Localization of bacteria within the epithelium of fresh whole tissue
The tissue we chose to use for this study, the corneal epithelium, is a multilayer of epithelial cells that covers the cornea of the eye. The cornea lends itself to imaging because it is both external and transparent. Epithelial cells on the corneal surface are viable (non-keratinized), and are normally bathed in fluid (tears) allowing the use of a water immersion objective with minimal disruption to normal physiology.
The healthy corneal surface does not normally bind bacteria or other microbes. Studying bacterial interactions with the corneal epithelium thus necessitates some form of manipulation to enable bacteria to adhere. As discussed above, superficial blotting with tissue paper can be used for that purpose, and it is minimally disruptive to tissue architecture 
. Subsequent epithelial traversal by the adherent bacteria can then be toggled on (or off), by use (or omission) of EGTA treatment after the blotting procedure.
With methods in hand to encourage bacteria to interact with the epithelium, the next goal was to develop methods for visualizing/localizing bacteria within the bacterially-challenged tissue that do not require processing (i.e. fixation, labeling, staining) of the tissue. Details of the methods utilized are outlined in the materials and methods. Briefly, bacteria were visualized using a multicopy plasmid expressing enhanced green fluorescent protein (GFP). Corneal epithelial cells within bacterially challenged, and then freshly excised unprocessed eyeballs, were imaged using one or more of three different techniques: 1) NAD(P)H autofluroescence (AF) (multiphoton, images cytoplasm of metabolically active cells). 2) A reflection technique (confocal, images both live and dead cells). 3) Use of transgenic mice expressing membrane-bound cyan fluorescent protein (CFP) (confocal, enables cell membranes to be visualized). All three methods enabled cells within all layers of the cornea (epithelium, stroma and endothelium) to be visualized (). Thus, it was possible to accurately localize the upper and lower limits of each layer, providing landmarks for quantifying the depth of microbe penetration using any of the three methods.
3D imaging of mouse corneas using freshly excised mouse eyeballs.
Each of the three imaging modalities had unique features for gaining more detailed information. For example, since NAD(P)H resides within the cytoplasm, the NAD(P)H AF method enables subcellular imaging; the cytoplasm fluoresces, while the nucleus and plasma membrane appear black in contrast (). Since its fluorescence intensity is proportional to cellular metabolic activity 
, this technique also provides information about the health of individual cells (). The reflection confocal and membrane-CFP methods each allow positive (rather than negative) visualization of epithelial cell boundaries, and thus can be used to ensure that both dead and live cells are accounted for (). Thus, when used in combination with NAD(P)H AF, confocal or membrane-CFP can enable live and dead cells to be distinguished from one another (e.g. ). The AF and reflection methods (in contrast to membrane CFP) do not require a specific type of genetically modified animal, which means that any animal can be studied, including wild-type and all genetically modified species. Membrane-bound CFP, on the other hand, provides a high signal-to-background ratio, and consequently a high image contrast of cell membranes ( and Video S1), compared to either AF or reflection ( respectively), making this an ideal method for determining if infecting bacteria are intracellular or extracellular.
Images from bacterially-challenged corneas (6 h) are shown in . The top panels show results for corneas that were first blotted and then treated with EGTA for 1 h prior to bacterial challenge. Both the confocal CFP membrane method ( and Video S2) and the NAD(P)H autoflurescence method ( and Video S3) revealed epithelial traveral by the bacteria; i.e. many bacteria were detected deep within the multilayered corneal epithelium. Consistent with our previously published findings showing that the the basal lamina between the epithelium and the stroma functions as a barrier to bacterial passage 
, few bacteria were noted in the underlying stroma. Also consistant with our previous results 
, corneas that were not subsequently treated with EGTA prior to bacterial challenge (blotted only, bottom panels and Video S4) were susceptible to bacterial adhesion only, without bacterial penetration beyond the epithelial surface.
3D imaging of bacterially challenged eyeballs.
Having established methods for enabling and also visualizing bacteria deep within unprocessed and unsectioned live tissue, we next explored if these methods could be used to provide more detailed spatial and temporal information about the traversal process and its impact. Use of the CFP membrane mouse and high magnification enabled the position of bacteria relative to cell membranes, and the status of those membranes, to be determined. For example, perusal of images in reveals that the CFP signal is relatively diffuse after bacterial traversal of blotted/EGTA-treated corneas relative to the very clearly localized CFP signal seen in the uninfected eye in . Blotting/EGTA treatment used alone without bacterial challenge did not cause CFP diffusion (data not shown), suggesting that it was the bacterial traversal per se
that impacted the appearance of the cell membrane marker; perhaps not surprising considering that P. aeruginosa
can elaborate multiple toxins 
and basolateral cell surfaces are highly susceptible to them 
. Use of the CFP-membrane mouse also revealed that P. aeruginosa
can induce the formation of membrane blebs, and can localize within them, a phenomenon which we have reported to occur when P. aeruginosa
infects cultured corneal epithelial cells 
. each show examples of membrane blebs; the higher magnification used in enabled a bacterium to be located within a bleb. While the significance of “bleb niche” formation in epithelial cells during P. aeruginosa
infection is still under investigation, this finding that they occur in actual infected fresh tissue, and not only in cells grown and infected in tissue culture wells, lends justification to that effort.
