Pseudomonas aeruginosa host-pathogen interactions in human corneal infection models

Purpose: To examine over time, the electron microscopic changes associated with Pseudomonas aeruginosa (PA) and human corneal tissue interactions in the context of microbial keratitis. Methods: Corneal stromal fibroblast monolayer and whole tissue models were made from human eye bank eyes and from residual tissue after corneal transplantation. In the whole tissue model (WTM), donor buttons were infected with PAO1 by scoring and intrastromal injection. Tissue was examined after 3, 9 and 24 hours post challenge by transmission electron microscopy (TEM) and scanning electron microscopy (SEM). In the cell culture model (CCM), corneal fibroblasts (CF) were infected in vitro with PAO1. Bacterial adherence and internalization were assayed at 3, 6 and 9h by SEM and TEM. Adherent and internalized bacteria were measured by the gentamicin protection invasion assay. A subset of infected fibroblasts was treated with gentamicin to study intracellular bacterial survival and cell viability using a lactose dehydrogenase assay (LDH). Results: In the WTM, bacteria were seen to penetrate the epithelium at the scored sites only. At 3h bacteria were seen in the stroma and by 9h distinct intrastromal bacterial colonies were observed. Clusters of intracellular bacteria were observed in keratocytes in the intrastromal injection model. In CCM, SEM demonstrated bacteria adherent to the surface of CF and the saponin lysis assay demonstrated adherence and internalization in a doseand time-dependent manner. Bacterial internalization was detected as early as 3h. Intracellular bacteria survived and replicated without affecting cell viability. Conclusion: PAO1 bacterial can infect stromal keratocytes only when the epithelium and basement membrane are breached, or bypassed by direct injection. PA interaction with CF occurs very early leading to internalization of bacteria where they are protected and can multiply intracellularly without affecting CF viability for 24 hrs. This may have relevance to ideal timing of medical intervention.


Introduction
Microbial keratitis remains a significant public health problem as a major cause of visual disability around the world, and yet it is alarmingly under-diagnosed and under-treated 1 .
There is an estimated 1.5-2 million people that develop corneal ulceration every year in developing countries, 2 and as much as 40% of childhood and 20% of adult corneal blindness are caused by infectious keratitis. Corneal infections, including trachoma, were the leading cause of unilateral and bilateral corneal blindness in some areas 3 4 .
Contact lens wear accounts for up to 50.3% of cases of microbial keratitis 5 in the western world, and Pseudomonas aeruginosa is the most commonly isolated causative organism for contact lens associated microbial keratitis, accounting for up to 68.8% of isolates [6][7][8][9][10][11] . P. aeruginosa microbial keratitis is one of the most devastating, rapidly progressing corneal infections, which can cause corneal perforation, endophthalmitis and, ultimately, blindness.
Numerous live animal models of microbial keratitis have been developed to study the effects of bacterial colonization and virulence factors, the host responses to the pathogen and the interplay between bacterium and host. In the vast majority of these models, the corneal epithelium is bypassed either by scoring it down to the epithelial basement membrane, or by directly injecting bacteria into the stroma. This is because in the healthy cornea, the intact corneal epithelium is a formidable barrier to invasive microbial keratitis, and only when the bacteria reach the corneal stroma, does a corneal ulcer develop. A number of animal studies have investigated the ultrastructural host-pathogen interactions that occur in microbial keratitis, but these have not been replicated in human tissue.
Although these animal models are fairly representative of the pathology, they are limited by the significant differences in the morphology and dimensions between, for example, murine and human corneas 12 . In addition, the pharmacokinetics of substances is different in humans from all laboratory animal species, and there are considerable differences in metabolic rates, body temperature, elimination routes, carrier proteins/protein binding, and defence mechanisms 13 .
Apart from interactions with the host immune cells, invading organisms affect the resident corneal epithelial cells and keratocytes. The response of these cells to the microbes is the forerunner for the immune response that follows. In vitro and ex-vivo models do not allow the study of the immune cell response, but this is an advantage when it comes to studying specific responses of the resident cells, as the picture is not compounded by the infiltrating inflammatory cells. In the current study, we designed in vitro and ex-vivo models of microbial keratitis using human whole tissue and cell culture environments to undertake an ultrastructural examination of host-pathogen interactions in a sequential manner.

