Rapid antibiotic susceptibility testing and species identification for mixed samples

Abstract

Antimicrobial resistance is an increasing problem on a global scale. Rapid antibiotic susceptibility testing (AST) is urgently needed in the clinic to enable personalized prescriptions in high-resistance environments and to limit the use of broad-spectrum drugs. Current rapid phenotypic AST methods do not include species identification (ID), leaving time-consuming plating or culturing as the only available option when ID is needed to make the sensitivity call. Here we describe a method to perform phenotypic AST at the single-cell level in a microfluidic chip that allows subsequent genotyping by in situ FISH. By stratifying the phenotypic AST response on the species of individual cells, it is possible to determine the susceptibility profile for each species in a mixed sample in 2 h. In this proof-of-principle study, we demonstrate the operation with four antibiotics and mixed samples with combinations of seven species.

Fig. 1. Schematic representation of the AST workflow with timeline.

a A cartoon of the microfluidics setup with the mixed species loaded on the chip. b Time-lapse phase-contrast images of the cells in the traps when grown in media with (top) and without (bottom) antibiotics. c Fluorescence images of the bacteria with ssDNA probes targeting the ribosomal RNA of specific bacteria for species identification. d Analysis of time-lapse stacks and species ID using deep learning for segmenting and tracking cells. e Detection of AST profiles for individual pathogens at a given antibiotic concentration. Part of Fig. 1a created using www.biorender.com.

Fig. 2. Analysis.

a Omnipose network showing network structure, inputs, and outputs to the neural network and reconstruction of output to generate cell masks. b Quantified performance of cell segmentation in mother-machine devices. Average precision vs IOU threshold plot for mixed-species dataset. The IOU threshold defines valid matching between predicted mask and ground-truth masks, 0.5 indicating half the pixels were correctly matched and 1 indicating pixel perfect match for every cell. Average precision (TP / (TP + FP + FN)) is calculated from the valid matches (TP), no valid matches (FP), and the ground-truth masks that have no valid match (FN). c Few examples of phase-contrast images, ground-truths and network predictions of U-net and Omnipose. d Overview of tracker network. e Confusion matrix of the edge predictions. f Tracks of cells in one channel of the mother-machine device.

Fig. 3. Species stratified responses to antibiotic treatments.

a–d AST profiles with normalized growth rates for the four antibiotics used. The species stratified responses (mean and SEM), as well as the pooled response (without species stratification), are shown for each antibiotic. In all AST profile plots, S and R represent the Susceptible and Resistance, respectively. The experiment is performed once per antibiotic, although several non-reported experiments were performed for calibration.

Fig. 4. Combinatorial FISH.

a Overview of the combinatorial FISH probing for the multi species identification. A cartoon illustrating the different bacterial species with their ribosomal RNA (left). Illustration of the specific sequences with the multiple adapters targeting the ribosomal RNA of individual bacteria and its hybridization to the target rRNA (middle). Detection probes with different fluorophores. Hybridization of detection probes to the adapter sequences along with unique sequences that are targeted to the species specific rRNA (Right). b Example images (Scale bar 20 µm) of mixed species loaded in the microfluidic chip and probed using combinatorial FISH for species identification. After the hybridization step, cells were imaged in different channels (PhC, Alexa 488, Cy3, Cy5, and Texas Red). The bacterial species are marked in white (Escherichia coli), magenta (Klebsiella pneumoniae), cyan (Pseudomonas aeruginosa), brown (Enterococcus faecalis), yellow (Acinetobacter baumannii), navy blue (Proteus mirabilis) and red (Staphylococcus aureus). The experiment with all seven species mixed was performed a single time. However, similar experiments with all adaptors and probes mixed are reported in Fig. 5.

Fig. 5. Species-wise AST profiles for experiments with 2 species using the combinatorial FISH method.

a P. aeruginosa and A. baumannii were treated with Gentamicin, b K. pneumoniae, and S. aureus treated with Ciprofloxacin, c E. coli, and P. mirabilis treated with Nitrofurantoin, d E. coli and E. faecalis treated with Vancomycin. e–h Biological repeats of 5a-5d, respectively. In all AST profile plots, S and R represent Susceptible and Resistance, respectively. Normalized growth rates ± SEM as a function of time are shown for each species detected in the experiments.