High-content high-resolution microscopy and deep learning assisted analysis reveals host and bacterial heterogeneity during Shigella infection

Ana T. López-Jiménez; Dominik Brokatzky; Kamla Pillay; Tyrese Williams; Gizem Özbaykal Güler; Serge Mostowy

doi:10.7554/eLife.97495.1

eLife assessment

This manuscript describes an AI-automated microscopy-based approach to characterize both bacterial and host cell responses associated with Shigella infection of epithelial cells. The methodology is compelling and should be helpful for investigators studying a variety of intracellular pathogens. The authors have acquired valuable findings regarding host and bacterial responses in the context of infection, which should be followed up with further mechanistic-based studies.

https://doi.org/10.7554/eLife.97495.1.sa3

Significance of findings

valuable: Findings that have theoretical or practical implications for a subfield

landmark
fundamental
important
valuable
useful

Strength of evidence

compelling: Evidence that features methods, data and analyses more rigorous than the current state-of-the-art

exceptional
compelling
convincing
solid
incomplete
inadequate

During the peer-review process the editor and reviewers write an eLife assessment that summarises the significance of the findings reported in the article (on a scale ranging from landmark to useful) and the strength of the evidence (on a scale ranging from exceptional to inadequate). Learn more about eLife assessments

Abstract

Shigella flexneri is a Gram-negative bacterial pathogen and causative agent of bacillary dysentery. S. flexneri is closely related to Escherichia coli but harbors a virulence plasmid that encodes a Type III Secretion System (T3SS) required for host cell invasion. Widely recognized as a paradigm for research in cellular microbiology, S. flexneri has emerged as important to study mechanisms of cell-autonomous immunity, including septin cage entrapment. Here we use high-content high-resolution microscopy to monitor the dynamic and heterogeneous S. flexneri infection process by assessing multiple host and bacterial parameters (DNA replication, protein translation, T3SS activity). In the case of infected host cells, we report a reduction in DNA and protein synthesis together with morphological changes that suggest S. flexneri can induce cell-cycle arrest. We developed an artificial intelligence image analysis approach using Convolutional Neural Networks to reliably quantify, in an automated and unbiased manner, the recruitment of SEPT7 to intracellular bacteria. We discover that heterogeneous SEPT7 assemblies are recuited to actively pathogenic bacteria with increased T3SS activation. Our automated microscopy workflow is useful to illuminate host and bacterial dynamics at the single-cell and population level, and to fully characterise the intracellular microenvironment controlling the S. flexneri infection process.

Introduction

Shigella flexneri is a Gram-negative bacterium and the etiologic agent of bacillary dysentery, an intestinal infectious disease that kills more than 150 thousand people worldwide per year, disproportionally affecting children under 5 years old from low- and middle-income countries (Global Burden of Disease Collaborative Network., 2020). After the ingestion of a small number of bacteria (10-100), S. flexneri rapidly colonises the human intestinal epithelium. For this, S. flexneri uses a Type 3 Secretion System (T3SS) that injects bacterial effectors into the host cell to modulate key aspects of the infection process and to evade innate immune defences by the host (Schnupf & Sansonetti, 2019). Following uptake by host cells, S. flexneri breaks its vacuolar compartment to access the cytosol (Ray et al., 2010). Here, the virulence factor IcsA enables S. flexneri to hijack host actin and polymerise comet tails that render bacteria motile to disseminate to neighbouring cells and spread in the tissue (Bernardini et al., 1989; Egile et al., 1999).

Septins are highly-conserved GTP-binding proteins that form heteromeric complexes (hexamers, octamers) that assemble into non-polar filaments, bundles and rings (Martins et al., 2023; Soroor et al., 2021; Szuba et al., 2021; Woods & Gladfelter, 2021). Septins are involved in many physiological cellular processes (eg. cytokinesis, cilia formation), and play key roles during infection (Mostowy & Cossart, 2012; Robertin & Mostowy, 2020; Spiliotis & Nakos, 2021). Septins form collar-like structures at the plasma membrane during bacterial invasion (Boddy et al., 2018; Mostowy et al., 2009; Robertin et al., 2023), and ring-like structures surrounding membrane protrusions in cell-to-cell bacterial spread (Mostowy et al., 2010). In the cytosol septins are recruited to actin-polymerising bacteria, including S. flexneri and Mycobacterium marinum, as cage-like structures (Mostowy et al., 2010). In the case of S. flexneri, septins are recruited to μm-scale curvature at bacterial poles and the division site, which are enriched in the negatively charged phospholipid cardiolipin (Krokowski et al., 2018). Recent work has shown that IcsA promotes the recruitment of septins to S. flexneri, and that lipopolysaccharide (LPS) inhibits S. flexneri-septin cage entrapment (Lobato-Márquez et al., 2021; Mostowy et al., 2010). Septin cages prevent bacterial actin tail motility and therefore are considered as an antibacterial mechanism of host defence (Krokowski et al., 2018; Mostowy et al., 2010).

Intracellular infection is usually portrayed as a uniform, synchronised and highly orchestrated process. However, infection assays performed in controlled laboratory environments with isogenic populations display a high level of complexity (Avraham et al., 2015; Toniolo et al., 2021). It is particularly unknown how different subsets of bacteria (eg. motile vs. non motile bacteria) can colonise distinct intracellular microenvironments and become targeted by multiple cell-autonomous immunity mechanisms (eg. septin cages, autophagy). It has been proposed that this diversity of intracellular niches impacts bacterial physiology (Day et al., 2023; Gutierrez & Enninga, 2022; Santucci et al., 2021). For example, it may affect individual bacterial ability to acquire nutrients, to be exposed to varying antibiotic concentrations, or to elicit different responses and signalling pathways within the host cell. While the concept of phenotypic heterogeneity in prokaryotic clonal populations is well established (Ackermann, 2015), the concept is emerging also in eukaryotic cells where non genetic differences may for instance explain individual cell sensitivity and responses by cell-autonomous immunity (McDonald & Dedhar, 2024). It is therefore paramount to study host and bacterial cells at the single-cell level, to assess the individual responses that collectively contribute to the infection process.

Here, to dissect the complexity of S. flexneri infection in epithelial cells, we use high-content high-resolution microscopy coupled to automated image analysis. High-content microscopy enables the automated acquisition of thousands of microscopy images and therefore is useful to capture the heterogeneity of individual host and bacterial cells during the infection process for quantitative analysis (Aylan et al., 2023; Brodin & Christophe, 2011; Deboosere et al., 2021; Dramé et al., 2023; Fisch et al., 2019a; Lensen et al., 2023; Pylkkö et al., 2021). We analyse multiple parameters, such as morphological host cell features, the de novo synthesis of DNA and proteins in both host and bacterial cells, as well as activation of the bacterial T3SS. To investigate the antimicrobial potential of septin-mediated immunity, we develop a deep-learning based approach to automatically identify septin-S. flexneri interactions. Together, we highlight the power of high-content microscopy and artificial intelligence to reveal insights on both host and bacterial cells, and illuminate fundamental biology underlying the S. flexneri infection process.

Results

Capturing the heterogeneity of S. flexneri infection using high-content microscopy

To understand the complexity and reveal the heterogeneity of S. flexneri infection in human epithelial cells, we dissected the infection process using high-content high-resolution microscopy (Figure 1A, B). We observed a mean of 11.5% of infected cells measured by fluorescence at 3h 40 min post infection (hpi) at MOI 100:1 (Figure 1C). These infected cells presented different burdens of bacteria following an exponential decay distribution (Figure 1D), reflective of the formation of infection foci (Ortega et al., 2019). With growing evidence of the cytosol being a non-homogeneous compartment containing concentration gradients, biomolecular condensates, protein complexes and concentrates (van Tartwijk & Kaminski, 2022), we tested whether S. flexneri may be exposed to different intracellular microenvironments. For this, we measured the distance of bacteria to the centroid of cells (Figure 1E) and observed a median distance of 8.74 µm. The median distance to the centroid of the cell significantly decreased with increasing bacterial burden (Figure 1E), suggesting that the microenvironment of bacteria is diverse and changes as bacteria replicate intracellularly and populate the central parts of host cytosol after uptake at the host cell periphery. To dissect the heterogeneity in host cells, we quantified an increase of 36.9% and 22.3% in the cellular and nuclear area of infected cells, respectively (Figure 1G, I), which was dependent on the infection burden (Figure 1H, J). To test whether this increase reflected an increase in volume, we measured the total amount of Hoechst fluorescence in the nucleus of the cell as a proxy for nuclear volume (Figure 1-figure supplement 1A) and observed a 29.4% increase in infected cells. To test if this effect was due to morphological rearrangements during infection, we treated cells with Latrunculin B to inhibit actin polymerisation. While cells treated with Latrunculin B decreased their cellular area (but not their nuclear area) (Figure 1-figure supplement 1B, C), the drug did not abrogate the difference in size between infected and uninfected cells. Taken together, these results highlight an increase in cellular volume during infection that is dependent on the infection burden.

High-content imaging captures the heterogeneity of *S. flexneri* infection
(A) Representative high-resolution microscopy image of HeLa cells infected with *S. flexneri* expressing mCherry acquired with the high-content microscope ZEISS CellDiscoverer7. Scale bar, 10 µm. (B) Enlarged view of insets in A, highlighting a *S. flexneri* bacteria polymerising an actin tail (1) and entrapped in a septin cage (2), scale bar, 1 µm. (C) Percentage of infected cells at 3 h.p.i. of 7 independent experiments. Graph represents mean = 11.49% ± SD = 4.36%. (D) Percentage of infected cells harbouring different bacterial doses, graph represents mean ± SD. (E) Distance of the bacteria to the centroid of the host cell. Graph represents the median (8.74 µm) and interquartile range (6.51 – 11.05 µm). Values from n = 4990 cells from 7 independent experiments. (F) Distance of individual bacteria to the centroid of the host cell depending on its bacterial load. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test. (G) Area of cells depending on infection. Graph represents the median area (uninfected: 3031 µm², infected: 4148 µm²) and interquartile range (unifected: 2308 – 4169 µm², infected: 3062 – 5752 µm²). Values from n = 9469 uninfected cells and n = 1031 infected cells from 7 independent experiments. (H) Area of infected cells depending on its bacterial load. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test. (I) Nuclear area of cells depending on infection. Graph represents the median area (uninfected: 1551 µm², infected: 1897 µm²) and interquartile range (unifected: 1209 - 1950 µm², infected: 1486 - 2333 µm²). Values from n = 9469 uninfected cells and n = 1031 infected cells from 7 independent experiments. (J) Nuclear area of infected cells depending on its bacterial load. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test.

