Breeding values used to select sugarcane parents for cross pollination in the Australian sugarcane improvement program have been obtained using an index called Net Merit Grade. This approach used Best Linear Unbiased Prediction methodology to estimate breeding values for sugarcane parents by applying a simple mixed model to analyse Net Merit Grade (a derived trait from combining cane yield and sugar content) rather than the two component traits individually. A number of limitations of this system have been recognised: i) site-specific spatial variation and interplot competition effects were not modelled; ii) only additive genetic effects were estimated without partitioning of non-additive effects for full-sib families; iii) each regional program was analysed independently, despite genetic linkage through common parents and families across the regions; and iv) additive genetic by environment interactions were not accounted for. These limitations potentially biased breeding value predictions and reduced gains when selecting parents and specific cross combinations to initiate each new breeding cycle. This thesis documents the development of a new approach for the Australian sugarcane improvement program by addressing these limitations to estimate the breeding values of parents more accurately.
Sugarcane full-sib family trials are typically large leading to spatial variability within trials. In addition, sugarcane families are grown in unguarded plots where families compete with their neighbours for resources resulting in interplot competition effects. Methods to model spatial variation and interplot competition were applied to 66 sugarcane family trials. These analyses demonstrated that significant spatial variation was present in 88% and 79% of trials for cane yield and sugar content respectively, while interplot competition effects were present in 86% of trials for cane yield. The presence of these effects in family trials substantially affected the estimates of breeding values of sugarcane parents. Removal of spatial effects resulted in additional genetic gains of up to 5% from parent selection, while modelling of interplot competition resulted in a further 3.9% increase in gains from parent selection. However, the competition model failed to converge with more than four years of family data from one region, and hence, could not be applied to the remainder of the analyses as these involved larger amounts of data across regions.
Data from each region were previously analysed independently, therefore genetic correlations between parents and families among regions were not exploited to improve breeding value predictions. To facilitate across-region analyses, the minimum amount of historical family data and pedigree information required to obtain accurate breeding values was determined. The accuracy of additive variance and breeding value estimates improved as more pedigree information and historical data were included in the analyses. However, inclusion of more than four generations of pedigree and seven years of historical data did not significantly improve the accuracy of breeding value estimates, and required substantially greater computational resources (computer time, power and memory).
Sugarcane family data were then combined across regions to investigate the importance of additive genetic by environment interactions using a Factor-Analytic mixed model. The magnitude of additive genetic by environment interactions was small relative to the additive genetic effects of sugarcane parents across regions for both cane yield and sugar content. Parental rankings were stable across regions, and the accuracy of breeding value predictions increased by up to 70% for cane yield and 50% for sugar content when family data were pooled across regions and additive genetic by environment interaction effects were accounted for across regions.
These results led to the development of an integrated approach to obtain improved estimates of breeding value by: i) modelling site-specific spatial variation; ii) partitioning genetic effects of sugarcane parents into additive and non-additive genetic effects; iii) combining family data across regions using a Factor-Analytic mixed model to exploit the genetic correlations among trials in different regions; and iv) estimating breeding values for cane yield and sugar content separately rather than for Net Merit Grade. Further, as Net Merit Grade no longer reflects current sugarcane production systems, the improved breeding values were incorporated into a new economic selection index.
This new approach to analysing sugarcane family data explained over 75% of the phenotypic variation in both cane yield and sugar content, while the previous approach explained less than 40% of the variation, leading to greater accuracy of breeding value prediction. Application of the new methodology suggests that additional genetic gains of up to 29% and 10% for cane yield and sugar content, respectively, can be achieved when parents are selected on breeding values estimated using the new approach.
Additional gains for the Australian sugarcane industry of $2.34 per tonne of sugar produced could be achieved if parents are selected using a new economic selection index combined with breeding values estimated using the new approach outlined in this thesis. Consequently, sugarcane breeders can now select parents with greater confidence.