Objectives: To assess the effectiveness of bioinformatic detection of resistance genes in whole-genome sequences in correctly predicting resistance phenotypes.
Methods: Genomes of a collection of well-characterized commensal Escherichia coli were sequenced using Illumina HiSeq technology and assembled with SPAdes. Antibiotic resistance genes identified by PCR, SRST2 analysis of reads and ResFinder analysis of SPAdes assemblies were compared with known resistance phenotypes.
Results: Generally, the antibiotic resistance genes detected using bioinformatic methods were concordant, but only ARG-ANNOT included sat2. However, the presence or absence of genes was not always predictive of the phenotype. In one strain, trimethoprim resistance was due to a known mutation in the chromosomal folA gene. In cases where the copy number was low, the aadA5 gene downstream of dfrA17 did not confer streptomycin or spectinomycin resistance. Resistance genes were found in the genomes that were not detected previously by PCRs targeting a limited gene set and gene cassettes in class 1 or class 2 integrons. In one isolate, the aadA1 gene cassette in the estX-aadA1 cassettes pair was outside an integron context and was not expressed. The qnrS1 gene, conferring reduced susceptibility to fluoroquinolones, and the blaCMY-2 gene, encoding an ESBL, were each detected in a single isolate and mphA (macrolide resistance) was present in six isolates surrounded by IS26 and IS6100.
Conclusions: WGS analysis detected more genes than PCR. Some were not expressed, causing inconsistencies with the experimentally determined phenotype. An unpredicted chromosomal folA mutation causing trimethoprim resistance was found.