Genome-wide selection in soybeans and optimization of phenotyping for grain yield

Abstract

In a breeding program, several factors influence the selection of cultivars, mainly due to the high number of genotypes under evaluation and the reduced experimental capacity in the initial phases of the program. In this context, the present study was divided into four parts. The first one aimed to identify the core locations for evaluation and selection of soybean genotypes in the macro-regions 1 (M1) and 2 (M2), in generations with low seed availability. The data set consisted of 22 soybean genotypes grown in 23 sites for three years. The GGL + GGE and G analyses versus the GE analysis were used. The locations Chapada-RS and Maracaju-MS were the most representative sites and discriminant macro-regions 1 and 2, respectively. Identification of the core location is fundamental to evaluation, since it is where the number of test sites can be summarized to a single site by soybean growing macro-region. The second study aimed to evaluate the experimental accuracy of different statistical methods used to analyze the assays with large numbers of soybean genotypes. The grain yield data from 324 soybean genotypes, evaluated in six replicates, were used. The data were analyzed by using the randomized block design, triple lattice design, and Papadakis method. The experimental accuracy indicators of the Papadakis method were more favorable when compared to those of the randomized block and triple lattice designs. Two replicates could be used when analyzing the data without reducing experimental accuracy: a randomized complete block design or the Papadakis method. In the third study, the productive performance, adaptability, and stability of modern soybean cultivars were evaluated in multi-environment assays. A total of 46 cultivars were evaluated in eight environments, in the adaptation micro-regions 102, 201, and 202, during the 2014/2015 harvest. Genotype × complex environment interactions occurred with changes in the ranking of cultivars between the sites. Among the genotypes evaluated, the cultivar NA 5909 RG, parental to the RILs in the genome-wide selection (GWS) assay, was considered to be among the genotypes with higher mean productivities, and it also showed high adaptability and stability. The fourth study had three objectives: to evaluate the accuracy of genomic selection in soybean, to identify the effect of intra-population structure on the accuracy of genomic selection, and to compare the efficiencies of the phenotypic and genomic selections in soybean. The BayesB model with cross validation was used for analyzing the phenotype data from the 324 soybean genotypes. The accuracy of GS for phenotypic characters with genotypic data of 5403 SNP molecular markers was also evaluated. The results indicated that the genotypic accuracy was similar, irrespective of consideration of the population structure. It was observed that the population structure did not significantly affect the accuracy of the models for the traits evaluated. It was verified that with this methodology it is possible to halve the selection time and increase the selection efficiency by 123% for grain yield.