicoDArT markers to determine the optimum number of subpopulations and indicated the presence of nine clusters for SNPs and eight clusters for SilicoDArTs (Figs. 5 and 6).

Fig. 5.

(A) BIC versus the number of clusters, indicating the optimal number of subpopulations (red arrow) determined by Discriminant Analysis of Principal Components (DAPC) for SNP markers. (B) Table comparing the distribution of accessions assigned to clusters based on STRUCTURE analysis using DAPC for SNP markers. (C) Graphical representation of the nine clusters identified using DAPC for SNP markers

Fig. 6.

(A) BIC versus the number of clusters, indicating the optimal number of subpopulations (red arrow) determined by Discriminant Analysis of Principal Components (DAPC) for SilicoDArT markers. (B) Table comparing the distribution of accessions assigned to clusters based on STRUCTURE analysis using DAPC for SilicoDArT markers. (C) Graphical representation of the nine clusters identified using DAPC for SilicoDArT markers

For SNPs, DAPC results showed that the optimal number of principal components (PCs) was nine (Fig. 5). The identified clusters (1–9) contained 27, 79, 13, 36, 20, 13, 20, 43, and 7 accessions, respectively. Overall, 89.77% of the accessions assigned to Subpopulation 1 in STRUCTURE were grouped into clusters 2 (60.23%), 3 (14.77%), and 6 (14.77%) (Fig. 5B, Supplementary Table S3). Similarly, 94.12% of accessions assigned to Subpopulation 2 were classified into clusters 1 (26.47%), 4 (26.47%), and 8 (41.18%). Meanwhile, 85.29% of the accessions assigned as admixture in STRUCTURE were distributed among clusters 2 (35.29%), 5 (11.76%), and 7 (27.94%) (Fig. 5B, Supplementary Table S3).

For SilicoDArTs, DAPC results indicated that the optimal number of PCs was eight (Fig. 6, Supplementary Table S4). The identified clusters (1–8) contained 49, 25, 20, 11, 23, 92, 22, and 13 accessions, respectively. Overall, 83.21% of the accessions assigned to Subpopulation 1 in STRUCTURE were grouped into clusters 3 (10.95%), 6 (55.47%), and 7 (16.79%) (Fig. 6B, Supplementary Table S4). Similarly, 79.17% of accessions assigned to Subpopulation 2 were classified into clusters 1 (54.17%) and 5 (25.00%). Meanwhile, 75.51% of accessions categorized as admixture in STRUCTURE were assigned to clusters 2 (32.65%), 6 (24.49%), and 8 (18.37%) (Fig. 6B, Supplementary Table S4).

Cluster analysis

Neighbor joining analysis identified four clusters for SNP markers and five clusters for SilicoDArT markers (Fig. 7).

Fig. 7.

Neighbor-joining cluster analysis of 258 durum wheat accessions using (A) SNP markers and (B) SilicoDArT markers

For SNP markers, clusters 1, 2, 3, and 4 contained 64, 56, 127, and 11 accessions, respectively (Fig. 7A, Supplementary Table S5). Overall, 31 out of 39 (79.49%) accessions from Algeria were assigned to cluster 1, while the remaining eight (20.51%) were in cluster 3. Similarly, 39 out 45 (86.66%) accessions from France were assigned to cluster 3, while five were in cluster 1 and one was in cluster 2 (Fig. 7A, Supplementary Table S5). Among the 54 accessions from Italy, 48 (88.88%) were assigned to cluster 3, while two, three, and one accession(s) were in clusters 1, 2 and 4, respectively. Of the 51 accessions form Portugal, 37 (72.55%) were assigned to cluster 2. Accessions from Spain were distributed across all four clusters, with 19, 10, 13, and 8 accessions assigned to clusters 1, 2, 3, and 4, respectively (Fig. 7A, Supplementary Table S5).

Cluster 1 contained 31 and 19 accessions from Algeria and Spain, respectively, accounting for 78.13% of the accessions (Fig. 7A, Supplementary Table S5). Cluster 2 included 37 accessions from Portugal and 10 from Spain, representing 83.93% of the accessions. Cluster 3 comprised 39, 48, and 13 accessions from France, Italy, and Spain, respectively, making up 78.74% of the accessions. Cluster 4 contained only 11 accessions, including eight from Spain, two from Portugal, and one from Italy (Fig. 7A, Supplementary Table S5). The three accessions from Egypt were assigned to clusters 2 and 3, while the two accessions from Israel were assigned to clusters 1 and 3. Accessions from Jordan and Morocco were placed in clusters 3 and 2, respectively. Accessions from Tunisia and Turkey were distributed among clusters 1, 2, and 3 (Fig. 7A, Supplementary Table S5).

Clusters 1 and 2 primarily contained landraces, with 51 and 40 accessions, respectively. Most cultivars (36 out of 43) were assigned to cluster 3, while the remainder were distributed between clusters 1 and 2. Accessions with uncertain improvement status were mostly assigned to clusters 2 and 3 (11 and 16 out of 34, respectively). Among the breeding lines, 48 out of 54 were placed in cluster 3, while the rest were assigned to clusters 1 and 2. Both near isogenic lines were assigned to cluster 3 (Fig. 7A, Supplementary Table S5).

