Stochastic simulation was used to test the hypothesis that optimum-contribution selection with genomic relationships using marker loci with low minor allele frequency (MAF) below a predefined threshold (referred as TGOCS) to control inbreeding maintained more genetic variation than pedigree relationships (POCS) at the same rate of true genetic gain (∆G_true). Criteria to measure genetic variation were the number of segregating QTL loci (quantitative trait loci) and the average number of founder alleles per locus. Marker alleles having a MAF below 0.025 were used in forming the genomic relationships in TGOCS strategy. For centering in establishing genomic relationships, when the allele frequency of marker loci with low MAF set to 0.5 the TGOCS strategy maintained 66% fewer founder alleles than POCS and there were 30% fewer QTL segregating. This TGOCS strategy maintained 61% fewer founder alleles than GOCS and 28% fewer segregating QTL loci. When the allele frequency of marker loci with low MAF was set to observed allele frequency these figures were 8%, 2%, 5% and 2%, respectively. Using marker loci with low MAF in the TGOCS strategy was inferior to both GOCS and POCS. Both TGOCS and GOCS were affected by the same constraint that is LD (linkage disequilibrium) between markers and QTL. Therefore, POCS is a more efficient method to maintain genetic variation in the population until a better way to use genomic information in optimum-contribution selection is identified.

Background In this study, we tested whether genotyping both live and dead animals (GSD) realises more genetic gain for post-weaning survival (PWS) in pigs compared to genotyping only live animals (GOS).

Methods Stochastic simulation was used to estimate the rate of genetic gain realised by GSD and GOS at a 0.01 rate of pedigree-based inbreeding in three breeding schemes, which differed in PWS (95%, 90% and 50%) and litter size (6 and 10). Pedigree-based selection was conducted as a point of reference. Variance components were estimated and then estimated breeding values (EBV) were obtained in each breeding scheme using a linear or a threshold model. Selection was for a single trait, i.e. PWS with a heritability of 0.02 on the observed scale. The trait was simulated on the underlying scale and was recorded as binary (0/1). Selection candidates were genotyped and phenotypes before selection, with only live candidates eligible for selection. Genotyping strategies differed in the proportion of live and dead animals genotyped, but the phenotypes of all animals were used for predicting EBV of the selection candidates.

Results Based on a 0.01 rate of pedigree-based inbreeding, GSD realised 14 to 33% more genetic gain than GOS for all breeding schemes depending on PWS and litter size. GSD increased the prediction accuracy of EBV for PWS by at least 14% compared to GOS. The use of a linear versus a threshold model did not have an impact on genetic gain for PWS regardless of the genotyping strategy and the bias of the EBV did not differ significantly among genotyping strategies.

Conclusions Genotyping both dead and live animals was more informative than genotyping only live animals to predict the EBV for PWS of selection candidates, but with marginal increases in genetic gain when the proportion of dead animals genotyped was 60% or greater. Therefore, it would be worthwhile to use genomic information on both live and more than 20% dead animals to compute EBV for the genetic improvement of PWS under the assumption that dead animals reflect increased liability on the underlying scale.

A premise was tested that genotyping both surviving and dead pigs will realise more genetic gain in post-weaning survival (PWS) than genotyping only surviving animals. Stochastic simulation was used to estimate the rate of true genetic gain in different genotyping scenarios that differed in varying proportions of genotyping dead animals. Selection was for only PWS that had heritability of 0.02. Mortality was assumed 10%. The trait was controlled by 7,702 biallelic quantitative trait loci distributed across a 30 Morgan genome. We used 54,218 biallelic single nucleotide polymorphisms (SNPs) that were used in genomic prediction. Genotyping both surviving and dead animals realised 12 to 24% more genetic gain than genotyping only surviving animals. The power of detecting SNP effects increased when animals of extreme phenotypes are genotyped. Therefore, genotyping both surviving and dead pigs realised more genetic gain than genotyping only surviving animals.