Epithelial cells reduce de novo DNA and protein synthesis upon infection

In order to understand the physiological state of enlarged epithelial cells infected with S. flexneri, we monitored host DNA replication and protein translation using Click-chemistry. Using a 1 h pulse of the clickable thymidine analogue Edu, we specifically visualized and quantified de novo DNA synthesis in the population of epithelial cells with high-content microscopy (Figure 2A, Figure 2-figure supplement 1). We identified the subpopulation of cells in S-phase by their high level of Edu incorporation. Consistent with this, this subpopulation of cells was fully abrogated in the presence of the S-phase blocker Aphidicolin (Pedrali-Noy et al., 1980) (Figure 2-figure supplement 2A, B). Approximately one third of uninfected cells (34.45%) were found in S-phase (Figure 2B, C), in agreement with S-phase lasting around one third of the cell cycle (∼8h out of ∼22h). Infected cells presented a similar proportion of cells in S-phase (35.89%), indicating that S. flexneri infection did not block transition of cells from G1 to S at the timepoint tested. We then assessed the amount of de novo DNA synthesis incorporated in cells in S-phase during infection. Surprisingly, we observed a reduction of Edu incorporation in infected cells (Figure 2D), which was dependent on the bacterial burden (Figure 2E). These results indicate that S. flexneri infection slows down DNA synthesis of epithelial cells.

*S. flexneri* infection reduces host DNA synthesis and host protein synthesis. (A) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing mCherry at 3 h.p.i, after 1h incubation with Click-IT Edu. Asterisks label the nucleus of infected cells with reduced Edu incorporation. Contour of selected cells is delimited with a dashed line. Scale bar, 10 µm. (B) Edu acquisition in uninfected cells is bimodal. Graph represents the normalised median of the ratio between the fluorescence intensity of Edu and Hoechst (1), and interquartile range (0.5924 – 31.22). Values from n = 8941 cells from 3 independent experiments. An arbitrary threshold of 5 is selected to define cells in S-phase. (C) Frequency of cells in S-phase (as defined in B) depending on infection. Graph represents mean ± SD (uninfected: 34.45% ± 9.78%, infected: 35.89% ± 4.69%). Student’s t-test. (D) Edu acquisition in cells in S-phase depending on infection. Graph representes the normalised median of the ratio between the fluorescence intensity of Edu and Hoechst (uninfected: 75.14, infected: 44.73), and interquartile range (uninfected: 27.89 – 131.0, infected: 22.05 – 93.36), from 3065 uninfected and 552 infected cells. Mann-Whitney U test. (E) Edu acquisition in cells in S-phase depending on bacterial load. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test. (F) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing mCherry at 3 h.p.i, after 1h incubation with Click-IT AHA. Asterisks label the infected cells with reduced AHA incorporation. Scale bar, 10 µm. (G). AHA acquisition in host cells depending on infection. Graph representes the normalised median of AHA fluorescence intensity (uninfected: 1, infected: 0.86), and interquartile range (uninfected: 0.7850 – 2.694, infected: 0.6207 – 1.241), from 36 uninfected and 233 infected cells. Mann-Whitney U test. (H) AHA acquisition in infected cells depending on bacterial load. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test.

To assess protein translation, we used a 1h pulse of the clickable methionine analogue AHA, which specifically labels de novo synthesized proteins as AHA incorporation is reduced after treatment with the inhibitor of translational elongation Cycloheximide (Ennis & Lubin, 1964) (Figure 2-figure supplement 1C, D). During infection, we observed a reduction of protein translation in infected cells compared to non-infected cells (Figure 2F,G), which was also dependent on the bacterial burden (Figure 2H).

Collectively, the cell size increase together with the reduction of de novo DNA and protein synthesis in infected epithelial cells suggests that S. flexneri induces an arrest in the host cell cycle.

Design of a deep learning pipeline to automatically identify bacteria interacting with SEPT7

S. flexneri have been shown to interact with septins during infection. These interactions include septin ring-like structures surrounding phagocytic cups, autophagosomes, actin tails and protrusions, all being morphologically diverse structures that are normally scored by hand (Krokowski et al., 2018; Lobato-Márquez et al., 2023; Mostowy et al., 2010). To study septin-associated bacteria using high-content microscopy, we designed a tailored analysis pipeline for the automatic and unbiased identification of SEPT7-bacteria interactions (Figure 3). SEPT7 is a core component of septin hexamers and octamers, and is historically used to identify septin structures (Mostowy et al., 2010). The analysis pipeline includes two deep learning models based on Convolutional Neural Networks (CNN), which are effective in solving classification tasks in computer vision (Krizhevsky et al., n.d.; Lecun et al., 2015; Nielsen, 2015). As septins can surround individual and also multiple bacteria simultaneously (Figure 4-figure supplement 1A), we focused on individual bacteria. For this, we trained the first classification model to discriminate between “single” bacteria and “clumps”. The second classification model was designed to assign those single bacteria into “SEPT7-positive structures” or alternatively “negatives”, according to the morphology of SEPT7 recruitment around them.

Microscopy pipeline to automatically identify SEPT7-*S. flexneri* interactions
(A) Diagram representing the steps involved in the imaging and analysis pipeline. After infection, fixation and staining, high number of images are automatically acquired using the high-content microscope. Scale bar, 10 µm. These are then processed to segment bacteria based on their fluorescence, so that a square field containing the bacteria in the centre is cropped and saved for subsequent analysis. A deep learning model based on CNN is applied using Jupiter notebooks and the Python Keras library hosted on Colab to sort individual, isolated bacteria. Several processing steps are applied using Python to remove noise (band-pass filtering by difference of Gaussian and mean filters), normalise the data across datasets for comparison, and remove signal away from the bacteria which is irrelevant for the identification of SEPT7 assemblies. Scale bar, 1 µm. Finally, a similar second deep learning model based on CNN is applied to identify SEPT7 interacting with bacteria.

Training of a deep learning algorithm to identify single isolated bacteria cells and SEPT7 associated *S. flexneri*
(A) Table summarizing the annotated dataset used for the training process. The labels of the two classification classes used were “Single bacteria” and “Clump”. The annotated images were randomly split into two groups, one used as training dataset (comprising 80% of the images) and another used as validation dataset (with the remaining 20%). A separate annotated dataset was used for testing. (B) Representative Airyscan images of bacteria used for each class. Scale bar, 1 µm. (C) Architecture of the deep learning algorithm used for training. Each raw describes the characteristics of the sequential transformation steps applied. Conv2D stands for 2D Convolution. (D, E) Accuracy and Loss as metrics to represent the training process over subsequent epochs (entire passing of the training data through the algorithm). Accuracy increases and loss decreases for both training and validation datasets, indicating a good fit of the model to the data. Vertical dashed grey line indicates Early Stopping, or epoch value for the minimum validation Loss. (F) Confusion matrix performed on an annotated test dataset not used previously for the training or validation process, indicating the percentage of predictions that were correct or wrong for each class. (G) Precision, recall and f1 score as metrics that summarize the performance of the model or classifier. (H) Table summarizing the annotated dataset used for the training process. The labels of the two classification classes used were “Septin” and “Negative”. The annotated images were randomly split into three groups, one used as training dataset (comprising 75% of the images), another used as validation dataset (with 15%) and a last one used as test dataset (with 10%). Due to the data being very imbalanced (15% natural frequency of SEPT7 associated bacteria), the images comprised in the Negative class were under-sampled as indicated in the table. (I) Representative Airyscan images of bacteria used for each class. Scale bar, 1 µm. (J) Architecture of the deep learning algorithm used for training. Each raw describes the characteristics of the sequential transformation steps applied. Conv2D stands for 2D Convolution, Batch_norm stands for Batch normalisation. (K, L) Accuracy and Loss as metrics to represent the training process over subsequent epochs. Accuracy increases and loss decreases for both training and validation datasets, indicating a good fit of the model to the data. Vertical dashed grey line indicates Early Stopping, or epoch value for the minimum validation Loss. (M) Confusion matrix performed on the test dataset not used previously for the training or validation process, indicating the percentage of predictions that were correct or wrong for each class. (N) Precision, recall and f1 score as metrics that summarize the performance of the model or classifier.

The first classification model (discerning “single” versus “clumped” bacteria) together with its training and performance metrics is summarized in Figure 4. A large dataset of annotated bacteria (Figure 4A, B, Figure 4-figure supplement 1B) was used to train a Sequential model with a typical CNN architecture that was experimentally tested and optimised accordingly for best results (Figure 4C). This model provided a good fit of the training data (high accuracy, low loss), with similar predictive power for the validation dataset (Figure 4D, E). Performance of the model was further assessed with a confusion matrix where a separate annotated test dataset was challenged and correct number of predictions scored, showing that 96% of SEPT7 structures and 83% of clumps were correctly identified (Figure 4F). Finally, effective performance of the model was ensured with high values achieved in additional metrics testing for type I and II errors (precision, recall, F1-score) (Figure 4G).

The second classification task (identification of “SEPT7 positive” vs “negative” bacteria) was more challenging due to their intrinsic heterogeneity plus the presence of ubiquitious SEPT7 filaments that populate the vicinity of bacteria (but do not associate with bacteria per se). To address these challenges, we increased the amount of data and complexity of the second classification model (Figure 4H-J, Figure 4-figure supplement 1C) compared to the first (Figure 4A-C). This included more than 1200 images of S. flexneri associated to SEPT7, both from infections with wild-type bacteria or expressing AfaI to increase bacterial invasion. The training process achieved a high accuracy and low loss for the training and validation dataset (Figure 4K, L). Correct prediction of manually annotated SEPT7 positive structures was 80% as shown in the confusion matrix (Figure 4M), with performance metrics Precision, Recall and F1 score also around 0.8 (Figure 4N).

Characterization of the morphological heterogeneity of SEPT7 recruitment to S. flexneri

The training of the classification model to identify bacteria associated to SEPT7 required a large dataset of images. We observed diverse morphological conformations of SEPT7 signal around bacteria (Figure 5 A-C). We differentiated five frequent subcategories from a total of 855 cases. The first corresponds to most cases (66.7%), where SEPT7 forms rings that entrap bacteria transversally (Figure 5A, B, a). We also observed two categories where SEPT7 surrounds bacteria more homogeneously either in a tight (18.2%) or loose association (6.1%) (Figure 5A, B, b-c). Finally, we observed partial SEPT7 recruitment, with only one (5.0%) or two (4.0%) bacterial poles being targeted (Figure 5A, B, d-e). Given this diversity, we combined the entire dataset to summarise the probability of SEPT7 interactions with bacteria (Figure 5-figure supplement 1). As bacteria are found in any orientation inside epithelial cells, we included only those instances where the z-projection of the bacterial axis measured more than 2.5μm, ensuring they were positioned horizontally when imaged. Representations shown in Figure 5-figure supplement 1A, C and graphs in Figure 5-figure supplement 1B, D can be understood as a probability map of the area of influence of SEPT7 when associating with intracellular S. flexneri.