For SilicoDArTs, clusters 1, 2, 3, 4 and 5 contained 34, 150, 31, 30, and 13 accessions, respectively (Fig. 7B, Supplementary Table S5). All accessions from Algeria were assigned to clusters 2 (21 accessions) and 4 (18 accessions). Overall, 91.11% of accessions from France were placed in cluster 2, while the remaining were assigned to clusters 3 and 4 (Fig. 7B, Supplementary Table S5). Similarly, 90.74% of accessions from Italy were assigned to cluster 2, with the rest distributed among clusters 1, 3, and 5. Among the accessions from Portugal, 74.51% were assigned to clusters 1 (25 accessions) and 3 (14 accessions). Accessions from Spain were distributed across all clusters, with 6, 18, 10, 7, and 9 accessions in clusters 1 to 5, respectively (Fig. 7B, Supplementary Table S5).

Cluster 1 contained 25 accessions from Portugal, six from Spain, and one from Italy (Fig. 7B, Supplementary Table S5). Cluster 2 included accessions from all countries. Cluster 3 was composed mainly of accessions from Portugal (14) and Spain (10). Cluster 4 contained 18 accessions from Algeria and seven from Spain, with the remaining accessions from France (three), Portugal (one), and Tunisia (one) (Fig. 7B, Supplementary Table S5). Cluster 5 included nine accessions from Spain and two each from Italy and Portugal (Fig. 7B, Supplementary Table S5).

Overall, 38 out of 43 cultivars (88.37%) were assigned to cluster 2 (Fig. 7B, Supplementary Table S5). Additionally, 20 out of 34 accessions (58.82%) with uncertain improvement status and 50 out of 54 breeding lines (92.59%) were placed in cluster 2 (Fig. 7B, Supplementary Table S5). Landraces were predominant in clusters 1, 3, 4, and 5, comprising 25 out of 34, 24 out of 31, 26 out of 30, and 10 out of 13 accessions, respectively (Fig. 7B, Supplementary Table S5).

Kinship analysis

Based on the kinship matrix, the 258 accessions were grouped into three clusters for SNP markers and four clusters for SilicoDArT markers (Fig. 8, Supplementary Table S6).

Fig. 8.

Heatmap plot of the kinship matrix displaying the relationships among 258 durum wheat accessions based on (A) SNP markers and (B) SilicoDArT markers

For SNPs, clusters 1, 2, and 3 comprised 40, 83, and 135 accessions, respectively (Fig. 8A, Supplementary Table S6). All accessions from Algeria were assigned to clusters 2 and 3 (six and 33 accessions, respectively). Accessions from France were evenly distributed between clusters 2 and 3 (23 and 22 accessions, respectively). Most accessions from Italy were assigned to clusters 2 and 3 (34 and 17 accessions, respectively) (Fig. 8A; Supplementary Table S6). Accessions from Portugal and Spain were primarily placed in clusters 1 and 3 (Fig. 8A, Supplementary Table S6).

Cluster 1 primarily comprised accessions from Portugal, accounting for 65% (26 accessions), along with eight accessions from Spain, three from Italy, and one accession each from Egypt, Tunisia, and Turkey (Fig. 8A, Supplementary Table S6). Cluster 2 mainly included accessions from France and Italy, representing 68.67% of the total, as well as accessions from all remaining countries, except Morocco. Cluster 3 contained accessions from all countries (Fig. 8A, Supplementary Table S6).

For SilicoDArTs, clusters 1, 2, 3, and 4 contained 24, 42, 26, and 166 accessions, respectively (Fig. 8B, Supplementary Table S6). Most accessions from Algeria (94.87%) were assigned to clusters 1 and 4, with 12 and 25 accessions, respectively. A total of 41 accessions from France were assigned to cluster 4, while the remaining four accessions were assigned to cluster 1. Accessions from Italy were mainly assigned to clusters 2 (35 accessions) and 4 (15 accessions), with two accessions assigned to both clusters 1 and 3. All accessions from Portugal were assigned to clusters 3 (19 accessions) and 4 (32 accessions), while 76% of accessions from Spain were assigned to cluster 4, with the remainder distributed among the other clusters (Fig. 8B, Supplementary Table S6). Overall, 50% of accessions in cluster 1 were from Algeria, with the remainder from France, Italy, Spain, Turkey, and Tunisia. Cluster 2 predominantly contained accessions from Italy (83.33%), while the remaining were from Algeria and Spain. In cluster 3, 73.08% of accessions were from Portugal, with the remainder from Italy, Spain, Tunisia, and Turkey. Cluster 4 included accessions from all countries (Fig. 8B, Supplementary Table S6).

