BBQ methods: streamlined workflows for bacterial burden quantification in infected cells by confocal microscopy

Accurate bacterial burden quantification (BBQ) within macrophages plays a crucial role in understanding host–pathogen interactions; for example, in evaluating the efficacy of antimicrobial responses, or in studying the role of host or bacterial factors in pathogenesis. Several approaches have been employed to assess intracellular bacterial load, each with its own strengths and weaknesses. Existing methods include colony forming unit (CFU) count from cell lysate or different microscopy-based approaches including manual counting of bacteria per cell, total fluorescence or integrated density of fluorescence per field of view, bacterial area per cell or luciferase luminescence quantification. While these techniques have provided valuable insights, they often present limitations in terms of precision, automation, and unbiased quantification.

CFU count has long been a standard method for estimating relative bacterial load in a population of infected cells (Sutton, 2012; Jiang et al., 2021; Boamah et al., 2023; Mittal et al., 2023). It involves plating dilutions of lysates from infected macrophages on agar plates, allowing bacterial colonies to grow, and subsequently counting the colonies. Although CFU count provides a quantitative measure of bacterial burden, it requires time-consuming culturing steps and may underestimate the true bacterial load due to variations in bacterial growth conditions and recovery rates. This is particularly true in the case of mycobacteria, which can take 1–3 weeks to display a reliable colony count due to their slow growth (Welch et al., 1993). Another complication is that mycobacteria tend to aggregate and therefore rarely appear as single bacilli in the lysates of infected cells and these aggregates will appear as a single colony thus leading to an underestimation of the bacterial burden. In addition, the CFU counting itself can introduce human error that increases with the amount of samples to analyze (Sutton, 2012).

The use of bioluminescent bacteria and the luminescence readout can provide a precise relative quantification of viable bacteria during macrophages infection. In the context of mycobacteria, it was proven efficient to assess the fitness of bacteria in host cells (Bedard et al., 2023; Raykov et al., 2023) The main advantages and limitations are discussed elsewhere (Andreu et al., 2010; Arafah et al., 2013), but in term of imaging the broad light emission spectrum from blue to yellow wavelength (Nijvipakul et al., 2008) may limit its use for quantitative multiplex microscopy approaches.

In general, microscopy provides a direct observation of the bacteria in the sample, so it gives a more precise and reliable quantification. At least in the context of mycobacterial studies, some available workflows using quantitative imaging approaches were also developed to study the growth dynamics of mycobacteria in host cells (Arafah et al., 2013; Barisch et al., 2015), or the phagosome maturation dynamics (Barisch et al., 2015; Schnettger and Gutierrez, 2017; Arévalo et al., 2023; Aylan et al., 2023b). Microscopy has been widely used for studies mycobacterial infection of macrophages as it provides a more rapid evaluation of bacterial burden compared to CFU count (Lerner et al., 2017; Mahamed et al., 2017; Aylan et al., 2023a; Golovkine et al., 2023; Malaga et al., 2023; Raykov et al., 2023). The main limitation is to have a bacterial strain constitutively expressing a fluorescent marker, but either mycobacterial expression plasmids for fluorescence proteins or already transfected strains are now easily available. Several methods are described in the literature to quantify bacterial burden by fluorescence microscopy. First, manual counting of bacteria per cell offers a direct assessment of bacterial burden at the single-cell level. This method involves visually inspecting microscopy images and manually enumerating the bacteria residing within each macrophage (Payros et al., 2021). While it allows for relatively precise quantification, manual counting is labor-intensive, subject to human error and bias, and impractical for analyzing large datasets. Additionally, it can be challenging to distinguish between closely clustered bacteria frequently encountered when working with mycobacteria for example. The quantification of the bacterial area per cell is another approach recently employed to estimate bacterial burden (Aylan et al., 2023a). By segmenting bacteria within macrophages and measuring the area occupied by bacterial fluorescence, it is possible to assess the relative bacterial load per cell. While this method offers good insight into the intracellular bacterial distribution, it does not account for variations in bacterial size or the three-dimensional architecture of infected cells. Quantifying bacterial burden based on total fluorescence or integrated density of fluorescence per field of view offers a precise and relative quantitation approach (Golovkine et al., 2023; Malaga et al., 2023). By summing the fluorescence intensity of all pixels within a defined region, this method provides a measure of the relative bacterial load compared to the background fluorescence of an uninfected cell. It is thus very suitable to monitor the growth or survival of bacteria between different infection conditions for example. However, the non-specific background fluorescence or auto-fluorescence from the host cell can introduce variability and affect the accuracy of quantification.

