Genomic Prediction and Association Mapping of Curd-Related Traits in Gene Bank Accessions of Cauliflower

Abstract

Genetic resources are an important source of genetic variation for plant breeding. Genome-wide association studies (GWAS) and genomic prediction greatly facilitate the analysis and utilization of useful genetic diversity for improving complex phenotypic traits in crop plants. We explored the potential of GWAS and genomic prediction for improving curd-related traits in cauliflower (Brassica oleracea var. botrytis) by combining 174 randomly selected cauliflower gene bank accessions from two different gene banks. The collection was genotyped with genotyping-by-sequencing (GBS) and phenotyped for six curd-related traits at two locations and three growing seasons. A GWAS analysis based on 120,693 single-nucleotide polymorphisms identified a total of 24 significant associations for curd-related traits. The potential for genomic prediction was assessed with a genomic best linear unbiased prediction model and BayesB. Prediction abilities ranged from 0.10 to 0.66 for different traits and did not differ between prediction methods. Imputation of missing genotypes only slightly improved prediction ability. Our results demonstrate that GWAS and genomic prediction in combination with GBS and phenotyping of highly heritable traits can be used to identify useful quantitative trait loci and genotypes among genetically diverse gene bank material for subsequent utilization as genetic resources in cauliflower breeding.

Keywords: GenPred; Genomic Selection; Shared Data Resources; cauliflower; gene bank; genome-wide association study; genomic prediction; genotyping-by-sequencing.

Figure 1

(A) Discriminant analysis of principal components plot for the five inferred clusters using the k-means algorithm (Jombart and Ahmed 2011). (B) Boxplots for number of days to budding for each DAPC-inferred cluster. Letters above boxplots display Tukey-test results. Clusters with the same letter are not significantly differentiated from each other. Values within boxplots display the mean time to budding for each cluster.

Figure 2

LD decay in the whole population (A) and clusters 1–5 (B–F). The dashed horizontal line indicates the average background LD of all chromosomes of a respective population. The dashed vertical line indicates the maximum distance between linked markers and is used as reference point for the LD decay.

Figure 3

Effect of increasing the number of markers, included in a five-fold cross-validation with 10 replications using a standard GBLUP model, on prediction ability. Values represent averages of 100 runs. 10, 25, 50, 100, 250 and 500 markers, respectively, were sampled randomly for each run.