Combining genome-wide association and genomic prediction to unravel the genetic architecture of carotenoid accumulation in carrot

Abstract

Carrots (Daucus carota L.) are a rich source of provitamin A, namely, α- and β-carotene. Breeding programs prioritize increasing β-carotene content for improved color and nutrition. Understanding the genetic basis of carotenoid accumulation is crucial for implementing genomic-assisted selection to develop high-carotenoid lines. While previous studies identified loci (Y2, Y, Or, and REC) associated with carrot color and carotenoid content, this study employed genome-wide association (GWA) in a diverse panel of 738 carrot accessions. We discovered a novel locus with a candidate gene encoding phytoene synthase, a key enzyme in carotenoid biosynthesis. The Y2, Y, Or, and REC loci are mostly fixed in orange varieties, yet considerable variation in carotenoid concentration persists. This suggests a multigenic trait influenced by the environment. GWA of carotenoid concentration identified a quantitative trait locus for total carotenoids and α-carotene. We explored the accuracy of genomic prediction (GP) models to predict carotenoid concentration. We determined the optimal number of plants and plots required for accurate carotenoid phenotyping, finding ≥5 plants per plot and three plots per site as the minimum effective sample per accession. GP models achieved accuracies ranging from 0.06 to 0.40 depending on the carotenoid measured and environment the carrots were assayed. Additional studies in breeding programs will clarify the potential of genomic-assisted selection for high-carotenoid carrots.

FIGURE 1

Genome‐wide association (GWA) mapping loci associated with orange carrot color. The Manhattan plot depicts −log10(p‐value) for single nucleotide polymorphism (SNP) markers across the carrot genome. The three loci identified in this study include the REC, PSY, and Y2 loci.

FIGURE 2

Genome‐wide association mapping of α‐carotene and total carotenoid content in carrot. (A) Manhattan plots illustrating −log10 (p‐value) for single nucleotide polymorphisms (SNPs) associated with total carotenoid (triangle) and α‐carotene (circle) content with combined data from WI_2018_605PI and CA_2019_730PI trials (n = 603). (B, C) Boxplots comparing total carotenoid (B) and α‐carotene (C) concentrations (µg/g dry tissue) between genotypes at a significant locus (p < 0.05).

FIGURE 3

Influence of marker density and model choice on genomic prediction accuracy. (A) Comparison of prediction accuracy for different marker densities (82,625 to 81 markers) when averaged across five genomic prediction models. Both the WI_2018_605PI (gray) and CA_2019_730PI (orange) were tested for all combinations of genomic prediction model and marker number. (B) Average prediction accuracy and standard error across various traits using a fixed marker density of 2582 markers. PI, plant introduction. BGLR, Bayesian generalized linear regression; BRR, Bayesian ridge regression.

FIGURE 4

Average prediction accuracy (correlation of genomic estimated breeding values [GEBVs] and observed phenotypes) when the Bayesian ridge regression model was applied using 2582 markers in the CA_2019_730PI (A) and WI_2018_605PI (B) trials. Additions to models included adding population structure and the color of the core of a carrot. Each carotenoid and the total concentration of carotenoids were tested for all combinations of input data. PI, plant introduction.

FIGURE 5

Variation in carotenoid measurements depending on location, sample size, and plots per inbred. (A) Boxplots describing β‐carotene concentrations in carrots grown in Wisconsin (WI) in 2022 and California (CA) in 2022 or 2023. Significant differences within a line between location or year (HSD.test, p‐value < 0.05) are indicated by different colored plots. (B) Boxplots describing α‐carotene concentrations in carrots grown in Wisconsin in 2022 and California in 2022 or 2023. Significant differences between location or year (HSD.test, p‐value < 0.05) are indicated by different colored plots. (C) Percentage of the largest confidence interval (n = 2) as more samples were added. Sampling without (square shape) and with replacement (circle) was tested. (D) Confidence interval width for estimated average carotenoid concentration based on different numbers of plots per phenotype (n = 10 samples/plot). (E) β‐Carotene concentration variability among plots of inbred Ns5154 grown in California in 2022. The average concentration is represented by a gray square, and individual samples by black dots.