Contribution of genomic selection to improve palm oil Yieldin ElaeisguineensisJacq.

Abstract

Genomic selection (GS) is expected to increase the annual genetic progress and lead palm oil production up to the growing world demand. Genetic improvement for hybrid performances has a major role to play to meet this demand while minimizing environmental impacts. A modified reciprocal recurrent scheme is used to select the most performing hybrids commercialized as hybrid cultivars or used for the most performing individuals as hybrid ortets in clonal selection. The current study empirically evaluated the interest of using genomic data from A × B hybrid individuals for the genomic approach applied to oil palm (Elaeis guineensis Jacq.). The efficiency of GS for clonal selection was first evaluated using a training set comprising almost 300 Deli × La Mé crosses phenotyped for eight palm oil yield components and the validation set 42 Deli × La Mé ortets. Genotyping-by-sequencing (GBS) revealed 15,054 single nucleotide polymorphisms (SNP). The effects of the SNP dataset (density and percentage of missing data) and two GS modeling approaches, across-population SNP genotype models (ASGM) and population-specific effects of SNP alleles models (PSAM), respectively ignoring considering the parental origin of alleles, were assessed. Secondly, we investigated the effect of two strategies to optimize the GS accuracy in oil palm hybrid: genotyping strategy for the training population, i.e., genotyping only the hybrid parents or also a sample of hybrid individuals, and modeling of markers ASGM and PSAM. For that purpose, genomic data of both parents and hybrid individuals were used for calibration and predictions were done using ASGM and PSAM. The training set was constructed with around 350 hybrid crosses, including around 15,000 to 23,000 individuals phenotyped, depending on trait. Validation was realized in an independent set of 213 hybrid crosses. GBS was applied on the parents of the training and validation sets and on around 400 training hybrid individuals, yielding 21,458 SNPs. The results showed prediction accuracies ranging from 0.08 to 0.70 for ortet candidates without data records, depending on trait, SNP dataset and modeling. ASGM with a mean prediction of 0.45 was better (on average slightly more accurate, less sensitive to SNP dataset and simpler) than PSAM with a mean prediction accuracy of 0.43, although PSAM appeared interesting for a few traits. With ASGM, the number of SNPs had to reach 7,000, while the percentage of missing data per SNP was of secondary importance, and GS prediction accuracies were higher than those of PS for most of the traits. Prediction accuracies ranged from 0.15–0.89 for hybrid crosses depending on trait, model and genotyping strategy. Prediction accuracies increased on average by 5% when training was done with genomic data of hybrid individuals and parents compared with only parental genomic data. Prediction accuracies increased on average by 3% with ASGM compared to PSAM. In our dataset, the mean prediction accuracy over traits of the best GS approach, i.e., ASGM with hybrid individuals’ genotypes, reached 0.53. Ultimately, this work makes possible two practical applications of GS, that will increase genetic progress by improving ortet preselection before clonal trials: preselection at the mature stage on all yield components jointly using ortet genotypes and phenotypes, and genomic preselection on more yield components than PS, among a large population of the best possible crosses at nursery stage. In addition, this work revealed that genomic data of the training hybrid individuals and GBLUP are useful to increase prediction accuracy; with ASGM the recommended modeling approach for that purpose. Further studies should investigate the factors controlling the relative performance of ASGM and PSAM approaches in oil palm, and focus on the optimal number of hybrid individuals to genotype to maximize the selection response per unit cost.