Genomic prediction of starch content and chipping quality in tetraploid potato using genotyping-by-sequencing

Abstract

Genomic prediction models for starch content and chipping quality show promising results, suggesting that genomic selection is a feasible breeding strategy in tetraploid potato. Genomic selection uses genome-wide molecular markers to predict performance of individuals and allows selections in the absence of direct phenotyping. It is regarded as a useful tool to accelerate genetic gain in breeding programs, and is becoming increasingly viable for crops as genotyping costs continue to fall. In this study, we have generated genomic prediction models for starch content and chipping quality in tetraploid potato to facilitate varietal development. Chipping quality was evaluated as the colour of a potato chip after frying following cold induced sweetening. We used genotyping-by-sequencing to genotype 762 offspring, derived from a population generated from biparental crosses of 18 tetraploid parents. Additionally, 74 breeding clones were genotyped, representing a test panel for model validation. We generated genomic prediction models from 171,859 single-nucleotide polymorphisms to calculate genomic estimated breeding values. Cross-validated prediction correlations of 0.56 and 0.73 were obtained within the training population for starch content and chipping quality, respectively, while correlations were lower when predicting performance in the test panel, at 0.30-0.31 and 0.42-0.43, respectively. Predictions in the test panel were slightly improved when including representatives from the test panel in the training population but worsened when preceded by marker selection. Our results suggest that genomic prediction is feasible, however, the extremely high allelic diversity of tetraploid potato necessitates large training populations to efficiently capture the genetic diversity of elite potato germplasm and enable accurate prediction across the entire spectrum of elite potatoes. Nonetheless, our results demonstrate that GS is a promising breeding strategy for tetraploid potato.

Fig. 1

Heat map of the genomic relationship matrix for the 755 offspring in the MASPOT population (purple marking in the left panel) and the 63 individuals in the test panel (green marking). The matrix is obtained from 171,859 markers. Rows and columns represent each individual. The absence of obvious high intensity off-diagonal clusters indicates absence of population structure

Fig. 2

Density histograms depicting the phenotype distributions for the MASPOT population (red) and the test panel (blue). Chipping quality a was determined as assessment of frying colour on a scale from 1 (poor) to 9 (best), while starch content b was measured as percentage

Fig. 3

Parent-offspring regressions for the offspring in the MASPOT population for chipping quality (a) and starch content (b) for estimating pedigree heritabilities. The average phenotypic value of each parent pair is plotted versus the average offspring value. Error bars depict the standard deviation. Narrow sense heritability is estimated as the slope of the curves

Fig. 4

Distribution of genomic estimated breeding values and observed phenotype values for BayesC model for the MASPOT population for chipping quality (a) and starch content (b). Red colour indicates predictions made with eightfold cross-validation. Blue colour indicates predictions made with leave-sibs-out cross-validation. The slopes of the regression lines indicate the degree of bias of the predictions

Fig. 5

Distribution of genomic estimated breeding values and observed phenotype values for BayesC model for the test panel for chipping quality (a) and starch content (b). Red colour indicates predictions made with all 171,859 SNPs. Blue colour indicates predictions made with GWAS selected SNPs. Triangle indicates model trained with the MASPOT population. Circle indicates model trained with the combined set (MASPOT population and test panel) with eightfold cross-validation. The slopes of the regression lines indicate the degree of bias of the predictions