Genetic variation and genome-enabled selection of white lupin for key seed quality traits

White lupin (Lupinus albus L.) requires selection for low quinolizidine alkaloid (QA) content and other seed quality traits to become an important high-protein crop. There is limited information on trait variation, genetic architecture, genotype × environment interaction (GEI), relevant genomic areas, and opportunities for genomic selection (GS).

A reference population of sweet-seed breeding lines possessing the pauper locus underwent multi-year evaluation for seed weight, protein content and oil content in two regions (Italy and Chile) and evaluation for content of 13 QAs by a gas chromatography-mass spectrometry method in Italy. A second population including landrace genotypes of worldwide origin was evaluated for protein and oil content and seed weight in Italy. We found substantial genetic variation for all traits. Only 24% of the breeding lines displayed total QA content below 200 mg/kg. Lupanine was the main QA, followed by 13α-hydroxylupanine and 13α-angeloyloxylupanine. GEI across regions was large for protein content, moderate for oil content, and low for seed weight, while being always low across cropping years within region. Genotyping-by-sequencing provided 33,473 SNPs for breeding lines and 41,116 SNPs for landrace genotypes. A genome-wide association study highlighted the polygenic control of total QA content and other traits, identified candidate genes and, particularly for protein content, showed inconsistency for significant SNPs across regions or reference populations. Landrace genotypes exhibited weak population structure partly related to phenology and geographic origin. Our results indicated that region-specific selection for seed weight, protein content and oil content is favoured by high broad-sense heritability, high consistency between parent and progeny values, low GEI, absence of high inverse correlations between traits, and high to moderately high intra-population GS predictive ability (0.41-0.80). The application of GS models defined in one region for selection in the other region, or that of GS models trained on the genetically broader landrace population for selection of breeding lines, proved convenient for seed weight, possible with limitations for oil content, and inconvenient for protein content. High predictive ability (0.66) emerged for total QA content.

Our results highlighted opportunities and limits for phenotypic and genome-enabled selection that can help define efficient breeding strategies.