Background
Neuromyelitis optica (NMO) is a severe, generally relapsing disease of the central nervous system (CNS) characterized by optic neuritis and transverse myelitis with longitudinally extensive spinal cord lesions [
1,
2]. The identification of an NMO-specific autoantibody (NMO IgG) and aquaporin 4 (AQP4) as an antigenic target of this antibody defined NMO as a distinct disease with unique pathogenic and pathological characteristics [
3]. AQP4, the principle water channel in the CNS, is densely expressed on perivascular astrocytic endfeet and is crucial for bidirectional water transport and normal CNS homeostasis [
4,
5]. The AQP4 expression pattern and distribution of NMO-specific lesions [
6] suggest that astrocytes are a cellular target of NMO IgG and that NMO is a primary astrocytopathy [
7].
Immunopathological analyses of active NMO lesions define a unique vasculocentric pattern of complement activation and granulocytic infiltration involving both eosinophils and neutrophils [
8,
9]. Characteristic IgG deposition and complement activation on the adluminal surface of the vasculature corresponds to the location of the astrocyte endfeet that envelop the blood vessels [
8]. Evidence from ex vivo and in vitro studies is currently interpreted in support of a model for NMO pathogenesis wherein NMO IgG gains entry into the CNS, binds to AQP4 on astrocytic foot processes, and induces complement activation and deposition of the terminal membrane attack complex, resulting in astrocyte injury and death that leads to recruitment of eosinophils and neutrophils into the lesions [
4,
10]. In this model, complement-mediated astrocyte death is the key driver of chemokine, cytokine, and toxic effector production in lesions that results in the recruitment of macrophages that then induce demyelination and the death of oligodendrocytes and neurons [
11]. This model defines granulocytic recruitment as a
consequence of complement-mediated astrocyte death. However, recent evidence from human tissue indicates that many NMO lesions are non-destructive but highly inflammatory, with prominent activation of parenchymal microglia and perivascular macrophages, infiltration of neutrophils, and degranulation of infiltrated eosinophils in the absence of astrocyte death, terminal complement deposition, or overt tissue destruction [
9,
12]. This suggests that alternative mechanisms may be responsible for granulocytic recruitment in early NMO lesions.
Astrocytes are central mediators of general CNS homeostasis, participating in and controlling key metabolic cascades that are vital for normal neuronal function. Astrocytes are also active participants in the pathogenesis of numerous CNS diseases, modulating local inflammatory responses, controlling blood–brain barrier function, and serving as a source of chemokines and cytokines [
13,
14]. Such astrocyte-initiated inflammatory responses set the stage for leukocyte-mediated feedback loops that elicit profound neuropathology during infection, inflammation, autoimmunity, and trauma. Recently, we observed that stimulation of primary rat astrocyte cultures with serum or IgG isolated from NMO patients resulted in the release of the potent pro-granulocytic chemokine CCL5, with essentially no release stimulated by serum from MS or systemic lupus erythematosus (SLE) patients [
7]. These data suggest that astrocytes respond directly to NMO patient-derived IgG, and that the stimulated chemokine response is disease-specific and pro-granulocytic. Based on these observations, we hypothesize that the astrocytic inflammatory response to stimulation by NMO IgG represents one of the earliest pathogenic events in NMO, preceding severe and irreversible cell death and tissue damage.
Methods
Histopathology analysis
Histopathology was performed on archival formalin-fixed paraffin-embedded autopsy-derived CNS tissue from 23 patients clinically and pathologically diagnosed with NMO or NMO spectrum disorder. Five-micrometer-thick sections were stained with hematoxylin and eosin (H&E), luxol fast blue, and periodic acid–Schiff or Bielschowsky silver impregnation. Immunohistochemistry was performed using primary antibodies against proteolipid protein (PLP) (1:500, Serotec), glial fibrillary acidic protein (GFAP) (1:100, Dako), and AQP4 (1:250, Sigma). C9neo was detected using monoclonal clone B7 (1:200) or polyclonal anti-C9neo (1:200), both a gift of Prof. Paul Morgan, Cardiff, UK.
A topographical map was made in order to define regions of interest based on the following: (1) stage of demyelinating activity (active demyelination, inactive demyelination, remyelination, periplaque white matter, or normal appearing white matter); (2) the extent of tissue damage, graded as none, mild (tissue vacuolation with mild microglial reaction), moderate (damaged and disorganized parenchymal cell components with obvious macrophage infiltration), or marked (prominent parenchymal cell loss or cystic lesions); (3) the nature of the astrocytic reaction based on GFAP staining and hypertrophy of astrocytic processes or the presence of dystrophic astrocytes [
15]; (4) the presence or absence of complement deposition; and (5) the loss of AQP4 expression.
Eosinophils and neutrophils were identified based on morphological characteristics using H&E-stained sections. Eosinophil infiltration was measured semi-quantitatively in regions of interest and categorized as follows: mild = 1–3 cells per high power field (HPF) (40× objective lens); moderate = 4–10 cells/HPF; or marked >10 cells/HPF. Neutrophil infiltration was categorized as follows: mild = 1–3 cells/HPF; moderate = 4–20 cells/HPF; or marked >20/HPF.
