Purified IgG from aquaporin-4 neuromyelitis optica spectrum disorder patients alters blood-brain barrier permeability

Background Neuromyelitis optica spectrum disorders (NMOSD) is a primary astrocytopathy driven by antibodies directed against the aquaporin-4 water channel located at the end-feet of the astrocyte. Although blood-brain barrier (BBB) breakdown is considered one of the key steps for the development and lesion formation, little is known about the molecular mechanisms involved. The aim of the study was to evaluate the effect of human immunoglobulins from NMOSD patients (NMO-IgG) on BBB properties. Methods Freshly isolated brain microvessels (IBMs) from rat brains were used as a study model. At first, analysis of the secretome profile from IBMs exposed to purified NMO-IgG, to healthy donor IgG (Control-IgG), or non-treated, was performed. Second, tight junction (TJ) proteins expression in fresh IBMs and primary cultures of brain microvascular endothelial cells (BMEC) was analysed by Western blotting (Wb) after exposition to NMO-IgG and Control-IgG. Finally, functional BBB properties were investigated evaluating the presence of rat-IgG in tissue lysate from brain using Wb in the rat-model, and the passage of NMO-IgG and sucrose in a bicameral model. Results We found that NMO-IgG induces functional and morphological BBB changes, including: 1) increase of pro-inflammatory cytokines production (CXCL-10 [IP-10], IL-6, IL-1RA, IL-1β and CXCL-3) in IBMs when exposed to NMO-IgG; 2) decrease of Claudin-5 levels by 25.6% after treatment of fresh IBMs by NMO-IgG compared to Control-IgG (p = 0.002), and similarly, decrease of Claudin-5 by at least 20% when BMEC were cultured with NMO-IgG from five different patients; 3) a higher level of rat-IgG accumulated in periventricular regions of NMO-rats compared to Control-rats and an increase in the permeability of BBB after NMO-IgG treatment in the bicameral model. Conclusion Human NMO-IgG induces both structural and functional alterations of BBB properties, suggesting a direct role of NMO-IgG on modulation of BBB permeability in NMOSD.


BACKGROUND
Neuromyelitis optica spectrum disorders (NMOSD) is a severe autoimmune disease of the central nervous system (CNS) that mainly affects the optic nerve and the spinal cord. The discovery of a serum antibody, termed NMO-IgG, directed against the aquaporin4 channel expressed at the CNS interfaces, has enhanced the understanding of NMOSD which is now considered an autonomous entity with distinctive pathophysiology, different from multiple sclerosis [1].
Contrary to the extended evidence concerning NMO-IgG involvement in NMOSD tissue lesion formation, the mechanisms for antibody penetration into the CNS, still awaits elucidation. Over the last few years, four main hypotheses have been proposed to explain the passage of NMO-IgG, from either serum or cerebrospinal fluid (CSF) into CNS. First, NMO-IgG has been observed to access the brain from the blood (blood-CNS barrier) through fenestrated endothelial cells in circumventricular organs such as the area postrema where AQP4 protein is highly expressed [2,3].
Although, NMO-IgG deposition was initially thought to be restricted to the area postrema [2], a recent study, using a monoclonal murine AQP4-antibody with high antigen affinity, found a wide diffusion in the CNS [4]. Second, in vitro models have 5 shown that either the NMO-IgG itself [5,6], or other components from the serum of NMOSD patients (such as matrix-metalloproteinase 2/9 protein, antibodies against brain endothelial cells, or glucose-regulated protein-78) may alter the blood brain barrier (BBB) at the glio-vascular unit (blood-CNS barrier) [7][8][9]. Third, NMO-IgG might gain access to the CNS via the CSF using the paravascular pathway (CSF-CNS barrier) [10] by which the end-feet of astrocytes would be directly exposed to CSF circulating NMO-IgG, allowing their entrance into the CNS [11]. In fact, higher levels of NMO-IgG have been found in the CSF of NMOSD patients at relapse compared to remission phases [12]. Finally, extravasation of the antibody through meningeal and small parenchymal vessels has been recently proposed as new route for antibody entry into the CNS [4].
In the field of NMOSD, the impact of NMO-IgG on BBB has mainly been evaluated using in vitro assays [5][6][7][8]. However, the use of ex vivo models could provide a more precise representation of the antibody behaviour when reaching the BBB, and a better understanding of the underlying pathophysiology.
Herein, an ex vivo approach consisting of fresh brain microvessels isolated from rat brain to model the blood-CNS barrier was used for the first time to assess the impact of NMO-IgG on the BBB. A first analysis was performed to determine whether IgG from AQP4-positive NMOSD patients (NMO-IgG) may orchestrate BBB alteration by inducing a specific cytokine secretome profile in isolated brain microvessels (IBM). Then, using both IBM and an in vivo animal model simulating CSF-CNS barrier, the structural modifications of the BBB induced by NMO-IgG were investigated.
Finally, to evaluate whether such structural modifications are sufficient to induce a breakdown of the barrier an in vitro bicameral model was set-up to simulate the blood-CNS barrier. 6

