Annexin A2 depletion exacerbates the intracerebral microhemorrhage induced by acute rickettsia and Ebola virus infections

Intracerebral microhemorrhages (CMHs) are small foci of hemorrhages in the cerebrum. Acute infections induced by some intracellular pathogens, including rickettsia, can result in CMHs. Annexin a2 (ANXA2) has been documented to play a functional role during intracellular bacterial adhesion. Here we report that ANXA2-knockout (KO) mice are more susceptible to CMHs in response to rickettsia and Ebola virus infections, suggesting an essential role of ANXA2 in protecting vascular integrity during these intracellular pathogen infections. Proteomic analysis via mass spectrometry of whole brain lysates and brain-derived endosomes from ANXA2-KO and wild-type (WT) mice post-infection with R. australis revealed that a variety of significant proteins were differentially expressed, and the follow-up function enrichment analysis had identified several relevant cell-cell junction functions. Immunohistology study confirmed that both infected WT and infected ANXA2-KO mice were subjected to adherens junctional protein (VE-cadherin) damages. However, key blood-brain barrier (BBB) components, tight junctional proteins ZO-1 and occludin, were disorganized in the brains from R. australis-infected ANXA2-KO mice, but not those of infected WT mice. Similar ANXA2-KO dependent CMHs and fragments of ZO-1 and occludin were also observed in Ebola virus-infected ANXA2-KO mice, but not found in infected WT mice. Overall, our study revealed a novel role of ANXA2 in the formation of CMHs during R. australis and Ebola virus infections; and the underlying mechanism is relevant to the role of ANXA2-regulated tight junctions and its role in stabilizing the BBB in these deadly infections.

Rickettsioses represent devastating human infections [1][2][3][4][5][6][7]. These arthropod-borne diseases are caused by obligatory intracellular bacteria of the genus Rickettsia (R.). A vaccine is not available for rickettsioses. Disseminated EC infection and EC barrier dysfunction are the central pathophysiologic features of human lethal spotted fever group rickettsial (SFGR) infections [8]. Typically, SFGR infection is controlled by appropriate broad-spectrum antibiotic therapy if diagnosed early. Nevertheless, SFGR infections present with nonspecific signs and symptoms rendering early clinical diagnosis difficult [7,9]. Untreated or misdiagnosed SFGR infections are frequently associated with severe morbidity and mortality [1,8,[10][11][12]. A fatality rate as high as 32% has been reported in hospitalized patients with Mediterranean spotted fever [12]. It has been forecasted that increased ambient temperatures under conditions of global climate change is a driver in rickettsial epidemiology, leading to more widespread distribution of rickettsioses [4]. Comprehensive understanding of rickettsial pathogenesis is urgently needed.
CMH can be acutely caused by pathogen-associated inflammation (AICMHs) [60,66]. One of the underlying pathology of AICMHs is the acute dysfunction of the blood-brain barrier (BBB) [67] which is the interface between circulating blood and the central nervous system (CNS) and is composed of brain microvascular endothelial cells (BMECs), pericytes, and astrocytes [68,69]. BBB properties are primarily determined by junctional complexes between the BMECs, i.e. adherens junctions (AJs) and tight junctions (TJs) [70][71][72]. Although ECs are susceptible to rickettsial and Ebola virus infections, the mechanisms of how the stability of the BBB is maintained during these infections is far from understood.
Annexin A2 (ANXA2) is a member of the large annexin family of Ca 2+ -regulated and phospholipid-binding proteins, which associates with cell membrane dynamics, cell-cell interactions, and cell adhesion [73][74][75][76][77][78][79]. ANXA2 can be monomeric, found mainly in the cytosol, or forming heterotetramer complex with S100A10 [75,80]. S100A10 is a unique member of S100 protein family that has been known to bind to ANXA2. The interaction between ANXA2 and S100A10 yields a heterotetramer complex ANXA2-S100A10, enabling ANXA2 to translocate across the EC membrane and perform a variety of functions, facilitating plasmin-based fibrinolytic activities on vascular luminal surfaces [74,75,81]. Recently, we identified host ANXA2 as a novel receptor for SFGR and staphylococcus aureus adhesions to ECs [82]. However, there was no difference in rickettsial adhesion to or invasion into white blood cells between the wild-type (WT) and ANXA2-knockout (KO) mice.
Here we report an observation that focal CMHs lesions exist in the cerebra of R. australis infected AXNA2-KO mice but not R. australis infected WT mice. We hypothesize cell-cell junction in the BBB is destabilized in AXNA2-KO mice rendering them susceptible to AICMH. In order to comprehensively investigate this possibility, we performed a proteomic analysis using the whole brain lysate and brain-derived isolated endosomes. We identified a variety of differentially expressed (DE) proteins that were relevant to vascular integrity. Functional group annotation and network analysis based on the identified DE proteins revealed a variety of protein functional group changes, such as cell-cell junction, stress fiber, MHC II protein complex binding, and stress response. These identified functional groups support that a structural impairment of the BBB might be involved. Consistently, immunofluorescence (IF) of brain tissue of R. australis infected mice revealed dramatic disruption and disorganization of TJ proteins ZO-1 and occludin in ANXA2-KO mice, but not WT mice. We then chose to investigate whether a similar ANXA2 dependent disorganization of TJs is present in the context of Ebola virus, another pathogen known to attack ECs and present with severe vascular leakage. Interestingly, ANXA2-KO mice challenged by Ebola virus also exhibited CMHs and aberrant TJs whereas WT mice showed no signs of bleeding into CNS, indicating this pathology is not specific to rickettsia infection. Collectively, these data suggest that ANXA2 is required for the integrity of TJs in response to acute rickettsia and Ebola infections.