Temporal and spatial tracking of EGTA-enabled bacterial traversal.
One clear advantage of imaging live versus fixed tissues is that temporal information can be obtained using the same sample. For example, individual bacteria and cells within the epithelium of fresh tissue can be tracked over time for their location and/or their viability ( and Video S5). Video S5 shows that the bacterium located within the membrane bleb in is exhibiting swimming motility inside the bleb, confirming another phenomenon that we have reported with cultured cells, but this time for a cell located within the corneal epithelium of a live eyeball. (high magnification z -stack) shows several bacteria in the basal (deep) layers of the corneal epithelium of intact eyeballs mostly located between epithelial cells. In this image moving bacteria are shown in red, while white bacteria are stationary since they were captured twice (sequentially in CFP and GFP channels) in the same z-plane (temporally spaced images were overlaid). In addition to providing insights into the temporal details of bacteria-cell interactions, these methods could also be used to monitor individual host cell responses to bacterial challenge over time (e.g. by morphology or by tracking NAD(P)H AF), and with respect to the position of individual bacteria.
Lack of significant bacterial penetration into the stroma as a result of an intact basal lamina likely explains why blotting/EGTA treatment is not sufficient to enable susceptibility to overt (clinically visible) disease despite bacterial traversal through the epithelium 
. We next explored the usefulness of these imaging methods when there is overt disease, i.e. when the cornea becomes optically “opaque” (non-transparent) due to the infiltration of leukocytes, e.g. neutrophils and monocytes 
. Disease was induced by scratching the cornea through to the level of the anterior stroma (damages basal lamina barrier) prior to challenging with bacteria 
. It remained possible to detect cellular NAD(P)H AF and to localize GFP-labeled P. aeruginosa,
even when the infecting bacteria were located deep within the stromal region of these opaque corneas (, Video S6). The images collected revealed interesting insights illustrating the power of this technology compared to fixed/sectioned tissue. For instance, the images shown in and Video S6 were collected at a region of the cornea where the basal lamina was not directly impacted by the scratching process. This z-stack was collected with the optical axis of the microscope parallel to, but laterally shifted from, the optical axis of the eye and as a result of the curvature of the eye, the image captured in (taken from the stack) shows the epithelium at the top and the interface between the epithelium and the stroma (basal lamina region) towards the bottom. This enabled direct visualization of the “filtering” effect of the basal lamina, i.e. bacteria can be seen deep in the epithelium arranged in circular patterns surrounding cells/nuclei with concentrations tapering off at, and under, the basal lamina. While the images collected directly under the basal lamina showed a distinct bacteria free zone, deeper regions of the stroma again revealed the presence of bacteria, this time arranged in a different pattern. Here, bacteria were found radiating out laterally in single file, end-to-end lines, corresponding with orientation of the stromal collagen fibrils (, Video S6). Images collected from adjacent regions showed that these bacterial “trains” originated at the scratched region where bacteria enter the corneal stroma directly through the damaged basal lamina, and then eminated away from the area in multiple directions. Whether P. aeruginosa
utilizes a form of surface-associated motility along collagen fibrils to disseminate through the cornea is to be determined. This is of interest, considering that twitching motility is a critical virulence factor for P. aeruginosa
in this corneal infection model 
Imaging of bacteria and cells within opaque infected mouse corneas.
Impact of MyD88 on bacterial traversal
To gain further mechanstic insights into how the corneal epithelium normally defends itself against bacterial traversal, the role of MyD88, a central adaptor protein for TLR signaling, was examined. This was done by comparing wild-type to MyD88 knockout (−/−) mice.
After image acquisition (z-stacks), image stacks were analysed using a custom program which permited quantitative z-axis profiling of bacterial traversal from any of the three signals that localized the structure of epithelium in addition to bacterial fluorescence (). To take tissue surface irregularities into account (the corneal surface is curved), an image stack (a field) was divided into approximately 1000 sub-volumes for analysis. In each sub-volume, the algorithm located the apical and basal sides of the epithelium. The epithelium was sub-divided into ten bins along the z-axis relative to its local thickness. Background subtraction was performed on the corresponding bacterial fluorescence trace for each sub-volume. The remaining fluorescence intensity of bacteria, which served as a proxy for the number of bacteria, in each z-bin was integrated. The sum of local profiles across all sub-volumes provided the final fluorescence intensity as a function of traversal distance for a particular field.