Materials and Methods
Human corneal tissue and cell culture Donor human corneal buttons (hCB), whole corneas and corneo-scleral rims remaining after endothelial and penetrating keratoplasty procedures were collected. Ethics Committee approval was obtained (ethics number 06/Q1602/56). The study adhered to the tenets of the Declaration of Helsinki. Corneal tissue samples were processed within 5 days of collection. Donor age range was 63-87 years with a mean age of 68 years. The causes of death were cancer, cardiac, infections, pulmonary, neurological and others.
For the whole tissue model, whole corneas or corneal buttons obtained after Descemet's stripping endothelial keratoplasty (DSEK; epithelium and anterior stroma without Descemet's membrane and endothelium) were used. For the cell culture model, primary human corneal fibroblasts (keratocytes) (hCF) were isolated as described previously 14 .
Briefly, the epithelium was scraped off the corneo-scleral rims, the rims divided into small sections of 2-3 mm each and the samples were incubated in 5 ml culture medium with 1 mg/ml collagenase (from C. histolyticum, Sigma-Aldrich, Dorset, UK) for 3h at 37 o C.
Thereafter the suspension was centrifuged at 72 g for 1 min, and the pellet was suspended in 10 ml of culture medium in 25ml flasks. The flasks were incubated for 3-4 days until the cells were confluent at the bottom of the flask, trypsinized, resuspended and expanded in fresh medium for 3-4 passages. For TEM and SEM studies, hCF were grown to confluence on collagen pre-coated transwells (ThinCerts™, transparent pore size 0.4 µm; Greiner Bio-one, Stonehouse, UK) in the final passage. Polymerase chain reaction (PCR) was used to confirm the absence of Mycoplasma contamination.

Bacterial Culture and infection of human corneal tissue
P. aeruginosa wild type strain PAO1 Holloway1C Stanier131 was obtained from NCIMB (Aberdeen, UK) and stored at -80°C. Prior to each experiment, bacteria were grown overnight on nutrient agar (Oxoid, UK) in an incubator at 37 o C, 5% (v/v) CO 2 . From this a bacterial suspension was prepared to attain an optical density (OD) (Hitachi U-1100 spectrophotometer, Hitachi, Tokyo, Japan) equal to an infective dose of 10 7 CFU/mL. Ex-vivo model. Two methods of infection were used: the intra-stromal or the transepithelial methods. In the intra-stromal method, 50 µL of infective solution were injected into the corneal stroma from the endothelial side using a 30 gauge needle. The whole corneas were incubated in culture medium [DMEM F12 medium with 2% FBS] at 37 o C for 24h. Thereafter, the corneas were washed 4 times in warm sterile PBS and processed for electron microscopy. In the trans-epithelial method, two buttons were used. One button was scarified by a 27 gauge needle to produce 4-5 linear epithelial scores that were deep enough to bypass the epithelium and score the underlying stroma. The other button was left with an intact epithelium and used as control.
TEM studies: A five-millimetre corneal trephine was inserted into the centre of each corneal button (scored and unscored) through at least half the thickness of the button so that the hollow of the trephine acted as a well in which the infective medium was placed. The corneas with the trephines were then immersed in culture medium, such that the cornea was submerged in the medium but the trephine acted as a barrier avoiding contact between the infective medium in the well and the culture medium outside of it. This was incubated at 37 o C for 9h.
SEM studies: Two buttons were used. One button was scored and divided into two halves.
One half was infected and the other not. The unscored button was similarly treated yielding one half that was infected and the other not. Infection was achieved by immersing the buttons in infected medium while the non-infected controls were immersed in clean culture medium.
Cell culture model. For SEM and TEM studies 1 mL of infective solution was added to each transwell containing a confluent monolayer of hCF. The plates were incubated in 5% (v/v) CO 2 at 37 o C for 3 to 24h. The cells were fixed and processed for electron microscopy. For association and invasion studies, 1 mL of infective solution was added to each monolayercontaining well, and the plates were incubated in 5% (v/v) CO 2 at 37 o C for 3 to 24h.

