An Image Processing Tool for Automated Quantification of Bacterial Burdens in Zebrafish Larvae

Zebrafish larvae are used to model the pathogenesis of multiple bacteria. This transparent model offers the unique advantage of allowing quantification of fluorescent bacterial burdens (fluorescent pixel counts: FPC) in vivo by facile microscopical methods, replacing enumeration of bacteria using time-intensive plating of lysates on bacteriological media. Accurate FPC measurements require laborious manual image processing to mark the outside borders of the animals so as to delineate the bacteria inside the animals from those in the culture medium that they are in. Here, we have developed an automated ImageJ/Fiji-based macro that accurately detect the outside borders of Mycobacterium marinum-infected larvae.


Introduction (Introduction + Results + Discussion 750 words)
The zebrafish larva has come into its own as a model for bacterial pathogenesis and drug discovery (1)(2)(3)(4).In addition to its genetic amenability, its optical transparency allows real-time visualization of infection.Husbandry is facile; larvae can be maintained for two weeks in petri dishes without feeding or media changes, allowing the addition of drugs to be tested to the media.
Determining in vivo bacterial burdens is a core component of infectious diseases research.The classical method -enumerating bacteria from lysed animals or tissues following plating on bacteriological media -is laborious and time-consuming.Visible colonies may take days to weeks to appear, depending on the pathogen.Furthermore, serial assessments cannot be made in the same animals.
We overcame this problem by developing a microscopical method to determine relative burdens of fluorescent bacteria by enumerating fluorescence pixel counts (FPC) within the outlines of larval bodies (5)(6)(7).96-well plates containing the larvae are imaged using an inverted fluorescence microscope with a motorized stage, allowing hundreds of measurements to be completed within an hour.Serial imaging of the larvae allows monitoring of infection progression and testing drug efficacy.It can be used to study nonculturable pathogens such as Mycobacterium leprae rendered fluorescent using dyes (8).
However, the throughput of the FPC method is hindered by the need for a manual image analysis step: fluorescent bacteria shed from skin and other debris create bright spots outside the body, which can interfere with the thresholding process during image analysis (Supplemental Material 1, Supplemental Figure 1).Precise analysis of bacteria within the animal requires time-consuming, labor-intensive manual blacking out of the space outside of the larva.Furthermore, the manual method requires a threshold value determined by each user, making for inter-user variability in the analysis.
Here, we have developed an Image J/Fiji-based macro to automatically identify the larval body contour and bacterial spots within it and validated it using zebrafish infected with fluorescent Mycobacterium marinum (Mm).

Results and Discussion
We developed a Graphical User Interface (GUI)-based Image J/Fiji macro (9) (Fig. 1A).This macro is designed to allow users to collect various features of bacterial foci from the dataset (Fig. 1B; Supplemental Material 2).An input image is taken using a conventional widefield fluorescence microscope utilizing both fluorescence and dim transmitted light to simultaneously delineate the larval body and the fluorescent bacteria within it (Fig. 1C, left).The macro initially determines the outer border of the zebrafish larva from the 96-well image (Fig. 1C, middle).While this process is not perfect, it is sufficient for most of the images that we tested to proceed to the next step.Once the larval "object" is determined, fluorescence bacterial foci within the larval object are detected, and output parameters are measured (Fig. 1C, right).The individual image processing steps are summarized in Supplemental Material 3. Additionally, the macro allows users to explore the "Segmentation sensitivity" to identify the optimal segmentation for their images.Typically, running the macro with a lower Segmentation sensitivity results in segmenting the outer border of the larva closer to the larval body (Fig. 1D; Supplemental Material 3).
To validate the new macro, we compared its outputs with those using our manual threshold-based FPC analysis, using the same image dataset.We selected datasets comprising Mm-infected zebrafish larvae, treated or not with the mTORC inhibitor rapamycin, which increases bacterial burdens (10) (detailed Materials and Methods are in Supplemental Material 4).We datasets where the manual FPC differences spanned ~ one log for untreated larvae with the corresponding treated FPC's being ~ half to one log higher (Fig. 1E, F and G).The outputs from the manual method and the macro were highly correlated (Fig. 1H, I and J).They were also similar except in the last case, the manual FPC was a little higher in the treated group (Fig. 1G).Re-examination of the original images revealed that manual thresholding but not the automated macro had included bacteria with low intensity fluorescence (Supplemental Material 5, Supplemental Figure 2).Nevertheless, the automated macro identified the significantly higher bacteria burdens in the treated animals (Fig. 1G).Because manual FPC requires fluorescence thresholding based on identifying the lowest signal intensity that eliminates background pixels, it is subject to interoperator and interexperiment variability as illustrated in Supplemental Material 5.The automated macro has the benefit of removing subjectivity.This automated, high throughout macro should be applicable for all fluorescent microbes studied in zebrafish larvae (7,8,(11)(12)(13)(14).The codes and instructions for the macro are freely available on GitHub (https://github.com/JaneliaSciComp/Zebrafish_96well_segmentation_mesure_FPC).were derived from (E), (F) and (G), respectively.R 2 and p-value were analyzed with the simple linear regression and F-test, respectively.

Figure 1 :
Figure 1: The newly developed image processing macro and its validation