Durum wheat (Triticum turgidum subsp. durum) is a major staple crop in the Mediterranean basin, and Italy and Spain are among the main producers (Eurostat, 2023). The crop is mainly used to produce pasta, which is consumed not only in the Mediterranean regions but worldwide (Broccanello et al., 2023). In addition to its economic importance, durum wheat is also important for its nutritional value. It is high in protein and dietary fiber, and it contains essential vitamins and minerals that play a very important role in the prevention of diabetes and cardiovascular diseases (Garutti et al., 2022). To ensure the sustainability of the crop and increase its yield and quality, plant breeders constantly selected high-yielding varieties with improved grain quality traits. In Italy, more than 300 durum wheat varieties are currently registered to the National Register of Varieties by several representative institutions and companies (Sian, 2023), emphasizing the intense breeding work conducted in Italy over the last decades. In the past, numerous studies were conducted to evaluate the impact of durum wheat breeding on yield and grain quality and define a new plant ideotype capable of maximizing yields in different environmental conditions. The genetic gain was systematically evaluated by comparing the performances of historical varieties released over different points of breeding time (Pecetti and Annicchiarico, 1998; Motzo et al., 2004; De Vita et al., 2007), and the results showed a genetic gain for durum wheat of about 0.6% per year, similar to the value reported in many cases of bread wheat (Calderini and Slafer, 1999). The superiority of the modern wheat cultivars in terms of grain yield has been attributed largely to changes in the harvest index, with small or negligible increases in total biomass production and the number of grains per unit of land (Austin et al., 1980; Austin, 1989; Siddique et al., 1989; Slafer and Andrade, 1993; Brancourt-Hulmel et al., 2003). These findings demonstrate how modern breeding has been successful in exploiting crop diversity for genetic improvement in both bread and durum wheat. However, the recent estimates of genetic gain for both species are stagnant, probably because modern intensive breeding practices have exploited the residual fraction of the available crop diversity and/or as a result of climate change. The narrow genetic base of elite germplasm compromises long-term genetic gain and increases the genetic vulnerability to unpredictable environmental conditions.

Efficient genetic diversity management is therefore required in breeding programs and low-cost genotyping platforms that generate thousands to millions of data points provide effective means for crop genetic research studies (Ganal et al., 2012). For wheat, a large amount of single nucleotide polymorphisms (SNPs), generated using different genotyping platforms, was available thanks to the recent release of the high-quality reference genome of bread and durum wheat. This opened new possibilities for untangling the genetic architecture of complex traits by genome-wide association study (Rabbi et al., 2021; Eltaher et al., 2022; Esposito et al., 2022) and to perform other genomic studies, for instance, the analysis of selective sweeps within or across species (Liu et al., 2019; Semagn et al., 2021; Soriano et al., 2021). In addition, recent studies on wheat and other crops have shown that Genome-Wide Association Studies (GWAS) coupled with haplotypic block (HB) analysis improved analysis based on a single marker not only in terms of statistical significance (better p-values) but also in estimating the allelic effects (Hao et al., 2012; Lu et al., 2012; N’Diaye et al., 2017; Ledesma-Ramírez et al., 2019; Li et al., 2019; Sehgal and Dreisigacker, 2019; Shokat et al., 2020).

The genetic architecture of complex traits can provide insights into the underlying genetic mechanisms that control the expression of these traits and develop new cultivars with improved traits. Moreover, to guarantee the plant variety protections at the end of the selection process, the use of molecular markers to discriminate morphological traits could be used in distinctness, uniformity and stability (DUS) testing of new varieties as a complement to, or replacement of, morphological observations listed in International Union for the Protection of New Varieties of Plants (UPOV) guidelines (Jamali et al., 2019; Yu and Chung, 2021).

In our previous study (Taranto et al., 2020) a large panel of Italian durum wheat accessions that includes landraces (LR), old (OC) and modern cultivars (MC) was subjected to genotyping using the Illumina iSelect 15K wheat SNP array. The study was carried out with the aims of i) assessing the genetic diversity and population structure in a large collection of durum wheat accessions (over 250) released since the early 1900s using genome-wide high-density SNP array and ii) understanding the history of Italian durum wheat breeding by identifying molecular signatures of divergence and selection. Relatively small differences in genetic diversity were observed among accessions whereas an increase in linkage disequilibrium (LD) and in changes in the allelic frequencies in DNA regions that control important agronomic traits, were found. In the LD analysis, to exclude possible bias due to the different sizes of sub-population and to minimize the sampling effect, we standardized the number of accessions of LR, OC and MC to 41, for a total of 123 accessions. Using this balanced core set of genotypes, in this paper, we present the results obtained by performing GWAS and LD HB analysis on several durum wheat traits, including grain quality and yield-related traits, morphological characters described in the UPOV test guidelines (UPOV, 2012), and grain morphometric parameters. The goals were to: i) identify phenotype-genotype associations for the 34 traits analyzed by GWAS; ii) evaluate the breeding effect by analyzing the temporal trend of LD haplotype blocks moving from the landraces to modern cultivars; and iii) identify specific candidate genes for each of the 3 breeding groups (i.e., LR, OC, MC).