Clustering methodologies comparison

Following Kabbaj et al. [35], a qualitative score was assigned to each accession based on STRUCTURE, DAPC, Neighbor joining, and kinship. The scoring criteria were as follows: ‘solid’ when all four methods agreed on the assignment, ‘good’ when three methods agreed, ‘fairly good’ when two methods agreed, and ‘bad’ when only DAPC made the assignment. For SNPs, of the 258 accessions, 155 (60.08%) were scored as ‘solid’, 60 (23.26%) as ‘good’, 40 (15.50%) as ‘fairly good’, and three (1.16%) as ‘bad’ (Table 3, Supplementary Table S7). For SilicoDArTs, 165 out of 258 accessions (63.95%) were scored as ‘solid,’ 75 (29.07%) as ‘good,’ 17 (6.59%) as ‘fairly good,’ and one (0.39%) as ‘bad’ (Table 3, Supplementary Table S8). Overall, clustering methodologies showed 83.34% and 93.02% agreement for SNPs and SilicoDArTs, respectively. For SNPs, clusters 1, 3, 5, 6, 7, and 9 were the most ‘solid’, with no ‘fairly good’ or ‘bad’ assignments and only a few ‘good’. In contrast, for SilicoDArTs, clusters 2, 3, 4, 5, and 7 were the most ‘solid’, showing no ‘fairly good’ or ‘bad’ assignments and a few ‘good’ assignments. Detailed assignments for each accession are provided in Supplementary Table S7 and Supplementary Table S8.

Table 3.

Reliability of the assignment of 258 durum wheat accessions based on comparisons among four genetic clustering methods: Discriminant Analysis of Principal Components (DAPC), STRUCTURE, Neighbor-joining, and kinship

	SNP					SilicoDArT
Cluster ID	Number of Accessions	Bad	Fairly Good	Good	Solid	Number of Accessions	Bad	Fairly Good	Good	Solid
1	27	0	0	7	20	49	0	5	4	40
2	79	3	34	20	22	25	0	0	9	16
3	13	0	0	0	13	20	0	0	3	17
4	36	0	4	10	22	11	0	0	1	10
5	20	0	0	5	15	23	0	0	4	19
6	13	0	0	0	13	92	1	10	52	29
7	20	0	0	1	19	25	0	0	1	24
8	43	0	2	13	28	13	0	2	1	10
9	7	0	0	4	3
Total	258	3	40	60	155	258	1	17	75	165
%	100	1.16	15.50	23.26	60.08	100	0.39	6.59	29.07	63.95

A qualitative score was assigned to each accession based on STRUCTURE, DAPC, Neighbor joining, and kinship. The scoring criteria were as follows: ‘solid’ when all four methods agreed on the assignment, ‘good’ when three methods agreed, ‘fairly good’ when two methods agreed, and ‘bad’ when only DAPC made the assignment. Assignment details are available in Supplementary Table S7 and Supplementary Table S8

Analysis of Molecular Variance (AMOVA)

The AMOVA using SNP markers for subpopulations assigned based on STRUCTURE analysis indicated that 12.37% of the variation was between populations, 74.86% was between samples within populations, and 12.77% was within samples. In contrast, the AMOVA using SNP markers for clusters assigned based on DAPC revealed that 25.95% of the variation was between populations, 61.30% was between samples within populations, and 12.76% was within samples. These results demonstrate that the DAPC model effectively captured one-fourth of the total variance by stratifying the 258 accessions into nine clusters.

Discussion

Compared to its wild relatives, durum wheat has been widely reported to have lost genetic diversity during domestication. Additionally, several studies indicate that the genetic distances between landraces and elite cultivars of this species are significant, suggesting an under-exploitation of landraces in breeding programs. Many reports highlight that landraces can be valuable sources of favorable alleles and genes that could be introduced into elite cultivars [5–7, 15, 36–38]. However, very limited germplasm has been consistently utilized in durum wheat improvement efforts [5–7, 15, 36–38].

In this context, investigating the genetic diversity of landraces, particularly from primary and secondary centers of wheat diversity, and comparing them to modern cultivars can help identify parental lines with diverse and desirable alleles for breeding programs [5]. Furthermore, examining wheat populations from different geographical origins provides insights into their diversity. It also helps eliminate redundant genotypes and facilitates the preservation of valuable genetic resources for the efficient selection of germplasm with specific traits. A comprehensive understanding of available genetic resources in gene banks, including their geographical origin, morphological and genetic diversity, and population stratification, offers valuable insights to breeders.

Recently, DArTseq, which provides SilicoDArT and SNP markers, has proven useful for investigating wheat diversity [39–42]. This platform has demonstrated cost-effectiveness and efficiency in exploring genetic diversity in crops and identifying markers associated with various traits, including responses to abiotic and biotic stresses [1, 2]. In this study, we investigated the efficiency of SilicoDArT and SNP markers in assessing the diversity of a collection of 258 spring tetraploid wheat accessions, which included 125 landraces, 54 breeding lines, 43 cultivars, 34 accessions of uncertain improvement status, and two near-isogenic lines from eleven countries in the Mediterranean basin.

