Imaging host–pathogen interactions using epithelial and bacterial cell infection models

Bacterial pathogens cause a myriad of diseases and interact with host cells in a variety of ways. They may colonize extracellular niches, attach to epithelial cell surfaces, breach host barriers, establish intracellular infections or combine several of these approaches during the process of infection (Alberts et al., 2002; Hurwitz et al., 2015; Ribet and Cossart, 2015). The combination of tissue culture with microscopy is well suited to recapitulate and document these complex microenvironments, and is an excellent tool for the in vitro study of host–pathogen interactions. Importantly, experiments aimed at capturing high-resolution images of host–pathogen interactions involve the integration of multiple disciplines and techniques. This Review highlights the cellular systems, and imaging tools and techniques that allow researchers to better understand the biology of bacterial infection, as well as potential targets for disease mitigation. The scope of the article includes studies focusing on cell culture models of human epithelial barriers and bacteria relevant to disease and pathology. Although critical for some research questions, we do not discuss animal models, tissue explants or cellular systems generated using epithelial–immune cell co-cultures (Gordon et al., 2019; Grivel and Margolis, 2009; Jelicks et al., 2013; Leevy et al., 2007; Papazian et al., 2016). Our discussion is organized around two main technical issues – choosing the biological components of the experimental system and assessing which microscopy-based tools are well-suited to label and image various cell compartments or biological processes. A variety of imaging techniques will be addressed, including both light- and electron-microscopy methods, which cover the spatial scale and resolution required to visualize entire cellular barriers down to the cellular ultrastructure (Fig. 1). The discussion will be complemented with examples of recent studies that have used these methods to generate highly informative micrographs that provide data relevant to researchers studying a wide range of subjects related to host-pathogen interactions.

Selecting and establishing an appropriate experimental model system is an important step, and an often-challenging initial hurdle, when designing an imaging study that involves the host–pathogen interface. Although there is no single in vitro model that can be considered a gold standard, there are approaches that might be more relevant depending on the specific research question. Therefore, the selection of epithelial cell type, the culture technique and quality control measures often requires a great deal of consideration before embarking on an imaging project. A selection of common options for building an infection model (e.g. epithelial cell type and culture technique), including relevant advantages and disadvantages of each, is presented in Table 1. The discussion below will focus on the quality control measures that can be used to assess the selected model system.

For any epithelial and bacterial cell infection model, the experiments typically require a balance between having a sufficient population of bacteria to study and not sacrificing the viability and/or integrity of the host model. Cellular viability and toxicity are commonly measured to assess potential cell damage caused by the exposure to the pathogen, and various methods to evaluate these parameters have been reviewed elsewhere (Stoddart, 2011). These readouts can be a good way to titrate parameters of the infection that may, in excess, inordinately affect host cell health to a point where the model is no longer representative of an in vivo barrier. For example, increasing the number of bacteria being added to the infection model relative to the number of host cells (i.e. increasing the multiplicity of infection) can increase epithelial cell cytotoxicity (Hertle and Schwarz, 2004). Barrier integrity can be assessed by measuring the transepithelial electrical resistance (TEER) across the cell layer(s) (Srinivasan et al., 2015). Higher TEER measurements are indicative of a confluent epithelial layer with high barrier integrity; conversely, lower TEER measurements signal breaks in the epithelial barrier, resulting from holes or weak cell–cell junctions (Srinivasan et al., 2015). Importantly, average or optimal TEER values can be cell line specific due to variability in the junctional complexes (Srinivasan et al., 2015). The integrity of the epithelial cell barrier can also be assessed by staining adherens junction proteins, such as E-cadherin and β-catenin, and tight junctional proteins, such as zonula occludens-1 (ZO-1, also known as TJP1), claudin-1 and occludin (Barrila et al., 2017; Buckley et al., 2018; Carvalho et al., 2005; Uotani et al., 2019). While viability, toxicity, TEER and junctional protein staining can be determined for all epithelial cell types, cell-type-specific experiments may also be used to validate your model, such as labelling markers of differentiation (Fig. 2A; Table S1). Finally, it is important to note that some experimental conditions, such as prolonged infection or use of highly virulent strains, may require additional or more frequent quality control assessments to ensure the viability and integrity of the infection model is maintained throughout the experiment (Anderson et al., 2008; Starner et al., 2006; Walker et al., 2017). The efforts taken to develop, optimize and evaluate infection models will ensure that an appropriate, biologically relevant system is selected for further microscopic investigation.

