Mechanisms of vascular damage by systemic dissemination of the oral pathogen Porphyromonas gingivalis

Several studies have shown a clear association between periodontal disease and increased risk of cardiovascular disease. Porphyromonas gingivalis (Pg), a key oral pathogen, and its cell surface‐expressed gingipains, induce oedema in a zebrafish larvae infection model although the mechanism of these vascular effects is unknown. Here, we aimed to determine whether Pg‐induced vascular damage is mediated by gingipains. In vitro, human endothelial cells from different vascular beds were invaded by wild‐type (W83) but not gingipain‐deficient (ΔK/R‐ab) Pg. W83 infection resulted in increased endothelial permeability as well as decreased cell surface abundance of endothelial adhesion molecules PECAM‐1 and VE‐cadherin compared to infection with ΔK/R‐ab. In agreement, when transgenic zebrafish larvae expressing fluorescently labelled PECAM‐1 or VE‐cadherin were systemically infected with W83 or ΔK/R‐ab, a significant reduction in adhesion molecule fluorescence was observed specifically in endothelium proximal to W83 bacteria through a gingipain‐dependent mechanism. Furthermore, this was associated with increased vascular permeability in vivo when assessed by dextran leakage microangiography. These data are the first to show that Pg directly mediates vascular damage in vivo by degrading PECAM‐1 and VE‐cadherin. Our data provide a molecular mechanism by which Pg might contribute to cardiovascular disease.


Introduction
The effect of oral health on systemic disease is currently a highly debated topic with increasing evidence indicating that blood-borne oral microbes significantly contribute to conditions such as cardiovascular disease (CVD) [1,2], rheumatoid arthritis [3], diabetes [4] and Alzheimer's disease [5]. The association between periodontitis and cardiovascular disease has been studied for the last three decades and is now generally accepted in the field [6].
Periodontitis is a chronic multifactorial inflammatory disease caused by the dysbiosis of indigenous microorganisms and characterised by the destruction of tooth-supporting structures leading to tooth loss [7,8]. Porphyromonas gingivalis (Pg), an oral anaerobic microbe, is considered the keystone pathogen that promotes the development of severe periodontitis by favouring oral microbial dysbiosis and dysregulation of the host immune response [9]. Pg resides in dental plaque and produces several virulence factors that enable this microbe to colonise the host, evade the immune system and contribute to disease progression and tissue destruction [9,10]. The predominant virulence factors appear to be gingipains, cysteine proteases with significant proteolytic activity [11].
The presence of oral bacteria has been detected in peripheral blood following tooth extractions and even regular oral hygiene procedures such as tooth brushing, giving rise to a transient bacteraemia in healthy individuals [12]. However, the likelihood of bacterial entry to the circulation is greatly increased in patients with chronic gingivitis or periodontitis where bacterialmediated loss of gingival epithelial integrity facilitates bleeding, allowing pathogenic oral bacteria facile movement from the periodontal pocket into the bloodstream [13][14][15]. Indeed, oral microbiome profiling performed on patients with atherosclerotic CVD undergoing bypass surgery revealed that Pg was the most abundant species found in nondiseased vascular tissue [16]. Furthermore, Pg DNA was detected in cardiac valves of patients with CVD and deep periodontal pockets [17].
Pg is able to bind to and invade endothelial cells cultured in vitro [18,19]. This microbial-host interaction mediates increased gene expression of several chemokines (e.g. CXCL8, CCL2), adhesion molecules (CD54, CD62E, PECAM-1) and inflammatory factors by endothelial cells via various mechanisms [20][21][22]. Paradoxically, in an attempt by the pathogen to subvert the host immune response, the proteins of these pro-inflammatory genes are degraded by Pg-derived gingipains [20,23,24]. Adhesion molecules localised at the cellular junctions of adjacent endothelial cells selectively regulate vascular permeability [25]. Dysregulated adhesion molecule expression may initiate vascular pathology due to abnormally elevated vascular permeability leading to oedema, chronic inflammatory and vascular damage [26]. We recently showed that Pg can cause vascular damage in an in vivo zebrafish larvae model of systemic infection in a gingipain-dependent manner [27], suggesting that these bacterial proteases may be important factors in the development of CVD. Here, we hypothesised that gingipain-dependent cleavage of endothelial cell junction adhesion molecules is important in mediating the vascular damage. By infecting transgenic zebrafish models that have fluorescent labelling of endothelial cell junctions with a wildtype and isogenic gingipain-null Pg mutant, we identify that gingipains dramatically alter adhesion molecule cell surface abundance leading to increased vascular permeability in vitro and in vivo. These data significantly enhance the growing evidence for the role of oral bacteria in CVD and indicate a potential mechanism by which oral bacteria can contribute to cardiovascular disease.