Septin assemblies associated to *S. flexneri* are heterogeneous
(A) Selected examples of the training dataset to showcase the diverse morphology of septin assemblies associated to bacteria. Frequent morphologies include (a) ring like structures around the bacteria, (b) smooth tight recruitment around the entire bacterial surface, (c) loose structures more distant from the bacterial surface, (d) septin recruitment to both bacterial poles, (e) septin recruitment to a single bacterial pole. Scale bar, 1 µm. (B) Intensity profiles of the cytosolic bacterial mCherry and the SEPT7 recruitment, as indicated in the white and orange dashed lines, respectively. (C) Pie chart representing the relative frequency of 855 *S. flexneri*-SEPT7 associations depicted in A.

SEPT7 is recruited to S. flexneri with increased T3SS activity

Septins have previously been shown to recognise growing bacterial cells in vitro and during infection of HeLa cells, using pharmacologic and genetic manipulation (Krokowski et al., 2018; Lobato-Márquez et al., 2021). To test the metabolic state of SEPT7-associated S. flexneri at 3h 40 min post infection (hpi) we used our high-content imaging and analysis pipeline (Figure 6). We measured S. flexneri de novo synthesis of DNA and proteins using a 1 h pulse Click-IT Edu and Click-IT AHA. Only active bacteria incorporated these analogues, as the antibiotic Nalidixic acid (inhibitor of the bacterial DNA gyrase (Sugino et al., 1977)) and Rifampicin (inhibitor of the bacterial RNA polymerase (Ezekiel & Hutchins, 1968)) significally reduced their incorporation, respectively (Figure 6-figure supplement 1A-D, Figure 6-figure supplement 1E,F). During infection, bacteria associated with SEPT7 showed heterogeneus levels of DNA synthesis and protein translation similar to SEPT7-negative bacteria (Figure 6A-H).

*S. flexneri* associated with SEPT7 are metabolically active
(A) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing mCherry at 3 h.p.i, after 1h incubation with Click-IT Edu. Scale bar 10 µm. (B) Enlarged view of insets in A, highlighting a *S. flexneri* associated or not to SEPT7 structures. Scale bar, 1 µm. (C) Frequency of the number of SEPT7-associated bacteria identified using deep learning from experiments in A, from n= 1024 from 3 independent replicates. Graph represents mean ± SD (9.23% ± 2.62%). (D) Normalised ratio between Edu signal and mCherry in individual bacteria. Graph representes the median (negatives: 1, septin associated: 1.059), and interquartile range (negatives: 0.8185 – 1.207, septin associated: 0.8573 – 1.487). Mann-Whitney U test. (E) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing mCherry at 3 h.p.i, after 1h incubation with Click-IT AHA. Scale bar 10 µm. (F) Enlarged view of insets in E, highlighting a *S. flexneri* associated or not to SEPT7 assemblies. Scale bar, 1 µm. (G) Frequency of the number of SEPT7-associated bacteria identified using deep learning from experiments in E, from n = 1872 bacteria from 3 independent replicates. Graph represents mean ± SD (7.01% ± 2.41%). (D) Normalised AHA signal in individual bacteria in A, B. Graph representes the median (negatives: 1, SEPT7 associated: 1.055), and interquartile range (negatives: 0.6986 – 1.345, SEPT7 associated: 0.7837 – 1.497). Mann-Whitney U test.

S. flexneri pathogenicity relies on its T3SS. To understand effector secretion at the individual bacterial level, we used S. flexneri expressing the Transcription-based Secretion Activity Reporter (TSAR) (Campbell-Valois et al., 2014). This reporter comprises a short lived fast-maturing superfolder GFP (GFPmsf2) that is expressed in conditions where the T3SS is active, such as in broth upon addition of congo red (Figure 7-figure supplement 1A, B) (Parsot et al., 1995) and during infection (Figure 7). We observed heterogeneity in TSAR signal (Figure 9A,B). T3SS activation was higher in bacteria found in cells with a low bacterial burden (less than 10 intracellular bacteria) (Figure 7A,C). This suggests the highest level of activation occurs in bacterial pioneers that colonise neighbouring cells, in agreement with (Campbell-Valois et al., 2014). We then assessed T3SS activation in S. flexneri associated with SEPT7 using our deep learning analysis pipeline (Figure 7D-H). We detected a significant increase in the percentage of TSAR-positive S. flexneri and the overall intensity of TSAR signal compared to SEPT7-negative bacteria (Figure 7G,H). These include SEPT7-associated protrussions and cage-like structures (Figure 7E, Figure 7-figure supplement 1C). These data underscore the antimicrobial role of septins in associating with actively pathogenic bacteria.

*S. flexneri* associated with SEPT7 have an active T3SS
(A) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing the TSAR reporter at 3 h.p.i. Scale bar 10 µm. Arroheads highlight TSAR+ bacteria. Selected infected cells contours are delimited with dashed lines. (B) Distribution of TSAR signal intensity in individual bacteria. Graph represents the normalised median of the ratio between TSAR and mCherry (1), and interquartile range (0.7827 – 1.369). Values from n = 3654 bacterial cells from 3 independent experiments. An arbitrary threshold of 1.5 is selected to define TSAR+ *S. flexneri*. (C) Intensity of TSAR signal in individual bacteria depending on infection load of host cells. Graph represents the normalised median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test. (D) Representative microscopy image of HeLa cells infected with *S. flexneri* expressing the TSAR reporter at 3 h.p.i. Scale bar 10 µm. (E) Enlarged view of insets in D, highlighting a *S. flexneri* associated or not to SEPT7 assemblies. Scale bar, 1 µm. (F) Frequency of the number of SEPT7-associated bacteria identified using deep learning from experiments in F, from n= 5030 bacteria from 4 independent replicates. Graph represents mean ± SD (9.044% ± 5.582%). (G) Percentage of TSAR+ *S. flexneri* depending on their association to SEPT7. Graph represents mean ± SD (negatives: 2.214 ± 1.603, septin associated: 17.74 ± 5.138). Student’s t-test. (H) Normalised ratio between TSAR signal and mCherry in individual bacteria. Graph representes the median (negatives: 1, SEPT7 associated: 1.122), and interquartile range (negatives: 0.7674 – 1.286, SEPT7 associated: 0.9047 – 1.511). Mann-Whitney U test.

Discussion

Unravelling the intrinsic heterogeneity of host and bacterial cells underlying the infection process is paramount to design effective therapeutic strategies. High-content and high-throughput image analysis are ideal tools to capture this heterogeneity when coupled to automatic phenotypic image analysis (Eliceiri et al., 2012; Mattiazzi Usaj et al., 2016; Pegoraro & Misteli, 2017). Here we use high-content high-resolution microscopy and automated image analysis to quantitatively dissect the heterogeneity of S. flexneri in epithelial cells at the single-cell and population level at unprecedented subcellular detail.

We discover that S. flexneri induces important physiological changes in epithelial cells after 3h40min post infection. These include a marked decrease of DNA synthesis in cells in S-phase, a collective shutdown of protein translation, and a larger cellular and nuclear area, all in a bacterial-burden dependent manner. Collectively, these observations suggest S. flexneri infection induces host cell stress and a halt in the normal progression of the cell cycle. Consistent with this, S. flexneri has been shown to induce genotoxic stress (Bergounioux et al., 2012), inhibit translation and induce formation of stress granules (Zhang et al., 2024), and block the G2/M phase transition (Iwai et al., 2007; Wang et al., 2019). In the natural S. flexneri niche of the human gut, such an arrest would prevent maturation and shedding of infected epithelial cells, thereby promoting efficient gut colonisation (Iwai et al., 2007).

To understand how bacteria are targeted by cell-autonomus immune defences, such as septin cage entrapment, it is necessary to investigate phenotypic characteristics of targeted bacteria and the heterogeneous recruitment of cell-autonomous immune factors. While traditional image segmentation tools fail to enable this task, rapidly evolving developments in the field of computer vision are democratising the use of deep learning and CNNs for multiple applications (Berg et al., 2019; Hung et al., 2020; Stringer et al., 2020; von Chamier et al., 2021; Weigert et al., 2020) including infection biology (Davidson et al., 2021; Fisch et al., 2019b, 2021; Touquet et al., 2018). However, the analysis of complex datasets still requires a tailored approach, as shown here for bacterial interactions with the septin cytoskeleton (which is ubiquitious in the cell and forms diverse assemblies whose precise identifcation requires high resolution). We successfully implement a CNN based approach for the automated classification of bacteria associated with SEPT7 assemblies, enabling the analysis of thousands of bacterial cells while saving time and preventing human bias.

A large dataset of images depicting S. flexneri – SEPT7 interactions was acquired to train the deep learning algorithm and to characterize the diversity of these structures. We observe that the most frequent septin assembly is in the form of rings transversally surrounding the bacterial surface. We also detected alternative morphologies, including homogeneous SEPT7 labelling surrounding the bacterial surface (tight vs loose), as well as the discrete recruitment of SEPT7 to one or two bacterial poles. While previous work has shown that septin cage assembly around S. flexneri can start with the discrete recruitment of septins to one or two bacterial poles (Krokowski et al., 2018), it is not yet known if the different septin assemblies we describe here correspond to different stages or fates of the same structure or whether it marks separate biological events.

We use deep learning analysis to interrogate intracellular bacteria and show that S. flexneri associated with SEPT7 present heterogenous DNA and protein synthesis. In addition, we show that bacteria associated with SEPT7 have increased T3SS activity (as compared to bacteria not associated with SEPT7) indicating that SEPT7 is recruited to actively pathogenic bacteria. T3SS activity has been shown to be higher in motile bacteria (Campbell-Valois et al., 2014) which reinforces the antimicrobial role of septins in counteracting actin tail motility. Altogether, our deep learning tool to analyse septin-bacteria interactions opens the door to future studies on interactions between bacteria and components of cell-autonomous immunity.

