Integrating rumen microbiome profiles to predict feed efficiency and methane emissions in Hereford beef cattle

Abstract

Microbiome plays a key role in traits like feed efficiency and methane emission production. In recent years, several studies have investigated the relationship between the host genome, microbiome profiles, and complex traits to improve the profitability and sustainability of the industry. In this study, we aimed to evaluate the predictive ability of models for individual feed intake, efficiency, and methane emissions using genomic and microbiome information.

A dataset of 537 Hereford bulls and steers with dry matter intake (DMI), body weight (BW), and residual feed Intake (RFI) information, recorded in 70-day post-weaning trials, was analyzed. A subset of 123 animals had daily methane production (DMP) measured during the same trials. All animals were genotyped using 87k SNP panels, and the rumen microbiota was sequenced via enzyme restriction-reduced representation sequencing. We evaluated the predictive ability of models incorporating the host genome (G), microbiome (M), or both (GM) for RFI, DMI, BW, DMP (g/d), and methane yield (MY) traits using Bayesian regression models. To assess the predictive ability of the models, we used a 10-fold cross-validation with five replicates.

Correlation coefficients for G and GM models showed similar or slightly improved predictive ability for DMI (0.72 – 0.73), BW (0.48 – 0.52), DMP (0.87 – 0.87) and MY (0.68 – 0.69). However, the predictive ability for RFI was lower with GM model (0.15 – 0.14). Increasing model complexity resulted in little change in prediction error, with a slight decrease observed for DMI and BW, but an increase for methane-related traits, likely due to the smaller sample size. Our results suggest that while predictive ability for BW improved with microbial data, its inclusion did not improve the prediction of the other traits. This could be due to the indirect genetic effects mediated by the rumen microbiome, which should be evaluated using larger datasets.