Estimation of dam line composition of 3-way crossbred animals using genomic information

Abstract

Background In genomic prediction including data of 3- or 4-way crossbred animals, line composition is usually fitted as a regression on expected line proportions, which are 0.5, 0.25 and 0.25, respectively, for 3-way crossbred animals. However, actual line proportions for the dam lines can vary between ~ 0.1 and 0.4, and ignoring this variation may affect the genomic estimated breeding values of purebred selection candidates. Our aim was to validate a proposed gold standard to evaluate different approaches for estimating line proportions using simulated data, and to subsequently use this in actual 3-way crossbred broiler data to evaluate several other methods. Results Analysis of simulated data confirmed that line proportions computed from assigned breed-origin-of-alleles (BOA) provide a very accurate gold standard, even if the parental lines are closely related. Alternative investigated methods were linear regression of genotypes on line-specific allele frequencies, maximum likelihood estimation using the program ADMIXTURE, and the genomic relationship of crossbred animals with their maternal grandparents. The results from the simulated data showed that the genomic relationship with the maternal grandparent was most accurate, and least affected by closer relationships between the dam lines. Linear regression and ADMIXTURE performed similarly for unrelated lines, but their accuracy dropped considerably when the dam lines were more closely related. In almost all cases, estimates improved after adjusting them to ensure that the sum of dam line contributions within animals was equal to 0.5, and within dam line and across animals the average was equal to 0.25. Results from the broiler data were much more similar between methods. In both cases, stringent linkage disequilibrium pruning of genotype data led to a relatively low accuracy of predicted line proportions, due to the loss of too many single nucleotide polymorphisms. Conclusions With relatively unrelated parental lines as typical in crosses in pigs and poultry, linear regression of crossbred genotypes on line-specific allele frequencies and ADMIXTURE are very competitive methods. Thus, linear regression may be the method of choice, as it does not require genotypes of grandparents, is computationally very efficient, and easily implemented and adapted for considering the specific nature of the crossbred animals analysed.

A look under the hood of genomic-estimated breed compositions for brangus cattle: What have we learned?

“…Nevertheless, they corresponded approximately to the probability of alleles identical by descent when confined to long chromosome chunks. We noted that alternative approaches existed, such as path analysis ( Wu et al, 2020 ) and the breed-of-origin (BOA) approach ( Calus et al, 2022 ), that could better handle the situation when ancestral breeds were correlated. In the present study, however, the correlation between Angus and Brahman was low (0.05–0.10).…”

Validation with single-step SNPBLUP shows that evaluations can continue using a single mean of genotyped individuals, even with multiple breeds

“…However, we are confident that the three breeds used here are sufficiently divergent for the usefulness of fitting multiple J-factors to be observable. This is based on the fact that these breeds have been selected from real breeding lines that are divergent, such that the average F ST value between the three breeds is 0.24, which is greater than between simulated populations that separated 50 generations ago (see Calus et al [ 16 ]). It would be interesting to consider other breeding designs and population structures, especially with multiple different types of crossbreds included, to investigate if they would benefit from additional J-factors.…”

“…This study reuses data from a broiler three-way cross experiment that was previously analysed [ 13 – 16 ]. The main advantages of this dataset are that (1) all the phenotyped animals are genotyped, (2) the pedigree and the observed breed composition of all the crossbred animals were previously derived by Calus et al [ 13 ], and (3) the three purebred lines have been shown to be distantly related [ 15 , 16 ]. Furthermore, Duenk et al [ 14 ] analysed the same dataset using models based on the breed-of-origin (BOA) of alleles, which enabled us to compare our results to those obtained with that model.…”

“…This study reuses data from a broiler three-way cross experiment that was previously analysed [13][14][15][16]. The main advantages of this dataset are that (1) all the phenotyped animals are genotyped, (2) the pedigree and the observed breed composition of all the crossbred animals were previously derived by Calus et al [13], and (3) the three purebred lines have been shown to be distantly related [15,16]. Furthermore, Duenk et al [14] analysed the same dataset using models based on the breed-oforigin (BOA) of alleles, which enabled us to compare our results to those obtained with that model.…”

“…We derived the breed fractions from a previous BOA analysis [13], based on the approach of Vandenplas et al [19]. We decided on this method because the previously derived BOA were available and because this is probably one of the most accurate ways of determining the actual breed composition [20].…”