Changes in the relative performance of genotypes (sires) across different environments, which are referred to as genotype–environment interactions, play an important role in dairy production systems, especially in countries that rely on imported genetic material. Importance of genotype by environment interaction on genetic analysis of milk yield was investigated in Holstein cows by using random regression model. In total, 68 945 milk test-day records of first, second and third lactations of 8515 animals that originated from 100 sires and 7743 dams in 34 herds, collected by the Iranian animal breeding centre during 2007–2009, were used. The different sires were considered as different genotypes, while factors such as herd size, herd milk average (HMA), herd protein average and herd fat average were used as criteria to define the different environments. The inclusion of the environmental descriptor improved not only the log-likelihood of the model, but also the Bayesian information criterion. The results showed that defining the environment on the basis of HMA affected genetic parameter estimations more than did the other environmental descriptors. The heritability of milk yield during lactating days reduced when sire · HMA was fitted to the model as an additional random effect, while the genetic and phenotypic correlations between lactating months increased. Therefore, ignoring this interaction term can lead to the biased genetic-parameter estimates, reduced selection accuracy and, thus, different ranking of the bulls in different environments.

Optimisation of pig breeding programs aims to increase the genetic gain in pig populations and to decrease the rate of inbreeding in the pig nucleus population. Genomic selection is a potential breeding strategy that can increase genetic gain and is also expected to decrease the rate of inbreeding in livestock breeding programs. Pigs are selected based on multiple correlated traits in the nucleus population. It might be difficult to improve response to selection in favourable direction for individual breeding objective traits because of an interplay between complex correlation structure and the economic value of each trait. On top of that, genomic selection might also shift genetic gain towards hard-to-measure traits. More work is needed on how genomic selection benefits the overall merit of breeding objectives and individual breeding objective traits.

Post-weaning survival (PWS) is an important breeding objective trait in the sire line of pigs. The benefits of genomic selection for PWS depend on the structure of the reference population, which should have both genotypes and phenotypes. Animal breeders might not be interested in genotyping dead pigs because dead pigs cannot be selection candidates. However, genotyping dead and live pigs might increase the genetic gain for PWS in comparison to genotyping live animals alone. While improving genetic gain, it is also important to reduce the rate of inbreeding because pigs are selected in a closed elite herd. Genomic markers might also increase genetic diversity because genomic relationships are more accurate than pedigree relationships. With the availability of genomic marker information, it is also easier to account for the dominance effect than pedigree information in the genetic evaluation model in the presence of dominance. Therefore, the broad objective of this thesis was to investigate the benefits of using genomic selection in pig breeding programs. This thesis explored multiple new avenues of using genomic information to increase the rate of genetic gain and decrease the rate of inbreeding in pig breeding programs.

In chapter 3, a premise was tested that the overall pig breeding objective achieves additional genetic gain in genomic selection compared to pedigree selection, but some traits achieve larger additional genetic gain than other traits. Results in a deterministic simulation study showed that genomic selection scenarios based on different sizes of reference populations increased overall response in the breeding objectives by 9% to 56% and 3.5% to 27% in the dam and sire lines, respectively, compared to pedigree selection. In the dam line of pigs, reproductive traits such as sow mature weight, number born alive, and sow longevity achieved 123% to 403%, 73 % to 351%, and 58% to 278% larger genetic gain in genomic selection compared to the pedigree selection respectively. In comparison, backfat thickness, average daily gain, and feed conversion ratio achieved 6% to 14%, 4% to 11%, and 7% to 9% smaller genetic gain in genomic selection than pedigree selection, respectively. In the sire line of pigs, post-weaning survival, drip loss, and middle portion percentage achieved larger genetic gain in genomic selection than the pedigree selection. Achieving larger genetic gain for reproduction traits in the dam line and post-weaning survival and meat and carcass quality traits in the sire line increased the overall merit of pig breeding objectives in genomic selection compared to the pedigree selection.

In chapter 4, a premise was tested that genotyping both live and dead animals realises more genetic gain for PWS (assuming a PWS of 90%) in pigs compared to the scenario where only live animals are genotyped. Stochastic simulation was conducted to compare genetic gain in the scenarios of either genotyping live and dead animals or genotyping live animals only. Genetic gain for PWS in these genotyping strategies was compared at 1% pedigree inbreeding in optimum contribution selection. Results showed that genetic gain with genotyping all live animals was 52% higher than pedigree selection. On top of that, genetic gain with genotyping live and 20 to 100% of dead animals was 14 to 33% higher than genotyping only live animals. Genotyping live and dead animals increased the accuracy of the genomic breeding values of live animals compared to genotyping only live animals, which resulted in a larger genetic gain for PWS.

