Clinically undetected polyclonal heteroresistance among Pseudomonas aeruginosa isolated from cystic fibrosis respiratory specimens

Background: Pseudomonas aeruginosa infection is the leading cause of death among patients with cystic fibrosis (CF) and a common cause of difficult-to-treat hospital-acquired infections. P. aeruginosa uses several mechanisms to resist different antibiotic classes and an individual CF patient can harbour multiple resistance phenotypes. Objectives: To determine the rates and distribution of polyclonal heteroresistance (PHR) in P. aeruginosa by random, prospective evaluation of respiratory cultures from CF patients at a large referral centre over a 1 year period. Methods: We obtained 28 unique sputum samples from 19 CF patients and took multiple isolates from each, even when morphologically similar, yielding 280 unique isolates. We performed antimicrobial susceptibility testing (AST) on all isolates and calculated PHR on the basis of variability in AST in a given sample. We then performed whole-genome sequencing on 134 isolates and used a machine-learning association model to interrogate phenotypic PHR from genomic data. Results: PHR was identified in most sampled patients (n = 15/19; 79%). Importantly, resistant phenotypes were not detected by routine AST in 26% of patients (n = 5/19). The machine-learning model, using the extended sampling, identified at least one genetic variant associated with phenotypic resistance in 94.3% of isolates (n = 1392/1476). Conclusion: PHR is common among P. aeruginosa in the CF lung. While traditional microbiological methods often fail to detect resistant subpopulations, extended sampling of isolates and conventional AST identified PHR in most patients. A machine-learning tool successfully identified at least one resistance variant in almost all resistant isolates by leveraging this extended sampling and conventional AST.