by Anna Heydecke, Hong Yin, Eva Tano, Susanne Sütterlin
Prediction of antibiotic resistance from whole genome sequence (WGS) data has been proposed. However, the performance of WGS data analysis for this matter may be influenced by the resistance mechanism’s biology. This study compared traditional antimicrobial susceptibility testing with whole genome sequencing for identification of extended-spectrum beta-lactamases (ESBL) in a collection of 419 Escherichia coli isolates. BLASTn-based prediction and read mapping with srst2 gave matching results, and in 381/419 (91%) isolates WGS was congruent with phenotypic testing. Incongruent results were grouped by potential explanations into biological-related and sequence analysis-related results. Biological-related explanations included weak ESBL-enzyme activity (n = 4), inconclusive phenotypic ESBL-testing (n = 4), potential loss of plasmid during subculturing (n = 7), and other resistance mechanisms than ESBL-enzymes (n = 2). Sequence analysis-related explanations were cut-off dependency for read depth (n = 5), too stringent (n = 3) and too loose cut-off for nucleotide identity and coverage (n = 13), respectively. The results reveal limitations of both traditional antibiotic susceptibility testing and sequence-based resistance prediction and highlight the need for evidence-based standards in sequence analysis.