Improving grain production while maintaining environmental sustainability under climate change remains the main breeding goal in the coming years, especially in vulnerable environments, such as the Mediterranean Basin (Ceglar et al., 2021). Understanding the effects produced by durum wheat breeding activities on the main traits of agronomic interest could help the selection of new varieties to face climatic conditions and better exploit the genetic variability of this species (De Vita and Taranto, 2019). Dissecting the genetic architecture of complex quantitative traits using high-density SNP markers in wheat could have practical implications in durum wheat molecular breeding for improving yield potential and grain quality but also for providing tools to accelerate plant variety protection and registration (Arriagada et al., 2020; Mulugeta et al., 2023).

In the current study, we exploited the ML-GWAS approach to study the genetic basis of 34 morpho-phenological and agronomic traits and identify QTNs to be exploited for marker-assisted selection and/or in UPOV protocol for durum wheat varietal protection. In addition, we defined haplotype blocks, using the LD-based approach (Qian et al., 2017), to identify genomic regions and/or candidate genes associated with agronomic traits under study and the impact of breeding programs on their architecture.

Most of the traits investigated in our panel showed normal distributions, typical of traits controlled by multiple QTLs and highly vulnerable to environmental factors. The six ML-GWAS methods confirmed the genetic complexity, revealing different traits-associated QTNs across all wheat chromosomes. The BLUP prediction method, a parameter to minimize the environmental bias allowing the estimation of the true individual genetic value (Robinson, 1991), increased the consistency of the six ML-GWAS models reducing the number of associated QTNs. Among all models used, the pLARmEB detected the highest number of associations, confirming previous studies on soybean (Zhang et al., 2018; Fang et al., 2020), maize (An et al., 2020), rape (Khan et al., 2019) and common bean (Delfini et al., 2021).

Comparing our GWAS findings with those of previous studies, we found that some important genes controlling flowering time and plant height in durum wheat were in LD decay with QTNs identified in this study. The Q.Ph-2B (Kukri_c36879_83) was in LD decay with candidate genes affecting PH such as diminuito/dwarf1 gene, Flowering-promoting and gras transcription factors involved in steroid synthesis (Klahre et al., 1998) and in gibberellin signal transduction pathways, respectively (Kovi et al., 2011; Cheng et al., 2022). In the same region, Maccaferri et al. (2011) found a QTL for HD. The Q.Hd-7B was identified in the proximity of QTLs for HD and PH previously identified by Maccaferri et al. (2014). The MTA Ra_c73292_443, associated with the heading date, was also found as a marker harboring the 5B.3 hotspot able to differentiate wheat Mediterranean landraces based on adaptative traits (i.e., flowering time) (Yannam et al., 2023).

Several QTNs found in the present study were related to kernels, traits usually poorly investigated in the literature. Among these, the vitreousness of the kernels is one of the grain qualitative parameters most appreciated by the pasta industry and the market. Vitreous kernels are believed to have higher protein content and to have higher grain quality, compared to non-vitreous kernels, also called ‘yellow berry’ (Dexter et al., 1988; Dexter et al., 1989; Samson et al., 2005; Sieber et al., 2015; Fu et al., 2018). Yellow berry is expressed by the presence of starchy spot areas in a usually vitreous grain conferring a less compact structure with numerous open spaces and physically discontinuous protein matrix in kernels (Dexter et al., 1989, Turnbull and Rahman, 2002; Samson et al., 2005). As a result, the yellow berry is a primary factor in the marketing of durum, influencing the milling and end-product quality of durum wheat. Poor nitrogen availability is considered the most critical factor in the determination of yellow berries (Morris and Beecher, 2012), however, few genetic studies have been conducted to elucidate the genetic basis of this trait in wheat. We found a strong association for yellow berry trait on chromosome 6A (Q.Yb-6A) in a region containing a nadh-ubiquinone oxidoreductase gene, involved in starch metabolism in rice (Hu et al., 2018). This gene could be considered a good candidate to be explored to understand its role in wheat. Previously, Pshenichnikova et al. (2008) detected QTLs for grain vitreousness on chromosomes 3A, 5D, and 6A using the bread wheat ITMI population. Dhaliwal et al. (1986) studied the inheritance of yellow berry among the progeny of six bi-parental crosses, and the Chinese Spring monosomics CIMMYT breeding line had the major dominant genes on chromosomes 1A and 7A. Prashant et al. (2011) mapped the yellow berry on chromosomes 5D and 2D in a genetic linkage map of hexaploidy wheat using SSR markers. Two QTLs for yellow berry tolerance were also reported by Ammiraju et al. (2002) in a bread wheat RIL population on chromosomes 5D and 6B, respectively. Thus, our result may support the fine mapping to localize the candidate gene responsible for yellow berry in durum wheat.