Scanning Electron Microscopy (SEM)
For the cell culture model, the cells were rinsed with sterile PBS and then processed for each. The transwell membranes were detached and subjected to dehydration using graded concentrations of ethanol from 30%, 50%, 70%, 95% and up to 100% (v/v) solutions, 10min for each step and 20min for the final step. The membranes were critical-point dried and mounted on to metal stubs for the final sputter coating step. Coated specimens were viewed using the Quanta 200 scanning electron microscope (FEI, Hillsboro, USA) and several images were obtained for each specimen at magnifications of x3200 and up to x24000. A similar method was used for the whole tissue model using hCB instead of hCF membranes.

Transmission Electron Microscopy (TEM)
Human corneal tissue (hCB and hCF) were processed in a manner similar to that described for SEM above, until the ethanol dehydration step. The samples were incubated overnight, on a rotor (2rpm), in spur resin at a 50:50 ratio with acetonitrile. After that, the samples were placed in fresh spur resin and polymerised in an electric oven at 60°C for 20-24h. In preparation for microscopy, sample blocks were cut into ultrathin sections (about 90 nm thick) using glass knives on a Leica Reichert Ultracut E microtome (Leica, Wetzlar, Germany).
Several sections per sample were fixed on to copper grids and stained using lead nitrate for 5min in the presence of NaOH crystals. The grids were briefly rinsed in distilled water and air dried. Sample viewing was done using either a H7000 transmission electron microscope (Hitachi, Japan) at 75Kv or a Tecnai T12 BioTwin transmission electron microscope (Hillsboro, OR, USA) at 100kV. Images were captured using a Megaview III Soft Imaging System (SIS) camera.

Assessment of bacterial association to and invasion of corneal fibroblasts
To quantify bacterial association to hCF, monolayers were challenged in triplicate in 24 well plates with multiplicity of infection (MOI) of bacteria to cells of 0.0001-1000 (Table 1). Total bacterial association was quantified over time using the viable counting and saponin lysis method, as described previously 15 . Briefly, hCF monolayers were challenged in triplicate with a MOI of 0.001, 0.1 and 100 and incubated at 37 o C for 3, 6, 9 and 24h. Cell monolayers were gently rinsed with warm PBS and attached bacteria were released using a saponin solution [10 mL PBS containing 1% (w/v) saponin powder and 1% (v/v) decomplemented foetal bovine serum]. The saponised lysates were inoculated on agar plates, incubated at 37 o C for 24h and colony-forming units (CFU) were quantified using a ProtoCOL model 60000 automated colony counter (Synoptics Ltd, Cambridge, UK).
Bacterial invasion was quantified as described previously 16 . Briefly, confluent hCF monolayers were challenged in triplicate with a MOI of 100 and incubated at 37 o C for 3h.
Cell monolayers were rinsed with PBS and then 1ml of medium containing 200 µg/ml of gentamicin (Sigma-Aldrich, Dorset, UK) was added to the wells and incubated at 37 o C for 90min. Monolayers were washed with PBS and intracellular bacteria were released using a saponin solution, and bacteria were quantified by viable counting on nutrient agar plates.
Bacterial association and invasion were assessed simultaneously in different wells. Wells treated with gentamicin yielded invaded bacteria only and wells not treated with gentamicin yielded both attached and invaded bacteria.

Confocal Microscopy
To analyse actin stress fibres, hCF monolayers were infected for 3h with Green Fluorescent Protein

Immunoprecipitation and western blot analysis
The RhoA/Rac1/Cdc42 Activation Assay Combo Biochem kit TM (Cytoskeleton, Inc) was used according to the manufacturer instructions. hCF monolayers were grown in DMEM containing 1% (v/v) dFCS and at 70% confluence, the medium was changed to serum-free medium. Serum-starved monolayers were infected with wild-type PAO1 for 1h (MOI=100) and total protein was extracted. Equal amounts of total protein extract was used to immunoprecipitate RhoA, Rac1 and Cdc42 total protein independently using rhotekin-RBD and PAK-PBD affinity beads, loaded onto a 12% (w/v) SDS-PAGE gel and transferred onto PVDF membranes (GE Healthcare). GTPase activation was examined with mouse monoclonal anti-RhoA, -Rac1 and -Cdc42 antibodies supplied in the kit and used at the recommended dilutions. Goat anti-Mouse IgG (H+L)-HRP Conjugate (Bio-Rad) was used as a secondary antibody at the recommended dilution.
To assess Src-kinase protein expression, total protein was extracted from hCF monolayers infected for 6h with PAO1 wild-type strain (MOI=10). Total protein (5μg) was separated by 12% (w/v) SDS-PAGE and transferred onto PVDF membranes. Polyclonal Src antibody ab7950 and anti-Src[pY418] (Abcam) were used to detect non-phosphorylated and phosphorylated Src, respectively. Goat anti-Rabbit IgG H&L HRP pre-adsorbed secondary antibody (Abcam) was used according to the manufacturer's recommendations. Immuno-detection was carried out in a Versadoc 4000 (BioRad) and data analysed with Quantity One 4.6.9 software (BioRad).