Microscopic techniques can be separated into two main categories: light microscopy and electron microscopy. A description of traditional light and electron microscopy techniques and general uses in host–pathogen imaging studies is provided in Table 2. When choosing the imaging tools and techniques for a host–pathogen interaction experiment, it is important to consider the spatial scale and resolution that will be required to answer the question posed (Fig. 1), and whether you will be working with fixed or live cells. Light microscopy allows both live-cell and fixed-cell imaging but is limited in both lateral (200 nm) and axial (500 nm) resolution due to the relatively long wavelength of visible light (Abbe, 1873; Rayleigh, 1896). Fluorescence capabilities are acquired with the addition of a broad-spectrum light source [i.e. visible and ultraviolet (UV) wavelengths], a series of filters that can fine-tune the source (excitation) and gathered (emission) light, depending on the target and a detector (Young, 1961). Live-cell imaging is less prone to preparation artefacts and can provide important information regarding cellular structures and processes in real-time, but is not compatible with multiple labelling techniques, as described below. Fixing your cells will prevent sample degradation, but the process can produce artefacts, such as the loss of antigenicity and changes to cell morphology (Hobro and Smith, 2017; Jamur and Oliver, 2010a). Compared to light microscopy, electron microscopy provides higher resolution by illuminating the sample with an electron beam, instead of photons, and because of this, it is used to investigate cellular ultrastructure (Gordon, 2014). Samples are fixed, chemically or environmentally, and treated with heavy metal coatings and stains for contrast, which excludes live-cell imaging and prevents the depth of field offered by some light microscopy methods (Winey et al., 2014).

Recently, a group of fluorescence microscopy methods have been developed that challenged the resolution limits of traditional light microscopy. In 2014, Betzig, Hell and Moerner were awarded the Nobel Prize in chemistry for their work in developing super resolved fluorescence microscopies. This group of methods, collectively termed super-resolution microscopy, uses specialized equipment or modifications to typical microscope arrangements, along with extensive post-acquisition image analysis and transformation, which results in resolution beyond the diffraction limit of visible light (Galbraith and Galbraith, 2011; Schermelleh et al., 2019). This means that cellular structures and processes can be seen at a resolution closer to that provided by electron microscopy, while still retaining the benefits of light microscopy, such as live-cell imaging and tools for specific labelling. Some of these super-resolution methods use physical grating, that is, controlled separation of light, to allow for non-uniform delivery of the excitation light source to improve resolution [e.g. structured illumination microscopy (SIM) (Langhorst et al., 2009)]. Alternately, dual objectives [e.g. dual objective multifocal plane microscopy (dMUM) (Ram et al., 2009)], or selective activation of fluorophores can be used [e.g, photo-activated localization microscopy (PALM) or stochastic optical reconstruction microscopy (STORM) (Henriques et al., 2011)]. A recent review (Vangindertael et al., 2018) discusses the theory behind the various super-resolution techniques in detail and addresses their potential use in life sciences. While not specifically focused on infection models, others have described how the biological questions can determine the selection of the specific super-resolution technique (Schermelleh et al., 2019). A striking example of the use of super-resolution microscopy in host–pathogen interactions is the imaging of host cytoskeletal proteins in a Shigella infection model (Krokowski et al., 2018; Mostowy et al., 2010). In the 2010 paper, the authors used mostly confocal microscopy to show the interaction of host cell cytoskeletal proteins called septins with intracellular bacteria, but also used STORM to capture high-resolution image of fluorescently labelled septin assemblies they refer to as ‘cages’ surrounding individual Shigella (Mostowy et al., 2010). In addition, they also provide a direct comparison between the resolving power of conventional and STORM methods (Mostowy et al., 2010). In subsequent work from the same group, the authors used fluorescent labelling and SIM to capture the overlap of the bacterial division protein filamentation temperature-sensitive protein Z (FtsZ) with septin 7 (SEPT7), which led them to conclude that the microscale change in membrane curvature that occurs during bacterial division can act as a trigger for host cytoskeletal arrangements that result in the entrapment of intracellular bacteria (Krokowski et al., 2018). Others have used STORM to visualize the nanoscale distribution and linear organization of the host ubiquitin on the surface of intracellular Salmonella Typhimurium (Van Wijk et al., 2017). These observations, which would not have been possible using traditional light microscopy, contributed to the hypothesis that specific ubiquitylation patterns, or ‘signatures’, may elicit specific host signalling pathways involved in bacterial clearance (Van Wijk et al., 2017). The main drawbacks to these methods, in general, is the cost and access to specialized equipment and/or software, and the high level of expertise required to use them. Accordingly, these highly technical and challenging methods have not been widely used in the studies that fall within the scope of this Review, but will likely become more prominent in the field in years to come as they become more accessible outside of core facilities or specialized research groups.

