Antimicrobial resistance is a growing public health threat predicted to cause up to 10 million deaths a year by 2050. To circumvent existing bacterial resistance mechanisms, discovering antibiotics with novel modes of action (MoAs) is crucial. While growth inhibition assays can robustly identify antibiotic molecules, they miss promising compounds with subinhibitory phenotypes and do not inform on drug MoA. Microscopy-based cytological profiling of drug-treated bacteria with hand-crafted image descriptors or more recently deep learning (DL) provides complementary information on the MoA. However, current approaches are limited by the need for fluorescent labelling and drug exposure at inhibitory concentrations. It also remains unclear if cytological profiling enables the detection of drugs with novel MoAs. Here, we demonstrate an approach based on supervised DL to identify antibiotic drug MoA from microscopy images without fluorescent labelling. We train a convolutional neural network to predict treatment conditions from brightfield images of Escherichia coli exposed to reference drugs covering multiple MoAs. Our method can detect drug exposure at subinhibitory concentrations and distinguishes individual drug treatments with high accuracy (86.21%). The learned representations implicitly capture MoA-specific phenotypes, enabling perfect MoA recognition (100%) without retraining using a model trained on only 644 images. Our approach can identify the MoA of previously unseen drugs with good accuracy (77.8±3.3%), as long as the MoA is represented by at least one of the training compounds. Finally, we show that our approach can detect if a drug has a novel MoA with an area under the curve of 0.77±0.19, facilitating microscopy-based identification of novel classes of antibiotics. Our methods and results pave the way towards an automated pipeline for antibiotic drug discovery based on imaging and DL. ### Competing Interest Statement The authors have declared no competing interest.

Single Molecule Localization Microscopy (SMLM) and its ability to resolve < 100 nm structures has generated an evergrowing demand in biomedical research. This technique is highly relevant when trying to gain better understanding of biological structures details or cellular machinery in infectious models. The Imaging and Modelling Unit led by C. Zimmer developed an open optical and computational method based on Zernike Optimised Localisation Approach (ZOLA) enabling 3D localization of single molecules using point spread function (PSF) engineering in the detection path. This technique offers different performances and trade-off depending on the required application. This unique flexibility is relevant when dealing with various types of samples and models as those presented to an Imaging core facility. We will present how the Unit of Technology and Services (UTechS) Photonic Bio Imaging (PBI), the imaging platform of Institut Pasteur in Paris has conducted the technological transfer of ZOLA 3D from a research laboratory to a Bio Safety Level 2 (BSL2) ISO 9001 core facility. This will make flexible 3D super-resolution imaging accessible to a wide range of biological projects, including the study of pathogens.

Histological analysis of skeletal muscle is of major interest for understanding its behavior in different pathophysiological conditions, such as the response to different environments or myopathies. In this context, many software programs have been developed to perform automated high-content analysis. We created MuscleJ, a macro that runs in ImageJ/Fiji on batches of images. MuscleJ is a multianalysis tool that initially allows the analysis of muscle fibers, capillaries, and satellite cells. Since its creation, it has been used in many studies, and we have further developed the software and added new features, which are presented in this article. We converted the macro into a Java-language plugin with an improved user interface. MuscleJ2 provides quantitative analysis of fibrosis, vascularization, and cell phenotype in whole muscle sections. It also performs analysis of the peri-myonuclei, the individual capillaries, and any staining in the muscle fibers, providing accurate quantification within regional sublocalizations of the fiber. A multicartography option allows users to visualize multiple results simultaneously. The plugin is freely available to the muscle science community.

Modern drug discovery approaches often use high-content imaging to systematically study the effect on cells of large libraries of chemical compounds. By automatically screening thousands or millions of images to identify specific drug-induced cellular phenotypes, for example, altered cellular morphology, these approaches can reveal ‘hit' compounds offering therapeutic promise. In the past few years, artificial intelligence (AI) methods based on deep learning (DL) [a family of machine learning (ML) techniques] have disrupted virtually all image analysis tasks, from image classification to segmentation. These powerful methods also promise to impact drug discovery by accelerating the identification of effective drugs and their modes of action. In this review, we highlight applications and adaptations of ML, especially DL methods for cell-based phenotypic drug discovery (PDD).

