Background. Mycobacterium abscessus complex (MABSC) is one of the most frequently nontuberculous mycobacteria isolated in clinical practice, especially in immunocompromised patients. The subspecies of MABSC (M. abscessus ssp. abscessus (MAA), ssp. massiliense (MAM), and ssp. bolletii (MAB)), are associated with specific antimicrobial susceptibility profiles and different clinical outcomes. The discrimination at subspecies level is crucial for the proper patient therapeutic management, but so far it can only be achieved by genotypic techniques. In this study, we explore the potential of FT-IR spectroscopy to differentiate MABSC subspecies.

Methods. 41 patient-derived and culture collections MABSC isolates (n=18 MAA, n=13 MAM, and n=10 MAB) were analyzed by the FT-IR spectroscopy-based IR Biotyper® system (IRBT, Bruker Daltonics, Bremen). Spectra were acquired from 3 independent biological replicates of strains grown on Löwenstein-Jensen medium, incubated for 72 h at 35±2 °C. The resolution power using different regions of the IR spectrum was investigated. Predictive models were created using PCA (principal component analysis) and LDA (linear discriminant analysis), using random 67% of the acquired spectra as training set, and the remaining 33% as testing set.

Results. The IRBT analysis showed that the wavenumber range which allowed the best resolution power is the one corresponding to the absorption of carbohydrates and lipids (1300-800 cm-1, 1500-1400 cm-1 and 3000-2800 cm-1). The LDA predictive models allowed a clear differentiation of MAM from MAA and MAB, while the latter two are separable but more similar to each other. It was also observed that within each MABSC subspecies, the colony"s phenotype (rough or smooth) is responsible for a relevant spectral variance. Therefore, this feature has to be considered when defining the classes to differentiate.

Conclusions.IR Biotyper could have a future role in rapidly discriminating between MABSC subspecies. Further studies with a much larger number of samples are needed to assess the real impact of this new technology application, and to develop prediction models which are robust and generalizable to all datasets.