Cell viability assessment
Cell viability was quantified using a Lactate dehydrogenase (LDH) assay (CytoTox 96® Non-Radioactive Cytotoxicity assay kit, Promega, Madison, WI, USA). Confluent fibroblast monolayers were challenged with bacteria for 90min before adding gentamicin. Cytotoxicity levels were calculated by dividing the average LDH release value of test samples by the average maximum LDH release value from lysed uninfected cells.

Statistical Analysis
Statistical analysis was done using SPSS version 22.0 (IBM, Armonk, NY, USA). For comparison between two groups, an unpaired Student's t-test was used. For non-normal distributions, the Mann-Whitney rank sum test was used. To determine the difference in the frequency of infection between hCB groups, a χ2 test was used.

Results
The corneal epithelium is a formidable barrier against infection SEM of corneal buttons showed that, unlike the smooth surface and the minimal exfoliation of superficial epithelial cells in control samples, the epithelial surface appeared much rougher in corneas infected with PAO1 bacteria for 3h ( Figure 1A-D). The number of exfoliating cells was also higher; in a given area of 60x60 µm, 21 exfoliating cells were seen in the infected cornea versus 11 in the control sample. With higher magnification, P.
aeruginosa bacteria were seen adhering to exfoliating epithelial cells ( Figure 1E), and some bacteria seemed to be penetrating the epithelial cell surface ( Figure 1F).
In a light microscopic section of scarified hCB, it was evident that the scores had bypassed the epithelial basement membrane into the stroma. Epithelial cells seemed to migrate to fill the gap (Figure 2A). In the TEM of the non-scarified hCB, the bacteria were seen in all layers of epithelium but not beyond the basement membrane (BM). They were seen to line up on the epithelial side ( Figure 2B) of the BM. In the scarified button, the area of the epithelial score showed epithelial cells and PAO1 bacteria within the score and invading the adjacent stroma ( Figure 2C and D). At the later time point of 9 hours bacteria appeared to aggregate within the stroma forming small colonies ( Figure 2E and F). At 24h, bacteria were found to accumulate inside stromal keratocytes, with occasional bacteria in the surrounding stroma ( Figure 2G-I).

P. aeruginosa bacteria adhere to and invade human corneal fibroblasts
Bacterial adherence was demonstrated by SEM. Bacteria were seen adhering to the surface of the fibroblasts, with a greater number of adherent bacteria observed with higher MOI (Figure 3B and C). This was confirmed by the saponin bacterial association assay. The bacteria showed a linear dose-dependent association with hCF monolayers over time (MOI 0.0001, 0.001, 0.01 and 0.1 - Figure 3A). This demonstrates that bacterial association to corneal fibroblasts is a normal, repeatable host-pathogen interaction not occurring by chance. Infection with MOI 10, 100 and 1000 led to a rapid saturation of the monolayers and monolayer destruction by 9-24h.