Markers distribution

The use of genotyping-by-sequencing (GBS) in wheat research has rapidly expanded as a reliable, efficient, and cost-effective methodology for assessing genetic diversity and population stratification. A higher number of SNP and SilicoDArT markers were observed in the B-genome than in the A-genome, consistent with numerous previous studies [17, 35, 37, 39, 41–50]. All Triticum species are reported to share an A-genome that is identical to the chromosomes of the diploid Triticum boeoticum [15, 51], while the B-genome is more divergent from its ancestral donors. This divergence may explain the higher level of polymorphism observed in the B-genome [15, 42, 51, 52].

The highest number of markers was found on chromosome 7A for SNPs and 2B for SilicoDArTs. These results are similar to those of Robbana et al. [47] and Laribi et al. [17], who reported the highest SNP counts on chromosome 7A for two different collections of Tunisian durum wheat accessions. Despite the involvement of homoeologous group 4 in numerous structural arrangements in wheat [53], it exhibited the lowest coverage for both SNPs and SilicoDArTs, consistent with the findings by Pascual et al. [54] in a collection of Spanish landraces. Several other studies have reported similar results [15, 17, 35, 42, 50].

In this study, SilicoDArT markers were more evenly distributed than SNPs, providing overall better coverage with higher density across all chromosomes; however, neither marker type exhibited an equal distribution of markers along all chromosomes. Similarly, Tyrka et al. [55] found that SilicoDArTs and SNPs were not homogeneously distributed among chromosomes, with SilicoDArT markers providing better genome coverage for 277 European varieties. These findings align with those reported by Tyrka et al. [55] for a collection of 168 breeding lines and three cultivars of common winter wheat using DArTseq markers (SilicoDArTs and SNPs). The authors noted an uneven distribution of both marker types in the wheat genome, with SilicoDArT markers offering superior genome coverage. Moreover, Yang et al. [56] found that SilicoDArTs were detected two to five times more frequently than SNPs across all chromosomes in an Australian bread wheat panel. The uneven distribution and diversity of markers among chromosomes may result from a combination of genetic drift and selective bias under limited gene flow, self-pollination, low effective recombination, and structural rearrangements [55, 57, 58].

Marker polymorphism and genetic diversity

The PIC value is an indicator of evolutionary pressure on the allele within a tested panel. In this study, the tested panel exhibited similar mean PIC values of 0.280 and 0.266 for SNPs and SilicoDArTs, respectively. These values are comparable to those reported by Baloch et al. [44] for a set of durum wheat from the Central Fertile Crescent (0.265 for DArTseq and 0.302 for SNPs) and the PIC value of 0.246 reported by Laribi et al. [17] for a collection of durum wheat accessions from Tunisia. However, the PIC value in our study is higher than the 0.188 reported by Ren et al. [59] for a worldwide collection of durum wheat accessions and the 0.165 reported by Robbana et al. [47] for a collection of durum wheat landraces from Tunisia. Additionally, the PIC value for SNPs in our study is higher than that reported by Dukamo et al. [41] for Ethiopian landraces (0.184) and the value of 0.203 reported by Alemu et al. [48] for an Ethiopian collection of durum wheat. Fufa et al. [42] found an average PIC of 0.19 for tetraploid wheat germplasm from Ethiopia based on SNP DArTSeq markers. The PIC values in our study are similar to those reported by others, including Kabbaj et al. [35], who found a PIC value of 0.32 in a global panel of durum wheat using SNPs from the Axiom 35K array. Similarly, Rufo et al. [37] reported a higher PIC value of 0.30 using SNP markers from the 15K SNP array for a collection of bread wheat landraces and cultivars.

In our study, SilicoDArT markers exhibited higher PIC values than SNP markers for both the A- and B-genomes, as well as for the whole genome. This finding aligns with Yang et al. [56], who noted slight differences between SNPs (0.24) and SilicoDArTs (0.25) on the A- and B-genomes for a collection of Australian bread wheat. The same PIC (0.280) was reported by Soriano et al. [60] for a panel of 387 durum wheat genotypes, including 183 landraces from 24 Mediterranean and Eastern European countries, along with 204 commercial varieties from 24 wheat-growing areas worldwide. Mazzucotelli et al. [15] reported PIC values of 0.27 for landraces and 0.28 for modern cultivars in a global durum panel. Mulugeta et al. [50] reported a PIC value of 0.27 in a collection of Ethiopian durum wheat using 8,178 SNPs from the Illumina Infinium 25K wheat SNP array.

For each marker type in this study, small differences in PIC values were observed between the A- and B-genomes, with the latter being slightly higher, although these values varied among chromosomes. Gene diversity was similar for both A and B genomes for SNP markers, although these values also differed among chromosomes. This variation in PIC and gene diversity values across chromosomes suggests the effects of selection pressure, adaptation, and introgression of diversity from other germplasm [46]. No significant differences in PIC and gene diversity between A- and B-genome values were reported by several authors in different collections of durum wheat accessions [48, 50, 55, 61]. Overall, the PIC values observed in this study for both SNP and SilicoDArT markers are comparable to previous reports in both durum and bread wheat. However, variations in PIC values across different studies can be attributed to factors such as the types and number of markers used, the number of accessions studied, and the species involved. The PIC values for both SNP and SilicoDArT markers in this study are considered suitable for investigating genetic diversity.