P. gingivalis invades human endothelial cells and localises to the perinuclear region through a gingipain-dependent mechanism
We initially confirmed Pg invasion of endothelial cells using an antibiotic protection assay. All wild-type laboratory strains examined (ATCC33277, W83, W50) invaded HDMEC at significantly greater levels than the clinical isolate (A245Br) (P ≤ 0.05, Fig. 1A). Wildtype W83 was not only capable of invading various types of endothelial cells including HDMEC (0.072% AE 0.032; Fig. 1A,B), HMEC-1 (0.48% AE 0.2; Fig. 1C) and HCAEC (0.21% AE 0.1; Fig. 1D), but was also shown to be significantly more invasive than the gingipain-deficient ΔK/R-ab mutant (P < 0.001) that displayed on average an 85% reduction in endothelial cell invasion for the three cell types tested ( Fig. 1B-D). Multichromatic confocal microscopy revealed that intracellular dwelling Pg were predominantly localised to the perinuclear regions, where they were colocalised within wheat germ agglutinin-positive membranes, suggesting residence within membranebound intracellular vesicles following internalisation, as previously shown in other model systems (Fig. 1E).

Endothelial cell permeability increases following Pg infection and is gingipain-dependent
We previously demonstrated that zebrafish systemically infected with Pg displayed significant oedema [27], suggesting that this bacterium may influence vascular permeability. To explore this possibility, we examined HDMEC monolayer permeability following Pg infection in vitro. Microscopically, Pg-infected HDMEC were found to lose cell-cell attachments compared with uninfected controls (Fig. 2A). Using a high molecular weight fluorescent dextran permeability assay, we observed a time-dependent increase in the levels of fluorescence passing through the endothelial monolayer and this was significantly increased in Pg-treated compared with untreated HDMEC at all time points examined for up to 5 h (Fig. 2B). Moreover, this increase in endothelium permeability was significantly reduced (P < 0.001) compared with untreated control levels or when Pg W83 was pre-incubated with KYT, a specific inhibitor of gingipain activity (Fig. 2 C) [28]. Similarly, endothelial permeability was also significantly reduced (P < 0.001) when monolayers were infected with the gingipain-deficient mutant ΔK/ R-ab compared with wild-type controls (Fig. 2D). Collectively, these data implicate gingipains as the primary permeabilising factor following Pg infection.
