Genomics-based breeding for Bread and Durum Wheat improvement

Bread and durum wheat are two related species with high genome similarity. Climate change influences negatively wheat yield due to increasing temperatures and decreasing rainfalls. In addition, given the constant growth of world population, demand for food, feed and fibre will increase in the next future in this scenario the most prominent challenge in bread and durum wheat breeding is to provide new high-yielding and well-adapted genotypes to counteract the negative effects of climate change. To provide some precious information to achieve this aim, this PhD thesis dealt with two genomics-based techniques applied to bread and durum wheat attempting to improve grain yield and other adaptive traits. In chapter 1, after a brief description of the origins and the breeding of wheat and durum wheat, two genomics-based breeding methods such as quantitative trait loci (QTLs) mapping and genomic selection (GS) were described. QTL mapping is used to identify genomic regions associated with a specific trait. This is the first step in a breeding program, followed by marker-assisted selection (MAS) to select individuals with desired trait values based on markers. Combining QTL mapping with MAS allows breeders to efficiently improve the trait of interest in their crops.GS is a new breeding method which is applied in two steps as well as MAS approach. The first step consists of model training using both phenotypic and genotypic information. The second step is concerned with the application of the model using genotypic information only. Therefore, in GS the selection of the new genotypes is carried out using the genomic estimated breeding values (GBEVs) predicted by the trained model. In chapter 2, several traits such as plant height (PH), juvenile growth habit (GH), heading date (HD), and total and fertile tiller number (TTN and FTN) were investigated by using the same bi-parental population at early (F2 and F2-derived F3 families) and late (F6 and F7, recombinant inbred lines, RILs) generations to detect QTLs and search for candidates. The population included 176 and 178 individuals in the early and late generations respectively. The two genetic maps spanned 2486.97 cM and 3732.84 cM in length, for the F2 and RILs, respectively. QTLs explaining the highest phenotypic variation were found on chromosomes 2B, 2D, 5A, and 7D for HD and GH, whereas those for PH were found on chromosomes 4B and 4D. Several QTL detected in the early generations (i.e., PH and tiller number) were not detected in the late generations as they were due to dominance effects. Some of the identified QTLs co- mapped to well-known adaptive genes (i.e., Ppd-1, Vrn-1, and Rht-1). Other putative candidates were identified for each trait, of which PINE1 and PIF4 may be considered new for GH and TTN in wheat. The use of a large F2 mapping population combined with NGS-based genotyping techniques could improve map resolution and allow closer QTL tagging. In addition, the detected QTLs can be applied in a MAS-based bread wheat breeding program. In chapter 3, several agronomical traits were evaluated using a univariate model in durum wheat; subsequently, different multivariate genomic prediction models were performed attempting to increase prediction accuracy (PA). The panel was phenotyped for 10 agronomic traits for two consecutive crop seasons in two different field conditions: high nitrogen and water (HIGH) input and low nitrogen and water (LOW) input. Multivariate GS was implemented using two cross-validation schemes: MV-CV1, testing the model for a target trait using only the markers, and MV-CV2, testing the model for a target trait using also phenotypic data from the other traits. The two MV-CVs were used in two different analyses: modelling the same trait in both HIGH and LOW input, and modelling grain yield together with the five most genetically correlated traits. We observed PA for all traits in HIGH was higher than in LOW input for each trait except for the yellow index. Among all traits, PA ranged from 0.34 (NDVI in LOW) to 0.74 (test weight in HIGH). Modelling the same traits in both HIGH and LOW, MV-CV1 produced improvements in PA only up to 12.45% (NDVI in LOW) compared to the univariate model. By contrast, MV-CV2 increased PA up to 56.72% (thousand kernel weight in LOW). After modelling grain yield together with correlated traits, MV-CV1 did not improve PA, whereas MV-CV2 significantly improved PA up to 17.97%. In this chapter, was proved the possibility to increase prediction accuracy for agronomic traits by modelling the same traits in two different field conditions using the MV-CV2 approach. In addition, was established the effectiveness of MV-CV2 when grain yield was modelled with additional traits. These outcomes suggest that multivariate genomic selection can be a innovative method in durum wheat breeding permitting to improve prediction accuracy as well as breeding efficiency. Chapter 4 was aimed to identify and classify the Target Population of Environments (TPEs) in the Italian peninsula, to detect the best-performing varieties for each TPE, to estimate the yield and yield-related trait gain of the bread wheat varieties over time, and to verify prediction accuracies for all traits across the years. A historical bread wheat dataset (from 1999 to 2015) was used to observe traits trends and perform GS analysis, then, 21 environmental parameters were utilized to define TPEs in Italy. Apparently, no variation for grain yield and heading date was recognized over time. By contrast, an increase in thousand kernel weight and a decrease in plant height was observed. Five TPEs were identified for the Italian peninsula such as I) North of Italy, II) Po Valley, III) East and West coast, IV) Apennines, and V) South of Italy and Sicily. Genotype by TPEs interaction analysis revealed the genotype adaptation for each TPE. Genomic selection showed some encouraging results in terms of PA for all traits training with the information from one year and testing the following one up to 0.96 for plant height (training with 1999/2000 and testing for 2000/2001). As expected, prediction accuracy decayed for all traits when the model was trained using the first year (1999/2000) and tested in the following years. Encouraging prediction accuracies were foundalso when the prediction of a whole target environment (Foggia) was performed. All the information derived from this chapter can be precious for wheat breeders which can set their breeding program using a specific adaptation strategy and suggest the potential application of historical datasets in genomic selection application. Finally, Chapter 5 presents a summary of the key findings and outcomes of the PhD work. The chapter includes an analysis of the results, research contributions, limitations, and suggestions for future research. The chapter ends with a general conclusion that synthesizes the main findings and discusses their implications and potential impact on bread and durum wheat breeding.