Maximizing the use of molecular markers in pine breeding in the context of genomic selection.

摘要

By 2030 demand for renewable energy, food and fiber is expected to double. To sustainably meet this increase in demand from the current land base, plant breeders need to develop higher yielding crops that require fewer inputs and better resist diseases and environmental change. Of particular importance is accelerating improvement in quantitative traits (QT), which show complex patterns of inheritance. Genomic selection (GS) provides an approach where molecular markers can be used directly in breeding programs regardless of the genetic architecture. While most GS studies have concentrated on prediction of breeding values, here this approach is extended to include non-additive variation and to maximize the use of molecular markers (SNPs) in pine breeding.;With a relatively dense panel of SNPs, a method to detect and correct errors in the pedigree information is presented, based on a marker-derived additive relationship matrix. The impact of pedigree errors on genetic parameter estimates and breeding value prediction is demonstrated. In addition, the performance of four published analytical methods for GS that differ in assumptions regarding the distribution of markers additive-effect is presented. Methods include: ridge regression--best linear 12 unbiased prediction (RR--BLUP), Bayes A, Bayes Cpi, and Bayesian LASSO. Furthermore, a modified RR--BLUP (RR-BLUP_B) that utilizes a selected subset of markers was developed and evaluated. All five methods for GS were evaluated for seventeen different traits of importance in pine breeding and with different predicted genetic architecture and heritabilities. While for QT no significant difference among methods was detected, for traits controlled by fewer genes, Bayes Cpi and RR-BLUP_B performed significantly better. Finally, the use of a dense panel of SNPs to partition the genetic variance into additive and non-additive components was evaluated. For tree height, use of the SNP-derived relationship matrices (additive and non-additive) in a statistical model including additive and non-additive effects performed best, not only to partition the genetic variances but also to improve considerably the breeding value prediction ability in trend, magnitude and top individual selection. This study indicates that markers can be used beyond prediction of additive effects, positively impacting the genetic gain of the breeding program.