Biologic containment
Infectious material and animals were handled in maximum-containment biosafety level 3 (for R. australis) and 4 (for Ebola virus) facilities at the Galveston Nationa Laboratory (GNL), University Texas Medical Branch at Galveston.

Ethics statement
All animal protocols were approved by the Institutional Animal Care and Use Committee of the University of Texas Medical Branch (protocol # 1702018 and protocol # 9505045G). The animal studies were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, USA. SFGR mouse infection model [83,84] ANXA2-KO on C57BL/6 background and C57BL/6 mice were used in this study. Animals (15 WT and 14 ANXA2-KO mice) were inoculated with 2LD50 dose (2 × 10 6 pfu per mouse) of R. australis via tail vein injection and observed daily. All procedures followed the approved IACUC protocol. Signs of ruffled fur, hunched posture, labored breathing and closed eyelids were identified as lethal illness (41, 42). The animals were observed for 10 days when most of the animals were all in lethal illness state. For time-dependent pathological study, mice were inoculated with 2 LD50 dose of R. australis and euthanized at day 2, 4, 5 post-infection (n = 5 for each time point). For mass spectrometry experiments, animals (WT or ANXA2-KO) were euthanized at 5 days p.i. and the brain samples were collected and digested into protein lysate for downstream analysis. The time point was selected for LC/MS because this was when mice started to showing up lethal illness.

Ebola virus mouse infection model[85]
To observe the effect of mice with ANXA2-KO compared to WT on Ebola hemorrhagic disease, C57BL6 6-12 week wild type (n = 5) and ANXA2-KO mice were inoculated intraperitoneally with 50 plaque-forming unit of mouse-adapted strain of Ebola Zaire (Mayinga) virus (provided by Thomas Ksiazek) in 200 uL of PBS. All procedures followed approved IACUC protocol. Mice were monitored multiple times daily for signs of illness and mortality p.i. Daily observations included evaluation of mice for clinical symptoms such as reduced grooming, ruffled fur, hunched posture, subdued response to stimulation, nasal discharge, and bleeding. Tissues and carcasses were collected for downstream assays.

Sample collection
Blood samples were collected via orbital sinus and serum was obtained after centrifuge and discarding the cellular content of the blood. Complete necropsies were performed on all mice to obtain the organs. For immunohistochemistry or H&E staining, organs were fixed in a 4% (vol/vol) formaldehyde. For mass spectrometry, brain tissues were lysed in protein lysing buffer with proteinase inhibitor and phosphatase inhibitor. Endosome isolation was perform using Minute TM endosome and cell fraction kit.

IF and H&E staining
The fixed samples were subjected to H&E staining or immunofluorescence (IF) with an antibody against the protein of interest. Mouse tissues were collected and fixed in 10% formaldehyde solution for 72 hrs. Fixed tissues were washed 4-5 times with PBS. For antigen retrieval for IF studies, tissue sections were incubated in pH 6.0 citrate buffer and heated in a steamer (Black and Decker, New Britain, CT) for 10 minutes. After washing in PBS, samples were incubated with proteinase K for 5 minutes at RT. Tissue sections were first permeabilised by incubating with Triton X-100 (0.1% v/v in PBS) for 10 min. When using mouse monoclonal IgG against VE-cadherin (1/500) (Meridian Life Science, Saco, ME), tissues were blocked with unconjugated AffiniPure Fab fragment goat anti-mouse IgG (H+L) at 20 μg/ml(115-007-003, Jackson ImmunoResearch, PA) for 1 hour at room temperature. Tissue sections were incubated with rabbit antibodies against ZO-1 (1/500) (Thermo Fisher Scientific, Rockford, IL), occludin (1/500) (Thermo Fisher Scientific), SFG rickettsiae (1/2000) (provided by Dr. David Walker), or Ebola virus (1/500) (provided by Dr. Thomas Ksiazek) overnight at 4˚C, followed by secondary antibody AlexaFluor 594 goat anti-mouse or AlexaFluor 488 or 594 goat antirabbit antibodies (1/1000) (Thermo Fisher Scientific). All tissue sections were followed by three final washes with PBS. Before mounting to the cover slide, the tissue sections were stained with DAPI. Normal rabbit and mouse IgGs were used as negative reagent controls (S1C Fig). Fluorescent images were taken with an Olympus BX51 microscope and analyzed using Olympus CellSens Standard software.

