Genomic evaluation in nellore cattle for reproductive traits: multiple ways to account for missing pedigrees

Abstract

Missing pedigrees are a common problem in most populations. Animals with unknown ancestors are usually treated as found- ers; however, this can underestimate inbreeding, not properly account for different base populations, and bias breeding values. We aimed to assess the use of unknown parent groups (UPG) or metafounders (MF) to model missing pedigrees in a beef cattle population. Phenotypic and genotypic data from the Nellore improvement programme of the Brazilian Breeders and Researchers Association were used. The pedigree contained 3.8M animals born between 1970 and 2022, of which 51,752 were genotyped. Records for scrotal circumference at 365days old (SC365, N=239,806), age at first calving (AFC, N=560,785) and accumulated cow productivity (ACP, N=269,330) were used. Four models were implemented: single-step GBLUP without explicitly dealing with missing pedigree (G0), with UPG (G1), with MF (G2) and with G accounting for group-specific allele frequencies (G3). UPG and MF were assigned based on commercial and registered herds (S1), uncertain paternity (S2) and patriarchs (S3). The accuracy and bias of predictions were assessed using the linear regression (LR) method. Linear, single-trait animal models were used for SC365 and AFC, and multi-trait for ACP. Heritability estimates ranged from 0.07 to 0.40. Compared to G0, accuracy was slightly higher in G2S2 and G2S3 (0.70 vs. 0.71) for SC365, G2S3 (0.49 vs. 0.51) for AFC, G1S2 for ACP (0.67 vs. 0.71). Bias was small in all the scenarios (≤0.06 SD), except of ACP that presented a great bias, including MF. Overall, G1 and G2 had similar accuracy, possibly because of the limited number of genotyped animals linked to MF. Centring the genomic relationship matrix by patriarchs' allelic frequencies resulted in similar accuracy and bias to the MF models. Replicating the study with a larger database containing more genotyped animals connected to MF could help improve the MF estimates, and thus, prediction accuracy and bias.