Genomic epidemiology, antimicrobial resistance and risk-tier stewardship of dairy Staphylococcus aureus from bulk-tank milk in subtropical China.

Bovine mastitis caused by Staphylococcus aureus increases antimicrobial use and may compromise milk quality, yet genomic evidence from subtropical China remains limited. This study aimed to characterise the genomic epidemiology, antimicrobial non-susceptibility and functional risk tiers of dairy-associated S. aureus in northern Fujian, China. We analysed 59 S. aureus isolates recovered from 162 bulk-tank milk samples collected from 14 farms during 2023-2024, using whole-genome sequencing, MLST/spa typing and core-genome SNP phylogenetics, together with disc diffusion susceptibility testing (CLSI M100) and annotation of resistance, virulence and plasmid replicons (CARD, VFDB and PlasmidFinder). An unsupervised multivariate workflow (principal component analysis followed by K-means clustering) integrated genomic and phenotypic features to assign functional-risk tiers. Seven MLST and eight spa types were detected, dominated by ST1-t114/t127 (40.7%) and ST9784-t189 (20.3%). Livestock-associated MRSA ST59-t437 carrying SCCmec IV and IEC-D was detected in raw milk from this region for the first time. Overall, 45.8% of isolates were non-susceptible to at least one antimicrobial, most commonly penicillin and ampicillin (28.8% each). ST9784-t189 more frequently carried lnu(A) and pLNU4-like plasmid sequences and showed higher clindamycin non-susceptibility than other lineages. PCA/K-means grouped isolates into low-, medium- and high-risk tiers (76.3%, 15.3% and 8.4%, respectively); medium/high tiers showed higher multidrug non-susceptibility and a greater enterotoxin gene burden and were more often associated with plasmid replicons including pKH1 and pAMalpha1. This single-province pilot study provides a genomic baseline for dairy-associated S. aureus in subtropical China and suggests that unsupervised multivariate grouping may support risk-aware surveillance and milk-quality assurance; larger multi-region datasets are needed to assess generalisability.