Better Multi-Breed Genomic Predictions for Tropical Bull Fertility Using a Breed-Adjusted Genomic Relationship Matrix

The pressing requirement for agricultural systems to adopt climate change adaptation strategies specifically designed for tropical environments underscores the significance of implementing sustainable bull breeding practices in beef cattle operations. We consider alternative approaches to generation of single- and multi-breed genomic predictions for a population of Brahman (N = 1051), Santa Gertrudis (N = 929) and UltraBlack (N = 844) bulls with genotypes at high-density and phenotypes for scrotal circumference (SCC; mean ± SD = 31.82 ± 4.32 cm), sheath score (SHS; 3.70 ± 1.98) and percent normal sperm (PNS; 63.34% ± 28.22%). We examined five genomic prediction models: three single breed and two multi-breed. The later contained a multi-breed genomic relationship matrix computed without (GRM_u) or adjusting (GRM_a) for breed-specific allele frequencies. Bias, dispersion and accuracy of the genomic predictions across the five models was computed based on cross-validation and using the LR method. The elements of the multi-breed GRM_u revealed anomalies including a multi-modal distribution of diagonal and off-diagonal elements with all diagonal values above one (range: 1.022 to 1.524) and averaging 1.163. Instead, GRM_a values were consistent with expectations: diagonals with a single mass around one (range: 0.844 to 1.391) and off-diagonal values with a single mass around zero. Estimates of heritability (h² ± SE) with the GRM_u model were 0.501 ± 0.037, 0.458 ± 0.037 and 0.362 ± 0.030, for SCC, SHS and PNS, respectively; while h² estimates using GRM_a were comparable for SCC (0.434 ± 0.037), and SHS (0.439 ± 0.038), and possibly lower for PNS (0.226 ± 0.032). Estimates of genetic correlation (r_g) were similar for both models, except for the r_g between SHS and PNS which moved from -0.095 ± 0.119 using GRM_u to -0.298 ± 0.113 using GRM_a. For all traits, the correlation between GEBV from GRM_u and GRM_a was > 0.90, indicating similar ranking of animals regardless of the relationship matrix used. However, a random cross-validation scheme showed that using GRM_a increased GEBV accuracies from 0.550 to 0.571 (or 3.7%) for SCC, from 0.496 to 0.509 (or 2.6%) for SHS and from 0.335 to 0.423 (or 26.2%) for PNS. Multi-breed genomic predictions for tropical bull fertility are feasible alternative to individual single-breed evaluation systems. Furthermore, the computational effort required to adjust the genomic relationship matrix for breed-specific allele frequencies is justified by the significant improvement in prediction accuracy, ultimately enhancing the selection of bulls in tropical environments. Collectively these results enable very early in-life selection for bull fertility traits, supporting genetic improvement strategies currently taking place within tropical beef production systems in northern Australia.