Discrimination of Escherichia coli, Shigella flexneri, and Shigella sonnei using lipid profiling by MALDI-TOF mass spectrometry paired with machine learning

Abstract

Matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS) has become a staple in clinical microbiology laboratories. Protein-profiling of bacteria using this technique has accelerated the identification of pathogens in diagnostic workflows. Recently, lipid profiling has emerged as a way to complement bacterial identification where protein-based methods fail to provide accurate results. This study aimed to address the challenge of rapid discrimination between Escherichia coli and Shigella spp. using MALDI-TOF MS in the negative ion mode for lipid profiling coupled with machine learning. Both E. coli and Shigella species are closely related; they share high sequence homology, reported for 16S rRNA gene sequence similarities between E. coli and Shigella spp. exceeding 99%, and a similar protein expression pattern but are epidemiologically distinct. A bacterial collection of 45 E. coli, 48 Shigella flexneri, and 62 Shigella sonnei clinical isolates were submitted to lipid profiling in negative ion mode using the MALDI Biotyper Sirius® system after treatment with mild-acid hydrolysis (acetic acid 1% v/v for 15 min at 98°C). Spectra were then analyzed using our in-house machine learning algorithm and top-ranked features used for the discrimination of the bacterial species. Here, as a proof-of-concept, we showed that lipid profiling might have the potential to differentiate E. coli from Shigella species using the analysis of the top five ranked features obtained by MALDI-TOF MS in the negative ion mode of the MALDI Biotyper Sirius® system. Based on this new approach, MALDI-TOF MS analysis of lipids might help pave the way toward these goals.

Keywords: MALDI; Shigella; identification; lipids.

Figure 1

Linear negative ion mode mass spectra of Escherichia coli (top panel), Shigella sonnei (middle panel), and Shigella flexneri (bottom panel).

Figure 2

Dendrograms showing hierarchical clustering of Escherichia coli, Shigella sonnei, and Shigella flexneri samples. Black indicates E. coli isolates, red indicates S. flexneri isolates, and green indicates S. sonnei isolates. Clustering of species using the naïve binary absence‐presence matrix.

Figure 3

Top 15 ranked features reported from “binda” (ranked from left to right, from top to bottom). p‐values come from the t‐test.

Figure 4

Bar plots showing accuracy values for species prediction (between Escherichia coli and combination of Shigella flexneri and Shigella sonnei) using top‐ranked features; all features and randomly selected features. The accuracy values come from the analysis being repeated 100 times of splitting training and testing data and random selection of peaks for control.

Figure 5

Barplots showing accuracy values for species prediction (between Shigella flexneri and Shigella sonnei) using top‐ranked features; all features and randomly selected features. The accuracy values come from the analysis being repeated 100 times by splitting training and testing data and random selection of peaks for control.

Figure 6

Classification metrics including precision, sensitivity, specificity, and F1

Figure A1

Workflows for the identification of shigellosis in a clinical microbiology laboratory. The routine workflow is represented by a black arrow while the lipid profiling identification workflow is represented by green arrows. Combined with a machine learning algorithm, lipid profiling by routine MALDI in the negative ion mode might have the potential to differentiate Escherichia coli from Shigella species.