ABSTRACT Macrophages are the major host cells of the protozoan parasite Leishmania in mammalian infection. These key innate immune cells display remarkable phenotypic plasticity ranging from pro-inflammatory M1 to anti-inflammatory M2 macrophages that can control infection and tissue homeostasis, respectively. It has been recognized that Leishmania exploits macrophage phenotypic plasticity to establish chronic infection. However, the current notion that these parasites simply trigger an M2-like phenotype seems over-simplified considering the immunopathology observed during leishmaniasis – in particular in response to Leishmania amazonensis - which is often characterized by a mixed Th1/Th2 immune response. Here we combined a series of systems-level analyses to shed new light on the phenotype of Leishmania -infected macrophages (LIMs) during short- and long-term infection, in vitro and in vivo . Immuno-metabolic profiling by RNA-seq, RT-qPCR, cytokine immunoassays, and real-time bioenergetic flux analysis of L. amazonensis -infected bone marrow-derived macrophages (BMDMs) revealed a highly complex and unique phenotypic and bioenergetic signature. In vitro LIMs were characterized by co-expression of both M1 and M2 markers at RNA and protein levels and increased expression of glycolytic genes that matched a progressive metabolic switch from a M2-like respiratory to a M1-like glycolytic energy production observed for both long-term in vitro and in vivo infected macrophages. Unlike in M1 macrophages, glycolytic gene expression did not correlate with increased expression of its key regulatory HIF-1α. In contrast, siRNA knock down experiments in primary BMDMs uncovered an essential role of the m 6 A reader protein IGF2BP2 in stabilizing m 6 A modified transcripts of the glycolytic pathway, contributing to HIF-1α-independent induction of glycolysis. In conclusion, L. amazonensis establishes a complex and unique phenotypic shift in infected macrophages in vitro and in vivo that combines M1-like and M2-like immuno-metabolomic characteristics and implicates differential mRNA stability in induction of aerobic glycolysis. Our data thus uncover epi-transcriptomic regulation as a novel target for Leishmania immune subversion to establish a host cell phenotype beneficial for intracellular parasite development and chronic infection.

The LabEx Milieu Interieur (MI) project is a clinical study centered on the detailed characterization of the baseline and induced immune responses in blood samples from 1,000 healthy donors. Analyses of these samples has lay ground for seminal studies on the genetic and environmental determinants of immunologic variance in a healthy cohort population. In the current study we developed in vitro methods enabling standardized quantification of MI-cohort-derived primary fibroblasts responses. Our results show that in vitro human donor cohort fibroblast responses to stimulation by different MAMPs analogs allows to characterize individual donor immune-phenotype variability. The results provide proof-of-concept foundation to a new experimental framework for such studies. A bio-bank of primary fibroblast lines was generated from 323 out of 1,000 healthy individuals selected from the MI-study cohort. To study inter-donor variability of innate immune response in primary human dermal fibroblasts we chose to measure the TLR3 and TLR4 response pathways, both receptors being expressed and previously studied in fibroblasts. We established high-throughput automation compatible methods for standardized primary fibroblast cell activation, using purified MAMPS analogs, poly I:C and LPS that stimulate TLR3 and TLR4 pathways respectively. These results were in turn compared with a stimulation method using infection by HSV-1 virus. Our “Add-only” protocol minimizes high-throughput automation system variability facilitating whole process automation from cell plating through stimulation to recovery of cell supernatants, and fluorescent labeling. Images were acquired automatically by high-throughput acquisition on an automated high-content imaging microscope. Under these methodological conditions standardized image acquisition provided for quantification of cellular responses allowing biological variability to be measured with low system noise and high biological signal fidelity. Optimal for automated analysis of immuno-phenotype of primary human cell responses our method and experimental framework as reported here is highly compatible to high-throughput screening protocols like those necessary for chemo-genomic screening. In context of primary fibroblasts derived from donors enrolled to the MI-clinical-study our results open the way to assert the utility of studying immune-phenotype characteristics relevant to a human clinical cohort.

[This corrects the article DOI: 10.3389/fimmu.2020.569331.].