Quantitative localization of bacteria within ex vivo blotted and bacterially challenged MyD88 deficient mouse corneas at 8 h time-point.
shows a quantitative comparison of wild-type to MyD88 knockout mice on traversal after tissue paper blotting to allow adherence (no EGTA treatment). The impact of MyD88 deficiency was found to be similar to EGTA treatment, i.e. it enabled bacteria to traverse deeply through the blotted, multilayered corneal epithelium with a significant number reaching the basal lamina by 8 h post-inoculation as compared to wild-type controls that showed no significant traversal ( and Video S7).
Impact of MyD88 deficiency on defense against epithelial traversal by bacteria after blotting.
Close inspection of the NAD(P)H AF images suggested that the epithelium of MyD88 knockout murine corneas was altered after the blotting and the subsequent 8 h of bacterial challenge. Many of the epithelial cells were rounded, and the total number of cell layers appeared reduced. While cell rounding was also seen after bacteria had traversed EGTA-treated corneas ( and ), the apparent “thinning” of the entire epithelial layer was observed only for challenged MyD88 knockout eyes. To explore the mechanism for these morphological changes, experiments were repeated using a shorter challenge time frame (4 h versus 8 h). The data showed unusually large numbers of bacteria adhering to the surface of MyD88 knockout corneas at 4 h, without significant traversal ( and Video S8). MyD88 mutation enhanced bacterial adhesion at time points earlier than it enabled traversal. Further, the underlying corneal epithelium appeared morphologically normal at the earlier (4 h) time point. Thus, changes to epithelial morphology noted at 8 h must have occurred after bacteria bound, and were probably caused by the bacteria, rather being a direct result of MyD88 mutation. Indeed, we might expect cells lacking MyD88 (important in innate defense responses) to be more sensitive to negative effects of bacteria than “normal” EGTA exposed cells.
Use of the reflection method in combination with NAD(P)H autofluorescence allowed the impact of 8 h of bacterial challenge on cell viability to be examined ( and Video S9). That data showed that the apparent “thinning” of the epithelial cell layer was actually due to loss of metabolic activity of many of the surface cells; i.e. some cells were visible by the reflection method, but not by NAD(P)H AF. While there were some bacteria located at the interface between non-active (visible by reflection) and metabolically active (visible by NAD(P)H autofluorescence), bacteria were also found located between and below metabolically active cells, suggesting that bacterial traversal was not simply a consequence of cell death, despite the noted changes to cell viability.
Since blotted MyD88 knockout corneas showed increased susceptibility to bacterial adhesion and bacterial traversal compared to wild-type, we next explored its impact in the absence of blotting. Surprisingly, the corneal epithelium of freshly excised MyD88 mutant eyeballs was susceptible to both bacterial adhesion (5752 ± 2601 cfu/mm2) and subsequent bacterial traversal () without the need for blotting, or any other form of manipulation. In contrast, bacteria only occasionally adhered to wild-type control mouse corneas (672 ± 319 cfu/mm2) under the same circumstances ().
Impact of MyD88 deficiency in the absence of blotting.
A potential mechanism for MyD88 knockout corneal susceptibility to bacterial adhesion and subsequent traversal would be if the lack of MyD88 disrupted epithelial junctional integrity. To explore that possibility, MyD88 knockout and wild-type mice were compared for corneal epithelial susceptibiity to fluorescein staining/penetration, a test used widely in epithelial cell culture, whole tissues, and in the eyes of human patients in routine clinical practice, to assess epithelial permeability/junctional integrity 
. Confocal microscopy was used to determine depth of fluorescein penetration. Even without subsequent EGTA treatment, the epithelium of blotted wild-type mouse corneas was susceptible to deep penetration by fluorescein (). In contrast, neither wild-type nor MyD88 knockout corneas labeled with fluorescein when not blotted (). Therefore, susceptibility of unblotted MyD88 knockout corneas to bacterial adherence and traversal was not due to disruption of epithelial tight junctions prior to bacterial exposure. Further, the data suggest that tight junctions do not act in isolation to modulate bacterial traversal in wild-type eyes. Whatever the case, these two models enable traversal by different mechanisms, which will provide additional tools for subequent studies of epithelial defense against infection.
MyD88 corneas do not label with fluorescein suggesting tight junctions are intact ex vivo and in vivo.