The overall gene diversity reported in this study (0.280) is higher than the values reported by Alemu et al. [48], Dukamo et al. [41], and Ren et al. [59], which were 0.246, 0.199, and 0.22, respectively. However, it was lower than the gene diversity values of 0.338 and 0.34 reported by Laribi et al. [17] and Mulugeta et al. [50]. The gene diversity observed in this collection suggests significant genetic variation, likely influenced by adaptation to different agro-ecological conditions, varying selection criteria, and the presence of seed mixtures along the Mediterranean Rim.

For SNPs, the lowest genetic diversity was observed on chromosome 7A (Nei’s = 0.264; PIC = 0.221). Similar findings were reported by Alemu et al. [48], who observed the lowest PIC and gene diversity scores were on chromosome 7A (PIC = 0.181; gene diversity = 0.217) in an Ethiopian durum wheat collection using SNP markers. Likewise, a diverse panel of landraces and cultivars showed the lowest PIC value on chromosome 7A (0.26). In contrast, homoeologous chromosome group 5 had the highest PIC and gene diversity for SNPs, consistent with the findings of Alemu et al. [48].

The B-genome exhibited the highest genetic diversity, a result previously reported by Ren et al. [59] in a worldwide collection of durum wheat germplasm. The expected heterozygosity in this study was 0.22 for SNPs and 0.40 for SilicoDArTs, which is higher than the range reported by Mulugeta et al. [50] (0.02 to 0.18 across all loci).

The AMOVA using SNP markers for subpopulations assigned based on STRUCTURE analysis revealed that 12.37% of the variation was between populations, 74.86% was among accessions within populations, and 12.77% was within accessions. Similarly, the AMOVA based on clusters assigned using DAPC analysis showed that 25.95% of the variation was between populations, 61.30% was among accessions within populations, and 12.76% was within accessions. These results indicate that the DAPC model effectively captured one-fourth of the total variance by stratifying the 258 accessions into nine clusters, similarly to the findings of Kabbaj et al. [35]. Furthermore, careful consideration should be given when selecting genotypes for further studies, ensuring that both individuals within and among populations are included.

Linkage disequilibrium

The determination of the extent, distribution, and pattern of linkage disequilibrium within a genome is crucial for understanding evolutionary forces acting on homoeologous chromosome pairs and identifying conserved regions during crop evolution. The extent of LD across the genome influences marker density and the resolution required for conducting genome-wide association studies (GWAS).

Our genome-wide LD analysis showed that the percentage of marker pairs with significant LD (p < 0.001) was 22.61% and 25.03%, with mean r² values of 0.076 and 0.006 for SNP and SilicoDArT markers, respectively. These results are consistent with the 24.6% reported by Ebrahimi et al. [40] for a durum wheat collection from the Central Fertile Crescent using both SilicoDArT and SNP markers. However, Roncallo et al. [62] reported a lower percentage of 13.4% for a worldwide durum wheat collection. Similarly, Mahboubi et al. [39] found lower percentages of 8% and 16.9% for SNP and SilicoDArT markers, respectively, in a collection of wheat genotypes from diverse origins. The percentage of significant LD marker pairs found in this study is lower than the values reported in previous studies, including 33% [37], 30.4% [50], 33.6% [17], 42% [63], and 43% [58, 64].

The average r² values for SNP and SilicoDArT markers across the whole genome were 0.23 and 0.19, respectively, which are close to the 0.2 suggested by Hill and Weir [34]. The r² values were slightly higher in the B-genome (0.24 for SNPs and 0.20 for SilicoDArTs) than in the A-genome (0.22 for SNPs and 0.18 for SilicoDArTs). Using SNP markers, Laribi et al. [17] reported similar r² values of 0.18 and 0.16 for the A- and B-genomes, respectively, with an average of 0.17. Similarly, Mulugeta et al. [50] found an r² value of 0.21 for the whole genome. Roncallo et al. [62] also reported an average r² of 0.196 for a global collection of accessions, although a higher r² of 0.302 was observed in Argentine durum wheat landraces.

Lower r² values have been reported using DArTseq markers. Fayaz et al. [65] found an r² of 0.12 in a collection of Iranian durum wheat landraces, while Roselló et al. [64] reported a similar value (r² = 0.115) for a Mediterranean durum wheat collection. Ebrahimi et al. [40] found an average r² of 0.14 across a global collection of durum wheat accessions, and Soriano et al. [60] reported average r² values of 0.12 and 0.11 for the A- and B-genomes, respectively, in a diversity panel of durum wheat genotypes.

