The aim of this study was to develop a new genetic evaluation model to estimate the genetic merit of boars for growth based on 1) performance of their crossbred progeny fattened in the test station and 2) their own performance or those of relatives from the on-farm testing system. The model was a bivariate random regression animal model with linear splines and was applied to Piétrain boars from the Walloon Region of Belgium mated with Landrace sows. Data contained 1) 12,610 BW records from the test station collected on 1,435 crossbred pigs from Piétrain boars and Landrace sows, and 2) 52,993 BW records from the on-farm testing system collected on 50,670 pigs with a breed composition of at least 40% Piétrain or Landrace. Since 2007, 56 Piétrain boars have been tested in the station. Data used to estimate variance components and breeding values were standardized for the age to take into account heterogeneity of variances and then pre-adjusted at 210 d of age to put all records on the same scale. Body weight records from the test station and from the on-farm testing system were considered as 2 different traits. The heterosis effect was modeled as fixed regression on the heterozygosity coefficient. As all test station animals were similarly crossbred, smaller variation in heterozygosity caused the sampling error of the regression estimate at 210 d to be larger in the test station than in on-farm data with estimates of 28.35 ± 14.55 kg and 9.02 ± 0.67 kg, respectively. Therefore, the most likely reason for the large differences in estimates was sampling. Heritability estimates ranged from 0.37 to 0.60 at 210 and 75 d, respectively, for test station BW and from 0.42 to 0.60 at 210 d and 175 d, respectively, for on-farm BW. Genetic correlation decreased when the age interval between records increased, and were greater between ages for test station than for on-farm data. Genetic correlations between test station and on-farm BW at the same age were high: 0.90 at 175 d and 0.85 at 210 d. For the 56 boars tested in the station, the average reliability of their EBV for ADG between 100 and 210 d was improved from 0.60 using only test station data to 0.69 using jointly test station and on-farm data. Based on these results, the new model developed was considered as a good method of detection of differences in growth potential of Piétrain boars based on test station and on-farm data.

Introduction Until recently, Piétrain boars in the Walloon Region were evaluated with performances recorded on their purebred progeny. However, these boars are mostly used in crossbreeding systems. Therefore, since 2007, a new genetic evaluation system has been developed in the Walloon Region. Piétrain boars are now evaluated on performances recorded on their crossbred progeny with Landrace sows. The aim of this study was to contribute to the new genetic evaluation system of these boars by the development of a genetic evaluation model for carcass quality. The objective was to develop a tool that allows selection of boars that produce carcass with a high lean meat percentage. Material and methodsData provided by the on-farm performance recording system, also recorded at a central test station, were utilized in this study. Performances were recorded on live animals by ultrasound with the Piglog 105. This strategy provided data recorded on animals from the test station, measured the week before slaughtering, and on boars themselves and other related animals on their farms of origin. The data file contains 60,546 records from pigs between 150 and 300 days of age, originated from 56 822 different animals. Animals used needed to have a breed composition of at least 40 % Piétrain or Landrace. Recorded animals were entire males, castrated males or females. Traits analysed in this study were backfat thickness (BF) and meat percentage (%meat). The model developed was a multitrait animal model. Fixed effects were sex, contemporary groups and heterosis, modelled as regression on heterozygosity. A clustering algorithm created contemporary groups containing at least three animals measured at the same location in an interval of maximum 75 days. Random effects were additive genetic, permanent environment and residual. Additive genetic and permanent environment effects were modelled by random regressions using linear splines with three knots at 175, 200 and 250 days. Variance components were estimated by restricted maximum likelihood (REML) on random samples of the dataset and then confirmed by a Gibbs sampling algorithm on the total dataset. Fit of the models was tested by computing residuals from a BLUP (Best Linear Unbiased Prediction) evaluation. The model that could explain the greatest proportion of the variation in each trait and thus with the smallest residuals was selected. The t-test of Student was used to test whether the means of residual distributions were significantly different from zero.Results Estimated heritabilities for BF and %meat were high and had a tendency to increase with age. Estimated heritability from 150 to 300 days increased from 0.56 to 0.75 for BF and from 0.55 to 0.69 for %meat. Genetic correlation was high between BF and %meat and varied from -0.90 to -0.93 between 150 and 300 days. Figure 1 shows the evolution of mean residuals of each trait with age between 150 and 300 days. According to this figure, mean residual is close to zero for both traits at any age. The means of residual distributions...