The MTA Ra_c69221_1167 (Q.Bp-5A) at 43.43 Mb on chr. 5A, associated with the black point disease resistance, was previously reported as an important player conferring resistance to Septoria and Powdery Mildew in winter wheat (Alemu et al., 2021). Our Q.Bp-5A appears to be different from those previously mapped to the same chromosome by other authors in bread wheat. Li et al. (2022) identified two QTLs for black point resistance on chr. 4A and 5A (2.1-274.2 Mb), explaining 3.3%-15.1% of the phenotypic variances, respectively. The QBp.caas-5AS for black point resistance detected by Liu et al. (2016) was at 110.7 Mb. Another six loci for black point resistance were detected on chromosome 5A but none overlapped with our Q.Bp-5A (Li et al., 2020; Lv et al., 2020).

Several phenotypic traits considered in this study are part of the UPOV technical protocols and used for DUS examination of candidate varieties for registration in the National/European catalogue. Currently, the UPOV test guidelines for numerous crops, describe the relevant traits to be evaluated, along with the recommended procedures for conducting the trials and the statistical analyses to use. In addition, UPOV guidelines establish which characteristics should be visually scored and which ones should be precisely measured. DUS testing is entirely independent from any evaluation of end-use value. UPOV characteristics for DUS testing primarily rely on morphological traits, chosen to reflect general genetic differentiation among varieties, strongly influenced by the environment and by visual assessment (Yu and Chung, 2021). The evaluation system also involves the use of a limited number of reference varieties to compare the new candidate variety. Therefore, adding the environmental influence, the operator’s subjectivity in visual scoring and the limited number of reference varieties, it seems necessary to find some alternative procedures to use in DUS testing. Since the DUS phenotypic evaluation has some limitations about the influence that environmental conditions have on the expression of these traits (Yu and Chung, 2021). The aim of the ML-GWAS analysis performed in this work was to identify QTN closely associated with DUS traits, and to integrate the use of DNA markers into UPOV test guidelines. Indeed, the analysis showed interesting results for kernel-related traits (i.e., area, length, width, coloration with phenol), flag leaf glaucosity of blade and sheath, spike color at maturity, lower glume shape, and pubescence of external surface.

Russo et al. (2014) using 136 F5 recombinant inbred lines, derived from a cross between modern durum wheat and T. dicoccum, identified six QTLs on chr. 1B, 2B, 3A, 3B, 4B and 7A, and 2 QTLs on chromosomes 3B and 4B for traits related to kernel morphology. Two QTLs reported by Russo et al. (2014) as associated with kernel length on chromosome 2B co-localized with those similar QTLs found in our study. In particular, the marker Kukri_c21135_1071 was ~ 5 Mb from Kukri_c74165_204 (Q.Kl-2B.2) and coincided also with QTL for thousand kernel weight (TKW) identified by Su et al. (2018).

The kernel coloration with phenol in durum wheat is the result of the conversion of phenol and other phenolic derivates into melanin pigments in the presence of tyrosinase (Taranto et al., 2017). Since the need to select advanced wheat breeding lines with a reduced aptitude for flour/semolina browning, a characteristic not appreciated by consumers, the kernel coloration with phenol was among the whole-seed assays also used in breeding programs to estimate the polyphenol oxidase activity (PPO) selecting those with low PPO activity in wheat (Sun et al., 2005; He et al., 2007; Taranto et al., 2015). Our results confirmed what was previously reported by Zhai et al. (2020) and Taranto et al. (2021; Taranto et al., 2022). The Q.Kcp-2A harbors the PPO locus, with the marker Ku_c13700_1196 in LD decay with Ppo-A1 and Ppo-A2 genes, confirming the markers associated with this trait as one of the most suitable genetic models to support DUS phenotypic protocols.

Unlike the previous one, which was already known in the literature, the associations found for flag leaf gloucosity (both lamina and sheath) represented a new result in durum wheat. In fact, the Q.FGSGls-2B.1 identified by the RAC875_c34512_685 marker mapped less than 1 Mb from the Epistatic inhibitors of glaucousness (Iw1) locus previously described by Yoshiya et al. (2011) in durum wheat using SSR markers and cloned by Huang et al. (2017) in bread wheat. However, it remains that, the chromosome region identified remains a valid result to be validated with further investigations.