There are many imaging tools that can be used to enhance the power and utility of light microscopy, and the information gained from labelling experiments often provides key clues to pathogenic mechanisms at the host–pathogen interface. Labelling techniques used with light microscopy can be broadly separated into those that can be used with fixed cells and those that can be used with live cells.

Methods used with non-living fixed cells commonly exploit fluorescently labelled antibodies to mark epithelial or bacterial cells, or use natural or synthetic compounds that bind specific cell structures (e.g. nuclei) or intracellular compounds (e.g. Ca²⁺) (Fig. 2A; Table S1). Antibody-based methods can be completed with traditional multidomain antibodies, or more recently developed single domain alternatives called nanobodies. Nanobodies are a unique form of immunoglobulin G (IgG) isolated from the sera of Camelidae, comprising a single heavy chain and a single variable domain, which favourably affects the size, solubility and affinity of nanobodies versus multidomain antibodies (Beghein and Gettemans, 2017), and reduces some labelling artefacts associated with larger multidomain antibodies (Beghein and Gettemans, 2017; Sograte-Idrissi et al., 2020). While nanobodies are not as widely available, or commonly used in traditional light microscopy, they are well suited for high- and super-resolution applications because of their spatial resolution in the nanometer range (Ries et al., 2012), as reviewed recently (Beghein and Gettemans, 2017). Fixed-cell methods can usually be completed without significant technical development prior to imaging and follow similar protocols in bacterial and host cells. A common method to label fixed cells is immunolabelling. These experiments use fluorescently labelled antibodies to detect proteins, and generally follow a standard procedure of fixation, permeabilization, blocking, and addition of primary and/or secondary antibodies that recognize the target antigen (Bratthauer, 2010). The need for a permeabilization step that allows antibodies to cross fixed membranes, and the reagent used to perform this step, will depend on the subcellular location of the antigen (Jamur and Oliver, 2010b) but can also be used as a method of examining specific host–pathogen interactions. For example, selective permeabilization with digitonin, which permeabilizes the plasma membrane but not intracellular membrane-bound organelles, has been used to determine whether intracellular Salmonella localize in the cytosol or inside a vacuole during infection of mucosal epithelial cells (Knodler et al., 2010).

With immunolabelling, primary antibodies can recognize a conserved antigen within the target and thus can be used across multiple cell lines (e.g. cytoskeletal or junctional protein antigens) or bacterial species and/or strains [e.g. highly conserved lipopolysaccharide core antigens (Pollack et al., 1987)]. Alternately, antibodies can be specific to a group of bacteria that share a common surface structure (i.e. serotype) (Aubey et al., 2019; Carvalho et al., 2005), or bind a specific protein such as a bacterial virulence factor (Uotani et al., 2019). Another strategy to detect or image specific cellular structures in fixed cells is epitope tagging; here, genetic constructs are generated that allow for the co-expression of polypeptide protein tags (with commercially available antibodies) and a gene of interest (Fritze and Anderson, 2000). Common tags include hemagglutinin (HA) or the Myc epitope. When fluorescently labelled antibodies are applied using indirect immunolabelling protocols (i.e. both primary and secondary antibodies are used), fluorophore-labelled secondary antibodies can be varied to accommodate different colour combinations in multiplexing experiments that simultaneously image multiple proteins within a single sample. When this approach is used, online tools can help to determine the spectral compatibility of dyes and probes before setting up experiments, including SpectraViewer from Thermo Fisher Scientific (https://www.thermofisher.com/ca/en/home/life-science/cell-analysis/labeling-chemistry/fluorescence-spectraviewer.html#!/) or Spectra Analyzer from Biolegend (https://www.biolegend.com/en-us/spectra-analyzer). After the application of the antibodies, counterstains that bind to specific structures, such as the nucleus or the plasma membrane, provide contrast and cellular context. However, it is important to keep in mind, when using these reagents in co-culture, that some common fluorescent dyes or stains bind to structures that are found in both host cells and bacterial cells. For example, wheat germ agglutinin (WGA)-conjugated dyes bind sialic acid residues, which are found on the surface of epithelial cells but also in some bacterial membranes (Severi et al., 2007). After the labelling protocols, the samples are then mounted, with the option of using anti-fade mounting media (Ono et al., 2001) to prolong the stability of the fluorescent signal for subsequent imaging.