DNA extraction and RT-qPCR
To quantify the rickettsia loading in the brain tissue, the DNeasy tissue kit (Qiagen, CA, 69506) was used to quantify DNA rickettsial DNA. Briefly, the brain samples were minced into pieces, and subjected to lysis buffer and proteinase k digestion for 10 minutes in 56 degrees Celsius; then DNA was precipitated in ethanol and purified using washing buffer. The purified DNA samples were stored in storage buffer in -20 C o . PCR was performed using the protocol as previously described [86]. Rickettsia-specific citrate synthase (CS) gene (gltA) was used as the target for rickettsia detection (gltA forward: GAGAGAAAATTATATCCAAA TGTTGAT; gltA reverse, AGGGTCTTCGTGCATTTCTT) [86].

ELISA
Plasma samples collected from ANXA2-KO and WT mice were used for ELISA to detect TNFα (Mouse TNFα Qantikine ELISA kit, MTA00B, R&D Systems; Minneapolis, MN) and IFNγ (Mouse IFNγ Qantikine ELISA, MIF00, R&D Systems). Standard curves were performed using the standard proteins according to the protocol provided in the ELISA kits. The ELISA plates were detected at 450nm.

LC-MS/MS
Proteins were acetone-precipitated and cleaned with 1 ml of ice-cold wash solution (tri-nbutyl phosphate/acetone/methanol (1:12:1 by volume) for 90 minutes and then centrifuged at 2800g for 15 minutes at 4˚C. The supernatant was removed and 1 ml ice-cold tri-n-butyl phosphate was added and incubated at 4˚C for 15 minutes and then centrifuged at 2800g for 15 minutes at 4˚C. The supernatant was discarded and 1 ml ice-cold acetone was added and incubated at 4˚C for 15 minutes and then centrifuged at 2800g for 15 minutes at 4˚C. The supernatant was discarded and 1 ml ice-cold methanol was added and incubated at 4˚C for 15 minutes and then centrifuged at 2800g for 15 minutes at 4˚C [87,88]. 50ug of protein was solubilized with 5% SDS, 50 mM TEAB, pH 7.55, in the final volume of 25 uL. The sample was then centrifuged at 17,000g for 10 minutes for debris removal. Proteins were reduced by making the solution 20mM tris(2-carboxyethyl)phosphine TCEP (ThermoFisher, #77720) and incubated at 65 � C for 30 minutes. The sample was cooled to room temperature. After 1 uL of 0.5 M iodoacetamide acid was added, the sample was allowed to react for 20 minutes in the dark. 2.75 ul of 12% phosphoric acid was added to the protein solution. 165uL of binding buffer (90% Methanol, 100mM TEAB final; pH 7.1) was then added to the solution. The resulting solution was loaded onto a S-Trap spin column (protifi.com) and passed through the column by a benchtop centrifuge (30-second spin at 4,000g). The spin column was washed with 400uL of binding buffer and centrifuged. The wash was repeated two more times. Trypsin was added to the protein mixture in a ratio of 1:25 in 50mM TEAB, pH = 8, and incubated at 37 � C for 4 hours.
Peptides were eluted with 80 uL of 50 mM TEAB, followed by 80 uL of 0.2% formic acid, and finally 80 uL of 50% acetonitrile, 0.2% formic acid. The combined peptide solution was then dried in a speed vac and resuspended in 2% acetonitrile, 0.1% formic acid, 97.9% water and placed in an autosampler vial [88].