Pg infection decreases endothelial cell surface adhesion molecule abundance but not gene expression in vitro The endothelial cell surface adhesion proteins PECAM-1 (CD31) and VE-cadherin (CD144) are crucial for maintaining cell-cell junction integrity and preserving a restrictive endothelial permeability barrier [25,29]. Gingipains have been shown to mediate the proteolytic degradation of several human proteins [20,30]; therefore, we examined whether these proteases were responsible for cleavage of PECAM-1 and VE-cadherin. Flow cytometric analysis revealed that cell surface abundance of PECAM-1 on both HMEC-1 and HDMEC was almost completely abolished upon W83 infection compared with uninfected controls when analysed by both flow cytometry (P < 0.05; Fig. 3A,C) and immunofluorescence microscopy (Fig. 3E). In contrast, endothelial cells infected with the gingipain-null mutant, ΔK/R-ab, showed no loss of PECAM-1, displaying levels similar to uninfected controls (Fig. 3A,C). Cell surface abundance of VE-cadherin on both HMEC-1 and HDMEC was reduced by sixfold upon infection with W83 (P ≤ 0.05; Fig. 3B,D). In contrast, cells infected with ΔK/R-ab displayed VE-cadherin levels similar to those of uninfected controls (Fig. 3B,D). Loss of VE-cadherin was observed predominantly at the cell-cell junctions by confocal microscopy, sites where this protein Strain and gingipain-dependent invasion of Pg into human endothelial cells. HDMEC were infected with wild-type strain Pg ATCC33277, W83, W50 or A245Br (A) or wild-type Pg W83 and ΔK/R-ab (B) at a MOI 100 for 90 min. HMEC-1 (C) and HCAEC (D) were infected with wild-type Pg W83 or ΔK/R-ab mutant also at a MOI 100 for 90 min. Bacterial cell invasion was determined by antibiotic protection assay and expressed as a percentage of primary bacterial inoculum recovered. Graphs show means AE SD (n = at least 3 individual experiments with each individual experiment performed in triplicate technical repeats; shown as filled circles). Data were analysed using one-way ANOVA with Tukey's post hoc comparison test (Fig. 1A) or Student's t-test (Fig 1B-D), *P < 0.05, **P < 0.01, ***P < 0.001. Representative maximum intensity Z projection images of HMEC-1 with intracellular (green) and extracellular (cyan) Pg W83. Cell nucleus (blue) and plasma membrane (red). Composite image shows intracellular dwelling Pg as orange (green and red colocalisation, white arrow), and extracellular (cyan) bacteria bound to the cell surface (E). All images show x-axis, y-axis and z-axis planes, and scale bars in images are all 5 μm.  4A) and not due to differences in endothelial cell death upon infection as cell viability was similar to controls following infection with either W83 or ΔK/ R-ab ( Fig. 4B-E). In addition, reduced abundance of the endothelial cell surface adhesion molecule, E-selectin (CD62E), a protein that plays a role in leucocyte adhesion rather than vascular permeability, was also observed (Fig. 5). The levels of cell surface E-selectin were increased on HDMEC at 4 and 8 h following treatment with TNFα, whereas immortalised HMEC-1 cells did not display E-selectin at any time point examined irrespective of cytokine treatment (Fig. 5A-D). Pg was able to invade both HDMEC and HMEC-1 at similar levels in the absence or presence of pro-inflammatory stimulation (Fig. 5E,F). In 8-h TNFα-treated HDMEC, infection with Pg significantly (P < 0.001) reduced E-selectin cell surface abundance by fivefold compared with uninfected controls (Fig. 5G,H).
An alternative explanation for the loss of endothelial cell surface adhesion molecule expression is that the bacterial infection may reduce gene expression. To test for this possibility, we checked for differences in endothelial mRNA expression by qRT-PCR following incubation with wild-type W83 and mutant ΔK/R-ab Pg. Incubation with either of these strains did not significantly alter gene expression of VE-cadherin, PECAM-1 or other tested cytokines (CXCL8, CCL2, ICAM-1) in HMEC-1 and HDMEC cultures when compared to the treatment with the potent pro-inflammatory cytokine, TNFα (Fig. 6A,B).

P. gingivalis reduces endothelial cell surface adhesion molecule abundance in an in vivo zebrafish infection model
We next used our established Pg zebrafish infection model [27] to determine whether the observed changes in adhesion molecule cell surface abundance seen in Transgenic zebrafish with endothelial cell-specific PECAM-1-EGFP expression tg (fli1a:PECAM1-EGFP)sh524 were inoculated systemically with either red-labelled wild-type W83 or ΔK/Rab Pg. Reporter fluorescence was quantified where the endothelium and Pg colocalised in the intersegmental vessels and the caudal vein adjacent to the yolk sac, anatomical sites that are free from any natural pigmentation to avoid problems with background fluorescence and where individual vessels were easily distinguished. In PBS-injected zebrafish, PCAM-1 expression was observed on the endothelial cell surface of the intersegmental vessels as well as the caudal artery and vein (Fig. 7A,D). Both W83 and ΔK/R-ab gingipain-null bacteria could be visualised within the vessels at these sites (Fig. 7B,E and C,F), binding to and in some instances having traversed the vascular barrier into surrounding tissue (Fig. 7Gii). Importantly, in W83-infected zebrafish, colocalisation of Pg at the endothelial surface was associated with a marked reduction in endothelial PECAM-1 fluorescence at these sites ( Fig. 7Gi,Gii). Consistent with our in vitro experiments, we observed that loss of PECAM-1-EGFP fluorescence was gingipain-dependent since infection with the gingipain-null mutant, ΔK/R-ab, was not associated with loss of PECAM-1-

Pg infection increases vascular permeability in vivo
We next tested whether Pg infection could also lead to increased vascular permeability in an in vivo setting.