To address these limitations, we propose two novel automated methods for BBQ within macrophages based on image analysis of confocal microscopy data. We first propose an improvement of the existing methods to collect the total fluorescence in infected cells (Lerner et al., 2017; Payros et al., 2021), by adding some filtering steps to exclude uninfected cells from the quantification, and by automatizing the process into the PyImageJ environment. Alternatively, we introduce a new method of quantification by calculating the total bacterial volume per cell, to circumvent potential biases in fluorescence-based readouts.

The goal of these workflows is to provide a reliable, unbiased and rapid method to quantify the bacterial load in infected cells. The example given here are RAW264.7 macrophages infected with YFP expressing strain of Salmonella enterica serovar Typhimurium (S. Tm). Before acquisition, the infected cells were fixed at the desired time post-infection and stained with DAPI (see Materials and Methods). Any other type of nuclei stain is expected to perform as well as DAPI.

The base image material used for this workflow are multi-channel, z-stack images acquired by confocal microscopy. The optimal organization of data; images and folders, for both workflows is illustrated in Fig. 1A. Each individual experiment is in their own folder or ‘parent folder’. The eventual multiple conditions would be in subfolders, containing the different fields of views or technical replicates. The workflows are presented as code contained in jupyter notebooks and can be adapted to many different types of images and indexing.

For the image analysis, the automated workflows were implemented using the recently released ‘PyImageJ’ library, that allows the use of ImageJ/Fiji into a Python environment (Rueden et al., 2022). Detailed documentation for the installation and its use are described here (https://py.imagej.net/en/latest/). This library allows the combined use of Fiji and python that permit the image treatment on ImageJ and simultaneous use of python-based segmentation or tracking software under the same ecosystem.

For nuclei and cell segmentation, we used the deep learning-based segmentation software Cellpose (Stringer and Pachitariu, 2022 preprint). This software has the major advantage of facilitating training of custom segmentation models, thus that can be adapted to any cell type where a difference in cell morphology can decrease segmentation efficiency. Cellpose can also be invoked using python commands, so side by side with Fiji in the same environment. We took advantage of this flexibility and interoperability to propose nearly fully automated image analysis workflows that can be adapted to multiple types of images. We propose one workflow to quantify the total bacterial fluorescence intensity per cell and a second workflow for the total bacterial volume per cell.

The code and process are described in detail in the associate notebooks available on GitHub (https://github.com/jaugenst/BBQ/). Each step of the workflow is represented in the notebooks as a cell or block of code that can be ran sequentially or all at once if desired and the proper data pre-processing performed.

All the steps described below are illustrated in the diagram in Fig. 1B. The code of the workflow is contained in the notebook ‘bacterial_burden’.

To isolate signals from single cells, the Voronoi network analysis method is used and relies first on the detection of nuclei from the image. To this end, the nuclei channel is selected from the original z-stack image and is compressed into a smoothed z-projection ‘max’, and finally saved. The nuclei signals are then detected and outlined using Cellpose. This software can provide as an output the coordinates for regions of interest (ROI) in a text file that can be converted in ROIs directly in ImageJ/Fiji.

If the use of DIC or transmitted light channel is possible, these channels can be used for cell segmentation which would outline the cells more precisely compared to Voronoi segmentation. Cellpose can be used to generate a model able to segmentate the cells from this type of channel (Fig. 1B) and as previously showed, can produce ROIs as an output directly usable in ImageJ/Fiji (Augenstreich et al., 2022 preprint). This overrides the need for a subsequent Voronoi segmentation and from this step it is possible to go directly to step 3.

The ROIs of the segmented nuclei in step 1 are used to perform the Voronoi segmentation in Fiji and adapted from a method described elsewhere (Payros et al., 2021) and that was also adapted from another study (Lerner et al., 2017). Briefly, it involves particle detection to collect ROIs’ centroids, and then perform the Voronoi segmentation. As an improvement, and to ensure proper and automatic ROI centroid detection, the ROIs are first reduced in size by 1 pixel to make sure that the subsequent particle detection can detect single objects. Indeed, 2 objects separated by only one pixel will frequently be detected as one, and watershed function is not always successful at separating objects. The ROIs are also used to remove the background signal and are filled to facilitate the transformation into a binary image. After the latter is performed, the centroids of nuclei are collected and the Voronoi segmentation is performed, and ROIs are created from each zone given by the segmentation.