Patients, IgG-patient purification, and NMO-IgG selection
De-identified serum specimens were obtained from six different relapsing NMOSD patients that were included in the French cohort of NMOSD (NOMADMUS), and stored at NeuroBioTec (Biological Resource Centre of the Hospices Civils de Lyon, France).
All patients were tested positive for AQP4-antibodies detected by cell-based assay [13], and all patients fulfilled the 2015 criteria for NMOSD [14]. Serum provided by the French blood service (Etablissement Français du Sang) from healthy blood donors were used as controls. Both NMO-IgG (called NMO-IgG 1-6 ) and healthy donors-IgG (called Control-IgG) were purified from the NMOSD and healthy donor serums, respectively, using chromatography over Protein-A Sepharose as previously described [15]. IgGs were then used at a final concentration of 2 µg/µl. NMO-IgG from the different patients were first tested on cultured astrocytes in order to select NMO-IgG for the experiments on fresh IBM and for the animal model (NMOrat). After 24 hours of treatment with NMO-IgG 1-6 (number of experiments=39), there was a significant decrease in cultured astrocyte AQP4 expression (mean 114.9±24.3) when compared to treatment with Control-IgG (n=12, mean 70.3±17.9, p<0.001), as measured by western blot (Wb). The effects of NMO-IgG 1-6 and Control-IgG were evaluated in duplicate or triplicate. NMO-IgG 1,2,6 were the most pathogenic, inducing a higher decrease in AQP4 expression than NMO-IgG 3,4,5 .
Among the most pathogenic NMO-IgG, NMO-IgG 1 was randomly selected to carry out experiments.

Intraventricular brain NMO-IgG infusion, and ethical procedures
Animal surgery was performed as previously described [15]. Briefly, rats were anesthetised and mounted in a stereotaxic frame. Two hundred µl of purified IgG (NMO-IgG 1 ) or NaCL (Control-IgG) were infused into the CSF during seven days at a 1 µL/hour using a sterile subcutaneous osmotic pump. After seven days, rats were anesthetised and received an intracardiac perfusion of 100 ml phosphate-buffered saline (PBS; 0.1 M, pH 7.4). Then, rats were sacrificed using pentobarbital, brain was then removed and frozen in isopentane at -30°C, and further stored at -80°C.
Brain tissue was embedded in Tissue-Tek OCT, and subsequently cut on a cryomicrotome [15]. Brain slices were used for immunohistochemical (IHC) study of tight junction proteins.
A total of 7 rats received NMO-IgG 1 (NMO-rats) and 9 rats received NaCl
Using oxygen-saturated buffers, cortices were homogenised in a Dounce-type glassglass homogeniser. The preparation was further homogenised in 5 vol/g tissue of 1% BSA-supplemented Kreps Ringer buffer (AKRB), and then filtered through a 500µm 8 mesh sieve (Netwell, Corning, Corning, NY). The filtrate was diluted with 1% AKRB (1:1) and homogenised again. Then, the homogenate was centrifuged for 10 minutes, the pellet was suspended in Krebs-Ringer buffer containing 17.5% 70 kDdextran, and further centrifuged for 27 minutes. Myelin was retained at the surface of the gradient and the resulting pellet was suspended in 1% AKRB. Pellet was consecutively filtered through 200 and 74-µm mesh sieves to eliminate larger vessels. The preparation was centrifuged for 15 minutes. Finally, the pellet was suspended and filtered on a 40µm mesh sieve (BD Bioscience, Erembodegem, Belgium). All procedures were performed at 4°C.
The fresh IBM retained were recovered in AKRB 0.1% and 2 µl were used to observe their morphology under the microscope. Immunofluorescence analysis showed the presence of Claudin-5, Occludin, and ZO-1 in fresh IBM preparation (Supplementary Finally, fresh IBM from each well were suspended in DMEM up to 1ml.
Fresh IBM were prepared for Wb or cytokine/chemokine analysis after 20 hours of incubation.