Here, we used transgenic zebrafish expressing a cyan fluorescent protein, mTurquiose2, in endothelial cells (tg(fli1a:mTQ2)sh321), to fluorescently label the vasculature and injected these with either fluorescein-labelled Pg or PBS as control followed by further injection with tetramethylrhodamine (a red fluorescent) dextran at 24 hpi. In PBS-injected larvae, the fluorescent dextran strictly localised to the lumen of the aortic arches, opercular artery, hypobranchial artery and other vessels in the vascular network, indicating that the high molecular weight dextran was not able to pass across the vascular barrier ( Fig. 9A,D). Following infection, Pg was observed both within blood vessels, their surrounding tissues and in particular around the oedematous heart (Fig. 9B,E). In these infected larvae, diffuse fluorescent dextran staining was observed throughout tissues, extending far beyond the confines of the aortic arches suggesting widespread dextran leakage from vessels ( Fig. 9C,F). Furthermore, dextran fluorescence was observed in the oedematous pericardiac region suggesting widespread vessel leakage in this area ( Fig. 9C,F). These data, along with our in vitro studies, provide compelling evidence that Pg is able to mediate vascular damage leading to loss of endothelial integrity and increased vascular permeability.

Discussion
Increasing evidence suggests that periodontal disease is associated with CVD and that this is influenced, in part, by bacteria that enter the circulation via diseased tissues in the oral cavity [1,6]. Further supporting evidence comes from in vitro studies showing that oral microbes, in particular Pg, can invade endothelial cells, but more compellingly from human studies where the DNA of several oral bacteria, including Pg, has been found in nondiseased vascular tissue from patients undergoing bypass surgery [16] as well as in atherosclerotic plaques [31][32][33]. However, the molecular mechanism driving this association is, as yet, unknown, with several divided theories proposed on how this relationship develops. Therefore, in this study we used both in vitro and in vivo approaches to decipher the mechanisms at play with particular emphasis on the role of gingipains, potent proteases known to play a key role in Pg-mediated disease processes [34].
Our initial in vitro studies using monolayers of endothelial cells from various vascular beds showed that Pg (strains ATCC33277, W83, W50, A245Br) were invasive but at low levels (~0.1%), which is in agreement with previous findings from other groups [18,19,35,36]. Collectively, these data show that endothelial cells are less susceptible to invasion than oral epithelial keratinocytes that can display from 2% to 10% cell invasion depending on the strain examined [37,38]. Moreover, the gingipain-null mutant ΔK/R-ab failed to invade endothelial cells, suggesting that bacterial gingipains are crucial for this process. Confocal microscopy analysis confirmed the presence of intracellular dwelling Pg enclosed within membrane-bound vesicles at perinuclear sites in endothelial cells, similar to that documented for gingival epithelial cells, most likely as a result of endocytosis [19,39,40]. Internalisation has several beneficial outcomes for the bacterium including protection from the host immune response, resistance to antibiotic therapy (many cannot pass the plasma membrane) and increased survival [39]. When infected with wild-type Pg, HDMEC within confluent monolayers became detached and the endothelium displayed significant loss of permeability in vitro that was prevented by inhibition of gingipain enzyme activity or upon infection with ΔK/R-ab, the gingipain-null strain. Similar observations were reported by Sheets et al. [24] who noted that bovine coronary artery endothelial cells and HDMEC exhibited loss of adhesion to tissue culture plastic upon culture with Pgderived culture extracts. These data clearly point to proteolytic degradation of cell surface molecules as the likely mechanism for loss of cell-cell contacts and adhesion. Indeed, our flow cytometric and confocal microscopy analysis revealed gingipain-dependent cleavage of PECAM-1, VE-cadherin and E-selectin (CD62E) on human endothelial cells when cultured as monolayers in vitro.
