Fine mapping and genomic prediction for detailed milk composition : Unraveling the genetics of milk composition towards implementing selective breeding

Abstract

Suitability of milk for processing into high-value products, such as cheese and butter, is affected by its protein and fatty acid (FA) composition. In addition, there are consumer concerns to some specific components of milk, whilst increasing preferences for others, mainly on health grounds. Therefore, economic and consumer pressures are driving interests in altering the detailed protein and FA composition of milk. Among potential strategies to alter detailed milk composition, genetic improvement provides cumulative effects carried over generations for a unit of investment. Selective breeding requires large-scale availability of data for accurate estimation of genetic parameters and prediction of genetic merits. Measurement for detailed milk protein and FA traits is currently limited to experimental scales due to costly and time-consuming analytical techniques. The PhD study aimed at improving accuracy of genetic parameter estimation and prediction of breeding values as well as understanding the genetic backgrounds of detailed milk protein and FA traits using efficient quantitative approaches. It is shown that improved accuracy of parameter estimation and genomic prediction is possible for scarcely recorded traits using multi-trait analyses. Advantages of existing multi-trait models is limited when genetic correlation between analyzed traits is weak. With a novel Bayesian model considering heterogeneous correlation structures over the genome, we show that despite weak genome-wide correlation, there exist genomic-regions explaining strong correlation and that it is possible to utilize such “local” correlations for accurate multi-trait genomic prediction. Combining datasets from different populations of a breed was another strategy investigated and shown to benefit genome-wide association (GWA) and genomic prediction for scarcely recorded traits. It is also demonstrated that existing linear genomic prediction models can be extended to incorporate GWA findings for further gain in prediction accuracy. Post- GWA analyses with multiple data sources including tissue-specific gene expression, ontology and pathway information can help refine GWA findings and provide potent information for genomic prediction models. Novel genomic regions and candidate genes established in the study contribute to the knowledge base on the complex genetic backgrounds of milk FA traits. The findings suggest that genomic selection for detailed milk composition is possible. Novel methods presented in the thesis will be of value for genomic prediction in other scarcely recorded traits of economic importance.