Application of random regression models to study growth curves and genotype x environment interactions in sheep

Abstract

Modelling sheep growth and producing estimated breeding values (EBV) for growth traits are widely used to optimize sheep production. The methods available model growth traits as a function of age, often with a set of fixed or random environmental effects.

Current methods to map growth rely on two prominent assumptions that: (1) The mean and covariance structure of the growth trait remains constant with time or age, and (2) each measurement of the growth trait is genetically different from, but correlated to, all other measurements of the growth trait.

However, this methodology is problematic. It has three main issues: (1) impracticality; (2) neglecting wide random variability in environmental effects; and (3) overparameterization.

Thus, this project used a random regression model to describe growth more accurately in sheep, thereby producing better selection criteria to choose the best breeding stock. These models included genetic effects, maternal genetic effects, and a host of conformation and meat quality data. However, random regression models can struggle to provide accurate results when data are sparse or unevenly distributed. Thus, the use of commercial data in this project allows investigation of model performance and assessment of the applicability of random regression models in commercial environments. Overall, the project found random regression models are highly sensitive to the distribution and variability of records across the age distribution. This results in high correlations between the parameters of the regression model which can result in inaccurate genetic parameters. In Chapter 2 random regression models were constructed for growth in Charollais and Suffolk sheep with constrained correlations resulting in heritabilities between (0.18-0.49) and (0.20-0.50) respectively. The inclusion of the constrained correlation was validated using a novel procedure. In Chapter 3, carcass information was incorporated using commercial mixed breed and Scottish Blackface research data with similar numbers of records. A selection index and a random regression model were compared. Model convergence for the random regression model was achieved by constraining the correlation. Genetic parameters could be calculated between weights and fat class or conformation. The selection index offered accurate information for slaughter weight and carcass weight. Chapter 4 assessed GxE effects in the RamCompare project using a sire model with a sire by flock interaction and a reaction norm model using a phenotypic deviation. It indicated GxE effects were present for birth weight, scan weight, weaning weight and muscle depth. It also showed that an assessment of GxE effects in commercial flocks can be conducted in datasets that lack environmental data.

The project contributes to a growing body of research on how to best model heritable traits and provide genetic information to commercial sheep flocks.