Models to estimate genetic parameters in crossbred dairy cattle populations under selection

Estimates of genetic parameters needed to control breeding programs, have to be regularly updated, due to changing environments and ongoing selection and crossing of populations. Restricted maximum likelihood methods optimally provide these estimates, assuming that the statisticalgenetic model used is correct.Generally, a model for analysis of milk production data assumes only additive genetic effects and random sampling. These assumptions are rarely met. In many animal populations genetic material from other populations is used. Crossing of lines or breeds often gives rise to non-additive effects. Furthermore, most of the data used for genetic analysis come from populations under selection. The subject of this thesis was to determine whether or not models for genetic evaluation of dairy populations should account for non-additive effects and selection, and how this should be done.The influence of non-additive effects on the estimation of heritabilities and breeding values was studied in Chapter 2. A population having progeny that descended from sires and dams with various fractions of genes from two breeds was simulated. Additive breed effects and non-additive effects from breed crosses, were simulated. Data on performance were analyzed using mixed models, that accounted for fixed additive genetic group and random sire effects. Three additive models, with genetic groups defined according to 1) breed composition of the progeny, 2) breed composition of the sire and dam, or 3) linear regression on breed fraction, were compared with a non-additive model, with a linear regression on breed fraction, heterozygosity and recombination in the genome of the progeny. Variance components were estimated using restricted maximum likelihood.Additive genetic variance and heritability were overestimated for an additive model with progeny groups. Additive models gave biased estimates for breed differences, group effects and breeding values. Breed differences were overestimated when sire groups were used. Estimates for each parameter were unbiased using the non-additive model.In Chapter 3, the same models were applied to data of cows with variable proportions of genes from the Dutch Friesian and the Holstein Friesian (HF) populations. The data set contained 92,333 first lactation records (305 days milk production) of cows from 675 young sires and 307,050 records of cows from 202 proven sires. Estimates for heterosis varied from 2.5% (fat yield) to 0% (protein percentage). Recombination effects varied from -1.9% (protein yield) to 1.5% (fat percentage). Additive models with progeny groups overestimated genetic variance by 6%. Models with sire groups overestimated additive genetic values of imported HF sires by 33%. Using a nonadditive model, heritability estimates were .38 for milk yield, .80 for fat percentage and .70 for protein percentage. It was concluded that a nonadditive model was preferable for estimation of genetic variance and prediction of breeding values in crossbred dairy populations.In the fourth chapter, the effect of selection on estimation of additive genetic variance was studied. A population of size 40 was simulated 100 times, for ten generations. Five out of twenty males were selected at each generation and each male was mated to four females and had two progeny. The additive genetic variance(σ 2a )before selection was 10 and the initial heritability was .5. The genetic variance was reduced to 6.72 after ten generations of selection, due to covariances among animals, inbreeding and gametic disequilibrium. Reduction of variance was lower in another population simulated with size 400 and ten percent of the males selected. Restricted Maximum Likelihood was used to estimate σ2a using an animal model. The estimate of σ2awas empirically unbiased, when all data and all relationships were used. Omitting data from selected ancestors caused biased estimates of σ2adue to the fact that not all gametic disequilibrium was accounted for. Inbreeding and covariances were adjusted for, when additional relationships between assumed base animals were considered. Bias from gametic disequilibrium decreased slightly with the use of more relationship information. Estimates from data based on later generations only, were biased by selection. Mean estimates of genetic variance depended on the assumed base population and were insensitive to the number of subsequent generations with data.A method to estimate genetic parameters conditional to selection occurring before formation of the base population was investigated in Chapter 5. For this, simulated data from the same populations as in Chapter 4 was used. The method assumes base parents as fixed and a conditional variance is based upon the Mendelian sampling of gametes from the base parents. Selection was for five generations but only animals of generations 4 and 5 were assumed to have performance records and parents known. Additive genetic and residual variance were assumed to be 10. When 20 out of 200 sires were selected per generation, estimated genetic variance was 8.58 when base animals were assumed random, and it was 6.03 when they were fixed. Residual variance was overestimated in the latter case. When males of generation 4 were not selected to have progeny, estimated genetic variance was 9.91. It was concluded that estimates for genetic parameters with the conditional model were not biased by selection of base animals. However, the procedure with fixed base parents was biased when descendants of base animals were selected to have progeny.Genetic variance of milk production traits was estimated with a conditional model to account for selection of sires. In the HF subpopulation, which had been selected more intensively, genetic variance for milk yield was estimated about 8% higher compared to a random models that assumes no selection.Estimates of heritability for milk production traits were found to be high with a sire model, after correction for non-additive effects (Chapter 3) and selection of parents (Chapter 5). Preliminary results with an animal model, which accounted for non- random mating of sires, did not show lower estimates. More research is suggested to determine whether the cause for high heritabilities is genetic or environmental.Main conclusions- By not accounting for non-additive effects in genetic evaluation of crossbred populations, biased estimates of breeding values and additive genetic differences between crossbred groups are found. Records of crossbred dairy cattle should therefore be adjusted for systematic additive and non-additive breed effects.- Estimation of crossbreeding parameters from field data can provide low standard errors, although sampling correlation may be high for certain mating designs.- Estimates of genetic variance based on data from selected generations only were biased by selection. Mean estimates of genetic variance depended mostly on the assumed base population and were insensitive to the number of subsequent generations with data. Additional relationships adjust genetic variance estimates for covariances among animals, and forsome of the gametic disequilibrium.- Estimates for genetic parameters with a conditional model are not biased by selection of base animals, but a bias will be introduced when descendants of base animals have been selected to have progeny.- Heritability estimates of milk production traits in crossbred dairy cattle data were found to be higher as parameters currently assumed for genetic evaluation.