RESUMEN
AIMS: Attenuated Total Reflection Fourier Transform Infrared (ATR-FT-IR) Spectroscopy and chemometric modelling, including soft independent modelling by class analogy (SIMCA), partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM), were applied to attempt to discriminate 60 clinical isolates of Enterococcus faecium and Enterococcus faecalis and hence evaluate the performance of the spectroscopic approach in identifying enterococci infections. METHODS AND RESULTS: The bacterial samples were identified by polymerize chain reaction (PCR) amplification and their ATR-FT-IR spectra acquired. Spectra were processed to the second derivative using the Savitzky-Golay algorithm and normalized using extended multiplicative signal correction employing the UnscramblerX (CAMO, Norway) software package. Multivariate classification models and their performance were evaluated using Cohen's Kappa coefficient. Principal component analysis (PCA) score plots showed separate clusters of spectra related to membership to E. faecium and E. faecalis, with this explained by bands assigned to PO2 (1230 cm-1 ), P-O-C (1114 cm-1 ), monosubstituted alkene (997, 987 cm-1 ) and C-O (1070, 1055, 1036 cm-1 ) corresponding to teichoic acids, polysaccharides and peptidoglycan from the cell wall in PCA and PLS-DA loading plots. The best classification model for E. faecium and E. faecalis is SVM, indicating via highest Kappa score. The classification coefficient between SIMCA, PLS-DA, SVM and PCR as reference method were 0·59, 0·9 and 1, respectively, shown as the Kappa scores. CONCLUSIONS: The main spectral differences observed between the two clinically relevant enterococci species were associated with changes in the teichoic acid content of cell walls. With regard to the binary classification method, SVM was found to be the best performing classification model, providing the highest correlation with the PCR results. SIGNIFICANCE AND IMPACT OF THE STUDY: The study shows that ATR-FT-IR spectroscopy in combination with chemometric modelling can be applied for the phenotypic identification and discrimination of clinically relevant and similar enterococcal species.