E-selectin is an adhesion molecule associated with tethering of leucocytes to the endothelium during inflammation. It was previously shown that Pg adhesion to human umbilical vein endothelial cells (HUVEC) that are derived from large blood vessels is mediated by E-selectin [24,41]. In contrast, our data indicate that Pg adhere and invade both E-selectin-deficient HMEC-1 and E-selectin-expressing HDMEC at similar levels, suggesting that E-selectin is not essential for invasion into microvascular endothelial cells. Gingipain-dependent degradation of other leucocyte adhesion molecules (ICAM-1/CD54, VACM-1/CD106) has also been observed [30], suggesting that Pg is able to modulate leucocyte recruitment at the endothelium surface. PECAM-1 and VE-cadherin act to form intercellular junctions that are crucial for maintaining a continuous endothelium and so loss of these cell-cell contacts will inevitably lead to loss of tissue integrity, increased permeability and endothelial dysfunction. In this study, loss of PECAM-1 and VE-cadherin cell surface abundance was identified in response to whole bacteria, whereas previous studies have used recombinant gingipains to show cleavage of endothelial cell surface proteins including N-cadherin and integrin β1 as well as VE-cadherin [24]. Gene expression analysis   [43,44], which may be due to bacterial strain differences or to use of cell lines or endothelial cells derived from large compared with microvascular vessels.
In vitro generated data using monolayer endothelium provide valuable information, but these experiments lack conditions such as flow, shear stress and presence of other cells that are important when examining systemic infection. We therefore set about confirming the in vitro data using our zebrafish Pg infection model [27]. Zebrafish have been extensively used for host-pathogen interactions as well as for cardiovascular studies [45] and have several advantages over murine models such as transparency and availability of fluorescently tagged proteins that allow for real-time analysis of cell-cell interactions. Moreover, the close homology between numerous zebrafish and human innate immune and cardiovascular-associated molecules means that this in vivo model system is also ideally placed to examine pathogen-mediated host responses that may impact on CVD risk. We previously showed that zebrafish present with oedema upon infection with wild-type but not the ΔK/R-ab gingipain-null mutant [27] and given our in vitro data on gingipain-dependent endothelial cell adhesion molecule degradation we hypothesised that gingipains would be responsible for cleavage of these molecules in vivo leading to vascular permeability. Using genetically engineered zebrafish whereby PECAM-1 or VE-cadherin was fluorescently labelled, we observed, for the first time, in vivo Pg-mediated loss of fluorescence upon infection with wild-type (W83) but not with the gingipain-null mutant strain, indicating gingipain-dependent degradation of these molecules. Loss of PECAM-1 and VE-cadherin on the endothelium was most evident when colocalised with W83 Pg and was apparent in several vascular regions of the zebrafish indicating that infection is widespread and that Pg adhesion and gingipain-mediated protein degradation are not confined to specific vascular beds. Even more important was that loss of adhesion molecule abundance was associated with increased vascular permeability as  [46]. Several in vivo studies using murine experimental models, in particular the hyperlipidaemic ApoE(null) mice, have shown that Pg infection directly influences atherosclerotic lesion formation or development [47,48] and several mechanisms have been suggested. However, visualising bacterial-host interactions at a cellular and even protein level are not possible in these models. To our knowledge, this is the first time that the effects of systemic Pg infection at the cellular level, in vivo, have been reported and our data clearly point to a role for Pg in mediating vascular damage. At present, it is unknown whether Pg acts as an initiating event, where it could be speculated that Pg-mediated endothelial cell damage leads to the exposure of the underlying vascular connective tissue leading to localised platelet activation and leucocyte recruitment. However, it could also be speculated that Pg acts to exacerbate the disease process by negatively modulating the immune response. Clearly, further studies are warranted to decipher the precise role of Pg in CVD.
