Genomic prediction of methane emissions in Danish Holstein using single step and multi-trait prediction models

Abstract

Enteric methane emissions from ruminants are a major contributor to global greenhouse gas emissions and pose a significant challenge to the sustainability of livestock production. To mitigate these emissions, breeding strategies have been mentioned as a promising tool, but prediction accuracies of methane emission traits are still limited by the size of datasets with records. Hence, using methane concentrations (MeC) in Danish Holstein cows as target trait, this study evaluated the predictive performance of pedigree-based BLUP (pBLUP) and single-step genomic BLUP (ssGBLUP) in univariate and multi-trait models, the latter including milk production traits. Previously, both ssGBLUP as well as multi-trait models have been shown to enhance prediction accuracies. The dataset included 1,744 primiparous (PP) and 2,989 multiparous (MP) cows from 15 Danish dairy farms, with over 600,000 daily records of MeC, fat yield (FY), and energy-corrected milk yield (ECM). Methane concentrations were measured using sniffers, and milk production data was acquired from milking robots and national milk recording data. At first, a pedigree-based variance component estimation revealed heritabilities between 0.17 (SE=0.03) for MeC in PP and MP cows to 0.38 (SE=0.06) for ECM in PP cows. Similarly, repeatabilities ranged from 0.32 (MeC, SE=0.01) to 0.81 (ECM, SE=0.01). Genetic correlations between MeC and production traits were positive but unfavorable, i.e., in a range from 0.15 (SE=0.13) between MeC and ECM in PP cows to 0.41 (SE=0.09) between MeC and ECM in MP cows, indicating a genetic antagonism between reducing emissions and maintaining milk yield. Prediction accuracies were generally higher for ssGBLUP compared to pBLUP models (up to 61.90% increase), and for MP cows compared to PP cows. Multi-trait models outperformed univariate models, particularly when phenotypic data for FY and ECM were available in both the reference and validation populations. The highest accuracy for MeC prediction in PP cows was 0.38 (ssGBLUP), while MP cows reached up to 0.51, both for the multi-trait model including both, ECM and FY. While incorporating FY and ECM improved MeC prediction, the unfavorable genetic correlations highlight the risk of compromising milk production when selecting for reduced emissions. Therefore, future breeding strategies should aim to expand methane phenotyping, develop methane traits independent of milk production, and implement multi-trait selection indices that balance environmental and economic goals. This study demonstrates the potential of multi-trait genomic prediction to enhance the genetic evaluation of methane emissions and supports its integration into sustainable dairy cattle breeding programs.