Inferring causal structures of gut microbiota diversity and feed efficiency traits in poultry using Bayesian learning and genomic structural equation models

Abstract

Feed and phosphorus efficiency are of increasing importance in poultry breeding. It has been shown recently that these efficiency traits are influenced by the gut microbiota composition of the birds. The efficiency traits and the gut microbiota composition are partly under control of the host genome. Thus, the gut microbiota composition can be seen as a mediator trait between the host genome and the efficiency traits. The present study used data from 749 individuals of a Japanese quail F2 cross. The birds were genotyped for 4k SNP and trait recorded for phosphorus utilization (PU) and phosphorus retention (PR), body weight gain (BWG) and feed per gain ratio (F:G). The gut microbiota composition was characterized by targeted amplicon sequencing. The alpha diversity was calculated as the Pielou’s evenness index (J’). A stable Bayesian network was established using a Hill-Climbing learning algorithm. Pielou’s evenness index was placed as the most upstream trait and BWG as the most downstream trait, with direct and indirect links via PR, PU, and F:G. The direct and indirect effects between J’, PU, and PR were quantified with structural equation models, which revealed a causal link from J’ to PU and from PU to PR. Quantitative trait loci (QTL) linkage mapping revealed three genome-wide significant QTL regions for these traits with in total 49 trait-associated SNP within the QTL regions. Structural equation model association mapping separated the total SNP effect for a trait into a direct effect and indirect effects mediated by upstream traits. Although the indirect effects were in general small, they contributed to the total SNP effect in some cases. This enabled us to detect some shared genetic effects. The method applied allows for the detection of shared genetic architecture of quantitative traits and microbiota compositions.

Mendelian randomization analysis of 34,497 German Holstein cows to infer causal associations between milk production and health traits

“…Another popular method to assess causality are structural equation models (SEM) [ 60 ] that enable the identification of causal relationships among traits and allow for the prediction of one trait by another “upstream” causal variable [ 61 ]. Recently, they have been extended to GWAS-SEM, where the total genetic effects for one trait are separated into direct and indirect genetic effects, i.e., genetic effects that are mediated by other upstream variables [ 62 – 64 ]. Thus, SEM provide detailed information about causal associations among traits.…”

Effects of wearable therapies on jump performance in sport horses

“…In consideration of the complex and interconnected nature of equine physiological and biomechanical responses, relationships between variables were explored via a Bayesian learning network (BLN). While a novel approach to this specific type of investigation in equines, Bayesian networks have been successfully applied in other specialties of animal and veterinary science, including nutrition ( 19 ), reproduction ( 20 ), lactation ( 21 ), genetics ( 22 ), and epidemiology ( 23 , 24 ). Bayesian networks possess several advantages over traditional linear models for analyzing highly variable responses that are difficult to interpret in isolation.…”