A crossbred reference population can improve the response to genomic selection for crossbred performance

Abstract

Background: Breeding goals in a crossbreeding system should be defined at the commercial crossbred level. However, selection is often performed to improve purebred performance. A genomic selection (GS) model that includes dominance effects can be used to select purebreds for crossbred performance. Optimization of the GS model raises the question of whether marker effects should be estimated from data on the pure lines or crossbreds. Therefore, the first objective of this study was to compare response to selection of crossbreds by simulating a two-way crossbreeding program with either a purebred or a crossbred training population. We assumed a trait of interest that was controlled by loci with additive and dominance effects. Animals were selected on estimated breeding values for crossbred performance. There was no genotype by environment interaction. Linkage phase and strength of linkage disequilibrium between quantitative trait loci (QTL) and single nucleotide polymorphisms (SNPs) can differ between breeds, which causes apparent effects of SNPs to be line-dependent. Thus, our second objective was to compare response to GS based on crossbred phenotypes when the line origin of alleles was taken into account or not in the estimation of breeding values.

Results: Training on crossbred animals yielded a larger response to selection in crossbred offspring compared to training on both pure lines separately or on both pure lines combined into a single reference population. Response to selection in crossbreds was larger if both phenotypes and genotypes were collected on crossbreds than if phenotypes were only recorded on crossbreds and genotypes on their parents. If both parental lines were distantly related, tracing the line origin of alleles improved genomic prediction, whereas if both parental lines were closely related and the reference population was small, it was better to ignore the line origin of alleles.

Conclusions: Response to selection in crossbreeding programs can be increased by training on crossbred genotypes and phenotypes. Moreover, if the reference population is sufficiently large and both pure lines are not very closely related, tracing the line origin of alleles in crossbreds improves genomic prediction.

Fig. 1

Schematic representation of the simulation steps. The crossbreeding program started in step 4 and consisted of five generations of purebred selection for crossbred performance; the random sample of individuals from the last generation of step 3 (Generation 2308) composed the purebred training set, and crossbred animals (AB*) in generation 2307 composed the crossbred training set; A_M and B_M represent the selected males of breeds A and B, A_F and B_F the selected females of breeds A and B; lines with arrows denote reproduction, while lines without arrows denote selection; the size of the reference population for scenarios with purebred training was 1000 within each pure breed, and 2000 for the scenarios with crossbred training; thus all scenarios had a total reference population size of 2000

Fig. 2

Phenotypic mean of crossbred animals. Scenario 1 separate training in both breeds A and B. Scenario 2 training on a combined set of animals from both breeds A and B. Scenario 3 training on crossbred animals with phenotypes and genotype probabilities and it was assumed that the alternate heterozygotes Aa and aA were identical in crossbred animals. Scenario 4 training on crossbred animals with phenotypes and genotype probabilities and it was assumed that the alternate heterozygotes could be distinguished in crossbred animals. Scenario 5 training on crossbred animals with phenotypes and genotypes and it was assumed that the alternate heterozygotes were identical in crossbred animals. Scenario 6 training on crossbred animals with phenotypes and genotypes and it was assumed that the alternate heterozygotes could be distinguished in crossbred animals; standard errors of phenotypic means ranged from 0.02 to 0.03

Fig. 3

Heterosis in crossbreds. Scenario 1: separate training in both breeds A and B; Scenario 2: training on a combined set of animals from both breeds A and B; Scenario 3: training on crossbred animals with phenotypes and genotype probabilities and it was assumed that the alternate heterozygotes Aa and aA were identical in crossbred animals; Scenario 4: training on crossbred animals with phenotypes and genotype probabilities and it was assumed that the alternate heterozygotes could be distinguished in crossbred animals. Scenario 5: training on crossbred animals with phenotypes and genotypes and it was assumed that the alternate heterozygotes were identical in crossbred animals. Scenario 6: training on crossbred animals with phenotypes and genotypes and it was assumed that the alternate heterozygotes could be distinguished in crossbred animals

Fig. 4

Cumulative response to selection for varying sizes of training populations. Training on crossbred animals with phenotypes and genotypes. Scenario 5: alternate heterozygotes Aa and aA were assumed identical in crossbred animals; Scenario 6: alternate heterozygotes could be distinguished in crossbred animals; TS stands for training population size of 500, 2000 and 8000; standard errors of phenotypic means ranged from 0.02 to 0.03