As with most in vivo experimental infection models, animals (mice or zebrafish) are systemically inoculated with high numbers of bacteria, much more than would be observed in humans, and it could be argued that these studies are therefore not completely physiologically relevant. However, it should be noted that many cardiovascular diseases are long-term, chronic conditions where the disease manifests over long periods with repeat exposure to transient levels of one if not several aetiological agents. Replicating this experimentally in vivo is extremely challenging and would require repeat low-dose inoculation or a long time frame, giving raise to ethical concerns over animal distress as well as experimental cost. Current, in vivo infection strategies aim to provide key information that collectively points to significant risk factors within the full knowledge of their limitations. In this regard, zebrafish provide clear benefits as an in vivo model to examine systemic host-pathogen interactions, and our data significantly add to others in implicating Pg in vascular pathogenicity. There is now an appreciation that many bacterial-mediated diseases are polymicrobial in nature, with periodontitis being a case in point where it is considered that several organisms within the subgingival plaque drive pathogenesis rather than one organism in isolation. It is therefore likely that other oral bacteria contribute to systemic disease. Indeed, in the oral context, studies have shown significant impacts on  [49] and it will be important to determine whether other microbes affect disease progression. Nevertheless, this study provides crucial evidence for the role of Pg, and gingipains in particular, in affecting endothelial function, reaffirming the likely role or oral microbes in influencing systemic disease outcomes.

Endothelial cell invasion quantification and visualisation
Antibiotic protection endothelial cell invasion assays were performed as described in Naylor et al. [51], with modification. HCAEC, HDMEC or HMEC-1 cells were seeded at 2 × 10 5 cells per well in a 24-well plate and cultured to confluence. Cell monolayers were washed with PBS and blocked for 1 h with 2% bovine serum albumin (BSA; Sigma-Aldrich) in cell culture medium. Endothelial cells were infected with Pg strains (ATCC33277, W83, W50 or A245Br) at a multiplicity of infection (MOI) of 100 in cell culture medium and incubated for 90 min at 37°C, 5% CO 2 . Serial dilutions were carried out to determine viable counts through colony-forming units (CFUs). Nonadhered bacteria were removed with PBS and the cell-adherent bacteria killed by incubation with 200 μgÁmL −1 metronidazole (Sigma-Aldrich) for 1 h at 37°C, 5 % CO 2 . Cells were then washed with PBS, lysed with dH 2 O, scraped, serially diluted on FA plates and incubated anaerobically for 3 days. CFUs were enumerated to determine the total ing 10% v/v FCS was incubated with the cells for 45 min at 37°C, 5% CO 2 . After washing, nuclei were counterstained with 7.5 μgÁmL −1 Hoechst ® (excitation 350 nm, emission 461 nm) (Thermo Fisher Scientific) for 10 min, rinsed and the endothelial plasma membrane stained with 5 μgÁmL −1 wheat germ agglutinin (WGA) (excitation 555 nm, emission 565 nm) (Thermo Fisher Scientific) for 10 min at 37°C. Chamber slides were analysed using a Zeiss LSM 880 inverted Axio Imager and AiryScan confocal microscope. Images were processed using AiryScan ® processing (Zeiss, Oberkochen, Germany), and composite and orthogonal views were created using FIJI ® software (National Institute of Health, Bethesda, MD, USA).