In chapter 5, a premise was tested that optimum-contribution selection with genomic relationships using only low MAF (minor allele frequency) markers below a predefined threshold to control inbreeding realises less rate of true inbreeding than optimum-contribution selection (OCS) with pedigree relationships. Genetic gain in genomic and pedigree-based OCS was fixed at a predefined value while comparing the rate of inbreeding. Results showed that pedigree-based OCS realised a lower rate of inbreeding than genomic-based OCS at the same rate of genetic gain. Genomic-based OCS fixed more favourable quantitative trait loci than pedigree-based OCS. In addition, genomic-based OCS selected more closely related animals than pedigree-based OCS. Therefore, pedigree-based OCS realised a lower rate of inbreeding than genomic-based OCS at the same rate of genetic gain.

Finally, in chapter 6, a premise was tested that genetic gain in pig breeding programs using dominance models that accounted for both random additive genetic and dominance effects was higher than additive models that included only random additive genetic effects under the presence of dominance. The stochastic simulation was conducted to compare models in thedam and sire line of pigs. In the sire line, similar additive genetic variances were estimated by the two models but with the additive model, the litter and residual variances were biased upward by 42% and 23%, respectively. When the model did not include a common litter effect in the dam line, the additive genetic variance was 10% smaller in comparison to the additive genetic variance estimated using the dominance model. Despite overestimating variance components using additive models, both models realised a similar rate of total true genetic gain. Since animals were selected based on additive genetic merit, the dominance model did not impact the rate of total true genetic gain. Therefore, the additive genetic model can be used for estimating breeding values if animals are selected based on additive genetic merit, even under the presence of dominance.

The results mentioned above showed the potential of genomic selection to increase genetic gain in pig breeding programs. This study investigated multiple new avenues of using genomic information for the genetic improvement in pigs. However, there are still many unanswered question. Use of genomics is beneficial for improving the accuracy of selection and genetic gain, it is not clear how to use genomics to control inbreeding. To take the advantage of genomics, more work is needed to investigate how to use genomics to control inbreeding. In addition, genomics can be useful for accounting non-additive genetic effects such as dominance and epistasis in the model. As more research emerges, use of genomics will be more useful for optimising the pig breeding programs because genomics opens up further opportunities to reveal the biology of traits.

The premise was tested that the additional genetic gain was achieved in the overall breeding objective in a pig breeding program using genomic selection (GS) compared to a conventional breeding program, however, some traits achieved larger gain than other traits. GS scenarios based on different reference population sizes were evaluated. The scenarios were compared using a deterministic simulation model to predict genetic gain in scenarios with and without using genomic information as an additional information source. All scenarios were compared based on selection accuracy and predicted genetic gain per round of selection for objective traits in both sire and dam lines. The results showed that GS scenarios increased overall response in the breeding objectives by 9% to 56% and 3.5% to 27% in the dam and sire lines, respectively. The difference in response resulted from differences in the size of the reference population. Although all traits achieved higher selection accuracy in GS, traits with limited phenotypic information at the time of selection or with low heritability, such as sow longevity, number of piglets born alive, pre- and post-weaning survival, as well as meat and carcass quality traits achieved the largest additional response. This additional response came at the expense of smaller responses for traits that are easy to measure, such as back fat and average daily gain in GS compared to the conventional breeding program. Sow longevity and drip loss percentage did not change in a favourable direction in GS with a reference population of 500 pigs. With a reference population of 1000 pigs or onwards, sow longevity and drip loss percentage began to change in a favourable direction. Despite the smaller responses for average daily gain and back fat thickness in GS, the overall breeding objective achieved additional gain in GS.

The responses to genomic selection in breeding programs for growing pigs were predicted using a selection index approach. Genomic selection increased overall predicted response by 2.6 (500 reference population) to 27.8% (5000 reference population) for a breeding objective consisting of backfat thickness (BFT), average daily gain (ADG), post-weaning survival (PWS) and feed conversion ratio (FCR) in growing pigs . Predicted response in PWS increased by 147% with genomic selection (5000 reference population) at the expense of the other traits like BFT, ADG, and FCR which had 14.5, 1.6, and 2.8% less genetic gain compared to the response in a conventional breeding program without genomic selection. The higher loss in genetic gain for BFT was due to a stronger genetic correlation with FCR in comparison to ADG. The predicted additional responses in the breeding objective is a guideline for the implementation of genomic selection in pig breeding programs.