To measure the total bacterial fluorescence per macrophage, the intracellular bacteria must be detected and their signal isolated in order to be measured. To do so, the bacterial channel is isolated, and a z-projection ‘sum’ is created and saved. This image is duplicated, smoothed, thresholded and transformed into a binary image. The bacteria are then detected, and ROIs are created around them and saved. This new ROI set is then applied back to the original z-projection image to clear any signals outside the bacteria.

The ROIs from the Voronoi segmentation are called and applied on the cleared bacteria z-projection and the total fluorescence (RawIntDen) is collected in each cell. The results table obtained is automatically exported as a CSV file and saved. It will display all the cells as a list of numbers and their corresponding total bacterial fluorescence value, which represent the bacterial burden (Fig. 1B). In the absence of bacteria, the cell will have an N/A value in the Fiji results table and no value displayed in the CSV table (Fig. 1B). Thus, the ratio of the number of cells containing bacterial fluorescence compared to the total number of cells detected would give the percentage of infected cells and rate of infection and can be easily calculated from the CSV file.

At the end of the workflow, the folders containing the different images of eventual different conditions of some experiments should now be populated with the images of isolated nuclei channel, the isolated bacterial channel, the cellpose output (png and txt), and a zip file containing the ROIs from the Voronoi segmentation, thus allowing control of the results of each different step if needed. Each folder should also contain a CSV file named after the folder containing all the results of measurements from each image of this folder concatenated into a single file.

It should be noted that in the absence of a bright field channel, the nuclei detection and Voronoi segmentation will be biased by the presence of bacteria on the coverslip away from cells and count the closest cell as infected. Special care with washing or killing of extracellular bacteria with antibiotic treatment during the chase period could help to minimize the bias.

We established an alternative method for the quantification of bacterial burden, which we propose to be an improvement of measurement of bacterial burden based on total fluorescence intensity. We propose that relying on bacterial fluorescence alone for bacterial burden quantification is potentially biased by different parameters. The main bias is that this quantification is based upon on the assumption that the expression of fluorescent proteins remains constant over time. Nevertheless, we provide evidence that expression of fluorescence proteins driven by a constitutive promoter change when bacteria are in different environments (Fig. 3).

To circumvent this bias, we propose a workflow that aims to measure an estimation of the total bacterial volume inside the infected cells that is independent of fluorescence intensities. The base image material for this workflow is similar to the one described for the first workflow, as it is designed for a multi-channel, z-stack image of cells infected with fluorescent bacteria or any microorganism. The main difference is in the data collection after either the Voronoi ROIs or the cell outlines from Cellpose are generated: instead of collecting the raw integrated density of fluorescence on a z-projection sum of the bacterial channel, the bacterial area is measured directly on the z-stack and used for further calculation. From the area measurement on each slice of the z-stack, it is possible to calculate estimated partial bacterial volumes, by multiplying the area with the interval between two slices (Fig. 2A). Ultimately, the sum of the partial volumes will estimate the total bacterial volume per cell in µm³. A diagram illustrating the workflow is given in Fig. 2B. The code of the workflow is contained in the notebook “bacterial_burden_volume” in the GitHub repository.

As described earlier, from the original z-stack image, the nuclei channel is selected and collapsed into a smoothed z-projection ‘max’, and finally saved. The nuclei signals are then detected and outlined using the Cellpose segmentation software. This software also makes it possible to obtain regions of interest (ROIs) from a text file that can be converted in ROIs directly in ImageJ/Fiji.

As explained above, if the use of DIC or transmitted light channel is possible, Cellpose is then used to segmentate the cells and produce ROIs as an output directly usable in ImageJ/Fiji. This overrides the need for a subsequent Voronoi segmentation and it is possible to go directly to step 3.

The detected nuclei in step 1 are used to perform the Voronoi segmentation in Fiji. The nuclei ROIs are first reduced in size by 1 pixel to make sure that for further particle detection can detect single objects. The ROIs are also used to remove background signals and are filled to facilitate the transformation into a binary image. After the latter is performed, the nuclei centroids are collected and used, the Voronoi segmentation is performed, and ROIs are created from each zone given by the segmentation.

The bacterial channel is first isolated, smoothed, thresholded and transformed into a binary image. The bacteria are then detected by particle detection function, and ROIs are created around them on each slice and saved. While the ROIs are being created, the “analyze particle” function also allows for the simultaneous collection of data by choosing the “summarize” option. It displays a table showing how many particles were detected per slice and the associated total area. The collection of data is iterated on each Voronoi ROI and the final summary table is saved.