Brain microvascular endothelial cells
Brain microvascular endothelial cells (BMEC) were isolated as previously described [17]. Briefly, microvessels were isolated from 3 rat brains following mechanical and enzymatic digestion of brain tissue, and plated in a T75 flask coated with collagen and fibronectin (microvessels from 1.5 cortex per T75 flask). BMEC were grown and purified with decreasing concentrations of puromycin for 5 days, followed by 3 days of culture in an insulin, transferrin, and sodium selenite-containing medium. The cells were cultured on plastic for BMEC experiments or harvested for seeding on filters to generate in vitro BBB models, as follows. After dissociation during 30 seconds with trypsin 0.05%-EDTA 0.02% solution at 37°C per T75 flask, the detached BMEC (approximately 3*10 6 cells/T75) were dissociated by pipetting up and down. Re-suspended BMEC were then added to the upper compartment of a precoated well (filters filled with endothelial cell media (ECM) at high density (160*10 3 cells/filter of 1.1cm 2 ). Endothelial cells expressed tight junction proteins within 3 days [17]. BMEC were then exposed to DMEM/BSA0.1% (termed non-treated BMEC), Control-IgG or NMO-IgG 1-6 for 20 hours at the concentration of 350 µg/ml.

Permeability analyses: Bicameral model
For antibody permeability analyses, BMEC were exposed to three different NMO-IgG in the upper compartment of the culture system. The acceptor compartment was then collected, and frozen for later NMO-IgG analysis by Enzyme-linked immunosorbent assays (ELISA). The apical part of the model simulates blood and the basolateral side simulates the CSF.
To test the integrity of the barrier, [14C] sucrose was added in the upper compartment at the end of the incubation with IgG, and sucrose transfer was measured over a 1-hour period. The radioactivity was measured over time in the lower compartment using a Perkin Elmer TRICARB, 4910TR liquid scintillation analyser, Singapore. Clearance curves were generated and permeability coefficients were calculated as described [18].

Enzyme-linked immunosorbent assays
The level of human IgG were quantified in the compartment acceptor by ELISA set according to the manufacturer's protocols (Human IgG ELISA Quantitation Set, Bethyl Laboratories, Inc, Montgomery, Alabama, USA). Samples were diluted at 1:15.
The optical density was measured at 450 nm with λ correction of 570 nm using a Spark spectrophotometric microplate reader (Tecan Trading AG, Switzerland). Based on the optical density value of each sample, the sample concentration was calculated in ng/mL and used to calculate clearance.

Cytokine Array analysis
Evaluation of cytokines and chemokines from fresh IBM supernatants were assayed by using Proteome Profiler Array kit (rat cytokine array panel A, R&D Systems, Minneapolis, MN, USA) according to the manufacturer's instruction. Fluorescence signal intensity was measured by ImageJ software.

Astrocyte cultures
As previously described, [15] primary glial cultures were isolated from 1-day-old rat pups (n=24) and dissociated cells were further diluted in DMEM to a density of 2.10 5 cells. Cells were seeded in 6-well plates and incubated at 37°C in a moist 5% CO 2 , 95% air atmosphere. In order to obtain pure astrocyte cultures and eliminate microglia and oligodendrocytes, cells were treated with cytosine arabinoside (AraC, 25nM, Sigma-Aldrich).

Immunohistochemistry procedures
Both cytospined IBM and brain slides were fixed in 4% paraformaldehyde for 10 minutes, then washed in PBS, and blocked for 1 hour in blocking solution (10% Normal Goat serum, PBS 1X, BSA 1%, Triton 0.3%). Then, the material was incubated overnight at 4°C with primary antibodies in blocking solution and further washed in PBS. Slides were labelled at room temperature with specific fluorescent secondary antibodies in blocking solution for 30 minutes, washed in PBS and incubated with 4',6-diamidino-2-phenylindole (DAPI). Slides were mounted in tamponed glycerol and stored at 4°C. Axio Imager Z1 Apotome technology (Zeiss, Oberkochen, Germany) was used to image the slides.

Immunoblotting
Fresh IBM, BMEC, or brain tissues were dissociated by ultrasound fragmentation in homogenisation buffer and phosphatases inhibitors, and then protein content was counted [15]. Protein samples were separated by sodium dodecyl sulphatepolyacrylamide gel electrophoresis (SDS-PAGE gels) and transferred to nitrocellulose membranes. Beta-actin was used as an internal standard. The membranes were treated with blocking buffer for 1 hour and incubated with the specified primary antibodies at 4°C overnight. Membranes were exposed to an anti-IgG antibody for 1h at room temperature and revealed using chemiluminescence, as described [15].
Quantification of the band intensity was obtained using ImageJ software.