Concluding remarks

Heterogeneity in a controlled lab environment is only a small representation of the broader heterogeneity underlying the clinical infection process, considering the genetic background of the patient, interactions with microbiota, etc. In the future, our pipeline can be applied to investigate septin-bacteria interactions among different Shigella spp. or clinically relevant strains, as well as other intracellular bacteria (eg. M. marinum, Staphylococcus aureus, Pseudomonas aeruginosa). In addition, our pipeline can be adapted to identify and score other cytoskeletal structures (such as actin tails) or membrane organelles (such as autophagosomes) associated with intracellular bacteria. Ultimately, we envision that our automated microscopy workflows will help to screen pharmacologic or genetic libraries in a multiparametric manner and understand their impact on host and bacterial cells at the single-cell and population level. In this way, novel host and bacterial factors underlying heterogeneity and bacterial pathogenesis could be identified, leading to transformative therapeutic strategies to combat bacterial infection in humans.

Materials and methods

Reagents

The following antibodies were used: rabbit anti-SEPT7 (1:100 dilution, #18991, IBL), Alexa-488-conjugated anti-rabbit antibody (1:500, #10729174, ThermoFisher Scientific), Alexa-647-conjugated anti-rabbit antibody (1:500, #A27040, ThermoFisher Scientific). The following dyes were used: Alexa-647-conjugated phalloidin (1:500, #10656353, ThermoFisher Scientific), Hoechst (1:500, #H3570, ThermoFisher Scientific), 0.01% (w/v) congo red (#C6767, Sigma-Aldrich). The following drugs were used: 50 μg/mL aphidicolin (#A0781 Sigma-Aldrich), 50 μg/mL cycloheximide (#2112 Cell Signalling), 2 μg/mL nalidixic acid (#N8878 Sigma-Aldrich), and 100 μg/ml rifampicin (#R3501 Sigma-Aldrich).

Bacterial strains and culture conditions

The bacterial strains and plasmids described in this study are listed in Sup. Table 1. Shigella flexneri 5a str. M90T mCherry was used throughout the manuscript unless indicated otherwise unless indicated otherwise. S. flexneri strains were grown in trypticase soy broth (TCS) agar containing congo red to select for red colonies, indicative of a functional T3SS. Conical polypropylene tubes (#CLS430828, Corning) containing 5 ml of TCS were inoculated with individual red colonies of S.flexneri and were grown ∼16 h at 37 °C with shaking at 200 rpm. The following day, bacterial cultures were diluted in fresh prewarmed TCS (1:50 or 1:100 v/v), and cultured until an optical density (OD, measured at 600 nm) of 0.6.

Cloning

Primers used in this study were designed using NEBuilder Assembly Tool (https://nebuilder.neb.com/#!/). To produce S. flexneri expressing AfaI and GFP, the pACAfaI plasmid was constructed (plasmids characteristics and primer sequences are located in Supplementary Table 1). Briefly, the afaI operon together with the amp^Rgene was amplified from S.flexneri afaI strain using the primers afaI-operon-fw and afaI-operon-rv. The ori p15A was amplified from pAC-P_lac using the primers p15A-ori-fw and p15A-ori-fw. Both DNA fragments were assembled using the Gibson assembly to generate the plasmid pAC-AfaI-AmpR. In a second step, the AmpR gene from pAC-AfaI-Amp was substituted by the CmR gene and generate pAC-AfaI-Cm. This was achieved by amplification of the CmR gene from pNeae2 using the primers CmR-fw CmR-rv and amplification of the backbone of pAC-AfaI-Amp with primers afaI-p15A-fw, afaI-p15A-rv, followed by Gibson assembly. Gibson assembly was performed at 50 °C for 30 min using the HiFi DNA Assembly Master Mix (#E2621L, New England Biolabs). Resulting plasmid pAC-AfaI-Cm was transformed into S. flexneri GFP by electroporation.

Mammalian cell culture

HeLa (ATCC CCL-2) cells were grown at 37 °C and 5% CO₂ in Dulbecco’s modified Eagle medium (DMEM, GIBCO) supplemented with 10% fetal bovine serum (FBS, Sigma-Aldrich).

Infection of human cells

Hela cells were seeded in 96-well microplates (PhenoPlates, Perkin Elmer) at a confluency of 10⁴ cells/well 2 days prior to infection. Alternatively, 9 × 10⁴ HeLa cells were seeded in 6-well plates (Thermo Scientific) containing 22 × 22 mm glass coverslips 2 days before the infection. Bacterial cultures were grown as described and cell cultures were infected with S. flexneri strains as described previously (Mostowy et al., 2010). Briefly, Hela cells were infected with S. flexneri by spin-inoculation at 110 g for 10 min at a multiplicity of infection (MOI, bacteria:cell) of 100:1. Then, microplates were placed at 37 °C and 5% CO₂ for 30 min. Infected cultures were washed 3 × with phosphate-buffered saline (PBS) pH 7.4 and incubated with fresh DMEM containing 10% FBS and 50 mg/mL gentamicin at 37 °C and 5% CO₂ for 3 or 4 h. In order to obtain more instances of S. flexneri associated to SEPT7 in Figure 4H, one of the datasets was obtained after infection with S. flexneri expressing the adhesin AfaI (Labigne-Roussel & Falkow, 1988) and GFP to increase host cell invasion at a MOI of 10 in the absence of spin inoculation.

Click chemistry

To label de novo DNA and protein synthesis, Click-iT™ EdU Cell Proliferation Kit for Imaging (#C10337 Thermofisher Scientific) and Click-iT™ AHA Protein Synthesis HCS Assay (#C10289 Thermofisher Scientific) were used, respectively, according to the manufacturer instructions. In the case of AHA, cells were cultured in glutamine, cystine and methionine-free DMEM (#21013024 Thermofisher Scientific) for 1 h prior to the treatment. 10μM EdU and 25μM AHA were supplemented to live bacterial or mammalian cell cultures for 1h. Cells were fixed for immunofluorescence as indicated below. Click azide/alkyne reactions were performed after permeabilization and prior to antibody staining as indicated below, according to the manufacturer’s instructions.

Immunofluorescence and fluorescence microscopy

Bacteria, coverslips or 96-well microplates containing adherent infected or uninfected cells were washed three times with PBS pH 7.4 and fixed 15 min in 4% paraformaldehyde (in PBS) at RT. Fixed cells were washed three times with PBS pH 7.4 and subsequently permeabilized 5min with 0.1% Triton X-100 (in PBS). Cells were then washed three times in PBS and incubated 1h with primary antibodies diluted in PBS supplemented with 0.1% Triton X-100 and 1% bovine serum albumin. Cells were then washed three times in PBS and incubated 45 min with anti-rabbit secondary antibodies diluted 0.1% Triton X-100 (in PBS), and Hoestch and Alexa-conjugated phalloidin where indicated. 96-well microplates were washed three times in PBS and preserved in 0.01% sodium azide (#10338380 Thermo Scientific) in PBS pH 7.4. Stained bacteria cultures and coverslips were washed three times with PBS pH 7.4 mounted on glass slides with ProLong™ Diamond Antifade Mountant (#P36970I Invitrogen). Fluorescence microscopy on stained bacteria was performed using a using a 63×/1.4 C-Plan Apo oil immersion lens on a Zeiss LSM 880 confocal microscope driven by ZEN Blacksoftware (v2.3). Microscopy images were obtained using z-stack image series taking 8–16 slices. Fluorescence microscopy on infected or uninfected cells was performed using a ZEISS Plan-APOCHROMAT 20× / 0.95 Autocorr Objective or a ZEISS Plan-APOCHROMAT 50×/1.2 water immersion lens coupled to a 0.5x tubelens on a Zeiss CellDiscoverer 7 with Airyscan detectors driven by ZEN Blue software (v3.5). Microscopy images were obtained using z-stack image series taking 32 slices. Confocal images were processed using Airyscan processing (Weinerfilter) using “Auto Filter” and “3D Processing” options.

Flow cytometry

5 x 10⁴ individual bacterial cells were analysed using flow cytometry with an LSRII flow cytometer (BD Biosciences). The data was analysed using FlowJo software, version 10.7.1. The median fluorescence of the bacterial population was determined and plotted for each staining.

Analysis pipelines

Number of bacteria per cell, as well as fluorescence of bacteria, nuclei and mammalian cells was analysed using Cellprofiler (v.4.0.7) (Carpenter et al., 2006). Custom made segmentation and deep learning pipelines for the classification of septin cages were written in ImageJ and Python using Keras.

Statistics

All graphs were plotted and statistical analysis were performed using Prism. n.s: non-significant, *: p-value < 0.05, **: p-value < 0.01, ***: p-value < 0.001).

Data availability

Due to the size of the data, representative datasets were deposited in Dryad (https://doi.org/10.5061/dryad.6wwpzgn5z). Materials and additional raw microscopy data can be obtained from the corresponding authors upon requests.

Code availability

Custom ImageJ macros, phyton scripts and Jupyter notebooks were annotated and deposited in Github (https://github.com/ATLopezJimenez/Toolset-high-content-analysis-of-Shigella-infection).

Acknowledgements

We thank members of the Mostowy and Brian Ho labs for helpful discussion. We thank François-Xavier Campbell-Valois for providing TSAR plasmids. We thank the instructors of the EMBL course Deep Learning for Image Analysis, in particular Prof. Pejman Rasti. A.T.L.J. was funded by the Swiss National Science Foundation Early Postdoc.Mobility Fellowship (P2GEP3_188277) and the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska - Curie grant agreement no. H2020-MSCA-IF-2020-895330. D.B. was supported by the Deutsche Forschungsgemeinschaft (DFG) Walter Benjamin Programme (BR 6637/1-1). G.Ö.G. is funded by the Human Frontier Science Program (LT000436/2021-L). Research in S. Mostowy laboratory is supported by a Wellcome Trust Senior Research Fellowship (206444/Z/17/Z) and European Research Council Consolidator Grant (772853 - ENTRAPMENT).

Supplementary Figure Legends

*S. flexneri* infection induces host cell size changes
(A) Hoestch fluorescence intensity of sum-projected nuclei as proxy of nuclear volume. Graph represents normalised median (uninfected: 1, infected: 1.294) and interquartile range (uninfected: 0.7588 – 1.287, infected: 0.9534 – 1.878) of n=2071 uninfected cells and n=195 infected cells from 3 independent experiments. Mann-Whitney U test. (B – C) Cell and nuclear area in presence and absence of infection and Latrunculin B. Graph represents the median and interquartile range. Kruskal-Wallis test and Dunn’s multiple comparisons test.