At the end of the workflow, the folders containing the different images of eventual different conditions of some experiment should now be populated with the isolated nuclei channel, the isolated bacterial channel, the Cellpose output (png and txt), and a zip file containing the ROIs from the Voronoi segmentation, thus allowing to control the results of the different step if needed. Each folder should also contain a series of CSV files containing the summary table of each image, and a CSV file named after the folder name containing all the volume calculation results from each image of this folder.

We proposed the need for development of a method to quantify bacterial burden by determination of the total bacterial volume per cell was because of the potentiality that bacterial expression level of the fluorescent reporter gene could change depending on stresses encountered by bacteria even if the gene is regulated by a constitutive promotor. We tested this hypothesis using a Salmonella enterica serovar Typhimurium (S. Tm) strain that constitutively expresses mCherry, a fluorophore typically known for its resistance to acidic pH (Doherty et al., 2010). We exposed S. Tm-mCherry to three different stresses that the bacteria are likely to encounter in a phagosome such as low pH, reactive oxygen species (ROS) or reactive nitrogen species (RNS). Single bacterial fluorescence analysis was performed as previously described (Lyu et al., 2023). We show that in an acidic pH, bacterial mCherry signal was lower than in a neutral pH (Fig. 3A). In contrast, mCherry expression was higher for bacteria cultivated in medium supplemented in H₂O₂ as an oxidative stress (Fig. 3A). In a similar fashion, a nitrosative stress induced by Spermine-NONOate (SPER/NO) increased fluorescence of mCherry compared to the control condition (Fig. 3A). The bacterial area was also measured and showed similar between the conditions, which demonstrates that the fluorescence variation is not the result of a difference in bacteria (Fig. 3B). Overall, these results demonstrate that the environment could directly bias constitutive expression level of reporter genes. Consequently, this will affect the reliability of bacterial burden measurements using fluorescence readouts and possibly luminescence readouts used to quantify live bacteria (Andreu et al., 2010).

Next, we assessed the precision of the bacterial volume calculation compared to the more classically used total-fluorescence-based quantification. The two workflows were used side by side on images of RAW264.7 cells infected with S. Tm expressing YFP. To induce variation in the bacterial burden, the cells were pre-treated with cytochalasin-D, which prevents phagocytosis, and thus, decreases bacterial burden. The direct cell segmentation by Cellpose from the Transmitted-light channel (T-PMT) was also chosen to exclude the extracellular bacteria on the coverslip and in part the ones that could still be adherent to the cells (Fig. 4A).

The cells were imaged at a fixed pinhole aperture of 1 airy unit, and the range of the z-stack was fixed at 10 µm. The same fields were acquired with different steps between the slices, 1 µm, 0.5 µm, which corresponds to the resolution in z at 1 airy unit, and 0.18 µm, which corresponds to the distance that match Nyquist Sampling. As shown, the calculation of the bacterial volume proved to be similar, with about 7% of volume variation on average between the sampling methods (Fig. 4B). This is reasonable in the light of acquisition time where one must adjust between sampling precision and the number of fields of views acquired. Here even a sampling at 1 µm returned comparable accuracy in volume calculation compared to thinner sampling (Fig. 4B). This accuracy was conserved on the other condition tested where the cells were treated with cytochalasin D (CytoD) before getting infected. And indeed, the expected decrease of bacterial burden by CytoD treatment was observed at any sampling distance with a marginal difference. Finally, the volume calculation was compared to total fluorescence quantification (Fig. 4B). Both quantification methods detected the decrease in a similar fashion thus performed similarly to detect a decrease in bacterial burden. Given the similarity in trends of the different curves, these results also highlight that a slicing of 1 µm that could be considered as an under-sampling gave an excellent approximation of bacterial burden, for both methods of quantification.

In order to validate even further the bacterial volume workflow, we assessed the bacterial burden of RAW264.7 cells infected with Mycobacterium tuberculosis, a bacterium that can persist for long periods of time in macrophages (Fig. 4C,D). After infection, the progression of bacterial burden was analyzed 1 h and 24 h after infection. First, total fluorescence measurement showed great variability between independent experiments but on average didn't detect any changes in bacterial burden (Fig. 4D). However, bacterial volume calculation showed a clear trend of a decreased bacterial burden at 24 h p.i. compared to 1 h p.i. on three out the four independent experiments (Fig. 4D). Indeed, observation of the micrographs indicated that the cells looked less loaded in bacteria but appeared brighter (Fig. 4C). The comparison of the fold changes at 24 h between the total fluorescence measurement and the volume measurement confirmed that the fluorescence readout systematically overestimated bacterial burden per cell. These data strongly suggest that the results obtained in vitro on S. Tm for variable expression of reporter genes due to environmental stresses (Fig. 3) are actually observed in macrophage infected with mycobacteria.