Fixed-cell samples can also be labelled using fluorescent in situ hybridization (FISH), which uses fluorescently labelled oligonucleotide probes that recognize a nucleic acid sequence of interest inside fixed cells (Spector et al., 2005; Liehr, 2017). This approach was used to tag Nontypeable Haemophilus influenzae (NTHi) in a co-culture system with primary epithelial cells, using a cyanine dye-labelled probe that recognized NTHi 16S RNA (Walker et al., 2017). Although quick and easy to perform in theory, this method is not as commonly used for imaging of host–pathogen interactions, likely due to a lower abundance of the target compared to protein antigens, as well as the potential technical challenges and time investment required for the optimization of the initial FISH protocol in bacteria, including probe design and sample preparation, as reviewed elsewhere (Frickmann et al., 2017; Moter and Göbel, 2000). The main advantage of this technique compared to antibody-based methods is that it can be used when there are no commercially available antibodies for the target or when the production of custom antibodies is cost prohibitive. A list of published rRNA-targeted probes for bacteria can be found on the open-source repository ProbeBase (https://probebase.csb.univie.ac.at/; Greuter et al., 2016).

Live-cell techniques most often involve the use of fluorescent microscopes (Table 2) and fluorescent protein labels. The most-used fluorescent protein is green fluorescent protein (GFP), which has many variants (see Box 1). After a suitable fluorescent protein has been selected, the corresponding gene can be transiently introduced into live bacteria using an autonomously replicating plasmid (Di Paola et al., 2017; Ernstsen et al., 2017; Knodler et al., 2010; Rajan et al., 2000; Starner et al., 2006; Sullivan and Ulett, 2018) (Fig. 3Aiii,Ciii) through standard transformation protocols (Froger and Hall, 2007; Gonzales et al., 2013). Plasmid-based systems are efficient to work with; they can provide a strong signal due to frequently producing multiple copies per bacterium and can often be used with more than one species. However, they commonly require the addition of antibiotics to the culture medium for plasmid maintenance, which may not be ideal for experiments involving several bacterial strains with variable antimicrobial sensitivity. This issue can be circumvented by empirically determining the time frame of plasmid stability following the removal of the antibiotic from the culture and completing the experiments within that window (Starner et al., 2006).

Alternatively, the gene encoding the fluorescent protein can be integrated into the genome by site-specific recombination. Examples of integration vectors include mini-Tn7 (Barnes et al., 2008; Choi et al., 2005) or mini-CTX (Hoang et al., 2000). Barnes and colleagues used the mini-Tn7 vector to demonstrate the adhesion of GFP-labelled Pseudomonas aeruginosa to the host extracellular matrix protein fibronectin (Barnes et al., 2008). One disadvantage to chromosomally expressed GFP, however, is that the signal is often not as strong due to the lower copy number compared to a plasmid-based approach. Once integrated, the gene can be under the control of (1) a constitutive promoter that provides continual expression, (2) an inducible promotor that provides on–off control, or (3) a gene-specific promoter that can also provide information on the expression of genes that, for example, influence virulence. The latter option was elegantly used to track the downregulation of the Salmonella type III secretion system (T3SS) genes and flagellar genes following host-cell internalization, and to show differential activation of individual genes depending on the location of the bacteria inside the cell and the time post-infection (Knodler et al., 2010) (Fig. 3Ciii).