Click-IT Edu and Click-IT AHA specifically label *de novo* DNA and protein synthesis in Hela cells
(A) Representative widefield image of HeLa cells after Click-IT Edu reaction in the presence or absence of Aphidicolin to inhibit eukaryotic DNA replication. Scale bar, 10 µm. (B) Quantification of *de novo* DNA synthesis in single HeLa cells in A as the ratio of nuclear Click-IT Edu signal to Hoechst. Graph represents the normalised median (untreated: 1, Aphidicolin: 0.7667) and interquartile range (untreated: 0.8474 – 6.571, Aphidicolin: 0.6816 – 0.8543) from 15828 untreated and 14943 treated cells from 3 independent experiments. Mann-Whitney U test. (C) Representative Airyscan confocal images of HeLa cells after Click-IT AHA reaction in the presence or absence of Cycloheximide to block eukaryotic translational elongation. Scale bar, 10 µm. (D) Quantification of cytoplasmic Click-IT AHA signal of the HeLa cells in A. Graph represents normalised median (untreated: 1, Cycloheximide: 0.3531) and interquartile range (untreated :0.6794 – 1.401, Cycloheximide: 0.2504 – 0.6637) of n=268 untreated and n= 258 Cycloheximide treated cells from 3 independent experiments. Mann-Whitney U test.

Additional examples of the annotated datasets for deep learning
(A) Representative images of multiple bacteria associated to SEPT7. Scale bar, 1 µm. (B) Representative images of the training and validation dataset to identify single bacteria versus clumps. Scale bar, 5 µm. (C) Representative images of the training and validation dataset to identify SEPT7-positive versus SEPT7-negative bacteria, after processing. Scale bar, 5 µm.

Probability maps of SEPT7 around *S. flexneri*
(A, C) Median of the signal for mCherry and SEPT7 in both datasets (SEPT7-positive and SEPT7-negative), centered at the bacterial midbody or bacterial pole, respectively. Only bacteria with width ≥ 2.5 µm were selected to ensure bacteria were mostly transversal to the acquisition axis. Scale bar, 1 µm. (B,C) Quantification of SEPT7 signal and *S. flexneri* mCherry signal in the midbody projection depicted in A and transversal dash line indicated in B.

Click-IT Edu and Click-IT AHA specifically label *de novo* DNA and protein translation in *S. flexneri*
(A) Representative Airyscan image of mCherry *S. flexneri* after Click-IT Edu reaction in the presence or absence of Nalidixic acid to block bacterial DNA replication. Scale bar, 1 µm. (B) Quantification of Click-IT Edu signal of bacteria in C measured by FACS. Graph represents normalised mean ± SD. One-way ANOVA and Tukey’s multiple comparisons test. (C) Representative Airyscan confocal images of HeLa cells infected with mCherry expressing *S. flexneri* after Click-IT Edu reaction in the presence or absence of Nalidixic acid to inhibit bacterial DNA nuclear replication. Scale bar, 10 µm. (D) Quantification of *de novo* DNA synthesis in single bacteria in E as the ratio of nuclear Click-IT signal to Hoechst. Graph represents normalised median (untreated: 1, Nalidixic acid: 0.0646) and interquartile range (untreated: 0.3589 – 1.924, Nalidixic acid: 0.0350 – 0.3080) of 9310 untreated and 4368 Nalidixic acid treated bacteria from 3 independent experiments. Mann-Whitney U test. (E) Representative Airyscan confocal images of HeLa cells infected with mCherry expressing *S. flexneri* after Click-IT AHA reaction in the presence or absence of Rifampicin to block prokaryotic translation. Scale bar, 10 µm; insets, 1 µm. (F) Quantification of cytoplasmic Click-IT AHA signal of individual *S. flexneri* cells in C. Graph represents normalised median (untreated: 1, Rifampicin: 0.8263) and interquartile range (untreated: 0.7520 – 1.350, Rifampicin: 0.6143 – 1.100) of n=2309 untreated and n=345 Rifampicin treated bacteria from 3 independent experiments.

TSAR labels T3SS activity specifically
(A) Representative Airyscan confocal images of *S. flexneri* harbouring the TSAR plasmid grown in broth in the presence or absence of congo red to induce the expression of the T3SS. Scale bar, 1 µm. (B) Quantification of the TSAR dependent GFP signal of bacteria in A measured by FACS. Graph represents the normalised mean ± SD (untreated: 1 ± 0, congo red: 155.4 ± 125.2) from 4 independent experiments. (C) Representative examples of TSAR-positive *S. flexneri* associated to SEPT7. C stands for “cage”, P stands for “protrusion”. Scale bar 5 µm.