The average r² values varied between chromosomes, ranging from 0.18 to 0.29 for SNPs and 0.09 to 0.25 for SilicoDArTs. Similar chromosome-specific variation has been reported by other studies, including Mulugeta et al. [50] (r² = 0.21–0.26), Roncallo et al. [62] (r² = 0.14–0.29), and Laribi et al. [17] (r² = 0.15–0.22).

The LD analysis in this study revealed distinct LD decay patterns between the homoeologous chromosomes of the A- and B-genomes for both SNP and SilicoDArT markers, confirming genetic differences between the two genomes across marker types. The whole-genome LD decay observed in this study was 7.50 Mb for SNPs and 0.35 Mb for SilicoDArTs. Mulugeta et al. [50] reported a comparable LD decay of 8.56 Mb. Higher LD decays of 9.6 Mb, 9.96 Mb, 11.8 Mb, and 14.6 Mb were reported by Wang et al. [63], Taranto et al. [38], Roncallo et al. [62] and Laribi et al. [17], respectively. Even greater LD decays of 29.5 Mb, 51.3 Mb, and 69.1 Mb were reported by Roncallo et al. [62], Bassi et al. [66], and Alemu et al. [48]. The variation in LD decay patterns observed across studies has been previously reported [17, 38, 40, 61, 62].

This variability can be attributed to several factors, including differences in the number and type of markers per homoeologous chromosome or group, marker density, population size, and the evolutionary forces acting on homoeologous chromosome pairs. Selection for specific traits and genetic drift may have further contributed to these differences [17, 38, 40, 50, 61, 65].

The LD half-decay at a fixed r² of 0.1 varied between chromosomes, ranging from 2.78 to 6.42 Mb (average: 4.69 Mb) for SNPs and from 1.08 to 2.82 Mb (average: 1.87 Mb) for SilicoDArTs. The smaller LD decay observed with SilicoDArTs compared to SNPs is consistent with previous studies. Similar findings were reported by Yang et al. [56] in an Australian bread wheat diversity panel, where LD decay at r² = 0.20 was 3 cM for SNPs and 0.68 cM for SilicoDArTs. Additionally, using SNP markers in a collection of Tunisian durum wheat accessions, Laribi et al. [17] reported an LD half-decay ranging from 2.9 to 7.3 Mb at r² = 0.1. These results indicate a rapid LD decay, likely due to varying selection pressures across genomic regions, and suggest that the accessions under investigation exhibit high genetic diversity. This average LD half-decay can serve as a reference for defining upstream and downstream genomic regions in GWAS for identifying candidate QTLs/genes associated with different traits [17, 61]. For instance, when several markers associated with a trait are closely linked, LD decay can reveal whether they represent the same or different signals, and the surrounding LD block can help define the region for candidate gene searches. Additionally, linkage disequilibrium is useful in building genomic prediction models to estimate breeding values, thereby accelerating selection without the need for phenotyping.

Population structure and dispersal routes

The analysis of population structure in 258 durum wheat accessions using SNP and SilicoDArT markers revealed a distinct division: landraces were primarily assigned to Subpopulation 1, while cultivars and breeding lines were mainly classified in Subpopulation 2. The Bayesian analysis further grouped accessions based on their geographical origin. Subpopulation 1 predominantly consisted of landraces from the Iberian Peninsula (Portugal and Spain), whereas Subpopulation 2 comprised breeding lines and cultivars from Algeria, France, and Italy (Eastern Mediterranean and North Africa). This structure aligns with previous studies indicating that durum wheat genetic diversity is structured according to the geographic origin of accessions [16, 67, 68]. These findings are also consistent with the historical dispersal pathways of durum wheat. The clustering of accessions from France and Italy with those from Algeria supports reports of wheat spreading to Sicily from North Africa [69–71]. In fact, North Africa was historically referred to as “Rome’s granary” or “breadbasket,” as it supplied Rome with substantial quantities of grain as tax payments during the Roman occupation [72]. Later, the Arabs further expanded and reinforced durum wheat cultivation by introducing new landraces from the Levant [6].

Durum wheat was also imported to France from North Africa, particularly Algeria, during the Arab expansion across North Africa, southern Turkey, Sicily, and Spain. During this period, trade routes along the North African coast and across the Sahara facilitated seed exchange with Sicily and Spain, from where durum wheat spread to the rest of Europe [5, 69, 71, 73]. From 1453 until the early 20th century, much of North Africa and the Eastern Mediterranean was part of the Ottoman Empire, further facilitating durum wheat movement between these regions. Later, French colonization of North Africa likely contributed to the dispersal of durum wheat between these areas.

Two major dispersal routes for durum wheat have been proposed: (1) the northern dispersal pathway, encompassing the northern and eastern Mediterranean basin, including Syria, Turkey, Bulgaria, Cyprus, Greece, and Italy, and (2) the southern dispersal pathway, covering North Africa (mainly Egypt, Algeria, and Morocco) and the Iberian Peninsula (Spain and Portugal) [16, 36, 50, 67–69, 71, 74]. Our findings align with previous research. For example, Soriano et al. [60] identified three distinct subpopulations: landraces from the eastern Mediterranean, northern Mediterranean, and western Mediterranean. Similarly, Mazzucotelli et al. [15], using a global durum wheat panel, found that accessions from the Fertile Crescent, North Africa, and southern Europe formed separate clusters. Maccaferri et al. [16] also reported a clear distinction between eastern and western Mediterranean landraces.

