Combining 16S rRNA sequencing data and genomic markers for genomic evaluation of methane in Holstein dairy cattle

Ruminant animals produce methane as a by-product of microbial fermentation in the rumen, with cattle being the largest contributors among ruminants. Methane emissions are influenced directly by the host’s genetics and indirectly through host-mediated effects on rumen microbiome composition. In this study, we investigated whether modeling both direct and indirect genetic effects could improve the genomic prediction of breeding values for methane emissions. We used genotype, methane phenotype, and 16S rRNA gene amplicon sequencing data (16S), including microbial taxa (16s-MT), predicted microbial genes (16s-PMG), and predicted metabolic pathways (16s-PMP) from Danish Holstein cows. Genomic best linear unbiased prediction (GBLUP) method was used for benchmarking a method (GMBLUP) combining results of a set of three linear models for the prediction of breeding values. In GMBLUP, the first model regresses methane emissions on three random animal effects (archaeal 16S, bacterial 16S, and polygenic effects) using corresponding microbiome and genomic relationship matrices. The second and third models separately regress archaeal and bacterial 16S on random animal effects defined by the genomic relationship matrix. We evaluated prediction accuracy using five-fold cross-validation. The GBLUP model achieved accuracy ranging from 0.12 to 0.21 across datasets, while GMBLUP showed improved accuracy, particularly for the 16S-PMP dataset (0.56). Using the top 10% most heritable archaeal and bacterial features improved prediction for the 16S-MT dataset (from 0.44 to 0.51), while reducing accuracy for 16S-PMP (from 0.56 to 0.40). These results highlight that incorporating microbiome information into genomic prediction models can enhance accuracy and support the development of breeding strategies aimed at reducing methane emissions.