References

1. Ackermann M
2015A functional perspective on phenotypic heterogeneity in microorganismsNature Publishing Group 13https://doi.org/10.1038/nrmicro3491
1. Avraham R.
2. Haseley N.
3. Brown D.
4. Penaranda C.
5. Jijon H. B.
6. Trombetta J. J.
7. Satija R.
8. Shalek A. K.
9. Xavier R. J.
10. Regev A.
11. Hung D. T
2015Pathogen Cell-to-Cell Variability Drives Heterogeneity in Host Immune ResponsesCell 162:1309–1321https://doi.org/10.1016/J.CELL.2015.08.027
1. Aylan B.
2. Botella L.
3. Gutierrez M. G.
4. Santucci P
2023High content quantitative imaging of Mycobacterium tuberculosis responses to acidic microenvironments within human macrophagesFEBS Open Bio 13:1204–1217https://doi.org/10.1002/2211-5463.13537
1. Berg S.
2. Kutra D.
3. Kroeger T.
4. Straehle C. N.
5. Kausler B. X.
6. Haubold C.
7. Schiegg M.
8. Ales J.
9. Beier T.
10. Rudy M.
11. Eren K.
12. Cervantes J. I.
13. Xu B.
14. Beuttenmueller F.
15. Wolny A.
16. Zhang C.
17. Koethe U.
18. Hamprecht F. A.
19. Kreshuk A
2019ilastik: interactive machine learning for (bio)image analysisNature Methods 2019 16:12 16:1226–1232https://doi.org/10.1038/s41592-019-0582-9
1. Bergounioux J.
2. Elisee R.
3. Prunier A. L.
4. Donnadieu F.
5. Sperandio B.
6. Sansonetti P.
7. Arbibe L
2012Calpain Activation by the Shigella flexneri Effector VirA Regulates Key Steps in the Formation and Life of the Bacterium’s Epithelial NicheCell Host & Microbe 11:240–252https://doi.org/10.1016/J.CHOM.2012.01.013
1. Bernardini M. L.
2. Mounier J.
3. D’Hauteville H.
4. Coquis-Rondon M.
5. Sansonetti P. J
1989Identification of icsA, a plasmid locus of Shigella flexneri that governs bacterial intra- and intercellular spread through interaction with F-actinProceedings of the National Academy of Sciences of the United States of America 86:3867–3871https://doi.org/10.1073/PNAS.86.10.3867
1. Boddy K. C.
2. Gao A. D.
3. Truong D.
4. Kim M. S.
5. Froese C. D.
6. Trimble W. S.
7. Brumell J. H
2018Septin-regulated actin dynamics promote Salmonella invasion of host cellsCellular Microbiology 20https://doi.org/10.1111/CMI.12866
1. Brodin P.
2. Christophe T
2011High-content screening in infectious diseasesCurrent Opinion in Chemical Biology 15:534–539https://doi.org/10.1016/J.CBPA.2011.05.023
1. Campbell-Valois F. X.
2. Schnupf P.
3. Nigro G.
4. Sachse M.
5. Sansonetti P. J.
6. Parsot C
2014A Fluorescent Reporter Reveals On/Off Regulation of the Shigella Type III Secretion Apparatus during Entry and Cell-to-Cell SpreadCell Host & Microbe 15:177–189https://doi.org/10.1016/J.CHOM.2014.01.005
1. Carpenter A. E.
2. Jones T. R.
3. Lamprecht M. R.
4. Clarke C.
5. Kang I. H.
6. Friman O.
7. Guertin D. A.
8. Chang J. H.
9. Lindquist R. A.
10. Moffat J.
11. Golland P.
12. Sabatini D. M
2006CellProfiler: Image analysis software for identifying and quantifying cell phenotypesGenome Biology 7:1–11https://doi.org/10.1186/GB-2006-7-10-R100/FIGURES/4
1. Davidson M. S.
2. Andradi-Brown C.
3. Yahiya S.
4. Chmielewski J.
5. O’Donnell A. J.
6. Gurung P.
7. Jeninga M. D.
8. Prommana P.
9. Andrew D. W.
10. Petter M.
11. Uthaipibull C.
12. Boyle M. J.
13. Ashdown G. W.
14. Dvorin J. D.
15. Reece S. E.
16. Wilson D. W.
17. Cunningham K. A.
2021Automated detection and staging of malaria parasites from cytological smears using convolutional neural networksBiological Imaging 1https://doi.org/10.1017/S2633903X21000015
1. Day N. J.
2. Santucci P.
3. Gutierrez M. G
2023Host cell environments and antibiotic efficacy in tuberculosisTrends in Microbiology https://doi.org/10.1016/J.TIM.2023.08.009
1. Deboosere N.
2. Belhaouane I.
3. Machelart A.
4. Hoffmann E.
5. Vandeputte A.
6. Brodin P
2021High-Content Analysis Monitoring Intracellular Trafficking and Replication of Mycobacterium tuberculosis Inside Host CellsMethods in Molecular Biology 2314:649–702https://doi.org/10.1007/978-1-0716-1460-0_29/COVER
1. Dramé M.
2. Garcia-Rodriguez F. J.
3. Buchrieser C.
4. Escoll P
2023High-content assay to measure mitochondrial function and bacterial vacuole size in infected human primary macrophagesSTAR Protocols 4https://doi.org/10.1016/J.XPRO.2023.102175
1. Egile C.
2. Loisel T. P.
3. Laurent V.
4. Li R.
5. Pantaloni D.
6. Sansonetti P. J.
7. Carlier M. F
1999Activation of the Cdc42 Effector N-Wasp by the Shigella flexneri Icsa Protein Promotes Actin Nucleation by Arp2/3 Complex and Bacterial Actin-Based MotilityJournal of Cell Biology 146:1319–1332https://doi.org/10.1083/JCB.146.6.1319
1. Eliceiri K. W.
2. Berthold M. R.
3. Goldberg I. G.
4. Ibáñez L.
5. Manjunath B. S.
6. Martone M. E.
7. Murphy R. F.
8. Peng H.
9. Plant A. L.
10. Roysam B.
11. Stuurmann N.
12. Swedlow J. R.
13. Tomancak P.
14. Carpenter A. E
2012Biological imaging software toolsNature Methods 2012 9:7 9:697–710https://doi.org/10.1038/nmeth.2084
1. Ennis H. L.
2. Lubin M
1964Cycloheximide: Aspects of inhibition of protein synthesis in mammalian cellsScience 146:1474–1476https://doi.org/10.1126/SCIENCE.146.3650.1474
1. Ezekiel D. H.
2. Hutchins J. E
1968Mutations affecting RNA Polymerase associated with Rifampicin Resistance in Escherichia coliNature 1968 220:5164 220:276–277https://doi.org/10.1038/220276a0
1. Fisch D.
2. Evans R.
3. Clough B.
4. Byrne S. K.
5. Channell W. M.
6. Dockterman J.
7. Frickel E. M
2021HRMAn 2.0: Next-generation artificial intelligence–driven analysis for broad host–pathogen interactionsCellular Microbiology 23https://doi.org/10.1111/CMI.13349
1. Fisch D.
2. Yakimovich A.
3. Clough B.
4. Wright J.
5. Bunyan M.
6. Howell M.
7. Mercer J.
8. Frickel E
2019Defining host–pathogen interactions employing an artificial intelligence workflowELife 8https://doi.org/10.7554/ELIFE.40560
1. Fisch D.
2. Yakimovich A.
3. Clough B.
4. Wright J.
5. Bunyan M.
6. Howell M.
7. Mercer J.
8. Frickel E
2019Defining host–pathogen interactions employing an artificial intelligence workflowELife 8https://doi.org/10.7554/ELIFE.40560
1. Global Burden of Disease Collaborative Network
2020Global Burden of Disease Study 2019 (GBD 2019)Institute for Health Metrics and Evaluation (IHME)
1. Gutierrez M. G.
2. Enninga J
2022Intracellular niche switching as host subversion strategy of bacterial pathogensCurrent Opinion in Cell Biology 76https://doi.org/10.1016/J.CEB.2022.102081
1. Hung J.
2. Goodman A.
3. Ravel D.
4. Lopes S. C. P.
5. Rangel G. W.
6. Nery O. A.
7. Malleret B.
8. Nosten F.
9. Lacerda M. V. G.
10. Ferreira M. U.
11. Rénia L.
12. Duraisingh M. T.
13. Costa F. T. M.
14. Marti M.
15. Carpenter A. E
2020Keras R-CNN: Library for cell detection in biological images using deep neural networksBMC Bioinformatics 21:1–7https://doi.org/10.1186/S12859-020-03635-X/FIGURES/1
1. Iwai H.
2. Kim M.
3. Yoshikawa Y.
4. Ashida H.
5. Ogawa M.
6. Fujita Y.
7. Muller D.
8. Kirikae T.
9. Jackson P. K.
10. Kotani S.
11. Sasakawa C
2007A Bacterial Effector Targets Mad2L2, an APC Inhibitor, to Modulate Host Cell CyclingCell 130:611–623https://doi.org/10.1016/J.CELL.2007.06.043
1. Krizhevsky A.
2. Sutskever I.
3. Hinton G. E.
2024Krizhevsky, A., Sutskever, I., & Hinton, G. E. (n.d.). ImageNet Classification with Deep Convolutional Neural Networks. Retrieved 13 February 2024, from http://code.google.com/p/cuda-convnet/
1. Krokowski S.
2. Lobato-Márquez D.
3. Chastanet A.
4. Pereira P. M.
5. Angelis D.
6. Galea D.
7. Larrouy-Maumus G.
8. Henriques R.
9. Spiliotis E. T.
10. Carballido-López R.
11. Mostowy S.
2018Septins Recognize and Entrap Dividing Bacterial Cells for Delivery to LysosomesCell Host & Microbe 24:866–874https://doi.org/10.1016/j.chom.2018.11.005
1. Labigne-Roussel A.
2. Falkow S.
1988Distribution and degree of heterogeneity of the afimbrial-adhesin-encoding operon (afa) among uropathogenic Escherichia coli isolatesInfection and Immunity 56:640–648https://doi.org/10.1128/iai.56.3.640-648.1988
1. Lecun Y.
2. Bengio Y.
3. Hinton G
2015Deep learningNature 2015 521:7553 521:436–444https://doi.org/10.1038/nature14539
1. Lensen A.
2. Gomes M. C.
3. López-Jiménez A. T.
4. Mostowy S
2023An automated microscopy workflow to study Shigella–neutrophil interactions and antibiotic efficacy in vivoDMM Disease Models and Mechanisms 16https://doi.org/10.1242/DMM.049908/307310/AM/AN-AUTOMATED-MICROSCOPY-WORKFLOW-TO-STUDY-SHIGELLA
1. Lobato-Márquez D.
2. Conesa J. J.
3. López-Jiménez A. T.
4. Divine M. E.
5. Pruneda J. N.
6. Mostowy S
2023Septins and K63 ubiquitin chains are present in separate bacterial microdomains during autophagy of entrapped ShigellaJournal of Cell Science 136https://doi.org/10.1242/JCS.261139/VIDEO-6
1. Lobato-Márquez D.
2. Moreno-Córdoba I.
3. Figueroa V.
4. Diáz-Orejas R.
5. Garciá-Del Portillo F
2015Distinct type I and type II toxin-antitoxin modules control Salmonella lifestyle inside eukaryotic cellsScientific Reports 2015 5:1 5:1–10https://doi.org/10.1038/srep09374
1. Lobato-Márquez D.
2. Xu J.
3. Güler G. Ö.
4. Ojiakor A.
5. Pilhofer M.
6. Mostowy S
2021Mechanistic insight into bacterial entrapment by septin cage reconstitutionNature Communications 2021 12:1 12:1–14https://doi.org/10.1038/s41467-021-24721-5
1. Martins C. S.
2. Taveneau C.
3. Castro-Linares G.
4. Baibakov M.
5. Buzhinsky N.
6. Eroles M.
7. Milanović V.
8. Omi S.
9. Pedelacq J.-D.
10. Iv F.
11. Bouillard L.
12. Llewellyn A.
13. Gomes M.
14. Belhabib M.
15. Kuzmić M.
16. Verdier-Pinard P.
17. Lee S.
18. Badache A.
19. Kumar S.
20. Mavrakis M
2023Human septins organize as octamer-based filaments and mediate actin-membrane anchoring in cellsJournal of Cell Biology 222https://doi.org/10.1083/JCB.202203016/213778
1. Mattiazzi Usaj M.
2. Styles E. B.
3. Verster A. J.
4. Friesen H.
5. Boone C.
6. Andrews B. J.
2016High-Content Screening for Quantitative Cell BiologyTrends in Cell Biology 26:598–611https://doi.org/10.1016/J.TCB.2016.03.008
1. McDonald P. C.
2. Dedhar S
2024Persister cell plasticity in tumour drug resistanceSeminars in Cell & Developmental Biology 156:1–10https://doi.org/10.1016/J.SEMCDB.2023.11.003
1. Mostowy S.
2. Bonazzi M.
3. Hamon M. A.
4. Tham T. N.
5. Mallet A.
6. Lelek M.
7. Gouin E.
8. Demangel C.
9. Brosch R.
10. Zimmer C.
11. Sartori A.
12. Kinoshita M.
13. Lecuit M.
14. Cossart P
2010Entrapment of Intracytosolic Bacteria by Septin Cage-like StructuresCell Host & Microbe 8:433–444https://doi.org/10.1016/J.CHOM.2010.10.009
1. Mostowy S.
2. Boucontet L.
3. Mazon Moya M. J.
4. Sirianni A.
5. Boudinot P.
6. Hollinshead M.
7. Cossart P.
8. Herbomel P.
9. Levraud J.-P. P.
10. Colucci-Guyon E
2013The Zebrafish as a New Model for the In Vivo Study of Shigella flexneri Interaction with Phagocytes and Bacterial AutophagyPLoS Pathogens 9https://doi.org/10.1371/journal.ppat.1003588
1. Mostowy S.
2. Cossart P
2012Septins: the fourth component of the cytoskeletonNature Reviews Molecular Cell Biology 13:183–194https://doi.org/10.1038/nrm3284
1. Mostowy S.
2. Nam Tham T.
3. Danckaert A.
4. Guadagnini S.
5. Boisson-Dupuis S.
6. Pizarro-Cerdá J.
7. Cossart P
2009Septins Regulate Bacterial Entry into Host CellsPLoS ONE 4https://doi.org/10.1371/journal.pone.0004196
1. Nielsen M. A
2015Neural Networks and Deep LearningDetermination press
1. Ortega F. E.
2. Koslover E. F.
3. Theriot J. A
2019Listeria monocytogenes cell-to-cell spread in epithelia is heterogeneous and dominated by rare pioneer bacteriaELife 8https://doi.org/10.7554/ELIFE.40032
1. Parsot C.
2. Ménard R.
3. Gounon P.
4. Sansonetti P. J
1995Enhanced secretion through the Shigella flexneri Mxi-Spa translocon leads to assembly of extracellular proteins into macromolecular structuresMolecular Microbiology 16:291–300https://doi.org/10.1111/J.1365-2958.1995.TB02301.X
1. Pedrali-Noy G.
2. Spadari S.
3. Miller-Faurès A.
4. Miller A. O. A.
5. Kruppa J.
6. Koch G
1980Synchronization of HeLa cell cultures by inhibition of DNA polymerase alpha with aphidicolinNucleic Acids Research 8https://doi.org/10.1093/NAR/8.2.377
1. Pegoraro G.
2. Misteli T
2017High-Throughput Imaging for the Discovery of Cellular Mechanisms of DiseaseTrends in Genetics 33:604–615https://doi.org/10.1016/J.TIG.2017.06.005
1. Pylkkö T.
2. Ilina P.
3. Tammela P
2021Development and validation of a high-content screening assay for inhibitors of enteropathogenic E. coli adhesionJournal of Microbiological Methods 184https://doi.org/10.1016/J.MIMET.2021.106201
1. Ray K.
2. Bobard A.
3. Danckaert A.
4. Paz-Haftel I.
5. Clair C.
6. Ehsani S.
7. Tang C.
8. Sansonetti P.
9. Van Nhieu G. T.
10. Enninga J.
2010Tracking the dynamic interplay between bacterial and host factors during pathogen-induced vacuole rupture in real timeCellular Microbiology 12:545–556https://doi.org/10.1111/J.1462-5822.2010.01428.X
1. Robertin S.
2. Brokatzky D.
3. Lobato-Márquez D.
4. Mostowy S
2023Regulation of integrin α5β1-mediated Staphylococcus aureus cellular invasion by the septin cytoskeletonEuropean Journal of Cell Biology 102https://doi.org/10.1016/J.EJCB.2023.151359
1. Robertin S.
2. Mostowy S
2020The history of septin biology and bacterial infectionCellular Microbiology 22https://doi.org/10.1111/CMI.13173
1. Santucci P.
2. Greenwood D. J.
3. Fearns A.
4. Chen K.
5. Jiang H.
6. Gutierrez M. G
2021Intracellular localisation of Mycobacterium tuberculosis affects efficacy of the antibiotic pyrazinamideNature Communications 2021 12:1 12:1–15https://doi.org/10.1038/s41467-021-24127-3
1. Schnupf P.
2. Sansonetti P. J.
2019Shigella Pathogenesis: New Insights through Advanced MethodologiesBacteria and Intracellularity American Society of Microbiology :15–39https://doi.org/10.1128/microbiolspec.bai-0023-2019
1. Soroor F.
2. Kim M. S.
3. Palander O.
4. Balachandran Y.
5. Collins R. F.
6. Benlekbir S.
7. Rubinstein J. L.
8. Trimble W. S
2021Revised subunit order of mammalian septin complexes explains their in vitro polymerization propertiesMolecular Biology of the Cell 32:289–300https://doi.org/10.1091/MBC.E20-06-0398/ASSET/IMAGES/LARGE/MBC-32-289-G004.JPEG
1. Spiliotis E. T.
2. Nakos K
2021Cellular functions of actin- and microtubule-associated septinsCurrent Biology 31:R651–R666https://doi.org/10.1016/J.CUB.2021.03.064
1. Stringer C.
2. Wang T.
3. Michaelos M.
4. Pachitariu M.
2020Cellpose: a generalist algorithm for cellular segmentationNature Methods 2020 18:1 18:100–106https://doi.org/10.1038/s41592-020-01018-x
1. Sugino A.
2. Peebles C. L.
3. Kreuzer K. N.
4. Cozzarelli N. R
1977Mechanism of action of nalidixic acid: Purification of Escherichia coli nalA gene product and its relationship to DNA gyrase and a novel nicking-closing enzymeProceedings of the National Academy of Sciences 74:4767–4771https://doi.org/10.1073/PNAS.74.11.4767
1. Szuba A.
2. Bano F.
3. Castro-Linares G.
4. Iv F.
5. Mavrakis M.
6. Richter R. P.
7. Bertin A.
8. Koenderink G. H
2021Membrane binding controls ordered self-assembly of animal septinsELife 10https://doi.org/10.7554/ELIFE.63349
1. Toniolo C.
2. Rutschmann O.
3. McKinney J. D
2021Do chance encounters between heterogeneous cells shape the outcome of tuberculosis infections?Current Opinion in Microbiology 59:72–78https://doi.org/10.1016/J.MIB.2020.08.008
1. Touquet B.
2. Pelissier L.
3. Cavailles P.
4. Yi W.
5. Bellini V.
6. Mercier C.
7. Cesbron-Delauw M. F.
8. Boumendjel A.
9. Aldebert D
2018High-content imaging assay to evaluate Toxoplasma gondii infection and proliferation: A multiparametric assay to screen new compoundsPLOS ONE 13https://doi.org/10.1371/JOURNAL.PONE.0201678
1. van Tartwijk F. W.
2. Kaminski C. F.
2022Protein Condensation, Cellular Organization, and Spatiotemporal Regulation of Cytoplasmic PropertiesAdvanced Biology 6https://doi.org/10.1002/ADBI.202101328
1. von Chamier L.
2. Laine R. F.
3. Jukkala J.
4. Spahn C.
5. Krentzel D.
6. Nehme E.
7. Lerche M.
8. Hernández-Pérez S.
9. Mattila P. K.
10. Karinou E.
11. Holden S.
12. Solak A. C.
13. Krull A.
14. Buchholz T. O.
15. Jones M. L.
16. Royer L. A.
17. Leterrier C.
18. Shechtman Y.
19. Jug F.
20. Henriques R.
2021Democratising deep learning for microscopy with ZeroCostDL4MicNature Communications 2021 12:1 12:1–18https://doi.org/10.1038/s41467-021-22518-0
1. Wang X.
2. Pernicone N.
3. Pertz L.
4. Hua D.
5. Zhang T.
6. Listovsky T.
7. Xie W
2019REV7 has a dynamic adaptor region to accommodate small GTPase RAN/Shigella IpaB ligands, and its activity is regulated by the RanGTP/GDP switchJournal of Biological Chemistry 294:15733–15742https://doi.org/10.1074/JBC.RA119.010123
1. Weigert M.
2. Schmidt U.
3. Haase R.
4. Sugawara K.
5. Myers G
2020Star-convex Polyhedra for 3D Object Detection and Segmentation in Microscopy2020 IEEE Winter Conference on Applications of Computer Vision (WACV) https://doi.org/10.1109/WACV45572.2020.9093435
1. Woods B. L.
2. Gladfelter A. S
2021The state of the septin cytoskeleton from assembly to functionCurrent Opinion in Cell Biology 68:105–112https://doi.org/10.1016/J.CEB.2020.10.007
1. Zhang Q.
2. Xian W.
3. Li Z.
4. Lu Q.
5. Chen X.
6. Ge J.
7. Tang Z.
8. Liu B.
9. Chen Z.
10. Gao X.
11. Hottiger M. O.
12. Zhang P.
13. Qiu J.
14. Shao F.
15. Liu X
2024Shigella induces stress granule formation by ADP-riboxanation of the eIF3 complexCell Reports 43https://doi.org/10.1016/J.CELREP.2024.113789