The separation of landraces from the Iberian Peninsula in this study, except for a few accessions from Algeria, Egypt, France, Italy, Morocco, Tunisia, and Turkey, supports findings by Moragues et al. [74], Martinez-Moreno et al. [5], and Kottek et al. [75]. These studies suggest that durum wheat was introduced to the Iberian Peninsula from North Africa and southeastern Europe, particularly via southern Italy, likely during the Roman period. Notably, many Spanish landraces, including Fanfarrón, Moruno, Recio, Rubio, or Raspinegro, were found to have a western Mediterranean origin, particularly from the Maghreb (Algeria, Morocco, and Tunisia) and Sicily [54, 76].

The extent of seed mixture and exchange

The admixture contained accessions from all countries except Turkey. Historically, all these countries were at some point part of the same empires (Roman, Ottoman, Arabic or Phoenician). Additionally, North African countries were colonized by France in modern history. Given these historical connections, the mixing of landraces is not surprising, as seed exchange frequently occurred among these regions. Over time, the transfer of germplasm may have contributed to the observed admixture of seeds.

This admixture could also be attributed to both natural and human selection during the domestication and spread of germplasm from its center of origin (the Fertile Crescent) [35, 39, 42, 48, 49, 59, 61, 77, 78]. Furthermore, farmers traditionally stored seeds for replanting in subsequent seasons and frequently exchanged seeds with other farmers, further facilitating gene flow [44]. Oliveira et al. [77] identified admixture in Mediterranean accessions and attributed it to polyphyly, suggesting that tetraploid wheat spread from multiple ancestral populations. They also suggested that the mixing of new and old genotypes in fields was a common occurrence in the Mediterranean basin, contributing to seed admixture. Similarly, studies have reported that most Italian landraces originated from natural or human selection over long periods from landraces introduced from North Africa and western Asia [38, 79].

Indeed, wheat landraces of diverse geographic origins were widely cultivated across the Mediterranean basin due to cultural and culinary preferences, leading to extensive seed exchange among and between communities. This exchange likely contributed to significant gene flow, as well as genetic drift and founder effects within the germplasm [35, 54]. The diverse environmental conditions across Mediterranean countries may have further contributed to genetic variation in local landrace populations. Occasional natural hybridization events may have also played a role, leading to increased variability among regional populations [35, 38, 54, 80]. However, the uneven representation of countries, particularly the limited number of accessions from Egypt, Israel, Jordan, Morocco, Tunisia, and Turkey compared to those from Algeria, France, Italy, Portugal, and Spain, only allows for preliminary theories about seed dispersal and exchange between these regions. Further studies including larger and more balanced numbers of accessions from all countries will enable more definitive conclusions. In this context, several accessions in this study with the same name were assigned to the same subpopulations. For example, Oued Zenati from Algeria and Spain, and Raspinegro from Portugal and Spain, clustered together. Meanwhile, other accessions were assigned to different subpopulations. For instance, Bidi 17 from France and Algeria was placed in two different subpopulations when analyzed using SNP and SilicoDArT markers. A similar pattern was observed for the Mahmoudi landrace from Algeria and Tunisia.

The historical use of limited genotypes in crossings

The admixture in this study included accessions with varying levels of genetic improvement, which can be attributed to the use of landraces from different geographical origins in breeding programs. For instance, Bidi 17 is a French cultivar derived from a selection within the Algerian landrace population Oued Zenati, which itself originated from the Bidi landrace, known to have multiple origins, including Algeria, Italy, and Tunisia. Similarly, the French cultivar Lez ((bidi-17/3/(bald-medeah) lumillo/mindum//bidi-17) has Bidi 17 in its genetic background. Another French cultivar, Montferrier, traces part of its genetic lineage to the Tunisian durum wheat accession BD-931. Likewise, Poinville (zenati-bouteille/wells//barrigon-yaqui/tehuacan-60/3/leeds), a French cultivar, incorporates multiple landraces from different countries in its genetic background, including Jenah-rhetifah and Mahmoudi from Tunisia, Bidi from Italy, and Zenati-bouteille from Algeria (detailed pedigree available at wheatpedigree.net).

Interestingly, most cultivars assigned to Subpopulation 1 based on Bayesian analysis using SNP markers were selected directly from landraces. For example, Aziziah-17-15 from Italy was derived from the Aziziah landrace of the Near East, Mekki 13 from Tunisia was selected from the Mekki landrace of the same country, Timilia SG1 was selected from the Timilia landrace in Italy, and Hedba 3 from Algeria was derived from the Hedba landrace of Algeria. These results suggest that some of these cultivars may have been misclassified as cultivars when they are, in fact, landraces. Similar cases of landraces and cultivars being mislabeled or wrongly classified have been reported by other studies [17, 35, 47, 81].