Overall, these results support the use of bacterial volume calculation for long-term analysis of bacterial burden quantification rather than fluorescence-based read-outs. Other reporter gene models, such as Luciferase gene expression, often rely on the same promotors and hence their expression levels are also likely to be affected by the bacterial environment. In our specific example of Mtb we did not test luminescence readouts, but commonly used luciferase reporter plasmids contain the same hsp60-derived promoter that is present in our DsRed expression plasmid backbone pMan-1 (Manzanillo et al., 2012). Consequently, our results suggest that this luciferase reporter gene expression might also be affected by the intracellular environment, independently of the survivability or fitness of the bacteria.

The present workflow can be used and expanded for many types of experiments, on both fixed and live samples, where Cellpose can perform adequately too (Stringer and Pachitariu, 2022 preprint). The workflow can be expanded to time-lapse imaging and could also be coupled with the existing workflows to analyze the recruitment of effectors to the phagosome, or the phagosome maturation level by quantitative imaging during mycobacterial infection (Barisch et al., 2015; Arévalo et al., 2023). Indeed, the volume calculation could theoretically be performed at the single cell level over time following the use of known tracking tool such as TrackMate that is available in Fiji (Ershov et al., 2022) and therefore in pyimagej environment, and could allow a chronologically accurate evolution of bacterial burden in a population of cells.

The workflow could also be theoretically applied to monitor multiple fluorescent strains in the case of a competition-infection study, as one would have to repeat the quantification part of the workflow on the channel of one detected strain, and then on the other. This brings to a potential limitation of the present workflow that could be the use of fluorescent proteins notoriously sensitive to acidic pH such as GFP (Kneen et al., 1998). While not tested here, an important decrease in fluorescence could potentially affect bacterial detection and underestimate the bacterial load, especially when studying bacterial mutant that would fail to inhibit phagosome maturation. The use of more stable proteins may be more adequate, such as mCherry or DsRed (Figs 3 and 4), or other available described mainly for mycobacteria such as the far-red E2-Crimson (Takaki et al., 2013). A recent study also explored the pH sensitivity of new bright green fluorescent protein (Campbell et al., 2022) and they could be used for robust bacterial detection and quantification.

In summary, we present automated workflows for the determination of bacterial burden in infected cells by total fluorescence quantification or bacterial volume calculation. This can allow accurate monitoring of bacterial growth in cells, or eventually bacterial clearance due to host cell response or drug treatment. It can also provide users with an easy measure of infectivity, that is a relevant readout for example for the study of mutant strains and their interaction with macrophages.

The code and process are described in detail in the associate notebooks available on GitHub (https://github.com/jaugenst/BBQ/) or provided in the supplementary material.

It is highly recommended to install any software and libraries required for this workflow in a virtual environment, using Anaconda, Pyvenv or environment modules. A detailed documentation of pyimagej is available online (https://py.imagej.net/en/latest/). To be noted that this workflow is not functional on MacOS as the “interactive mode” of initialization of Fiji is not compatible with the operating system. The workflow development was done in a jupyter notebook in the software jupyter-lab (https://jupyter.org/). Cellpose software was installed following the documentation and installation instructions available online (https://github.com/mouseland/cellpose). ChatGPT 3.5 (OpenAI) was occasionally used for code debugging and editing.

The cells used in this study were RAW-Lucia™-ISG (invivogen) or RAW264.7 cells (a kind gift from Dr Tamara O'Connor, Johns Hopkins University). The RAW-Lucia™-ISG cells were regularly tested for mycoplasma. They were cultivated in DMEM medium (Gibco) supplemented with 10% heat-inactivated fetal bovine serum (ThermoFisher Scientific). The day prior to the infection, the cells were detached using a cell scrapper and seeded into a 24-well plate in fresh medium that was supplemented in M-CSF 40 ng/ml for the case of infections with Mtb.