NanoLC MS/MS analysis
Instrument performance was verified by analyzing a standard peptide mix and a complex protein digest (HeLa) before the sample set was run between each experimental block and at the end of the experiment. The HeLa data files were analyzed in order to confirm that instrument performance remained consistent throughout the experiment. Peptide mixtures from digested brain tissue were analyzed by nanoflow liquid chromatography-tandem mass spectrometry (nanoLC-MS/MS) using a nano-LC chromatography system (UltiMate 3000 RSLCnano, Dionex), coupled on-line to a Thermo Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific, San Jose, CA) through a nanospray ion source (Thermo Scientific). A trap and elute method was used. The trap column is a C18 PepMap100 (300um X 5mm, 5um particle size) from ThermoScientific. The analytical columns is an Acclaim PepMap 100 (75um X 25 cm) from (Thermo Scientific). After equilibrating the column in 98% solvent A (0.1% formic acid in water) and 2% solvent B (0.1% formic acid in acetonitrile (ACN)), the samples (1 μL in solvent A) were injected onto the trap column and subsequently eluted (400 nL/min) by gradient elution onto the C18 column as follows: isocratic at 2% B, 0-5 minutes; 2% to 32% B, 5-100 minutes; 32% to 50% B, 100-108 minutes; 50% to 90% B, 108-109 minutes; isocratic at 90% B, 109-114 minutes; 90% to 2%, 114-115 minutes; and isocratic at 2% B, till 130 minutes.
All LC-MS/MS data were acquired using XCalibur, version 2.1.0 (Thermo Fisher Scientific) in positive ion mode using a top speed data-dependent acquisition (DDA) method with a 3 sec cycle time. The survey scans (m/z 350-1500) were acquired in the Orbitrap at 120,000 resolution (at m/z = 400) in profile mode, with a maximum injection time of 50 msec and an AGC target of 400,000 ions. The S-lens RF level was set to 60. Isolation was performed in the quadrupole with a 1.6 Da isolation window, and CID MS/MS acquisition was performed in profile mode using rapid scan rate with detection in the orbitrap (res: 35,000), with the following settings: parent threshold = 5,000; collision energy = 35%; maximum injection time 100 msec; AGC target 500,000 ions. Monoisotopic precursor selection (MIPS) and charge state filtering were on, with charge states 2-6 included. Dynamic exclusion was used to remove selected precursor ions, with a +/-10 ppm mass tolerance, for 60 sec after acquisition of one MS/MS spectrum.
Database Searching. Tandem mass spectra were extracted and charge state deconvoluted by Proteome Discoverer (Thermo Fisher, version 1.4.1.14). Deisotoping was not performed. All MS/MS spectra were searched using Sequest. Searches were performed with a parent ion tolerance of 5 ppm and a fragment ion tolerance of 0.60 Da. Trypsin was specified as the enzyme, allowing for two missed cleavages. Fixed modification of carbamidomethyl (C) and variable modifications of oxidation (M) and deamidation of asparagine and glutamine, were specified in Sequest. Scaffold (version Scaffold_4.8.7, Proteome Software Inc., Portland, OR) was used to validate MS/MS-based peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability. Peptide Probabilities from X! Tandem and Sequest were assigned by the Scaffold Local FDR algorithm. Peptide Probabilities were assigned by the Peptide Prophet algorithm [89] with Scaffold delta-mass correction. Protein identifications were accepted if they could be established at greater than 95.0% probability and contained at least 2 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm [89]. Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony.

Functional enrichment analysis
Raw mass spectrometry data was filtered to rule out low abundance protein. Specifically, proteins that are included in the analysis must have at least 4 total spectra count. A list of differentially expressed proteins was obtained by comparing the -spectral counts between ANXA2-KO and WT groups and calculating the fold change. Proteins with a fold change greater than or equal to 2 were considered upregulated; protein with fold change less than or equal to -0.6 were considered down-regulated. Selected DE proteins are listed in Tables 1-4. Next, the DE proteins were further annotated using The Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.8, which is also capable of doing functional enrichment analysis. Functional enrichment analysis helps us better correlate the molecular patterns with the gross pathology we have observed in the animal model. GO term and KEGG pathways were used to categorize the DE proteins (Fig 1). The Biological Networks Gene Ontology tool (BiNGO) [90] was used to perform the network analysis, which provides a direct structural visualization of the functional enriched groups. Protein-protein interaction (PPI) network was performed using Cytoscape GeneMania [91].

Absence of ANXA2 is associated with the incidence of CMH in R. australis infection
First, we designed a survival study to investigate the potential role of ANXA2 in lethel R. australis infection. WT (n = 14) and ANXA2-KO (n = 15) mice were inoculated with an ordinarily lethal dose of R. australis (2 x 10 6 ) via tail vein injection (i.v.) [82,84] and observed up to 10 days post-infection (p.i.). Accumulative survival data were obtained from three independent experiments (S1A Fig) and subjected to Kaplan-Meier (K-M) analysis. We found no difference in survival between WT (21.43%) and ANXA2-KO (13.33%) groups. However, the gross pathology of the brain surface (S1B Fig) observed apparent different color cerebral areas between infected WT mice and infected ANXA2-KO. To examine underlying associated pathology, brain tissue sections were subjected to histological examination with hematoxylin and eosin (H&E) staining (Fig 1), which revealed striking focal CMHs, in the cerebra of all lethally infected ANXA2-KO mice (Fig 1G-1L), but not in the two surviving infected ANXA2-KO mice. Conversely, such CMHs were absent from both surviving and lethally-infected WT mice (Fig 1D-1F).