Protein tyrosine kinases (PTK) are involved in bacterial internalisation into corneal fibroblasts
A molecular analysis of PAO1 invasion pathway was done to explain the temporal phenotypic changes induced as a response to hCF infection. Protein tyrosine kinases (PTKs) are considered to be important for bacterial invasion of mammalian cells 17 . Specifically, Src kinases are involved in the signalling pathway downstream the bacterial-stimulated integrins 18 . To determine the involvement of PTKs in PAO1 uptake by hCFs, genistein (50M) was used to block total PTKs, and PP2 (10M) to inhibit Src kinase during PAO1 invasion. There was a significant 2 log reduction in hCF invasion by PAO1 (P<0.0001) with both genistein-and PP2-treated hCF compared to untreated cells (P<0.0001) ( Figure 5A).
Human Src kinases are composed of a N-terminal SH3 and SH2 kinase domain and a Cterminal regulatory tyrosine residue (Y 529 ). When the C-terminal-tyrosine is dephosphorylated, a change in protein conformation promotes trans-autophosphorylation of Y 418 that allows kinase activation [19][20][21][22] . Phosphorylation and consequent activation of Src kinase during PAO1 hCF-invasion was confirmed by western blot analysis. Protein band densitometry relative to -actin revealed a significant 87.5% induction of total Src protein, 53% of which was phosphorylated, as a percentage of total Src kinase induction, in hCF infected with PAO1 and differently to non-infected monolayers, in which no Src protein could be detected ( Figure 5B).
RhoA, Rac1, and Cdc42 are Rho family small GTPases involved in signalling pathways for cytoskeleton actin rearrangements 23 24 . RhoA, Rac1 and Cdc42 proteins were immunoprecipitated from PAO1-infected hCFs and protein levels analysed by western blot. The three GTPases were identified in PAO1-infected hCF, but only RhoA showed the GTP bound activation state. Even though RhoA activation was observed in uninfected hCFs, protein band densitometry relative to uninfected condition revealed 52% induced RhoA-GTP in infected hCF compared to control (uninfected) monolayers ( Figure 6A). RhoA involvement in PAO1 hCF uptake was confirmed by analysing bacterial invasion following RhoA inhibition with Rhosin. RhoA inhibition significantly reduced PAO1 invasion of hCF in a Rhosin dosedependent manner (P<0.01 with 30M Rhosin and P<0.001 with 50M Rhosin) ( Figure 6B).
Confocal microscopy confirmed the disassembly of stress fibres promoted by PAO1 activation of RhoA during hCF invasion. The intensity of stress fibres detected after infection with PAO1 wild-type strain decreased and cell arches disappeared, compared to the long and well-developed stress fibres observed in control monolayers. Nyquitist sampling confirmed stress fibre disruption in PAO1-infected hCFs ( Figure 6C).

P. aeruginosa bacteria use corneal fibroblasts as reservoirs for protection and reinfection
In both the cell culture and the whole tissue TEM studies, bacteria were seen clustered inside corneal fibroblasts. In the subset of fibroblast monolayers that were treated with gentamicin 3h post challenge, all extracellular bacteria were destroyed but the number of bacterial recovered after 24hours had significantly increased suggesting that surviving PAO1 bacteria replicated inside corneal fibroblasts (Figure 7). Samples of supernatant taken after 24 hours showed no bacterial growth. Control samples showed no effect of gentamicin on non-infected cells as has been previously demonstrated.
The LDH assay showed that infected fibroblasts without antibiotic treatment released significantly more LDH (~50-100%) than uninfected cells (p<0.05). However, there was no significant difference between the levels of LDH released from hCF with intracellular bacteria treated with gentamicin and uninfected cells at 9 or 24h (P>0.05), suggesting that cell viability was not affected by the presence of intracellular bacteria (Figure 8).

