Improving imputation accuracy across tropically adapted beef cattle: an application for Brahman and Nellore using whole-genome sequencing data.

Combining information of different breeds is a cost-effective strategy to increase the size and genetic diversity of reference populations, which would improve imputation and/or genomic prediction accuracies in comparison with single-breed evaluations. Here, we have evaluated the impact of combining sequence information from two of the most relevant tropically adapted beef cattle breeds (Brahman and Nellore) on imputation accuracies to the sequence level. Whole-genome sequencing data of 279 (128 Brahman and 151 Nellore) animals were used in this study. Animals were chosen based on their contribution to the respective breed, attempting to reach high imputation accuracies by maximizing the genetic variability captured in the sequencing. Ten well-designed imputation scenarios from high-density single-nucleotide polymorphism (SNP) panel (∼777 K) to whole-genome sequence, implemented using the software FImpute3, were used to study different strategies to combine Brahman and Nellore sequencing data for imputation purposes. Animal and SNP imputation accuracies were assessed by the squared correlation between observed and imputed genotypes. The analysis of the genetic structure of the sequenced animals showed that Nellore and Brahman are genetically distinct cattle breeds with similar patterns of linkage disequilibrium. Compared to single-breed evaluations, the average imputation accuracy per animal improved from 0.89 to 0.91 in Brahman and from 0.94 to 0.96 in Nellore by utilizing a multibreed model. The overall average SNP-wise imputation accuracies were also improved (from 0.78 to 0.82 in Brahman and from 0.86 to 0.92 in Nellore) by combining sequence data from Nellore and Brahman, including a considerably better imputation for the known hard-to-impute genomic regions on chromosomes 5, 10, 12, 15, and 23. This study showed that higher accuracy of imputation to whole-genome sequencing can be achieved for both Brahman and Nellore using multibreed models in comparison to the standard single-breed evaluations, especially when restricting the analysis to a reference panel that is segregating in both breeds.