Inactivation of ANXA2 does not affect the proliferation of R. australis or serum levels of IFNγ and TNFα in mice
During the survival study, we observed mice becoming moribound at day 6 p.i. Therefore, we designed time-dependent pathological studies at days 2, 4, and 5 p.i. Mice were inoculated with an ordinarily lethal dose of R. australis (2 x 10 6 ). At 5 days p.i. (5 mice per group), both ANXA2-KO and WT mice were euthanized as designed. H&E examination revealed extensive focal hemorrhagic lesions, at levels of arteriole, capillary, and venule, in the cerebra of all infected ANXA2-KO mice on day 5 p.i., but not WT mice (Fig 2A-2D).
To examine whether ANXA2 plays a role in serum levels of IFNγ and TNFα in lethal R. australis infection, serum from mice at days 2, 4, or 5 p.i. were collected and proceeded to analyze the concentration of IFNγ and TNFα. However, no difference was found comparing WT and ANXA2-KO mice (Fig 2E and 2F). Meanwhile, on day 5 post-infection, the real-time qPCR analysis revealed no difference in bacterial loads in brain between WT (n = 4) and AXNA2-KO mice (n = 4) (Fig 2G). Immunofluorescent staining (IF) of rickettsia in the liver, brain and lung did not show any difference between WT and ANXA2-KO mice on day 5 p.i. (Fig 3). These data suggest ANXA2 depletion does not affect the overall proliferation of R. australis or serum levels of IFNγ and TNFα in the mice.

Proteomic analysis
In order to decipher the protein profile related to the CMHs observed in the ANXA2-KO mice post rickettsia infection, we performed a proteomic analysis of the whole brain protein lysate and isolated endosome. Isolated endosome protein pattern is important due to the involvement of ANXA2 in endocytosis and turnover of surface proteins and nucleotide [75]. Whole-brain lysate. We compared the protein profiles of brain lysates from ANXA2-KO and WT mice challenged with R.australis on day 5 p.i., using LC/MS analysis, which identified one hundred thirty DE, with 93 upregulated and 37 downregulated (ANXA2-KO versus WT).

PLOS NEGLECTED TROPICAL DISEASES
The top upregulated and downregulated proteins are listed in Tables 1 & 2. It is noteworthy that the hemoglobin levels were higher (4-fold) in the infected ANXA2-KO mice compared to WT mice, which may account for the different gross pathology findings between the two groups.Functional enrichment analysis and visualization (Fig 4 & S1 Table) revealed statistically significant functional groups that are potentially associated with CMH in ANXA2-KO mice. The noteworthy upregulated proteins that are associated with the cell junction structure integrity include heat shock protein 90 α (HSP90α), heat shock protein 90 beta, enolase 1, clathrin (heavy polypeptide); important downregulated proteins include myosin (heavy polypeptide 10). Interestingly, TJ protein ZO-1 from the brain lysate is not differentially expressed in WT versus ANXA2 KO mice as measured by LC/MS. Functional enrichment analysis was performed based on Gene Ontology (GO) Term and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. Noteworthy GO Term include cell-cell adherens junction, stress fiber, actin cytoskeleton, MHC class II protein complex binding, vesicle-mediated transport, and extracellular vesicle. In KEGG pathway, we identified a variety of signaling pathways, in which PI3K-Akt signaling pathway, cAMP signaling pathway, and protein digestion and absorption pathway may be relevant to the pathology of CMHs.  Due to the nature of endosomes, the proteins enriched in the endosomes are either subjected to degradation or recycled back to the plasma membrane [92]. Increased accumulation of proteins in the endosomal compartment suggests increased protein turnover rate and degradation; whereas reduction of protein enrichment in the endosomes might result in protein accumulation in another compartment such as cytosol or plasma membrane [92].
ANXA2 is known to be associated with the endosomal membrane. Endocytosis of several targets depends on the presence of tyrosine phosphorylation of ANXA2 [93][94][95][96][97][98]. The absence of ANXA2 may lead to a differential endosomal protein profile, which may shed light on the underlying mechanisms associated with the increased susceptibility of ANXA2-KO mice to CMH upon rickettsia infection.
Comparing the endosomes isolated from the mouse brains of WT mice and ANXA2-KO 5 days after rickettsial infection, we have identified 47 DE proteins, with 18 upregulated proteins, and 29 downregulated proteins in the infected ANXA2-KO mice. Noteworthy upregulated DE proteins include talin1, alanyl-tRNA synthetase, glutathione peroxidase 1, and calcium/calmodulin-dependent protein kinase IV. Important downregulated proteins are cofilin 2, heat shock protein 8, HSP90α, thimet oligopeptidase and seryl-aminoacyl-tRNA synthetase. Top proteins are listed in Tables 3 & 4. Functional enrichment analysis was performed based on these 47 DE proteins, generating a list of functional enriched clusters (Fig 5 & S1 Table). Noteworthy Go Term functional groups include cell-cell AJ, metalloproteinase, and MHC II protein complex binding. Significantly altered signaling pathways in KEGG are pyruvate metabolism, biosynthesis of antibiotics, metabolic pathways, carbon metabolism, valine, leucine and isoleucine degradation, synthesis and degradation of ketone bodies. Whole brain lysate-and endosomal DE protein-derived functional enrichment (GO CC) network (Fig 6) analysis identifies the nodes associated with cell junction; and the interactome of the DE proteins under cell junction category is revealed in Fig 7A & 7B.