The COVID-19 pandemic highlights the importance of efficient and safe vaccine development. Vaccine adjuvants are essential to boost and tailor the immune response to the corresponding pathogen. To allow for an educated selection, we assessed the effect of different adjuvants on human monocyte-derived dendritic cells (DCs) and their ability to polarize innate and adaptive immune responses. In contrast to commonly used adjuvants, such as aluminum hydroxide, Toll-like receptor (TLR) agonists induced robust phenotypic and functional DC maturation. In a DC-lymphocyte coculture system, we investigated the ensuing immune reactions. While monophosphoryl lipid A synthetic, a TLR4 ligand, induced checkpoint inhibitors indicative for immune exhaustion, the TLR7/8 agonist Resiquimod (R848) induced prominent type-1 interferon and interleukin 6 responses and robust CTL, B-cell, and NK-cell proliferation, which is particularly suited for antiviral immune responses. The recently licensed COVID-19 vaccines, BNT162b and mRNA-1273, are both based on single-stranded RNA. Indeed, we could confirm that the cytokine profile induced by lipid-complexed RNA was almost identical to the pattern induced by R848. Although this awaits further investigation, our results suggest that their efficacy involves the highly efficient antiviral response pattern stimulated by the RNAs' TLR7/8 activation.

Quantitative imaging of epithelial tissues requires bioimage analysis tools that are widely applicable and accurate. In the case of imaging 3D tissues, a common preprocessing step consists of projecting the acquired 3D volume on a 2D plane mapping the tissue surface. While segmenting the tissue cells is amenable on 2D projections, it is still very difficult and cumbersome in 3D. However, for many specimen and models used in developmental and cell biology, the complex content of the image volume surrounding the epithelium in a tissue often reduces the visibility of the biological object in the projection, compromising its subsequent analysis. In addition, the projection may distort the geometry of the tissue and can lead to strong artifacts in the morphology measurement.Here we introduce a user-friendly toolbox built to robustly project epithelia on their 2D surface from 3D volumes and to produce accurate morphology measurement corrected for the projection distortion, even for very curved tissues. Our toolbox is built upon two components. LocalZProjector is a configurable Fiji plugin that generates 2D projections and height-maps from potentially large 3D stacks (larger than 40 GB per time-point) by only incorporating signal of the planes with local highest variance/mean intensity, despite a possibly complex image content. DeProj is a MATLAB tool that generates correct morphology measurements by combining the height-map output (such as the one offered by LocalZProjector) and the results of a cell segmentation on the 2D projection, hence effectively deprojecting the 2D segmentation in 3D. In this paper, we demonstrate their effectiveness over a wide range of different biological samples. We then compare its performance and accuracy against similar existing tools.We find that LocalZProjector performs well even in situations where the volume to project also contains unwanted signal in other layers. We show that it can process large images without a pre-processing step. We study the impact of geometrical distortions on morphological measurements induced by the projection. We measured very large distortions which are then corrected by DeProj, providing accurate outputs.

Background: Microglial cells are tissue-resident macrophages of the central nervous system. They are extremely dynamic, sensitive to their microenvironment and present a characteristic complex and heterogeneous morphology and distribution within the brain tissue. Many experimental clues highlight a strong link between their morphology and their function in response to aggression. However, due to their complex “dendritic-like” aspect that constitutes the major pool of murine microglial cells and their dense network, precise and powerful morphological studies are not easy to realize and complicate correlation with molecular or clinical parameters. Methods: Using the knock-in mouse model CX3CR1, we developed a 3D automated confocal tissue imaging system coupled with morphological modelling of many thousands of microglial cells revealing precise and quantitative assessment of major cell features: cell density, cell body area, cytoplasm area and number of primary, secondary and tertiary processes. We determined two morphological criteria that are the complexity index (CI) and the covered environment area (CEA) allowing an innovative approach lying in (i) an accurate and objective study of morphological changes in healthy or pathological condition, (ii) an in situ mapping of the microglial distribution in different neuroanatomical regions and (iii) a study of the clustering of numerous cells, allowing us to discriminate different sub-populations. Results: Our results on more than 20,000 cells by condition confirm at baseline a regional heterogeneity of the microglial distribution and phenotype that persists after induction of neuroinflammation by systemic injection of lipopolysaccharide (LPS). Using clustering analysis, we highlight that, at resting state, microglial cells are distributed in four microglial sub-populations defined by their CI and CEA with a regional pattern and a specific behaviour after challenge. Conclusions: Our results counteract the classical view of a homogenous regional resting state of the microglial cells within the brain. Microglial cells are distributed in different defined sub-populations that present specific behaviour after pathological challenge, allowing postulating for a cellular and functional specialization. Moreover, this new experimental approach will provide a support not only to neuropathological diagnosis but also to study microglial function in various disease models while reducing the number of animals needed to approach the international ethical statements.