Endothelial cell permeability
Endothelial monolayer permeability was measured using a Transwell assay system. Millicell ® cell culture inserts (Merck Millipore, Watford, UK) were fibronectin-coated (10 μgÁmL −1 ) prior to seeding with HDMEC that were cultured until confluent. Pg W83 were treated with KYT-1 or KYT-36 (2 μM; Peptides International, Louisville, KY, USA) to inhibit gingipain activity as previously described [28]. HDMEC were infected in the absence or presence of Pg W83 (with or without inhibitors) at a MOI of 1000 for 1.5 h at 37°C in nonsupplemented MV medium. Bacteria were removed, inserts transferred to a new 12-well plate with 500 μL supplemented MV medium and 450 μL of MV supplemented medium added to the apical compartment of the insert together with 70 kDa fluorescent-labelled dextran, a molecular weight that does not readily pass through a confluent endothelial monolayer (Thermo Fisher Scientific; final concentration of 65 μgÁmL −1 ). Dextran leakage through the cell monolayer from apical compartment of the insert to the bottom well was monitored for up to 5 h by measuring the fluorescence intensity (excitation 494 nm, emission 521 nm; Tecan Ltd, Männedorf, Switzerland) of 250 μL medium aspirated from the bottom well; 250 μL of fresh medium was added to the bottom well for further readings. Inserts without cells were used as control wells.
In vitro PECAM-1 and VE-cadherin cell adhesion protein expression In vitro cell adhesion protein expression was analysed using flow cytometry and immunofluorescence microscopy. For flow cytometric analysis, HMEC-1 and HDMEC were seeded at 4 × 10 5 cells per well, cultured until confluent, rinsed and infected with MOI 1000 W83 or ΔK/R-ab Pg in nonsupplemented medium for 1 h. Addition of medium only was used as a control. Cells were washed, removed using 0.02% EDTA (Sigma-Aldrich) for 20 min, centrifuged and resuspended in 100 μL FACS buffer (0.1 % BSA and 0.1 % sodium azide in PBS). For PECAM-1 (CD31) expression, 0.06 µg per test of PE-Cyanine7-conjugated anti-human PECAM-1 (Clone MW59; Thermo Fisher Scientific) or its matching conjugated IgG control (IgG2254 Isotype Control; Thermo Fisher Scientific) was added and cells incubated for 45 min on ice. For VE-cadherin (CD144), 0.25 μg per reaction anti-human VE-cadherin antibody (Clone 123413, R&D Systems, Minneapolis, MN, USA) or IgG isotype control (Clone 11711, R&D Systems) was incubated for 45 min on ice, supernatant was removed and Alexa Fluor ® 488 goat anti-mouse antiserum (Abcam, Cambridge, UK) diluted at 1 : 100 in FACS buffer was added for 30 min on ice. Cells were washed and resuspended in 300 μL FACS buffer, and single-colour sample analysis was carried out using a LSR II flow cytometer (BD Biosciences, San Jose, CA, USA) for PE-CAM1 and FACSCalibur flow cytometer (BD Biosciences) for VE-cadherin analysis. The cell viability dye, TO-PRO-3 ® (1 mgÁmL −1 , Thermo Fisher Scientific), was added to each sample before analysis. Cell populations were gated using forward scatter (FSC) and side scatter (SSC) voltages and TO-PRO-3-positive cells excluded from the analysed population. Five thousand viable cells were collected for HDMEC analysis, while 10 000 live cells were collected for HMEC-1 analysis. Threshold for positively fluorescent cells was set using isotype-matched controls, and data were analysed using FlowJo software (BD Biosciences). The normalised median fluorescence index (nMFI) was calculated by dividing the median fluorescence intensity of the positive samples by that of the IgG controls.