Since bacteria bound to (and penetrated) unblotted MyD88 deficient corneas, MyD88 is required for defending the healthy corneal epithelial surface against bacterial adhesion, in addition to protecting it against bacterial traversal. How these findings reconcile with the established role of surface-associated mucins in protecting against bacterial adhesion is not yet clear.
The above studies were done using freshly excised eyeballs. We have reported that tear fluid, which bathes the corneal epithelium in vivo
, is protective against bacteria 
. In a more recent in vitro
study we showed that tear fluid can modulate epithelial immunity directly to protect cells against various bacterial virulence strategies, involving upregulation of both RNase7 and ST2 
. Thus, we performed experiments in which the blotting and bacterial challenge steps were both performed while the eye was in vivo
. Experiments were done both with and without blotting using 8 h of bacterial challenge, and both wild-type (control) and MyD88 knockout mice. With blotting, the in vivo
results mirrored the ex vivo
results, i.e. both wild-type and MyD88 knockout mouse corneas bound bacteria, with more extensive binding and also epithelial traversal by bacteria for the MyD88 mutant eyes (). As expected, bacteria did not associate at all with non-blotted wild-type corneas in vivo
(). MyD88-deficient corneas showed only low level bacterial adhesion and only shallow bacterial penetration beyond the surface (). This contrasted with results obtained for experiments done ex vivo
that yielded significant adhesion and penetration to the basal lamina. Indeed, ex vivo
data showed maximal GFP fluorescence intensity was 12,000 at a depth of 0.4 towards the basal lamina (); in vivo
it peaked at only 1,200 and at a more shallow level of 0.2 (). Thus, the unmanipulated MyD88-deficient corneas were less susceptible to bacterial colonization in vivo
, than they were when removed and challenged in vitro
. Whether this is explained by (MyD88-independent) protective biochemical factors in vivo
, or simply by physical removal of bacteria during blinking/tear flow, is yet to be established.
Impact of MyD88 deficiency on colonization in vivo.
Blotting/EGTA pretreatment versus MyD88 knockout mice as models for enabling traversal to be studied
While useful for studying defenses against traversal, these models also provide systems that could be used to enable the traversal process to be studied (e.g. identifying bacterial factors involved). The two approaches have different advantages; i.e. blotting/EGTA treatment can be used with mice of any background (i.e. wild-type, gene knockouts, or transgenic mice such as the CFP membrane mice), while use of MyD88 knockouts does not necessitate chemical or other treatment of the tissue to enable traveral.
Given the known functions of MyD88, its role in defense against both bacterial adhesion and bacterial traversal are likely to involve either TLR- or IL-1-mediated epithelial-, or resident macrophage or dendritic cell-, derived/initiated innate defenses. These could include antimicrobial peptides, or infiltrating phagocytes (in vivo
. Junctional integrity has also been shown to be responsive to TLR-, and MyD88-, mediated signaling in intestinal epithelia 
. While the data in show normal barrier function in naïve MyD88 knockout mice (and therefore that tight functions are functionally intact), it remains possible that regulation of junctional integrity in the face of bacterial challenge is compromised in the corneas of MyD88 knockout animals. Other possibilities are that mucins (or other epithelial surface-associated moieties that either bind or repel bacteria) are dependent on MyD88, or that they act in synergy with MyD88-mediated factors (such as antimicrobials) to exert their protective effects. Further research will be needed to sort through these and other potential possibilities for MyD88 involvement in epithelial defense against traversal by microbes.
In this report, we demonstrate that the combination of various imaging methods can be used to further our understanding of bacterial-host cell interactions in live unprocessed tissues. These relatively simple and robust non-invasive methods, used with strategies for manipulating tissue susceptibility to microbes, and a custom data analysis program, represent a set of novel, quantitative tools for studying both pathogenesis of, and defenses against, infection. Using them, we have shown directly that MyD88 plays a critical role in defense against epithelial colonization by a Gram-negative opportunistic pathogen, and demonstrating that host-pathogen interactions reported to occur in vitro
(e.g. membrane bleb-niche formation) can also occur when tissues are challenged intravitally. The data presented also suggest new insights into bacterial pathogenesis not previously reported (such as a spatio-temporal relationship between bacteria and collagen fibrils during infection), that can now be expanded upon using the developed methods. While there are clear advantages to using mice, including the availability of gene knockouts and other reagents, these methods could be used with eyes from other species or with other types of tissue. With the correct combination of resources, and if image stabilization can be achieved at the high magnification needed to detect infecting bacteria, these methods could be implemented in vivo
using anesthetized live animals as has been done for studying immune cell trafficking 
. Since these imaging methods are non-invasive, there could even be potential for translation to use in humans.