Endothelial TJ proteins disorganized in brains of ANXA2-KO mice after lethal SFGR infections
The overall functional enrichment analysis supports our hypothesis that deletion of ANXA2 affects the cell-cell junction structure during the rickettsial infection. However, LC/MS showed that the expression level of TJ protein ZO-1 in the brain was similar between infected ANXA2-KO and infected WT mice. We postulated that strutures of TJ might be disorganized instead of downregulation of the expression levels.
Occludin is an important transmembrane protein integral to TJs, which has been previously shown to directly interact with ZO-1 [99]. To investigate the structure of TJs, we applied IF assay to the brain tissue using antibody against ZO-1 and occludin. Remarkably, in all five ANXA2-KO mice, dramatic disruption and disorganization of ZO-1 and occludin were detected after D5 p.i. (Fig 8A). Interestingly, occludin disorganization was also observed in the livers of these ANXA2-KO mice on day 5 p.i. (Fig 8B). AJ protein VE-cadherins were also investigated. Animals in both WT and ANXA2-KO groups exhibited similar disintegrated structures of VE-cadherin in response to R. australis infection on day 5 p.i. (Fig 8C).

CMHs and disorganized EC TJs in ANXA2-KO mice subjected to Ebola virus infection
We then chose to investigate whether a similar ANXA2-dependent disorganization of TJs is also present in Ebola infection, given that both Ebola virus and rickettsiae are known to attack endothelial cells and disease progression is marked by severe vascular leakage [15,[26][27][28][29]100]. We first investigated whether the ANXA2 deletion-dependent CMHs could be observedin the context of Ebola virus infection via retrospective histology examination of brain tissue samples from WT (n = 5) or ANXA2-KO (n = 5) mice challenged by a mouse-adapted strain of Ebola Zaire virus (50 plaque-forming unit by the intraperitoneal route [85]). In this experiment, a 20% cumulative survival was observed in ANXA2-KO mice 12 days p.i., compared with an 80% cumulative survival in the group of WT mice at the same time (p = 0.08, Log-rank test) (S2 Fig). IF assay of Ebola virus in the liver, brain and lung tissues showed no difference between WT and ANXA2-KO mice on day 7-12 p.i. (Fig 9A). The sample size was not Visualization of the GO Term CC CC for whole brain lysate (A) and isolated endosome from the brain (B). Cytoscape BiNGO was used to generate the networks. Each node stands for a GO Term CC category, edges are present to connect functional related nodes. CMH-relevant functional groups (i.e., stress fiber and adherens junction) are enlarged and highlighted from the networks. The color bar indicates the range of p value corresponding to node (ANXA2 KO vs. WT in R. australis infection). https://doi.org/10.1371/journal.pntd.0007960.g006

PLOS NEGLECTED TROPICAL DISEASES
robust enough to allow a conclusion regarding survival between groups, as this was not the goal of the study. However, future studies comparing the survuival between WT and ANXA2-KO mice infected with Ebola virus using an increased number of mice in each group are planned.
Similarly, CMHs were detected using H&E staining in all ANXA2-KO mice (Fig 9C-9E), but not in WT mice (Fig 9B) during 7-12 days p.i. Disorganized TJ protein ZO-1 and occludin were also present in brain tissue from ANXA2-KO, but not WT mice (Fig 9F) infected with same dose of Ebola virus intraperitoneally. Furthermore, dramatic disruption and disorganization of occludin were also observed in the livers of ANXA2-KO mice post Ebola virus infection (Fig 9G).
Collectively, these data suggest disorganization ofthe TJ structures as an underlying mechanism of the CMHs observed in ANXA2-KO mice post-R. australis or Ebola virus infections.

Discussion
In this study, we report a novel finding that ANXA2-KO mice are more susceptible to rickettsia-or Ebola virus-induced CMHs, underlying mechanisms relevant to endothelial TJs-based BBB dysfunction. Furthermore, we performed a proteomic analysis on the whole brain lysate and isolated endosomes using LC/MS. The obtained list of differentially expressed proteins was subjected to function enrichment analysis, combining GO term and KEGG pathway.

