Hierarchical clustering as a rapid tool for surveillance of emerging antibiotic-resistance phenotypes in Klebsiella pneumoniae strains

Antimicrobial resistance is on the rise, and its early detection and surveillance are critical to implement effective control measures. The aim of this study was to develop a rapid hierarchical clustering bioinformatic tool for application on antibiotic susceptibility testing (AST) results (resistant, intermediate, sensitive) of a series of Klebsiella pneumoniae clinical isolates from Algeria and from France for surveillance of antibiotic-resistance phenotypes. A total of 1011 K. pneumoniae strains were collected from August 2008 to December 2012: 221 clinical isolates from western Algeria and 790 clinical isolates from Marseille, France. AST against a panel of 16 antibiotics was done for all isolates. Results of AST were introduced into MultiExperiment Viewer (MeV) software to perform hierarchical clustering, with resistant, intermediate and sensitive being translated to 1, 0 and −1 values, respectively. Hierarchical clustering results were compared to standard resistance phenotypes to evaluate the accuracy of the method. Based on the AST results, the 221 K. pneumoniae strains from Algeria could be separated into six phenotype groups as regards their resistance to β-lactam compounds: extended spectrum β-lactamase (ESBL) (68.3 %), ESBL associated with cephalosporinase (13.1 %), cephalosporinase (0.9 %), penicillinase (3.6 %) and wild-type (14.0 %). Hierarchical clustering by the MeV software applied to the AST results for all 1011 isolates generated clusters that were significantly representative of phenotypic classification and geographical origin, in less than 1 min. Moreover, adding to the dataset the AST results of a K. pneumoniae NDM-1 positive strain, the only strain resistant to imipenem in the series, immediately generated a new branch in the dendrogram. We have developed a rapid and simple hierarchical clustering tool for application on AST results that was able to survey qualitatively and quantitatively the prevalence of known and unknown phenotypes. This tool could be easily implemented in routine clinical microbiology laboratories.