Importantly, when labelling a bacterium with a fluorescent protein certain caveats and controls must be considered. First, a signal from a fluorescent protein may not mean a live bacterium, so alternative methods of accurately assessing viability such as LIVE/DEAD staining and estimating cell colony forming units are necessary. Second, maturation and degradation rates of the fluorescent proteins may vary between bacterial species, growth conditions or culture conditions, and standard curves of relative fluorescence units (RFU) versus time should be completed when the signal is used for quantification or assessment of viability, especially in the case of systems with inducible promoters. Finally, the production of large fluorescent proteins can become a metabolic burden so growth curves should be completed to ensure that the use of fluorescent proteins in live-cell systems is not generating experimental artefacts (e.g. decreased growth or virulence).

For host cells, similar approaches can be employed using expression vectors and/or transfection reagents intended for use in eukaryotic cells and either transient or stable transfection protocols (Colosimo et al., 2000). In general, the choice between transient or stable transfection will depend on the timeline of the experiment and how amenable the cells are to manipulation. For shorter experiments (12–72 h) involving cells growing as an undifferentiated monolayer, transient transfection methods can deliver the genetic information required for fluorescent protein production in a large proportion of cells in the culture and provide a fluorescent signal for few days post transfection, before it is lost owing to replication or degradation (Colosimo et al., 2000). This approach was used recently to label autophagic membranes of Listeria monocytogenes-infected cells growing in multiwell plates (Kortebi et al., 2017). The transfected protein was expressed up to 3 days post-infection and micrographs from these experiments aided in demonstrating that the bacteria entered a persistent state and became trapped in lysosome-like vacuoles in a subset of cells (Kortebi et al., 2017). In contrast, infection models that involve well-differentiated polarized cells (Rybakovsky et al., 2019) or cell systems that require weeks to establish, are better suited to stable transfection approach that produces cultures with inheritable fluorescence (Colosimo et al., 2000). Some aspects of the stable transfection protocol may require optimization, such as the method of transfection, the selection reagent, the culture medium or the plasmid concentration, which can become a rate-limiting step in the transfection of different cell types. However, the final product, now considered a new cell line, can be a valuable tool for ongoing research. For instance, stable transfection using a lentiviral vector was used to generate mCherry-expressing human gastric epithelial spheroids (Sebrell et al., 2018) that were employed in Helicobacter pylori infection studies (Sebrell et al., 2019). Of note, the non-profit plasmid repository Addgene includes information on both prokaryotic and eukaryotic vectors and contains a fluorescent protein guide (www.addgene.org/fluorescent-proteins/) that can aid in the selection of the appropriate tools. In addition, a small selection of stable human GFP- and/or red fluorescent protein (RFP)-expressing cells are now commercially available for use in situations where the transfection protocols cannot be performed and there is flexibility regarding the cell line that can be used.

In addition to labelling cellular structures, reagents are also available to track processes that are relevant to host–pathogen interactions in live host cells (Fig. 2; Table S1); these include the externalization of phosphatidylserine associated with early stages of apoptosis, assessed for instance by Annexin V staining (Fig. 2B), DNA degradation during late stages of apoptosis by TUNEL staining, the activation of caspase enzymes (e.g. caspase-1) linked to inflammation (Fig. 2B), fluctuations in Ca²⁺ that indicate signalling events (by using Fura-2) (Bain et al., 1998), the acidification of lysosomes (e.g. LysoTracker™) (Capasso et al., 2016), as well as cell location or its replication state by using CellTracker™ or CellTrace™ (Capasso et al., 2016). The use of these reagents can provide important information regarding the underlying mechanisms of infection. For example, studies examining the infection of host cell layer with either P. aeruginosa (Capasso et al., 2016) (Fig. 2B) or S. Typhimurium (Knodler et al., 2010) (Fig. 2C) both showed host-cell extrusion events associated with areas of pathogen accumulation, but the respective use of reagents to track cellular processes suggests distinct underlying mechanisms. In the P. aeruginosa infection, Annexin V staining was used to demonstrate that these bacteria preferentially bind to cell–cell junctions in host cells that are undergoing apoptosis (Capasso et al., 2016) (Fig. 2B). Importantly, control experiments showed that under their culture conditions, the bacteria were not inducing host cell death, merely taking advantage of its outcomes, as indicated by similar rates of apoptosis in uninfected controls (Capasso et al., 2016). In contrast, extrusion events in the S. Typhimurium infection model showed that host-cell extrusion was associated with an active inflammatory process involving the increased activation of caspase enzymes, as indicted with caspase-1 immunolabelling (Knodler et al., 2010) (Figs 3Ci–iii and 2C). This type of labelling, therefore, provides information that can be used to target separate and specific host-cell processes when developing anti-infective therapy for these pathogens.

