Holobiont effect accounts for more methane emission variance than the additive and microbiome effects on dairy cattle

Abstract

Rumen microbiota has been previously related to phenotypic complex traits of relevance in dairy cattle. The joint analysis of the host's genetic background and its microbiota can be statistically modelled using similarity matrices between microorganism communities in the different hosts. Microbiota relationship matrices (K) enable considering the whole microbiota and the cumbersome interrelations between taxa, rather than analyzing single taxa one at the time. Several methods have been proposed to ordinate these matrices. The aim of this study was to compare the performance of twelve K built from different microbiome distance metrics, within a variance component estimation framework for methane concentration in dairy cattle. Phenotypic, genomic and rumen microbiome information from simulations (n = 1000) and real data (cows = 437) were analyzed. Four models were considered: an additive genomic model (GBLUP), a microbiome model (MBLUP), a genetic and microbiome effects model (HBLUP) and a genetic, microbiome and genetic × microbiome interaction effects model (HiBLUP). Results from simulation were obtained from 25 replicates. Results from simulated data suggested that Ks with flattened off-diagonal elements were more accurate in variance components estimation for all compared models that included Ks information (MBLUP, HBLUP and HiBLUP). Multidimensional scaling (MDS), redundancy analysis (RDA) and constrained correspondence analysis (CCA) performed better in simulation to estimate heritability and microbiability. The models including Ks from the MDS, RDA and CCA methods were also between the most plausible models in the real data set, according to the deviance information criteria (DIC). Real data was analyzed under the same framework as in the simulation. The most plausible model in real data was HiBLUP. Estimates variated depending on K; methane heritability (0.15–0.17) and microbiability (0.15–0.21) were lower than the proportion of the phenotypic variance attributable to the host-microbiome holobiont effect (0.42–0.59), which we have defined here as “holobiability”. The holobiability including the genomic × microbiome interaction from the HiBLUP was between 0.01 and 0.15 larger than the holobiability explained from the sum of the genetic and microbiome effects without interaction between them, from the HBLUP, depending on K. The findings in this study support the potential of the joint analysis of genomic and microbiome information. Accounting for the hologenome effect (genomic and microbiome) could improve the accuracy in variance component estimation of complex traits relevant in livestock science.