Genomic signatures associated with antibiotic resistance and predictive modelling in Pseudomonas aeruginosa.
Abstract
Pseudomonas aeruginosa is a major opportunistic pathogen with a remarkable capacity to develop resistance to multiple antibiotics. Although numerous resistance determinants have been characterized, broader genomic features associated with antibiotic resistance remain incompletely understood. In this study, more than 3 million CDSs were extracted from 7,801 P. aeruginosa genomes and clustered into a non-redundant CDS set comprising 87,744 centroid CDSs. Functional characterization of the CDSs was conducted using the Comprehensive Antibiotic Resistance Database, the Virulence Factor Database and KEGG annotations, and genome-wide enrichment analyses were performed to identify CDSs and functions associated with antibiotic resistance. Resistant strains generally carried more resistance- and virulence-associated CDSs than susceptible strains. Several CDSs encoding efflux pumps and antibiotic-modifying enzymes, e.g. aad genes, tet(G) and cmlA9, were significantly enriched in resistant strains. Pathway enrichment analyses revealed antibiotic-specific functional signatures, including porphyrin metabolism, sulphur metabolism and folate-related pathways among aminoglycoside-resistant strains, and homologous recombination, DNA replication and mismatch repair pathways among levofloxacin-resistant strains. However, polysaccharide production was enriched among CDSs nominally significantly associated with susceptible strains. Random forest models demonstrated moderate to excellent predictive performance with external validation confirming robustness. Co-occurrence network analyses further identified associations among virulence genes, e.g. between lipopolysaccharide biosynthesis- and O-antigen-associated genes, as well as associations between resistance determinants and virulence-associated genes, such as β-lactam resistance and iron acquisition genes. These findings revealed antibiotic-specific gene and pathway signatures, enabled the prediction of antibiotic resistance using genomic features and revealed potential genomic linkages between canonical resistance determinants and other resistance-associated features.