In addition, time-lapse or real-time imaging in host–pathogen interaction is important as it captures the dynamic nature of the interactions during the infection process. Examples include the recording of ciliary beat frequency to determine whether H. influenzae exposure affects mechanical innate immune responses (i.e. mucociliary clearance) (Walker et al., 2017). Time-lapse imaging can also be used to visualize intracellular motility of various pathogens to examine the mechanism involved in host-cell transmigration (Golovkine et al., 2016), motility within host cells (Nieto et al., 2019), or the switch between dissemination and persistence (Kortebi et al., 2017; Nieto et al., 2019). It has also been used for the verification of intracellular replication (Lerner et al., 2016; Nieto et al., 2019). Finally, real-time imaging is an excellent technique for visualizing bacterial translocation across the surface of cultured cells, and has been used to demonstrate the cell–cell transmission of pathogenic Escherichia coli (Sanger et al., 1996; Velle and Campellone, 2017), which has the important consequence of expanding the range of bacterial colonization within a host.

For all techniques described above that use fluorescence microscopy, it is important to assess non-specific fluorescence (i.e. autofluorescence) that can arise from the cell culture media (e.g. riboflavin) or autofluorescent components that naturally occur in host tissues (e.g. collagen and tryptophan). Here, washing out, modifying or using specific imaging media (e.g. FluoroBrite^TM) (Di Paola et al., 2017) can help limit autofluorescence, as can the use of blocking and quenching reagents (e.g. animal serum, BSA and TrueBlack^® Lipofuscin) during sample processing.

Electron microscopy provides greater image resolution compared to light microscopy and, therefore, is commonly used to investigate the ultrastructure of cells and their components. The two main types of electron microscopy are transmission electron microscopy (TEM) and scanning electron microscopy (SEM) (Table 2), and both are commonly and effectively used to study host–pathogen interactions.

TEM requires samples to be fixed either chemically, with aldehydes and osmium tetraoxide, or by using an environment with cryogenic temperatures at high pressure, termed high pressure freezing and freeze substitution; afterwards, the samples are embedded in a plastic resin and sliced in thin sections for imaging (Bozzola and Russel, 1998). Importantly, the options at each step may introduce reagent- or method-specific artefacts, such as shrinkage and masking of antigens, which have been discuss extensively elsewhere (Maunsbach and Afzelius, 1999). While the preparation of thin sections is technically challenging, this method has resulted in very detailed and informative images of host–pathogen interactions. For instance, the specific location of Bordetella pertussis attachment to cell monolayers and the ultrastructure surrounding the site of host–pathogen interactions have been shown in detail by TEM (Guevara et al., 2016). TEM was an important tool in this study because it offered the resolution to show the interaction of individual bacteria with individual cilia on the highly differentiated epithelial cells (Guevara et al., 2016). From this, the authors were able to discern patterns of intimate attachment, including the clustering of multiple bacteria and the attachment to cilial ‘tufts’ (Guevara et al., 2016). Furthermore, the ‘cross-section’ view afforded by thin-sectioning allows for the visualization of areas of adherence, invasion, bacterial replication and host cytoskeletal changes all in the same sample, and thus provides a detailed overview of the infection process in this model (Guevara et al., 2016). One of the most common applications of thin-sectioning and TEM is the visualization of internalized bacteria. For instance, TEM has been used to demonstrate that internalized P. aeruginosa reside within an intracellular membrane vesicle (Capasso et al., 2016). Similarly, Streptococcus pneumoniae were found to reside within a vacuole after infection, but were also shown to be able to transmigrate between epithelial cells to spread infection (Weight et al., 2019) (Fig. 3Dii). TEM can also be used as a quality control protocol to confirm specific characteristics of an infection model that may be relevant to host–pathogen interactions, such as the presence of mucus granules, where H. influenza were shown to congregate during the early stages of infection (Marrazzo et al., 2016). With TEM, specific cellular structures can be labelled by gold-conjugated antibodies, using similar methods as those described above for light-microscopy-based immunolabelling experiments.

