Abstract
Breed characteristics are shaped by complex biological mechanisms which, when dissected at the molecular level, can lead to the identification of novel phenotypes useful for the development of phenomics-guided breeding strategies. Building on this, we investigated differences at the metabolome level between the two main maternal lines used in Italian heavy pig crossbreeding schemes: Italian Landrace (ILA) and Italian Large White (ILW), given their relevance in the production of protected designation of origin (PDO) products. We conducted untargeted plasma metabolomics on 205 pigs (53 ILA and 152 ILW), generating high-quality profiles of 576 metabolites. Differentially abundant metabolites were identified using Boruta, a random forest-based feature selection algorithm integrated into a computational pipeline designed to reduce stochastic variability. We also leveraged the unbalanced breed sample sizes to evaluate the influence of population composition and sampling design on metabolite selection. Specifically, ILW animals were either downsampled into three balanced but distinct subsets or included in unbalanced datasets of increasing size. Across all analyses, 59 metabolites differentiated the two breeds. Approximately, 30% were repeatedly identified across all three balanced subsets, and 60–70% appeared in at least two. Inclusion of larger, unbalanced datasets increased the number of selected metabolites by ∼25%, indicating improved sensitivity with larger sample sizes while balanced designs enhanced robustness. Biologically, ILW pigs showed elevated gamma-glutamyl amino acids, indicating enhanced glutathione-cycle activity, while ILA pigs exhibited higher hexosylceramides, consistent with greater fat deposition. These findings highlight breed-specific metabolic architectures and support the application of metabolomics in precision breeding for PDO-oriented systems.