Conserved Core and Species-Specific Signatures in the Milk Exosomal microRNA Targetome: A Preliminary Comparative In Silico Analysis of Human, Cow, Goat and Donkey Milk

Milk-derived extracellular vesicles (EVs) transport microRNAs (miRNAs) that are unusually stable and have been proposed to survive digestion and modulate gene expression in the consumer, although their dietary bioavailability and physiological relevance remain debated. How the predicted regulatory potential of these miRNAs differs among the milks of different animals most relevant to human nutrition has not been systematically compared. Here, we performed an integrative in silico analysis of publicly available small-RNA sequencing data from 29 milk and milk-cell samples of human, cow, goat, and donkey origin. miRNAs were quantified against human (hsa) miRBase references—thereby restricting the analysis to evolutionarily conserved miRNAs with human orthologs—and their predicted effect on the human transcriptome was modeled by integrating predicted (mirDIP database) and experimentally supported (TarBase v9 database) miRNA–target interactions into a per-gene, per-species weighted targeting score. Because miRNAs act predominantly as repressors, this score is read as a prediction of which genes would be post-transcriptionally down-regulated in a recipient. miR-148a-3p dominated the exosomal spectrum of all four species (human, cow, goat, and donkey; ≈21.5% of pooled abundance), and the twenty most abundant miRNAs accounted for roughly three quarters of the signal. Of 4577 robustly targeted genes, a 1809-gene conserved “pan-milk” core showed the highest cross-species targeting and was enriched for transcriptional regulation, PI3K–Akt, MAPK, and TGF-β/SMAD signaling, autophagy and—strikingly—the components of the RNA-interference machinery itself. Species-restricted gene sets recapitulated biologically plausible programs, including a human-biased neuronal/axon-guidance and chromatin module, a donkey-biased transcriptional, epithelial, and immune (CD47) module, and a ruminant lipid/cholesterol and insulin–mTOR module. Across categories, we observed a reproducible confidence–exclusivity trade-off. We emphasize that these results are computational predictions that assume dietary miRNA uptake and do not constitute experimental validation. We provide the complete targetome as a hypothesis-generating resource to prioritize candidate genes, pathways, and milk types for future functional, nutritional, and epigenetic investigation. Across the 29 samples from the four species, miRNA composition segregated by species (silhouette width = 0.82, a cluster-separation measure ranging from −1 to 1, with values near 1 indicating well-separated groups) and the category structure exceeded a permutation null, indicating that the between-species signal is robust to differences in dataset origin and milk state.