For immunofluorescence microscopy, fibronectin-coated (10 μgÁmL −1 ) glass coverslips were seeded with 5 × 10 4 endothelial cells. Pg was stained with 5 μM Red CMTPX CellTracker ® (Thermo Fisher Scientific) and diluted to an MOI 1000 in nonsupplemented cell culture medium to infect primary and immortalised endothelial cells. Cells were washed with PBS, fixed in 3.7% formalin for 10 min at room temperature, washed again and blocked for 1 h

In vitro gene expression analysis
Quantitative reverse transcription polymerase chain reaction (qRT-PCR) was used to analyse mRNA expression of endothelial immune molecules. HMEC-1 and HDMEC were seeded at 4 × 10 5 cells per well and cultured until confluent. Cells were either infected with P. gingivalis W83 or ΔK/R-ab mutant at a MOI 100 for 4 h, and unstimulated or TNFα-stimulated (25 ngÁmL −1 ) cells were used as controls. Total RNA extraction was carried out using ISO-LATE II RNA Mini Kit (Bioline, Meridian Bioscience Inc., Cincinnati, OH, USA) according to the manufacturer's instructions and RNA measured by NanoDrop (Thermo Fisher Scientific). High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific) was used to produce cDNA and qPCR performed using Qiagen Rotor-Gene (Qiagen, Hilden, Germany). Reactions consisted of 5 μL of TaqMan ® qPCR BioProbe Blue Master Mix (PCR Biosystems, London, UK), 3.5 μL RNase-free water, 0.5 μL human β-2-microglobulin (B2M) as an endogenous reference control (VIC/MGB Probe, Thermo Fisher Scientific), 0.5 μL target primer and 0.5 μL cDNA. Target primers used were PECAM-1 (Hs01065279_m1), VE-cadherin (Hs00901465_m1), CXCL8 (Hs00174103_m1), CCL2 (Hs00234140_m1) and ICAM-1 (Hs0014932_m1) (Thermo Fisher Scientific). All samples were run in triplicate; control samples excluded cDNA. The threshold cycle (Ct) for each test gene was normalised against their respective reference controls. Fold change in expression relative to unstimulated cells was calculated with ΔC t values of the sample and reference gene using the formula 2 ÀΔΔCt .
In vivo PECAM-1 and VE-cadherin cell adhesion protein expression Tg(fli1a:PECAM1-EGFP)sh524 and Tg(cdh5ubs8-/-;cd-h5TS) transgenic zebrafish were outcrossed with nacre zebrafish. Embryos were dechorionated and green fluorescent protein (GFP)-positive embryos selected at 1 dpf using a fluorescent dissecting microscope (Zeiss Axio Zoom.V16 fitted with an HXP 200C Illuminator and a Zeiss Axiocam 503 monocamera). Zebrafish embryos at 30 hpf were anaesthetised using tricaine (0.02% (w/v) 3-amino benzoic acid ester tricaine/MS-322 (Sigma-Aldrich), positioned in a solution of 3% (w/v) methylcellulose (Sigma-Aldrich) in E3 medium and 2 nL (5 × 10 4 CFU) of 5 μM Red CMPTX (CellTracker TM , Thermo Fisher Scientific) stained W83 or ΔK/R-ab Pg were injected systemically using a microcapillary needle (Scientific Laboratory Supplies) via direct microinjection into the Duct of Cuvier (the common cardinal vein); PBS was used as control. Live imaging of anaesthetised fish (5 zebrafish embryos per group) was carried out 24 h postinfection (hpi) using a spinning disc confocal microscope (PerkinElmer UltraVIEW VoX, PerkinElmer Inc., Waltham, MA, USA) running on an inverted Olympus IX81 motorised microscope. 488-nm and 561-nm lasers were used to visualise fluorescent PECAM-1 or CDH5/VEcadherin (green) and Pg bacteria (red) in zebrafish embryos. Micrographs were captured in two areas of the tail, adjacent to the cloaca (closer to the yolk) and 4 vessels further away from the cloaca (closer to the tip of the tail) using Volocity ® software (Quorum Technologies Inc., Puslinch, Ontario, Canada). Quantification of relative GFP fluorescence intensity (pixel value) was performed by batch processing of maximum intensity z projection of micrographs of whole images in the yolk and tail region using an adjusted script on FIJI ® Macros plug-in.

Statistical analysis
All data presented are from at least 3 independent experiments (unless specified). Results are expressed as the mean AE standard deviation (SD) except for flow cytometry data where the nMFI (normalised median fluorescence index) was used. Normality was determined using Kolmogorov-Smirnov analysis. The differences between two groups were assessed using either Student's t-test or Mann-Whitney U-test, while the differences between group data were assessed using one-way ANOVA followed by either Tukey's or Dunn's post hoc multiple comparison test depending on the nature of the data (parametric or nonparametric). All tests were carried out using GraphPad Prism v8.2 (GraphPad, San Diego, CA, USA), and statistical significance was assumed at P < 0.05.