Genome-based discovery of trait networks in dairy cattle

Structural equation modeling (SEM) is a powerful tool for modeling phenotypic networks. We propose SEM-GWAS as a complementary approach to multi-trait GWAS (MTM-GWAS) methods for inferring networks among traits, making possible to partition single nucleotide polymorphisms (SNP) effects on each trait into direct and indirect. Our specific aim was to illustrate the application of SEM-GWAS in dairy cattle by using phenotypic traits related to udder health, i.e., milk yield (MY), somatic cell score (SCS), lactose (%, LACT), pH and casein (CN, % of total milk N), in a cohort of 1,158 Italian Brown Swiss cows, as a case example. Animals were genotyped with the Illumina BovineSNP50 Bead Chip v.2. A Bayesian multi-trait genomic best linear unbiased prediction model was fitted to the five traits to obtain posterior means of model residuals, used as input for inferring putative causal networks and directions among traits via the Hill-Climbing algorithm. The inferred structure led to a set of structural equations, which were used to estimate SEM parameters. We found negative path coefficients for MY→LACT and LACT→SCS, and positive for SCS→pH and LACT→CN. Joint use of MTM-GWAS and SEM-GWAS enabled to identify six significant SNPs for SCS, and their effects were decomposed into direct and indirect, i.e., mediated by the up-stream phenotypes in the network (MY and LACT). Pathway enrichment analyses confirmed an overrepresentation (false discovery rate < 0.05) of pathways consistent with the traits biology (e.g., organic anion transmembrane transporter activity for pH and MY or Wnt signaling for SCS). In summary, SEM-GWAS may offer new insights on relationships among phenotypes and on the path of SNP effects, producing useful information for selective breeding of cows and for management decisions