Immunohistochemical reagents
The following IgGs were used for tissue sample immunodetection and lysates. The

Statistical analysis
All Wb analyses were performed by investigators blinded for treatment conditions. Data are presented as median (interquartile range [IQR]) and percentages. The nonparametric Kruskal-Wallis test was used to evaluate differences in more than two groups and U-Mann-Whitney test was used to compare two groups. All statistical analyses were performed using Prism 5.0 GraphPad Inc software.
Globally, exposition to purified IgG from AQP4-antibody positive NMOSD patients resulted in the production of an inflammatory secretome profile from brain 13 microvessels.

Purified-IgG induce Claudin-5 loss in brain microvessels
Assuming that the inflammatory environment induced by NMO-IgG has the capacity to alter the BBB, Claudin-5, Occludin, and ZO-1 protein expression analysis was performed in fresh IBM treated with NMO-IgG 1 (n=6) and Control-IgG (n=4) for 20 hours (Figure 2. Using the NMO rat model [15] an in vivo expression analysis of the same tight junction proteins was performed using Wb on fresh IBM obtained from these animals. Claudin-5 expression decreased by 16.2% in NMO-rat (n=1), compared to Control-rat (n=1) (Supplementary Table 1).
These findings were confirmed in primary BMEC cultures following 20-hour treatments with the six NMO-IgG 1-6 ( Figure 2.b Table 1.b). Immunoblotting showed that Claudin-5 expression decreased by at least 20% in five out of the six NMO-IgG conditions (NMO-IgG 1-5 ).
Tight junction proteins in situ, in the brain of NMO-rat were also studied. A decrease pattern of immunofluorescent staining intensity of periventricular Claudin-5 expression in NMO-rats (n=2) compared to Control-rats (n=2) was found ( Figure 3) but not for Occludin and ZO-1.Overall, these results indicate that purified IgG from AQP4-antibody positive NMOSD patients modify the expression of the tight junction protein Claudin-5.
When comparing the effects of each NMO-IgG 1-6 on either AQP4 expression in 14 astrocytes or on Claudin-5 on BMEC, expression changes in each cell type was not correlated (rho=-0.486, p=0.329) (Figure 4).
On the bases of this observation, we evaluated whether IgG from the six tested NMOSD patients were directed to BMEC. Using Wb on BMEC lysate, we did not find any binding pattern neither with patients' IgG (NMO 1-6 ) nor Control-IgG (Supplementary Figure 2).

NMO-IgG induce alteration of BBB functional properties in vivo and in vitro
The observation, in ex vivo and in vivo experiments, that NMO-IgG induce structural modification on tight junction proteins, suggests that NMO-IgG may lead to an alteration of BBB functional properties. We evaluate the diffusion of rat-IgG in periventricular brain areas using immunoblotting on the brains of NMO-rats (n=4) and Control-rats (n=6). One extra rat (Control-animal) did not receive saline intracardiac perfusion which allows blood elimination from brain vasculature and was thus used to identify rat IgG using Wb. Heavy