The first official durum wheat breeding program was established in Italy in the early 1900s, primarily focusing on the exploitation of genetically diverse landraces from Italy, North Africa, and western Asia [1, 38]. This program led to the release of the cultivar Senatore Capelli by breeder Nazareno Strampelli in 1915, derived from a selection with the North African landrace Jenah-rhetifah. Between 1930 and 1970, Senatore Capelli gained widespread popularity, covering 60% of Italy’s durum wheat-growing area for several decades and extending to several Mediterranean countries [2, 38, 73]. Due to its extensive use in breeding programs, Senatore Cappelli is present in the genetic background of almost all durum wheat varieties cultivated in the Mediterranean, particularly Italy [1, 38, 73, 82].

In this study, Vallega Zitelli populations from Italy and the French cultivar Villemur were obtained from a cross that included Senatore Capelli. Interestingly, Senatore Capelli collected from Spain and Andalucia 344 from Spain were previously identified as part of an admixture. This is expected, as Senatore Capelli spread from Italy and was incorporated into Spanish breeding programs, where it was crossed with local landraces to develop cultivars such as Ledesma and Andalucía 344, which became widely cultivated in Spain during the 1950s and 1960s [6].

Clustering of genotypes aids in selecting parental lines from different clusters to avoid inbreeding and increase heterozygosity in the offspring. Depending on the breeding program objectives, genotypes defined as admixture can be used in crosses to combine traits; however, if the objective is to develop pure lines, admixture should be avoided. In addition, tracking the performance of these genotypes across different environments can help assess their adaptation and genotype-by-environment interactions. Incorporating phenotypic data, such as morphological and agronomic traits, as well as responses to biotic and abiotic stresses, can further enhance our understanding of the diversity among accessions and identify markers associated with desirable traits. This, in turn, can help identify donor lines for crossing in breeding programs, supporting the preservation and strategic use of genetic resources.

Conclusion

Investigating the genetic diversity of durum wheat from its centers of origin is crucial for preserving valuable genetic resources and utilizing them to develop high-yielding cultivars with resistance to various abiotic and biotic stresses. This study examined the genetic diversity of accessions from the Mediterranean Basin using DArTseq markers, which effectively differentiated genotypes based on either their level of improvement or geographical origin. These markers can be used to conduct further research, such as genome-wide association studies, to identify beneficial alleles for abiotic and biotic stress resistance.

Our findings support the existence of two primary pathways for durum wheat dispersal: the northern route, encompassing the northern and eastern Mediterranean Basin, and the southern route, covering North Africa and the Iberian Peninsula. Additionally, this study provides insights into seed mixtures within Mediterranean countries and highlights both the extensive use of a limited genetic pool in breeding programs and the underutilization of potentially valuable genetic resources. The widespread reliance on a narrow set of genetic resources for developing modern cultivars has contributed to the emergence or re-emergence of several biotic stresses in the region. Harnessing the genetic variability of durum wheat landraces from these centers of origin could help broaden the genetic background of elite cultivars and facilitate the identification and introgression of beneficial alleles.

Abbreviations

AMOVA

Analysis of Molecular Variance

BIC

Bayesian Information Criterion

DAPC

Discriminant Analysis of Principal Components

DArT

Diversity Array Technology

GD

Gene diversity

LD

Linkage disequilibrium

MAF

Minor allele frequency

NJ

Neighbor-joining

PIC

Polymorphic Information Content

SNP

Single nucleotide polymorphism

Authors’ contributions

ML, AHY, SBM, SFH, and SES designed the research; ML performed the laboratory procedures, ML, AHY, and SBM conducted field work; ML and SMV analyzed the data, ML wrote the manuscript with input from SES; and all authors reviewed, edited, and approved the final version of the manuscript.

Funding

This work was supported by the CGIAR Research Program on Wheat (CRP-WHEAT) Precision Phenotyping Septoria Platform (Tunisia Grant: FIT BUS200029) and the United States Agency for International Development (USAID) Bureau for Resilience and Food Security as part of Feed the Future. Additional financial support was provided by the Natural Sciences and Engineering Research Council (NSERC) of Canada through a Discovery Grant (RGPIN-2021-04010) to SES. This study was also funded by Alberta Grains, Results Driven Agriculture Research (RDAR), Sask Wheat, and the Saskatchewan Ministry of Agriculture through a project grant (ADF20230343) to SES.

Data availability

The datasets generated for this study can be found in the text, figures, tables, and supplementary materials of the article. The SNP and SilicoDArT data used in this study have been deposited in the Genome Variation Map (GVM) of the National Genomics Data Center, Beijing Institute of Genomics, Chinese Academy of Sciences and China National Center for Bioinformation, under accession number GVM001028 (https:/ngdc.cncb.ac.cn/gvm/getProjectDetail?Project=GVM001028).

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.