Discussion
Early studies of PA microbial keratitis show progressive destruction and liquefaction of the cornea that rapidly leads to perforation unless intense antibiotic treatment is immediately initiated 25 . This has long been noticed in clinical practice, and patients presenting with suspected microbial keratitis are immediately started on intensive topical antibiotic treatment, without waiting for culture and sensitivity results, which when available are used to modify treatment if there is an insufficient response. Clinically it is known that infected corneal ulcers require immediate treatment but the exact duration is difficult to quantify.
This study was conducted to ascertain how rapidly bacteria interact with corneal tissue cells and propagate through the cornea. Although the surface epithelial cells desquamate to reduce the bacterial load and the epithelial basement membrane prevents further propagation of bacteria into the stroma, bacteria colonise the stroma and rapidly interact with corneal fibroblasts when there is a breech in defences. They adhere to and invade the cells as early as 3h and destroy them by 9h, or they can survive and replicate intracellularly without affecting cell viability, using fibroblasts as reservoirs for self-protection and re-infection.
The normal wound healing response observed in our whole tissue model demonstrated that this was a viable model for this study. When corneal buttons were infected with P. aeruginosa, they showed a rougher epithelial surface and greater levels of desquamating cells than uninfected buttons, which is consistent with findings in a rat model, where most internalized bacteria were found in cells that were readily desquamated from the cornea with rinsing 26 . This indicates that the internalization/desquamation sequence is a defence mechanism that the human cornea also employs to reduce the bacterial load. We also showed that P. aeruginosa successfully adhered to corneal epithelial cells, and managed to penetrate the entire epithelium but could not penetrate further into the stroma unless there was a breach in the epithelial basement membrane, or when injected directly into the stroma. This is consistent with studies utilizing mouse models showing similar findings 27 . Thus, although these bacteria can invade intact epithelium, they penetrate the basement membrane with difficulty.
In the stroma, P. aeruginosa bacteria rapidly interact with corneal fibroblasts, and are able to colonize the stroma quite efficiently. In our model, PAO1 bacteria adhered to corneal fibroblasts in a linear dose-and time-dependant manner, starting as early as 3h post challenge, and increasing progressively with time. Our results also showed that P. aeruginosa internalisation into corneal fibroblasts could be detected as early as 3h post challenge, where bacteria entered cells by endocytosis and resided in cytosolic vesicles.
Bacterial internalisation by corneal fibroblasts appears to be an important host-pathogen interaction, which could confer survival advantage to the bacteria. Internalised bacteria can survive and replicate inside invaded fibroblasts without affecting fibroblast viability for up to 24h. Apart from directly colonising the stroma, bacteria also cluster inside stromal keratocytes. This indicates that, although in a clinical scenario bacteria do not particularly need to invade corneal fibroblasts in order to traverse the stroma, they may well use this internalization to evade host defences and instilled antibiotics. This is sometimes noticed clinically, when ulcers remain active despite antibiotic treatment, and when a "responding ulcer" flares up on withdrawal or reduction in frequency of instillation of the antibiotic. In a retrospective review of therapeutic keratoplasty in Singapore 28 , P. aeruginosa accounted for 58.7% of refractory microbial keratitis requiring penetrating keratoplasty.
Interestingly, all patients were treated initially with gentamicin and cefazolin, and then other modalities were tried when the ulcers were progressively worsening. It is possible that in those patients, bacteria would have survived intracellularly, safe from gentamicin, and reached a critical threshold when they externalised, probably through cellular disruption of fibroblasts/kertocytes. The large number of externalised bacteria would overwhelm host defences and perpetuate infection.
In our model we were able to replicate in human tissue, observations made in animal models.
Although many investigators have used mouse models to study the pathology of microbial keratitis, the mouse cornea has many structural and morphological differences to the human cornea. It is much smaller, with a diameter of 2.3-2.6 mm, compared to 11-13 mm in humans, and much thinner, measuring 122-137 µm, compared to 550 µm on average in humans. The murine epithelium represents one third of the corneal thickness compared to one tenth in humans, and is composed of 13 layers of cells with multiple layers of squamous cells, compared to 5-6 layers in humans. There is also no distinct epithelial basement membrane and Bowman's layer, but rather a much less welldeveloped anterior limiting lamina separating the epithelium from the stroma. Finally, the anterior stroma does not show the dense and complex interweaving of collagen fibres seen in human corneas. 12 By using our human tissue derived models, the ultrastructural host-pathogen interactions can be accurately described in the context of human tissue and closely resemble changes in real-life clinical scenarios.
Our model has a number of limitations. A significant part of the morbidity associated with microbial keratitis is attributable to the host immune response, and our model lacks the presence of immune cells, which are pivotal in the pathogenesis of microbial keratitis. Our model is more reflective of changes occurring in infectious crystalline keratopathy, where the local host immune response is compromised. Further studies using these models with the introduction of immune cells to simulate microbial keratitis more closely, may provide more relevance. We used one laboratory strain of P. aeruginosa (PAO1), which might behave differently than clinical isolates. The models are also amenable to further studies using pathogenic clinical isolates from patients with corneal ulcers.
The sequence of events observed in this study provides a useful temporal insight into the ultrastructural changes that accompany the host-pathogen interactions in corneal cells. This information is of relevance to the progression of microbial keratitis and can provide some understanding of the pathogenesis of keratitis and timing of medical intervention.  [Original magnification; A= x200, C= x1600, D and E= x5000, B and F= x18500, G= x4200, H= x16500, I= x26500]