Metabolomic profiles of divergent pigs for feed intake: towards the genetic dissection of a complex trait using molecular phenotypes

The pig production sector is constantly searching for new phenotypic traits, to expand the traditional phenotype landscape, aiming to advance selection programs and develop innovative breeding strategies. Traditional phenotypes, such as production and performance traits, are influenced by complex molecular systems making it challenging to identify their genetic determinants. By utilizing the metabolome, defined as the full set of circulating small molecules, as intermediate phenotypes, we can bridge the gap between end phenotypes and the genetic background of the animals. Feed intake, a proxy for feed efficiency and growth rate, was the focus of this study in a Duroc pig line. We applied untargeted plasma metabolomics to analyze 290 pigs, forming two groups of animals with extreme and divergent values for feed intake (150 and 140 for high and low values, respectively), selected from a larger cohort. Initially, the genomic heritability was obtained for approximately 700 metabolites. Metabolomic profiles of the two groups of pigs were compared using a statistically robust machine learning approach based on over 1000 independent iterations of random forest models including cross-validation. A total of 29 metabolites were found to be differentially abundant (i.e. selected in > 1000 random forest runs with P < 0.05) between the two groups of pigs with high and low feed intake values, with most being amino acids (19) and the rest evenly distributed among nucleotide metabolites (5), lipids (3), carbohydrates (1), and unnamed metabolites (1). PCA showed no separation between the two groups when using the whole metabolomic profile, while using only the 29 differentially abundant metabolites showed a more marked separation. Considering their correlation with raw feed intake values, the 29 selected metabolites were among those with the highest Pearson r in the dataset, with absolute values ranging from 0.12 to 0.40 and a mean of 0.28 ± 0.07; values in the whole metabolomic profile ranged from 0 to 0.4 with mean 0.1 ± 0.08. The same was true for the ROC AUC metric, further confirming their role in discriminating between the two extreme groups. Some of these metabolites, which also have high heritability, have already been described as being involved in the hypothalamic mechanisms controlling food intake, along with other direct and indirect associations with feed intake and growth-related traits. These findings show that proxies for feed intake can be identified in the pig metabolome and help dissect it, along with other complex production traits. Pending further investigation and more in-depth studies, these results can be used to dissect complex traits at the molecular level and better define the genetic complexity underlying their variability.