In our previous study (Taranto et al., 2020) we compared the patterns of genetic variation observed in the Italian durum wheat germplasm, including LR, OC and MC, in order to better understand the effect of artificial selection and provide a list of genes/loci under selection associated with useful agronomic traits. In the present work, we corroborated the observed signature of selection using the LD haplotype blocks (HBs) partitioning, through the evaluation of haplotype diversity (i.e. allele type, frequency, length) within haploblocks across LR, OC and MC. The results revealed large and significant differences in the extent and pattern of LD among the three groups. As expected, moving from LR to MC, the number and size of LD HBs increased, confirming the increasing LD decay previously reported by Taranto et al. (2020). The observed pattern of LD decay was in line with the estimation over three main breeding periods (1915-1979, 1980-1999 and 2000-2020) made by Roncallo et al. (2021) in a worldwide durum wheat collection. The higher haplotype diversity as well as the higher number of haplotypes in MC than OC and LR indicated that modern breeding practices directly affected the composition and variation of gene pools over generations, diminishing effective population sizes, increasing inbreeding, and consequently increasing LD and HBs. The longest blocks were on chromosomes 1B, 2B and 5B for LR, OC and MC, respectively. Long-range LD blocks on 1B were also reported by Joukhadar et al. (2019) and Roncallo et al. (2021) and signatures of selection based on LD on 1B were observed by Maccaferri et al. (2014) confirming the selective pressure on genomic loci related to grain quality traits. The identification of different HBs among landraces and modern cultivars affecting agronomic performance within QTL hotspots were also reported by Royo et al. (2021) and Soriano et al. (2021).

A likely cause of the observed differences in length and number of HBs among the three genetic groups could be related to the breeding effect exerted on genes involved in response to abiotic and biotic stresses and gravitropism and phototropism; these latter are considered an adaptive response of crucial importance in plants (Lariguet et al., 2006; Kippes et al., 2020).

An enrichment of saur-like auxin-responsive genes was found in the HBs of LR and OC, although on different chromosomes. These are a family of auxin-responsive genes involved in plant growth and in response to internal and external signals (Stortenbeker and Bemer, 2019). It has been demonstrated that the overexpression of SAUR66-5B can increase the biomass and grain yields of transgenic wheat, as well as the nitrogen concentration and accumulation (Lv et al., 2022). In addition, the role in abiotic stress signaling of SAUR genes in wheat was also elucidated by Abhinandan et al. (2018). Our findings showed an enrichment of key genes as signal transducers in phototropism signalling in the HBs specific for MC. In particular, we found the phototropic-responsive genes PKS and NPH3, for which the role in Arabidopsis and cotton phototropism has been demonstrated (Lariguet et al., 2006; Christie et al., 2018; Grover et al., 2020; Kippes et al., 2020; Kimura et al. 2022). These genes could be good candidates to better understand the adaptative mechanisms moving from landraces to modern cultivars, together with the most renowned genes that we found in several HBs and QTNs such as TaGW2 genes (6A and 6B, Qin et al., 2014), Ppo (Taranto et al., 2021), Ppd (Arjona et al., 2018), dwarfism genes Rht-1 (Pearce et al., 2011), and Vrn (Royo et al., 2020).

The fact that most of the genes included in HBs are involved in adaptative mechanisms, highlighted the necessity to explore their allelic composition in the wild, domesticated, and modern germplasm with the aim to broaden the genetic variability and constitute new ideotypes to be used in durum wheat breeding programs.

Several phenotypic traits considered in this study are part of the international legal framework established by the UPOV guidelines for awarding Plant Breeders’ Rights PBR, similar to patent or intellectual property rights to new crop varieties. For some of them (i.e., kernel coloration with phenol and flag leaf glaucosity), strong associations were found through GWAS and LD haplotype block analysis, proving their power in detecting genetic variants associated with the traits of interest and for evaluating the impact of breeding programs. In particular, the increase in the number and length of haplotype blocks in MC suggested the high selective pressure exerted by breeders for the traits of agronomic interest.

We believe that the results reported here will not only expand the knowledge regarding the genetic architecture of many of these traits but also allow the employment of diagnostic/perfect genetic markers in current UPOV descriptive protocols, strongly influenced by environment and visual assessment, thus improving the plant variety protection and registration system.

Publicly available datasets were analyzed in this study. This data can be found here: Figshare repository (doi: 10.6084/m9.figshare.11835663).

FT and PV conceived the work, coordinated and supervised the activities. FF, SR, MS, LG and GT phenotyped the genetic materials. FT and SE performed statistical, QTNs and LD block analyses and data visualization. PV, FT, and SE contributed to interpretation and wrote the manuscript. All authors critically revised the manuscript. All authors read and approved the final manuscript.