Whole-brain lysate proteomic functional group
Cell-cell junction. The impermeability properties of BBB are primarily determined by BMEC AJs and TJs, which offer contracting forces to bridge the adjacent ECs via cadherins and nectins [101]. AJs directly and indirectly via via actin-associated proteins interact with TJs, together they grant the essential stability of the BBB. [99,101]. AXNA2 affects the stability of AJs. The depletion of ANXA2 has been shown to dissociate VE-cadherin from AJ [76]. The DE proteins specifically under AJs category include myosin heavy polypeptide 9 (non-muscle), septins, tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, HSP90α Cell junction associated protein-protein interactome from whole brain lysate (A) and brain derived isolated endosomes (B). Darker nodes are the source node, representing target genes associated with cell junctions. The gray nodes stands for the interacting genes for our target genes. https://doi.org/10.1371/journal.pntd.0007960.g007

PLOS NEGLECTED TROPICAL DISEASES
(cytosolic), aldolase A, enolase 1, heat shock protein 5.These proteins are not direct structural proteins of the AJs such as VE-cadherin and beta-catenin but contribute to developmental AJs assembly and. For example, inhibition of HSP90α is capable of attenuate the reduced VE-cadherin and beta catenin level induced by thrombin, LPS, VEGF, or TGF-β1, which all are known to increase endothelial permeability [102][103][104]. It is also suggested that HSP90α is a downstream effector of RhoA signaling mediated by LPS [104]. Our LC/MS analysisrevealed a PLOS NEGLECTED TROPICAL DISEASES 9.4 fold increase in HSP90α in ANXA2-KO mice (whole brain lysate) compared to WT 5 days p.i. . . Elevated levels of HSP90α could lead to the instability of the AJ in the presence of LPS [105]. There is only one published study supporting the association, showing that high glucose increases ANXA2 expresasion in ECs and enhances the association of HSP90α and ANXA2 in both ECs and rat aorta [106]. The functional interaction between ANXA2 and hsp90 in the context of pathogen infection is largely unknown.
Stress fiber. Stress fiber is formed by contractile actin filament bundle and a variety of proteins responsible for actin stabilization including septins, α-actin, tropomyosin, Rho-associated protein kinase (ROCK), fascin, and myosin 2, etc. [107]. It is formed in the cells in response to the environmental mechanical stress, such as ECs in the presence of fluid shear stress. Cells cultured on stiff substrate form stress fiber whereas those on the soft substrate

PLOS NEGLECTED TROPICAL DISEASES
have much less fiber. Physiologically, stress fiber is important for cell adhesion, mechanotransduction and AJ maturation [107][108][109]. Stress fiber plays a positive role in AJ maturation, but it is not always the case in the pathological environment. Such phenomenon is also seen in VEGF-and activated-neutrophil-treated ECs [110,111].
Our data report four DE proteins in the stress fiber category: septin 5, septin 7, tropomyosin 1 and tropomyosin 3. Three out of the four proteins were increased in ANXA2-KO group. The association of ANXA2-dependent integrity of TJs to stress fiber formation is elusive. One of the possibilities is relevant to the dynamic between cortical actin cytoskeleton and stress fibers. TJs, the core structures of the BBB that seal BMECs, are at the most apical of all junctional components [101,112]. ZO-1 has been identified as a functional adaptor to link other TJ components to the cortical actin cytoskeleton [113], which lies just beneath the plasma membrane. When stress fibers are formed in response to different stimuli, such as inflammation and oxidative stress, cortical actins are lost, potentially causing disturbances in the TJ aperatus [114,115]. In BHK-IR cells, insulin-mediated loss of stress fiber and resulting cell detachment depend on the presence of AXNA2 and its tyrosine phosphorylation state [77]. Silencing ANXA2 with siRNA in MOI Muller cells directly leads to the accumulation of stress fiber [73].

