Characterization of mycobacteria species by HPLC and pattern recognition

Multivariate methods of analysis are used to characterize 36 mycobacteria species following high-performance liquid chromatography (HPLC) of extracts of mycolic acids. After growth of the bacterial cultures, cell wall mycolic acids are saponified, extracted, then derivatized to the p-bromophenacyl esters. Reversed-phase HPLC with gradient elution separates 38 peaks in 10 min, plus an internal standard. A training set of 276 samples is composed and classification models are developed to predict the mycobacteria species. Classifications during the training phase are 91 and 95% accurate for single-pass and multilevel approaches, respectively. Validation is performed with a 549-sample set; successful classifications of known species are 87 and 92%, respectively. Bunching the predictions of MAIS complex species into one category improves the latter result to 97% correctly classified species. Fourteen of the 27 species (including M. tuberculosis) are classified without error in the validation data set.