Discussion
The present study evaluating the effect of purified IgG from AQP4-antibody positive NMOSD patients on BBB properties showed that: 1) NMO-IgG induce an inflammatory secretome profile in brain microvessels, 2) have a direct impact on the molecular structure of the BBB, and 3) induce a functional alteration of the BBB.
In line with previous studies, we found an upregulation of several cytokines and chimokines, notably IL1-ra, IL1-β, TNFα, CXCL10 and CXCL3. This inflammatory secretome has been found increased in the CSF or serum of NMOSD patients reflecting a cell recruitment to sites of inflammation or even BBB weakening [19][20][21][22][23]. Interestingly, we found an up-regulation of IL-6, proposed to be a surrogate diagnostic and prognostic marker in NMOSD [19,[24][25][26], enhancing plasmablast survival and AQP4-antibody secretion [27]. More recently, IL-6 receptor blockage treatments have been proposed as a promising therapeutic option for NMOSD [28,29]. Apart from its effect on plasmablasts, IL-6 could be involved in BBB alteration. Indeed, a recent study using astrocyte and endothelial cell co-cultures showed IL-6 production in astrocytes and AQP4 internalisation after exposure to NMO-IgG in the abluminal side of the culture (brain compartment) [5]. Authors also found a structural BBB impairment characterised by a decreased and discontinuous Claudin-5 immunoreactivity at tight junctions, and a molecular leakage through the BBB that was reverted after IL-6 soluble receptor blocking [5]. Our data suggest that isolated capillaries may be a source of IL-6 production other than astrocytes, as previously described [20,30]. In summary, the specific inflammatory secretome induced by NMO-IgG in brain microvessels likely modulates a subsequent pathophysiological process characterised by the structural weakening and functional impairment of the BBB. Whether endothelial cells or pericytes remaining embedded in the basal membrane that delimit the isolated microvessels are the main source of these cytokines remains to be established.
In order to study the impact of NMO-IgG on BBB structure itself, we use an original ex vivo BBB mimicking the close relationship between NMO-IgG and brain microvessels. Fresh IBM are known to maintain in vivo BBB properties, and have been previously used for the study of endothelial molecular transporters [31,32], drug pharmacokinetic or pharmacodynamics [33], and disease pathophysiology in animal models [34,35]. However, the application of this technique on NMOSDrelated models has not been previously performed. The decrease in expression of Claudin-5 in IBM when exposed to NMO-IgG in the absence of any previous contact with breaching substances, suggests that NMO-IgG directly contribute to the structural destabilization of the BBB from the brain compartment. Moreover, we further reproduced same findings with the use of BMEC cultured with NMO-IgG observing such decrease of Claudin-5 expression. Two previous studies found similar results with a down-regulation of Claudin-5 in immortalised human brain microvascular endothelial cells treated with the serum of NMO patients but not in controls [7,8]. Accordingly, a decrease in the transendothelial electrical resistance (TEER) was observed which suggested a functional disruption of the BBB. [7,8] Whether the NMO-IgG "by itself" or other unknown auto-antibodies present in NMOSD patients are able to directly open the BBB is not completely resolved.
Interestingly, we found that the decreased expression of Claudin-5 did not correlate with the loss of AQP4 in individual patients. This finding may suggest that other IgG, different from the specific AQP4-antibody, could be involved in BBB disruption as recently proposed [7][8][9]. In fact, GRP78 an antibody specific to the surface of endothelial proteins has been recently identified in the serum of patients with AQP4antibodies and lupus erythematous systemic [9]. Other study described antibodies against BMEC in the serum of NMOSD patients that could alter BBB properties through an autocrine secretion of antivascular endothelial growth factor (VEGF) by the BMEC. [7] However, when performing the Wb we did not observe any specific pattern of IgG binding to endothelial cells in the present study.
NMO-IgG may exert its pathological effect from the CSF compartment after passing through subpial spaces where vessels penetrate the brain parenchyma [4,11]. To validate this hypothesis in vivo, an intraventricular chronic infusion of NMO-IgG in rat brain was performed. This leads to a wide NMO-IgG distribution through CSF in the neural tissues [15], reaching the subarachnoid and subpial spaces [11]. By analysing brain microvessels from NMO-rats it was possible to detect not only the deleterious effect of NMO-IgG, concerning mainly the loss of Claudin-5 but also a functional alteration represented by the passage of rat-IgG from the blood to the CNS. AQP4-expressing astrocytic end-feet would be directly exposed to CSF NMO-IgG, thus triggering astrocyte damage and further BBB disruption [6].

Conclusion
Overall, the present study provides a global perspective on the pathophysiological role of NMO-IgG at the BBB level. NMO-IgG induce a distinctive inflammatory secretome profile that may modulate the subsequent alterations of the BBB properties. The major finding involved in structural weakening is driven by the decrease of Claudin-5 expression both from the blood and the CSF compartments.
The functional alteration is reflected by a higher permeability through brain microvessels allowing the passage of circulating molecules to the CNS when the antibody is present in the CSF compartment. To this regard, the NMO-IgG "by itself" may penetrate through the breaching barrier, distribute through the extracellular fluid and, finally, trigger the disease after binding to the AQP4 channel.

Declarations
Author's contributions AC-C, AR, NS, JF G-E, PG and RM contributed to model design, implemented the model, performed simulations, and were major contributors to writing the manuscript. CR, SR and SC contributed to perform simulations and the critical review of the manuscript.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Consent for publication
N/A.

Funding
The present study is supported by a grant from ARSEP foundation and a grant provided by the French State and handled by the "Agence Nationale de la  Abbreviations; IBM, isolated brain microvessels; NMO, neuromyelitis optica  Figure 1 Cytokine/chemokine expression in isolated brain microvessels after 9 hours of treatment with Figure 2 Tight junction protein expression detected using Western blotting in ex vivo and in vivo mod Tight junction protein expression in brain microvessels of periventricular areas after NMO-IgG Aquaporin-4 and Claudin-5 expression tested in astrocytes and brain microvascular endothel