Article and author information

Author information

Ana T. López-Jiménez
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
ORCID iD: 0000-0002-0289-738X
- Corresponding authors: ana.lopezjimenez@lshtm.ac.uk, serge.mostowy@lshtm.ac.uk
Dominik Brokatzky
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
Kamla Pillay
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
Tyrese Williams
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
Gizem Özbaykal Güler
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
Serge Mostowy
Department of Infection Biology, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, United Kingdom
ORCID iD: 0000-0002-7286-6503
- Corresponding authors: ana.lopezjimenez@lshtm.ac.uk, serge.mostowy@lshtm.ac.uk

Version history

Sent for peer review: March 6, 2024
Preprint posted: March 8, 2024
Reviewed Preprint version 1: June 4, 2024

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Not revised: This Reviewed Preprint includes the authors’ original preprint (without revision), an eLife assessment, public reviews, and a provisional response from the authors.

Reviewing Editor
Ethel Bayer-Santos
The University of Texas at Austin, Austin, United States of America
Senior Editor
Wendy Garrett
Harvard T.H. Chan School of Public Health, Boston, United States of America

Reviewer #1 (Public Review):

Summary:

In this study, López-Jiménez and colleagues demonstrated the utility of using high-content microscopy in dissecting host and bacterial determinants that play a role in the establishment of infection using Shigella flexneri as a model. The manuscript nicely identifies that infection with Shigella results in a block to DNA replication and protein synthesis. At the same time, the host responds, in part, via the entrapment of Shigella in septin cages.

Strengths:

The main strength of this manuscript is its technical aspects. They nicely demonstrate how an automated microscopy pipeline coupled with artificial intelligence can be used to gain new insights regarding elements of bacterial pathogenesis, using Shigella flexneri as a model system. Using this pipeline enabled the investigators to enhance the field's general understanding regarding the role of septin cages in responding to invading Shigella. This platform should be of interest to those who study a variety of intracellular microbial pathogens.

Another strength of the manuscript is the demonstration - using cell biology-based approaches- that infection with Shigella blocks DNA replication and protein synthesis. These observations nicely dovetail with the prior findings of other groups. Nevertheless, their clever click-chemistry-based approaches provide visual evidence of these phenomena and should interest many.

Weaknesses:

There are two main weaknesses of this work. First, the studies are limited to findings obtained using a single immortalized cell line. It is appreciated that HeLa cells serve as an excellent model for studying aspects of Shigella pathogenesis and host responses. However, it would be nice to see that similar observations are observed with an epithelial cell line of intestinal, preferably colonic origin, and eventually, with a non-immortalized cell line, although it is appreciated that the latter studies are beyond the scope of this work.

The other weakness is that the studies are minimally mechanistic. For example, the investigators have data to suggest that infection with Shigella leads to an arrest in DNA replication and protein synthesis; however, no follow-up studies have been conducted to determine how these host cell processes are disabled. Interestingly, Zhang and colleagues recently identified that the Shigella OspC effectors target eukaryotic translation initiation factor 3 to block host cell translation (PMID: 38368608). This paper should be discussed and cited in the discussion.

https://doi.org/10.7554/eLife.97495.1.sa2

Reviewer #2 (Public Review):

Summary:

Septin caging has emerged as one of the innate immune responses of eukaryotic cells to infections by intracellular bacteria. This fascinating assembly of eukaryotic proteins into complex structures restricts bacteria motility within the cytoplasm of host cells, thereby facilitating recognition by cytosolic sensors and components of the autophagy machinery. Given the different types of septin caging that have been described thus far, a single-cell, unbiased approach to quantify and characterise septin recruitment at bacteria is important to fully grasp the role and function of caging. Thus, the authors have developed an automated image analysis pipeline allowing bacterial segmentation and classification of septin cages that will be very useful in the future, applied to study the role of host and bacterial factors, compare different bacterial strains, or even compare infections by clinical isolates.

Strengths:

The authors developed a solid pipeline that has been thoroughly validated. When tested on infected cells, automated analysis corroborated previous observations and allowed the unbiased quantification of the different types of septin cages as well as the correlation between caging and bacterial metabolic activity. This approach will prove an essential asset in the further characterisation of septin cages for future studies.

Weaknesses:

As the main aim of the manuscript is to describe the newly developed analysis pipeline, the results illustrated in the manuscript are essentially descriptive. The developed pipeline seems exceptionally efficient in recognising septin cages in infected cells but its application for a broader purpose or field of study remains limited.

https://doi.org/10.7554/eLife.97495.1.sa1

Reviewer #3 (Public Review):

Summary:

The manuscript uses high-content imaging and advanced image-analysis tools to monitor the infection of epithelial cells by Shigella. They perform some analysis on the state of the cells (through measurements of DNA and protein synthesis), and then they focus on differential recruitment of Sept7 to the bacteria. They link this recruitment with the activity of the bacterial T3SS, which is a very interesting discovery. Overall, I found numerous exciting elements in this manuscript, and I have a couple of reservations. Please see below for more details on my reservations. Nevertheless, I think that these issues can be addressed by the authors, and doing so will help to make it a convincing and interesting piece for the community working on intracellular pathogens. The authors should also carefully re-edit their manuscript to avoid overselling their data (see below for issues I see there). I would consider taking out the first figure and starting with Figure 3 (Figure 2 could be re-organized in the later parts)- that could help to make the flow of the manuscript better.