SEM can also provide high-resolution ultrastructural information on host–pathogen interactions. Processing is, in general, less laborious than TEM, with the vast majority of samples following a standard protocol of chemical fixation, dehydration by critical point drying and then coating with a thin layer of electrically conductive metal, such as gold (Fischer et al., 2012). Importantly, because SEM generates images by scanning the surface of a sample with electrons rather than penetrating the sample, visualization is limited to the observation of extracellular host–pathogen interactions. Accordingly, SEM is often used to study the initial attachment of bacterial cells on a monolayer. For example, SEM studies were able to detect epithelial membrane folding as a response to S. pneumoniae interaction (Weight et al., 2019) (Fig. 3Di). This was an important observation both because S. pneumoniae is conventionally considered an extracellular pathogen, and the folds were different from the membrane ruffles previously seen during invasion by intracellular pathogens such as S. Typhimurium, suggesting a novel mechanism of invasion (Weight et al., 2019). In addition, SEM is a good tool for studying bacterial biofilms, as it can demonstrate both bacterial cell clustering and production of exopolymeric substances (EPS), both of which are hallmark features of biofilm formation (Flemming and Wingender, 2010). For example, SEM was used to show that H. influenzae forms biofilms more readily on the surface of epithelial cells from patients with primary ciliary dyskinesia as compared to healthy controls (Walker et al., 2017). Combined with TEM, SEM was also used to show that H. influenzae formed adherent biofilms on the apical surface airway epithelia of patients with cystic fibrosis (Starner et al., 2006) (Fig. 3Ai,ii). It is important to note that unlike light microscopy, and to some degree TEM, specific cell structures or processes cannot be labelled when samples are prepared according to the SEM protocol; however, more innovative techniques are continuing to push the limits of what both TEM and SEM can tell us about host–pathogen interactions.

Understanding host–pathogen interactions is critical to advancing our knowledge and treatment of bacterial infections. The studies referenced and discussed in this Review provide a small subset of the observations that have been previously made using microscopy, and undoubtedly, with the technological advances in imaging techniques, the research questions that can be answered with microscopy will continue to grow. One of the major limitations with traditional electron microscopy methods is sample thickness. As described above, imaging has been traditionally limited to the surface of intact samples in SEM and to serially sectioned samples in TEM. Recently this limitation has been addressed by techniques that automate the removal of precise layers from the surface of a sample using either an ultramicrotome, in a process known as serial block face (SBF), or a gallium ion source, described as focused ion beam (FIB) milling. Both techniques, and how they contribute to three-dimensional sample imaging, have been reviewed elsewhere (Peddie and Collinson, 2014). These techniques can be used with samples that have been prepared in a variety of ways (e.g. embedded, fixed, dried or cryo-preserved). Samples are alternately milled then imaged, resulting in a Z-stack of SEM or TEM images, which combined with post-acquisition image analysis can give rise to high-resolution three-dimensional images. A workflow for FIB processing of infected eukaryotic cells has been published elsewhere (Medeiros et al., 2018), as well as a detailed description of combining FIB with SEM to image tissues and epithelial cells (Drobne, 2013) and a workflow for SBF-SEM (Russell et al., 2017). Although these methods are labour intensive, and require optimization and extensive sample preparation, the use of these techniques with electron microscopy has the potential to generate highly informative images of pathogens interacting with the internal compartments of host cells in three-dimensional space.