Isolated endosome function group
Cell-cell junctions. AJ is a significantly enriched functional group. Ten proteins were found in this category. Endocytosis is known to mediate the stability of AJ [116]. Based on structural analysis using IF and drug intervention, Georgiou et al., suggested that impaired endocytosis leads to AJ instability [116]. AXNA2 is involved in the formation of early endosomes. AXNA2 depletion results in morphological changes in endosomes and alters their distribution [76]. Our proteomic data analyzing the protein difference between WT and ANXA2-KO mice upon rickettsial infection identified 10 proteins associated with AJ in the brain endosomal compartment. Interestingly, cadherin complex protein cadherin 2 and talin 1 were upregulated in ANXA2-KO after infections. Cadherin 2 is a direct structural protein of AJ, and it is involved in pericyte-EC adhesion, which is critical for the integrity of the BBB. Focal injection of antibody against cadherin-2 is sufficient to induce intracerebral hemorrhage [117]. The biological significance of cadherin-2 upregulation in the endosomal compartment is unknown. It has been shown, however, that cadherin-2 and other several other junction-associated proteins were subjected to rapid endocytosis in a drug-induced disruption of cell-cell contact model [118]. Talin 1 is a well-known AJ-associated protein located at the cytoplasmic side to the AJ [119]. Endosomal talin has been shown to regulate the function of endosomal integrin [120].
MHC II protein complex binding and stress response. Three major players are found in MHC II protein complex binding: HSP90α, heat shock protein 8 (HSPa8), and pyruvate kinase. HSP90α is noteworthy. In ANXA2-KO mice, it was drastically enriched in whole-brain but decreased in the endosomal compartment, indicating a reduced protein turnover rate or degradation. Given the direct association of ANXA2 and endocytosis, ANXA2 may be an upstream regulator of HSP90α responsible for its recycling upon stimulation. HSP90α is involved in the disruption of BBB function post cerebral ischemic stroke. Inhibition of HSP90α with ATP competitive inhibitor results in reduced activity of metalloproteinase 9 (MMP9), which plays an important role in BBB dysfunction, and rescues TJ protein expression abnormality [121]. HSP90α is also associated with inflammation induced by LPS [122]. It has been shown that the involvement of HSP90α is associated with PI3K and NF-κB pathways [123]. Therefore, a possible mechanism explaining the increased susceptibility of ANXA2-KO to R.australis-induced CMH is that ANXA2 depletion causes reduced degradation of HSP90α, which destabilizes the cell-cell junction and increases the hyperpermeability of BBB. So far AXNA2 is known to interact with HSP90α and such interaction is increased in the presence of high glucose [106], yet the role of HSP90α-AXNA2 interaction in bacterial infection is unknown. Further in vitro experiment is needed to confirm this mechanism.
HSPa8 is a noncanonical member of the heat shock protein 70 family [124]. The majority of HSPa8 resides in the cytosol and nucleus, but upon stimulation, it also translocates to the plasma membrane or exosome, contributing to antigen presentation via MHC II to CD4 T cells, clathrin-mediated vesicles transport, and chaperone-dependent autophagy [124,125]. In our data, similar to HSP90α, HSPa8 was found to be reduced in the endosomal compartment in ANXA2-KO group, whereas it was upregulated in the whole brain lysate (one-fold). However, the biological role of the association between AXNA2 and HSPa8 is unknown.
HSP90α, HSP8 and mitogen-activated protein kinase kinase 4(Map2k4) comprise another functional group, stress response. Map2k4 is a direct activator of c-jun activation protein kinase (JNK) [126]. Map2k4 was enriched in the endosomal compartment in ANXA2-KO mice brain compared to WT. Interestingly, it has been reported that ANXA2 knockdown by shRNA enhances the activation of JNK and p38 in response to oxidative stress [127].

Other considerations
An earlier study reported that the deletion of ANXA2 rendered mice susceptible to chemically-induced carotid arterial thrombosis [128]. To determine the possible involvement of thrombus formation, which is associated with EC injury [129], in cerebral microhemorrhage, we examined all H&E staining but no thrombus was visualized.
We previously showed that the exchange protein directly activated by cAMP (EPAC) plays a critical role during SFGR infections [84] and that EPAC regulates ANXA2-mediated vascular fibrinolysis [130]. Given the regulatory role of EPAC on ANXA2, we retrospectively reviewed archival H&E-stained brain sections of EPAC1-deficient mice [84,130] infected with an ordinarily lethal dose of SFGR, but no histological evidence of CMHs was found, suggesting EPAC is not associated with CMHs during infections.

Conclusion
This is the first report that correlates the special role of ANXA2 and CMHs in the context of rickettsial and Ebola virus infections. Our data suggest that ANXA2 does not affect the in vivo proliferation of R. australis or the survival of the mice infected with R. australis. Inflammatory cytokines TNFα and IFNγ are probably not essential to ANXA2-dependent CMHs in R. australis infection. LC/MS studies using the whole brain lysates and endosomal proteins from the brain of R. australis-infected mice has revealed that deletion of ANXA2 results in a series of differentially expressed proteins that are relevant to cell-cell junction category. Consistently, histology showed that deletion of ANXA2 abrogates the structural integrity of endothelial TJs in rickettsial and Ebola virus infections, suggesting an essential role of ANXA2 in stabilization of the TJs and BBB. The detailed mechanism is unclear, but HSP90α seems to be an interesting target for future study. Further in vivo experiments are necessary to characterize the role(s) of ANXA2 in the pathogenesis of Ebola infection.