Examining across year genotype by environment interactions for production and reproduction traits in Merino sheep

Variation in feed resource availability within production systems can cause genotype by environment interactions that change the ranking of the best animals to select between environments. Mediterranean environments have high variation in pasture growth between years that could cause genetic by environment interactions for sheep production traits. Therefore, we estimated heritabilities for live weight, fleece weight, fibre diameter and number of lambs weaned in six years from 2000 to 2005 and correlations between years comparing multivariate analysis and random regression analysis. We compared 3 methods: 1 multivariate analysis estimating (co)variances for traits in each year, 2 Random regression estimated (co)variances for intercept and slope for traits as repeated measurements fitted against average pasture growth in each year and 3. Random regression fitted against corrected average performance of animals in each year. Random regression was estimated with an order of polynomial of one for additive genetic variance and zero for permanent environmental effects. This combination of polynomials was the best fit based on Bayesian information criterion. We estimated heritabilities for each year and correlations between years using records from 3299 pedigreed Merino ewes managed at Katanning in Western Australia. There were 4651 records for adult live weight, 6750 for adult clean fleece weight, 6965 for adult fibre diameter, and 7774 for number of lambs weaned across all 6 years. Number of lambs weaned had more genotype by environment interactions than other traits, with fibre diameter and fleece weight having genotype by environment interactions between only a few years. Based on Bayesian information criterion values, multivariate analysis fit the data better for live weight, fleece weight and fibre diameter. Additionally, random regression estimated higher genetic correlations between years than multivariate analysis suggesting there was not enough flexibility in the random regression analysis, which used only first order polynomials, to fit differences between years. Pasture growth across years did not explain differences in performance for traits across years. Therefore, for number of lambs weaned, random regression using corrected average performance was a better fit than average pasture growth. For other traits, more years or a better indicator of variation in performance within and between years are required to use random regression for genotype by environment interactions.