Likewise, several research questions in host–pathogen interaction studies require the high magnification and resolution presented by electron microscopy but would also benefit from the specific labelling, spatial resolution and depth of field offered by fluorescence microscopy. In addition to super-resolution techniques discussed above, correlative light electron microscopy (CLEM) (reviewed by de Boer et al., 2015) will likely emerge as another key tool in the future of host–pathogen imaging. While not extensively used in studies that fall within the scope of this Review, CLEM can, in theory, provide detailed cellular context for structures and processes labelled by fluorescent proteins or be used to pre-select rare events for EM analysis. A recent methods paper describes a workflow that can be used to combine confocal light microscopy of intact epithelial monolayers with SEM to address interactions between polarized epithelial cells and bacterial pathogens (Kommnick et al., 2019). In a study discussed above, CLEM had also been used to obtain information about the ultrastucuture within infected host cells (Mostowy et al., 2010). Specifically, the authors used fluorescence light microscopy to identify the host SEPT6–GFP surrounding intracellular Shigella, and then employed SEM to determine the morphology of the filamentous septin assemblies within the host cell at high resolution (Mostowy et al., 2010). Importantly, the authors were able to use SEM to image the ultrastrucuture of the interior of the host cell because they first removed the host-cell membrane by treatment with Triton X-100 in a cytoskeleton stabilization buffer (Mostowy et al., 2010). Immunogold labelling was also used to identify the specific location of host-cell phosphorylated myosin light chain (pMLC) and SEPT6, in the samples prepared for SEM, which supported the conclusion that these specific host proteins form the cages that serve to compartmentalize the intracellular bacterium (Mostowy et al., 2010). In a more recent study, CLEM was used to correlate images of fluorescently labelled Mycobacterium tuberculosis and host-cell membranes with SEM images to locate replicating bacteria within the cytosol and autophagosomes of the host cell (Lerner et al., 2016). In that study, SBF-SEM was used to gain access to the inside of the host cell at the precise location of previously identified population of EGFP-labelled M. tuberculosis (Lerner et al., 2016). The results from the CLEM experiments showed that a subpopulation of M. tuberculosis is able to grow in LC3-positive autophagic compartments within host lymphatic endothelial cells, which are specialized epithelial cells that line lymphatic vessels (Lerner et al., 2016). The authors noted that finding replicating bacteria in this location was unexpected considering the established role of autophagy in the antimycobacterial host immune response (Lerner et al., 2016). Importantly, obtaining this specific observation required the incorporation of live-cell imaging, specific light-microscopy-based labelling, preselection of areas within the host cell that contain bacteria and ultrastructural information about host cell compartments, which could only be accomplished using CLEM techniques.

While the imaging approaches and recent technical advances discussed throughout this Review have contributed greatly to the advances in the field, we would also like to see more complexity added at the level of experimental design. This could include studies that use multiple bacterial strains (laboratory strains and/or clinical strains) (Hendricks et al., 2016; Starner et al., 2006; Weight et al., 2019), and/or epithelial cell lines (Weight et al., 2019), types (Guevara et al., 2016; Hendricks et al., 2016), or models (Carvalho et al., 2005), to demonstrate universal mechanisms that are independent of model-specific variables. Furthermore, we chose to focus on imaging studies of two-component models (i.e. one epithelial cell and one pathogen) because the relative simplicity of the cell culture techniques makes the models discussed here accessible to most research laboratories, and because this reductionist approach is an important first step in understanding the basic interactions between bacteria and host epithelial barriers. However, as our understanding of bacterial interspecies communication (Azimi et al., 2020) and community-based bacterial phenotypes (e.g. cheating and competition) (Dunny et al., 2008; Rainey and Rainey, 2003), as well as the collective role of bacterial communities in the expression of infectious disease (Chow et al., 2011; Libertucci and Young, 2019; Vayssier-Taussat et al., 2014) progresses, it will be important to expand these models to include more than one pathogen. Similarly, adding complexity by using host models with more than one mammalian cell type (Barrila et al., 2017; Noel et al., 2017; Papazian et al., 2016) will expand the utility of infection modelling that is based on cell culture systems. Importantly, as the complexity of both imaging and infection modelling continues to increase, collaborative studies between research groups with expertise in each of these areas will become even more important, allowing for the combination of multiple, powerful and complementary methods that will allow scientists to improve their knowledge of host–pathogen interactions.

The authors would like thank Dr Michaela Strüder-Kypke, manager of Advanced Light Microscopy Unit within the Molecular and Cellular Imaging Facility, University of Guelph, for her insightful comments and guidance during the writing of this manuscript.