Strengths:

The high-content analysis including the innovative analytical workflows are very promising and could be used by a large number of scientists working on intracellular bacteria.

The finding that Septins (through SEPT7) are differentially regulated through actively secreting bacteria is very exciting and can steer novel research directions.

Weaknesses:

The manuscript makes a connection between two research lines (1: Shigella infection and DNA/protein synthesis, 2: regulation of septins around invading Shigella) that are not fully developed - this makes it sometimes difficult to understand the take-home messages of the authors.

It is not clear whether the analysis that was done on projected images actually reflects the phenotypes of the original 3D data. This issue needs to be carefully addressed.

https://doi.org/10.7554/eLife.97495.1.sa0

Author response:

Reviewer #1 (Public Review):

Summary:

In this study, López-Jiménez and colleagues demonstrated the utility of using high-content microscopy in dissecting host and bacterial determinants that play a role in the establishment of infection using Shigella flexneri as a model. The manuscript nicely identifies that infection with Shigella results in a block to DNA replication and protein synthesis. At the same time, the host responds, in part, via the entrapment of Shigella in septin cages.

Strengths:

The main strength of this manuscript is its technical aspects. They nicely demonstrate how an automated microscopy pipeline coupled with artificial intelligence can be used to gain new insights regarding elements of bacterial pathogenesis, using Shigella flexneri as a model system. Using this pipeline enabled the investigators to enhance the field's general understanding regarding the role of septin cages in responding to invading Shigella. This platform should be of interest to those who study a variety of intracellular microbial pathogens.

Another strength of the manuscript is the demonstration - using cell biology-based approaches- that infection with Shigella blocks DNA replication and protein synthesis. These observations nicely dovetail with the prior findings of other groups. Nevertheless, their clever click-chemistry-based approaches provide visual evidence of these phenomena and should interest many.

We thank the Reviewer for their enthusiasm on the technical aspects of this paper, regarding both the automated microscopy pipeline coupled with artificial intelligence and the click-chemistry based approaches to dissect DNA replication and protein synthesis by microscopy.

Weaknesses:

There are two main weaknesses of this work. First, the studies are limited to findings obtained using a single immortalized cell line. It is appreciated that HeLa cells serve as an excellent model for studying aspects of Shigella pathogenesis and host responses. However, it would be nice to see that similar observations are observed with an epithelial cell line of intestinal, preferably colonic origin, and eventually, with a non-immortalized cell line, although it is appreciated that the latter studies are beyond the scope of this work.

The immortalized cell line HeLa is widely regarded as a paradigm to study infection by Shigella and other intracellular pathogens. However, we agree that future studies beyond the scope of this work should include other cell lines (eg. epithelial cells of colonic origin, macrophages, primary cells).

The other weakness is that the studies are minimally mechanistic. For example, the investigators have data to suggest that infection with Shigella leads to an arrest in DNA replication and protein synthesis; however, no follow-up studies have been conducted to determine how these host cell processes are disabled. Interestingly, Zhang and colleagues recently identified that the Shigella OspC effectors target eukaryotic translation initiation factor 3 to block host cell translation (PMID: 38368608). This paper should be discussed and cited in the discussion.

We appreciate the Reviewer’s concern about the lack of follow up work on observations of host DNA and protein synthesis arrest upon Shigella infection, which will be the focus of future studies. We acknowledge the recent work of Zhang et al. (Cell Reports, 2024) considering their similar results on protein translation arrest, and we fully agree that this reference should be more fully discussed in a revised version of the manuscript.

Reviewer #2 (Public Review):

Summary:

Septin caging has emerged as one of the innate immune responses of eukaryotic cells to infections by intracellular bacteria. This fascinating assembly of eukaryotic proteins into complex structures restricts bacteria motility within the cytoplasm of host cells, thereby facilitating recognition by cytosolic sensors and components of the autophagy machinery. Given the different types of septin caging that have been described thus far, a single-cell, unbiased approach to quantify and characterise septin recruitment at bacteria is important to fully grasp the role and function of caging. Thus, the authors have developed an automated image analysis pipeline allowing bacterial segmentation and classification of septin cages that will be very useful in the future, applied to study the role of host and bacterial factors, compare different bacterial strains, or even compare infections by clinical isolates.

Strengths:

The authors developed a solid pipeline that has been thoroughly validated. When tested on infected cells, automated analysis corroborated previous observations and allowed the unbiased quantification of the different types of septin cages as well as the correlation between caging and bacterial metabolic activity. This approach will prove an essential asset in the further characterisation of septin cages for future studies.

We thank the Reviewer for their positive comments, and for highlighting the strength of our imaging and analysis pipeline to analyse Shigella-septin interactions.

Weaknesses:

As the main aim of the manuscript is to describe the newly developed analysis pipeline, the results illustrated in the manuscript are essentially descriptive. The developed pipeline seems exceptionally efficient in recognising septin cages in infected cells but its application for a broader purpose or field of study remains limited.

The main objective of this manuscript is the development of imaging and analysis tools to study Shigella infection, and in particular, Shigella interactions with the septin cytoskeleton. In future work we will provide more mechanistic insight with novel experiments and broader applicability, using different cell lines (in agreement with Reviewer 1), mutants or clinical isolates of Shigella and different bacteria species (eg. Listeria, Salmonella, mycobacteria).

Reviewer #3 (Public Review):

Summary:

The manuscript uses high-content imaging and advanced image-analysis tools to monitor the infection of epithelial cells by Shigella. They perform some analysis on the state of the cells (through measurements of DNA and protein synthesis), and then they focus on differential recruitment of Sept7 to the bacteria. They link this recruitment with the activity of the bacterial T3SS, which is a very interesting discovery. Overall, I found numerous exciting elements in this manuscript, and I have a couple of reservations. Please see below for more details on my reservations. Nevertheless, I think that these issues can be addressed by the authors, and doing so will help to make it a convincing and interesting piece for the community working on intracellular pathogens. The authors should also carefully re-edit their manuscript to avoid overselling their data (see below for issues I see there). I would consider taking out the first figure and starting with Figure 3 (Figure 2 could be re-organized in the later parts)- that could help to make the flow of the manuscript better.

Strengths:

The high-content analysis including the innovative analytical workflows are very promising and could be used by a large number of scientists working on intracellular bacteria. The finding that Septins (through SEPT7) are differentially regulated through actively secreting bacteria is very exciting and can steer novel research directions.

We thank the Reviewer for their constructive feedback and the excitement for our results, including our findings on T3SS activity and Shigella-septin interactions_._ In accordance with the Reviewer’s comments, we agree to carefully re-edit our manuscript to avoid overselling our data in a future version of the manuscript. We will also consider to rearrange figures depending on new results.

Weaknesses:

The manuscript makes a connection between two research lines (1: Shigella infection and DNA/protein synthesis, 2: regulation of septins around invading Shigella) that are not fully developed - this makes it sometimes difficult to understand the take-home messages of the authors.

We agree that the manuscript is mostly technical and therefore some of our experimental observations would benefit from follow up mechanistic studies in the future. We highlight our vision for broader applicability in response to weaknesses raised by Reviewer 2.

It is not clear whether the analysis that was done on projected images actually reflects the phenotypes of the original 3D data. This issue needs to be carefully addressed.

We agree with the Reviewer that characterizing 3D data using 2D projected images has limitations.

We observe an increase in cell and nuclear surface that does not strictly imply a change in volume. This is why we measure Hoechst intensity in the nucleus using SUM-projection (as it can be used as a proxy of DNA content of the cell). However, we agree that future use of other markers (such as fluorescent labelled histones) would make our conclusions more robust.

Regarding the different orientation of intracellular bacteria, we agree that investigation of septin recruitment is more challenging when bacteria are placed perpendicular to the acquisition plane. In a first step, we trained a Convolutional Neural Network (CNN) using 2D data, as it is easier/faster to train and requires fewer annotated images. In doing so, we already managed to correctly identify 80% of Shigella interacting with septins, which enabled us to observe higher T3SS activity in this population. In future studies, we will maximize the 3D potential of our data and retrain a CNN that will allow more precise identification of Shigella-septin interactions and in depth characterization of volumetric parameters.

https://doi.org/10.7554/eLife.97495.1.sa4

Significance of findings

Strength of evidence

Abstract

Introduction

Results

Capturing the heterogeneity of S. flexneri infection using high-content microscopy

High-content imaging captures the heterogeneity of S. flexneri infection

Epithelial cells reduce de novo DNA and protein synthesis upon infection

Design of a deep learning pipeline to automatically identify bacteria interacting with SEPT7

Microscopy pipeline to automatically identify SEPT7-S. flexneri interactions

Training of a deep learning algorithm to identify single isolated bacteria cells and SEPT7 associated S. flexneri

Characterization of the morphological heterogeneity of SEPT7 recruitment to S. flexneri

Septin assemblies associated to S. flexneri are heterogeneous

SEPT7 is recruited to S. flexneri with increased T3SS activity

S. flexneri associated with SEPT7 are metabolically active

S. flexneri associated with SEPT7 have an active T3SS

Discussion

Concluding remarks

Materials and methods

Reagents

Bacterial strains and culture conditions

Cloning

Mammalian cell culture

Infection of human cells

Click chemistry

Immunofluorescence and fluorescence microscopy

Flow cytometry

Analysis pipelines

Statistics

Data availability

Code availability

Acknowledgements

Supplementary Figure Legends

S. flexneri infection induces host cell size changes

Click-IT Edu and Click-IT AHA specifically label de novo DNA and protein synthesis in Hela cells

Additional examples of the annotated datasets for deep learning

Probability maps of SEPT7 around S. flexneri

Click-IT Edu and Click-IT AHA specifically label de novo DNA and protein translation in S. flexneri

TSAR labels T3SS activity specifically

References

Article and author information

Author information

Ana T. López-Jiménez

Dominik Brokatzky

Kamla Pillay

Tyrese Williams

Gizem Özbaykal Güler

Serge Mostowy

Version history

Copyright

Peer review process

Editors