From genotype to antibiotic susceptibility phenotype in Enterobacteriaceae: a clinical perspective

Abstract

Predicting the antibiotic susceptibility phenotype from genomic data is challenging, especially for some specific antibiotics in Enterobacteriaceae. Here we aimed to assess the performance of whole genomic sequencing (WGS) for predicting the antibiotic susceptibility in various Enterobacteriaceae species using the detection of antibiotic resistance genes (ARGs), specific mutations and a knowledge-based decision algorithm. We sequenced (Illumina MiSeq 2x250b) 187 clinical isolates from species possessing (n=98) or not (n=89) an intrinsic AmpC-type cephalosporinase. Antibiotic susceptibility was performed by the disc diffusion method. Reads were assembled by A5-miseq and ARGs were identified from the ResFinder database using Diamond. Mutations on GyrA and ParC topoisomerases were studied. We assessed the prediction rates for amoxicillin, co-amoxiclav, piperacillin, piperacillin-tazobactam, ceftazidime, cefepime, meropenem, amikacin, gentamicin and fluoroquinolones. A total of 1,870 isolate/antibiotic combinations were considered. In 295 cases (15.8%) no attempt of prediction was made. Among the 1,575 attempts, 1,537 (97.6%) were correct (1,011 for predicting susceptibility and 526 for predicting resistance), 15 (0.9%) were major errors (MEs) and 23 (1.5%) were very major errors (VMEs). The concordance rates were similar between non-AmpC and AmpC-producing Enterobacteriaceae (907/935 [97.0%] vs 630/640 [98.4%], Chi2 test p=0.07), but more VMEs were observed in non-AmpC producing strains than in those producing an AmpC (20/935 [2.1%] vs 3/640 [0.5%], Chi2 test p=0.007). The majority of VMEs were putatively due to the overexpression of chromosomal genes. In conclusion, the inference of antibiotic susceptibility from genomic data showed good performances for non-AmpC and AmpC-producing Enterobacteriaceae species.