The first bacteria used for the study are Salmonella enterica spp Typhimurium (S. Tm) expressing YFP or mCherry. The S. Tm chromosomal integration of the yellow fluorescent protein (YFP) is constructed according to the method described here (Fan et al., 2019). Briefly, chloramphenicol-yellow fluorescent protein cassette (cat-P_tet-yfp) along with sequences homologous to the target region (3,095,232 bp to 3,095,506 bp) was integrated into the S. Tm chromosome by induction of Red recombinase, and the positive clones were selected by chloramphenicol and verified by sequencing. The strains expressing mCherry were generated by transformation with pZS-P_tet-mCherry-TGA-YFP-X-ECFP. The X stands for promoter regions of cadBA, hmpA and katG genes that are activated by low pH, nitrosative stress and oxidative stress respectively, and each construct was used in the stressful condition corresponding to their respective activation properties. The strains were cultivated in Luria-Bertani (LB) medium at 37°C. Oligonucleotides used for this study are reported in table S1.

The second bacterial strain used for this study is Mycobacterium tuberculosis H37Rv (ATCC) expressing DsRed, described here (Augenstreich et al., 2022 preprint). The bacteria were grown in liquid Middlebrook 7H9 medium supplemented with 10% oleic acid-albumin-dextrose catalase (OADC) growth supplement, 0.2% glycerol, 0.05% tween 80, and Zeocin 100 µg/ml.

For this study, overnight cultures of S. Tm containing their respective pZS-Ptet-mCherry-TGA-YFP-X-ECFP were 1:100 diluted in either LB (pH 7.0), acidic LB (pH adjusted to 4.5 by HCl) or LB supplemented with H₂O₂ at 500 µM (ThermoFisher Scientific), Spermine-NONOate at 0.5 mM (ThermoFisher Scientific) and grown for 24 h at 37°C. Cultures were harvested by centrifugation and resuspended in 20 μl LB. A volume of 1 μl of the resulting cultures were placed on a 1.5% agarose LB pad on a 12-well slide. The bacteria were then imaged by wide-field fluorescence microscopy (see procedure below).

For infections with S. Tm, the bacteria were pelleted and resuspended in PBS followed by measuring the OD₆₀₀ of the suspension. Bacteria for the infection were opsonized by taking 45 µl of bacterial culture, mixing with 20 µl of normal mouse serum (Jackson ImmunoResearch) and 135 µl of DMEM with 10% normal FBS (Gibco) for 20 min. This was followed by adding an additional 600 µl of DMEM with 10% normal FBS, and then adding the desired bacteria to the cells at an MOI of 10. The plate was spun at 100×g for 5 min then incubated at 37°C for 30 min. The cells were washed three times with PBS and fixed with PFA 4% for 1 h at room temperature. The cells were washed three times in PBS and mounted with Prolong mounting medium with DAPI (Molecular probes).

For infections with Mtb, the bacteria were pelleted and washed in PBS-tween 80 0.05%. The OD600 was measured, and the bacteria were added on the cells at an MOI of 10. The plate was then spun at 200×g for 5 min and incubated for 1 h. The cells were washed three times with PBS and incubated with fresh complete medium. At the designated time post-infection, the cells were then washed three times with PBS, fixed with PFA 4% for 1 h at room temperature. Finally, the cells were washed three times in PBS and mounted with Prolong mounting medium with DAPI.

For the bacterial imaging, the images were obtained using a BZ-X800 fluorescence microscope (Keyence) with a 100× oil objective. The single bacteria fluorescence was collected as performed before (Lyu et al., 2023). Briefly, the bacteria were segmented on the mCherry channel using Cellpose. The single bacteria mean fluorescence intensity and area in µm² was then measured. The data collection was processed in high throughput using the pyimagej library in jupyter-lab.

Imaging on the fixed infected cells was performed using a Zeiss laser scanning confocal microscope LSM-800, equipped with two gallium arsenide phosphide photomultiplier tube (GaAsP-PMT) detectors and a transmitted light photomultiplier tube detector (T-PMT), using the 63×/NA1.4 Oil objective. Each field of view was acquired as a Z-stack ranging 10 µm with a 1 µm step unless specified otherwise.

The graphics were generated using either the seaborn library on python, or GraphPad prism 10. The statistical comparisons were performed on GraphPad prism 10. For the comparative study of stress responses, the large datasets were analyzed by comparing the frequency distributions using Spearman's Rank Correlation Coefficient calculation. For the time course study, the means from each experiment and the fold change were compared by two-tailed paired t-test.

All the figures and drawings were made and assembled using Inkscape.

We sincerely thank Dr Serge Mazeres and Marc Augenstreich for their constructive feedback on the method validation.