Background: Diet is widely recognized as one of the main modifiable drivers of gut microbiota variability, and its influence on microbiota composition is an active area of investigation. Objective: The present work aimed to explore the associations between usual diet and gut microbiota composition in a large sample of healthy French adults. Methods: Gut microbiota composition was established through sequencing of the 16S rRNA gene in stool samples from 862 healthy French adults of the Milieu Interieur study. Usual dietary consumptions were determined through the administration of a food-frequency questionnaire. The associations between dietary variables and alpha- and beta-diversity indexes and relative taxa abundances were tested using Spearman correlations, permutational ANOVAs, and multivariate analyses with linear models, respectively. Results: Foods generally considered as healthy (raw fruits, fish) were positively associated with alpha-diversity, whereas food items for which a limited consumption is generally recommended (fried products, sodas or sugary drinks, fatty sweet products, processed meats, ready-cooked meals, and desserts) were negatively associated with alpha-diversity. Fruits, fried products, ready-cooked meals, and cheese contributed to shifts within microbiota composition (beta-diversity). Our results also highlighted a number of associations between various food group intakes and abundances of specific phyla, genera, and species. For instance, the consumption of cheese was negatively associated with Akkermansia muciniphila abundance. Conclusions: This large-scale population-based study supports that the usual consumption of certain food items is associated with several gut microbial features, and extends the mechanistic arguments linking Western diet to an altered microbiota composition. These results provide new insights into the understanding of complex diet-gut microbiota relations, and their implications for host health deserve further investigation because altered microbiota diversity was consistently linked to increased risk of several health outcomes. This trial was registered at clinicaltrials.gov as NCT01699893.

Cell-based phenotypic screening has proven to be valuable, notably in recapitulating relevant biological conditions, for example, the host cell/pathogen niche. However, the corresponding methodological complexity is not readily compatible with high-throughput pipelines, and fails to inform either molecular target or mechanism of action, which frustrates conventional drug-discovery roadmaps. We review the state-of-the-art and emerging technologies that suggest new strategies for harnessing value from the complexity of phenotypic screening and augmenting powerful utility for translational drug discovery. Advances in cellular, molecular, and bioinformatics technologies are converging at a cutting edge where the complexity of phenotypic screening may no longer be considered a hinderance but rather a catalyst to chemotherapeutic discovery for infectious diseases.

Pathogenic enterobacteria face various oxygen (O-2) levels during intestinal colonization from the O-2-deprived lumen to oxygenated tissues. Using Shigella flexneri as a model, we have previously demonstrated that epithelium invasion is promoted by O-2 in a type III secretion system-dependent manner. However, subsequent pathogen adaptation to tissue oxygenation modulation remained unknown. Assessing single-cell distribution, together with tissue oxygenation, we demonstrate here that the colonic mucosa O-2 is actively depleted by S. flexneri aerobic respiration-and not host neutrophils-during infection, leading to the formation of hypoxic foci of infection. This process is promoted by type III secretion system inactivation in infected tissues, favouring colonizers over explorers. We identify the molecular mechanisms supporting infectious hypoxia induction, and demonstrate here how enteropathogens optimize their colonization capacity in relation to their ability to manipulate tissue oxygenation during infection.

Early detection of tumors is today a major challenge and requires sensitive imaging methodologies coupled with new efficient probes. In vivo optical bioluminescence imaging has been widely used in the field of preclinical oncology to visualize tumors and several cancer cell lines have been genetically modified to provide bioluminescence signals. However, the light emitted by the majority of commonly used luciferases is usually in the blue part of the visible spectrum, where tissue absorption is still very high, making deep tissue imaging non-optimal, and calling for optimized optical imaging methodologies. We have previously shown that red-shifting of bioluminescence signal by Fluorescence Unbound Excitation from Luminescence (FUEL) is a mean to increase bioluminescence signal sensitivity detection in vivo. Here, we applied FUEL to tumor detection in two different subcutaneous tumor models: the auto-luminescent human embryonic kidney (HEK293) cell line and the murine B16-F10 melanoma cell line previously transfected with a plasmid encoding the Luc2 firefly luciferase. Tumor size and bioluminescence were measured over time and tumor vascularization characterized. We then locally injected near infrared emitting Quantum Dots (NIR QDs) in the tumor site and observed a red-shifting of bioluminescence signal by (FUEL) indicating that FUEL could be used to allow deeper tumor detection in mice.

