MMHelper: An automated framework for the analysis of microscopy images acquired with the mother machine

Abstract

Live-cell imaging in microfluidic devices now allows the investigation of cellular heterogeneity within microbial populations. In particular, the mother machine technology developed by Wang et al. has been widely employed to investigate single-cell physiological parameters including gene expression, growth rate, mutagenesis, and response to antibiotics. One of the advantages of the mother machine technology is the ability to generate vast amounts of images; however, the time consuming analysis of these images constitutes a severe bottleneck. Here we overcome this limitation by introducing MMHelper ( https://doi.org/10.5281/zenodo.3254394 ), a publicly available custom software implemented in Python which allows the automated analysis of brightfield or phase contrast, and any associated fluorescence, images of bacteria confined in the mother machine. We show that cell data extracted via MMHelper from tens of thousands of individual cells imaged in brightfield are consistent with results obtained via semi-automated image analysis based on ImageJ. Furthermore, we benchmark our software capability in processing phase contrast images from other laboratories against other publicly available software. We demonstrate that MMHelper has over 90% detection efficiency for brightfield and phase contrast images and provides a new open-source platform for the extraction of single-bacterium data, including cell length, area, and fluorescence intensity.

Figure 1

Overview of the analysis pipeline. The analysis pipeline is broken down into five major steps. (A) The imaging mode is detected, determining whether images are brightfield or phase contrast. (B) Channels are detected, assigned specific labels and ordered consecutively from left to right. (C) Bacteria are detected in each channel. (D) Channels are tracked throughout the image time-series. In these representative images, the mother machine device at t = 1 h has moved approximately 10 μm to the left with respect to t = 0 h, as indicated by the arrow. Our algorithm quantifies this frame shift and relabels each channel accordingly, for example the channel indicated by the arrow is recoloured in yellow. (E) After channel tracking, the detected bacteria in each channel are tracked accordingly and relabelled where necessary, each bacterium keeping the same unique colour through consecutive time points as indicated by the arrow.

Figure 2

Pipeline for channel detection. (A) The original image is filtered (Sobel for phase images and Frangi for brightfield) followed by thresholding to identify potential ridges. These ridges are then filtered by size to leave the masks of the channels. (B) A new mask is created with the centre of each channel filled and through a simple subtraction of the previous mask with the new one, the centre of each channel is extrapolated. These masks can appear irregular in shape due to the presence of the bacteria they host. Consequently, new profiles are determined by creating vectors around the perimeter to form an average channel shape. (C) The spacing between these channels is determined and, after interpolation to determine the location of missing channels, the average channel shape is stamped in place. Noteworthy, our algorithm performs well also with images where the main channel is not horizontal resulting in slightly staggered labels. (D) A yellow contour is drawn around each label to delineate the detected channels.

Figure 3

Pipeline for bacteria detection. (A) By using the masks for the detected channels, the corresponding original image for each channel is identified and the image inverted using background subtraction. (B) This is followed by scale space filtering and thresholding. As a result, markers are identified that can be used for a watershed transformation. (C) Each single element within each channel identified by the watershed transformation is given a unique label, represented by a different colour. The result of the watershed is filtered to remove non-bacterial particles. Bacterial splits are identified, using a combination of width and pixel intensity, and a mask of the detected bacteria produced using a combination of distance transformation and pixel intensity along the skeleton.

Figure 4

Overview of bacteria tracking. Individual bacteria detected in an experiment using (A) minimal medium, (B) antibiotic treatment, or (C) growth medium at t = 0 and at t = 1 h (channels at the left and right hand side of each panel, respectively). (D–F) Corresponding tracked bacteria are relabelled, where necessary (e.g. second channel from the left in D), at t = 1 h so that their label (i.e. contour colour) matches that at t = 0. When a division occurs each of the offspring is assigned a new unique label (e.g. first and second channel from the left in F).

Figure 5

Dynamics in single-bacterium parameters. Temporal changes in (A) area, (B) length, and (C) GFP fluorescence for three representative bacteria, and their progeny, growing in lysogeny broth. Data bifurcations indicate bacterial divisions, e.g. bacterium 3 divided at t = 3 h and its daughters divided at t = 5 h.

Figure 6

Comparison of MMHelper and Molyso performances. Kernel density estimation for precision and recall of channel detection from (A) brightfield and (B) phase contrast images via MMHelper (red) and Molyso (blue). The distribution of precision and recall values obtained via MMHelper on phase contrast images tightly clusters around a recall value of 1 and a precision value of 0.55. Therefore, we have zoomed this area in the dashed circle to facilitate its visualisation. (C,D) Corresponding kernel density estimation for precision and recall of bacteria detection. Insets: representative images of channel (A) and bacteria (C) detection.