A genetic evaluation of growth in sheep using random regression techniques

Repeated measures of live weight in growing animals are used to describe the path by which they travel from birth to maturity. A family of growth functions-the Gompertz is one in particular-has been used successfully to describe this journey with relatively few parameters (most importantly mature size and a rate parameter). However, using these functions to differentiate the genetic merit of individual animals to grow is problematic since the estimates of these parameters are highly correlated and are obtained with varying precision among animals. An alternative is random regression (RR) methodology. It allows environmental effects specific to the time of recording to be accounted for and can accommodate genetic differences in the shape of each animal's growth curve. At present, though, only linear models (polynomials) can pragmatically be fitted with RR. This may be limiting since a priori beliefs about the appropriate form of a growth function, such as the non-linear Gompertz equation, cannot be accommodated. This paper describes the application of RR techniques to describe growth on a population of Suffolk sheep and compares the genetic evaluation predicted from a RR model with that obtained from a more traditional method based on a Gompertz form. The RR model chosen as providing the best fit (P < 0.01) included additive genetic and permanent environmental (between repeat records of an individual) effects fitted to a fifth order polynomial, and dam effects fitted to a third order polynomial. Measurement error was modelled as six classes. The heritability varied at different points along the growth trajectory (from 0.09 at 15 days to 0.33 at 150 days), suggesting that live weight early in a lamb's life is a different trait to live weight later in life. There was genetic variation in the growth curves of individual animals, which was accounted for by fitting a RR model. Breeding values obtained by RR and a Gompertz approach were moderately to highly correlated (0.81 at 56 days, 0.91 at 150 days). If breeding value for live weight at 150 days of age were the selection criterion, similar individuals would be chosen with both methodologies. The 'better' properties and greater flexibility of the RR approach are discussed.