Using genomic data to estimate genetic correlations between countries with different levels of connectedness

Abstract

Genetic correlations (r_g) between countries are required for international evaluations. The estimation of those r_g is challenging or even unfeasible using only pedigree and phenotypes when poor connectedness between countries is structural in the data due to a limited number of bulls having recorded (grand-)offspring across countries. Genomic information could be used to estimate r_g between countries by capturing connectedness that is not traced by pedigree recordings. Indeed, populations that appear as (completely) disconnected through pedigree can, theoretically, be connected through genomic data. Thus, our study aimed to investigate if estimates of r_g between countries based on genomic information are more accurate compared to estimates based on pedigree data, considering different levels of genetic connectedness. A maternally affected trait mimicking weaning weight was simulated for two beef cattle populations of the same breed. Different levels of connectedness between populations were simulated by exchanging different proportions of top sires in the last five generations: 0% (completely disconnected), 2.5% (lowly connected), 5% (medium), and 20% (high). Genomic data in the form of individual SNP genotypes at medium density were stored in the last three generations and used only for the estimation process. r_g between populations were estimated using three different relationship matrices: i) a pedigree-based relationship matrix (A) including all phenotyped animals; ii) a genomic relationship matrix (G) including phenotyped and genotyped animals only from the last three generations; and iii) a combined pedigree and genomic relationship matrix (H) including all phenotyped and genotyped animals. With disconnected and lowly connected populations, estimates of direct and maternal r_g were, on average, close to the simulated values when using genomic data through G or H. With lowly connected populations, estimates of direct r_g were close to the simulated values when using A, but estimates of maternal r_g showed large variation. With more connected populations, estimates obtained with A, G, and H matrices were overall similar. For all scenarios, when using genomic data in the estimation process, estimates of r_g had smaller standard errors. Our results show that genomic data can help the estimation of r_g between countries and especially reduce their standard errors for populations that appear as completely disconnected or lowly connected through pedigree information, such as in beef and (small) dairy cattle populations.