The cytoskeleton, composed of actin microfilaments, microtubules, and intermediate filaments (IF), plays a key role in the control of cell shape, polarity, and motility. The organization of the actin and microtubule networks has been extensively studied but that of IFs is not yet fully characterized. IFs have an average diameter of 10 nm and form a network extending throughout the cell cytoplasm. They are physically associated with actin and microtubules through molecular motors and cytoskeletal linkers. This tight association is at the heart of the regulatory mechanisms that ensure the coordinated regulation of the three cytoskeletal networks required for most cell functions. It is therefore crucial to visualize IFs alone and also together with each of the other cytoskeletal networks. However, IF networks are extremely dense in most cell types, especially in glial cells, which makes its resolution very difficult to achieve with standard fluorescence microscopy (lateral resolution of ~250 nm). Direct STochastic Optical Reconstruction Microscopy (dSTORM) is a technique allowing a gain in lateral resolution of one order of magnitude. Here, we show that lateral dSTORM resolution is sufficient to resolve the dense organization of the IF networks and, in particular, of IF bundles surrounding microtubules. Such tight association is likely to participate in the coordinated regulation of these two networks and may, explain how vimentin IFs template and stabilize microtubule organization as well as could influence microtubule dependent vesicular trafficking. More generally, we show how the observation of two cytoskeletal components with dual-color dSTORM technique brings new insight into their mutual interaction.

During development, coordinated cell behaviors orchestrate tissue and organ morphogenesis. Detailed descriptions of cell lineages and behaviors provide a powerful framework to elucidate the mechanisms of morphogenesis. To study the cellular basis of limb development, we imaged transgenic fluorescently-labeled embryos from the crustacean Parhyale hawaiensis with multi-view light-sheet microscopy at high spatiotemporal resolution over several days of embryogenesis. The cell lineage of outgrowing thoracic limbs was reconstructed at single-cell resolution with new software called Massive Multi-view Tracker (MaMuT). In silico clonal analyses suggested that the early limb primordium becomes subdivided into anterior-posterior and dorsal-ventral compartments whose boundaries intersect at the distal tip of the growing limb. Limb-bud formation is associated with spatial modulation of cell proliferation, while limb elongation is also driven by preferential orientation of cell divisions along the proximal-distal growth axis. Cellular reconstructions were predictive of the expression patterns of limb development genes including the BMP morphogen Decapentaplegic.

The shape of cellular membranes is highly regulated by a set of conserved mechanisms that can be manipulated by bacterial pathogens to infect cells. Remodeling of the plasma membrane of endothelial cells by the bacterium Neisseria meningitidis is thought to be essential during the blood phase of meningococcal infection, but the underlying mechanisms are unclear. Here we show that plasma membrane remodeling occurs independently of F-actin, along meningococcal type IV pili fibers, by a physical mechanism that we term ‘one-dimensional’ membrane wetting. We provide a theoretical model that describes the physical basis of one-dimensional wetting and show that this mechanism occurs in model membranes interacting with nanofibers, and in human cells interacting with extracellular matrix meshworks. We propose one-dimensional wetting as a new general principle driving the interaction of cells with their environment at the nanoscale that is diverted by meningococci during infection.

Background: Skeletal muscle has the capacity to adapt to environmental changes and regenerate upon injury. To study these processes, most experimental methods use quantification of parameters obtained from images of immunostained skeletal muscle. Muscle cross-sectional area, fiber typing, localization of nuclei within the muscle fiber, the number of vessels, and fiber-associated stem cells are used to assess muscle physiology. Manual quantification of these parameters is time consuming and only poorly reproducible. While current state-of-the-art software tools are unable to analyze all these parameters simultaneously, we have developed MuscleJ, a new bioinformatics tool to do so. Methods: Running on the popular open source Fiji software platform, MuscleJ simultaneously analyzes parameters from immunofluorescent staining, imaged by different acquisition systems in a completely automated manner. Results: After segmentation of muscle fibers, up to three other channels can be analyzed simultaneously. Dialog boxes make MuscleJ easy-to-use for biologists. In addition, we have implemented color in situ cartographies of results, allowing the user to directly visualize results on reconstituted muscle sections. Conclusion: We report here that MuscleJ results were comparable to manual observations made by five experts. MuscleJ markedly enhances statistical analysis by allowing reliable comparison of skeletal muscle physiology-pathology results obtained from different laboratories using different acquisition systems. Providing fast robust multi-parameter analyses of skeletal muscle physiology-pathology, MuscleJ is available as a free tool for the skeletal muscle community.