All features of interest were captured as categorical data. Each feature was summarized in a contingency table and cross-classified according to the semi-quantitative assessment of granulocyte infiltration. To test for associations, the contingency tables were analyzed using log-linear regression models in the framework of generalized estimating equations that employed an “exchangeable” correlation structure in order to account for repeated observations among patients [
16]. Each contingency table summarized region-level data such that an individual patient could contribute multiple regions to the data set. Intra-patient regional correlations were controlled for in the generalized estimating equations. All analyses were performed using R statistical software package version 3.0.2.
Patient serum processing and IgG purification
Blood was drawn from patients or healthy volunteers and IgG was isolated from sterile-filtered, heat-inactivated serum samples as previously described [
7]. For the present study, results were generated using purified IgG from five different pools prepared since 2011 (Additional file
1: Table S1). Representative results were generated from NMO patient sera pooled from 5 males and 36 females ranging in age from 14 to 79, with a median age of 48 (Additional file
2: Table S2). Control sera were pooled from age- and sex-matched donors (Additional file
3: Table S3). All treatments with human IgG were at 100 μg/mL [
7].
Mouse primary mixed glial cultures
Mixed glial cultures were prepared from P1-P3 Balb/c mouse pups, as described [
17]. Cells were plated at 1.3 × 10
5 cells/cm
2 on poly-L-lysine hydrobromide. After 4 days in vitro, flasks were shaken to remove microglia and oligodendroglia. The astrocyte-enriched cultures were incubated for an additional 22 days and were then replated at 5.2 × 10
4 cells/cm
2 on poly-D-lysine. For all biochemical measurements, cells were stimulated starting at 31 days in vitro.
Immunostaining and imaging
Cells were immunostained with mouse anti-GFAP antibody (Millipore, MAB360) at 1:200 and anti-NFκB p65 antibody (Cell signaling, 8242) at 1:400, as described [
7]. Images were acquired using an LSM780 inverted confocal microscope (Carl Zeiss) and Zen software. Z-stacks were rendered into maximum intensity projections in ImageJ. All images were collected under identical conditions within a given experiment.
Microarray
RNA samples were assessed by Agilent for integrity, purity, and concentration. Samples passing quality control were analyzed on Illumina mouse WG-6 v 2.0 expression BeadChips in the Mayo Clinic Medical Genome Facility Gene Expression Core. Expression data were analyzed using Excel and MatLab [
7]. Heatmaps and hierarchical clusters were derived using Gitools v2.2.2 and pathway identification was performed using Ingenuity Pathway Analysis.
Immunoblotting
Cells were serum-starved overnight prior to stimulation with NMO IgG or control IgG then lysed in RIPA buffer containing protease/phosphatase inhibitors. Cell lysates (10–30 μg) were run on 4–15 % Criterion Tris–HCl gels (Biorad). After transfer, blots were probed using anti-IκB-α (Cell Signaling 9242), anti-phosphorylated IκB-α (Cell Signaling 2895), anti-p65 (Cell Signaling 8242), anti-NUP98 (Cell Signaling 2598), or anti-tubulin (Sigma T9026) antibodies.
ELISA
Following stimulation of cells, supernatants were collected, clarified, and stored as aliquots at −80 °C until analysis. Mouse CCL5, CCL2, CXCL1, and CXCL2 were detected in the supernatants using ELISA construction kits (Antigenix America).
Statistics
α = 0.05 and β = 0.2 were established a priori. Post hoc power analysis was performed for all experiments and significance was only considered when power ≥0.8. Statistical analysis was performed using SigmaStat (Systat Software). Normality was determined by the Shapiro–Wilk test and normally distributed data were checked for equal variance. Parametric tests were only applied to data that were both normally distributed and of equal variance. The Student–Newman–Keuls pairwise comparison test was used for all post hoc sequential comparisons. The figures show representative results from at least two separate experiments performed in triplicate using independent cell cultures and purified IgG. Over the course of this study, five different NMO patient pools were utilized for the preparation of purified IgG (Additional file
1: Table S1). As shown in Additional file
4: Figure S1, similar results were attained using different patient serum pools.
Study approval
All cell culture-based experiments were performed using materials approved by the Mayo Clinic institutional animal care and use committee. All studies were conducted in accordance with the United States Public Health Service’s Policy on Humane Care and Use of Laboratory Animals. The Mayo Clinic institutional review board approved the use of human materials. All subjects provided written informed voluntary consent after the nature and possible consequences of the study were explained.
Discussion
It is clear that NMO is associated with a unique granulocytic “footprint.” NMO patients often have CSF pleocytosis that includes the presence of polymorphonuclear leukocytes [
23‐
25], an increase in pro-granulocytic chemokines in the CSF [
25‐
28], and notable accumulation of granulocytes in lesions [
8,
29,
30] (Fig.
1). In the current conventional model of NMO pathogenesis, terminal complement deposition following binding of NMO IgG to astrocyte endfeet precipitates damage and concomitant granulocyte recruitment into the CNS. This model considers granulocyte recruitment as a downstream
effect of terminal complement complex formation and tissue injury. However, terminal complement deposition is not a universal feature of all NMO lesions, and therapeutics targeting complement inhibition are efficacious in only some patients [
31], suggesting that complement-mediated tissue destruction in NMO lesions may represent only one possible pathogenic mechanism. We contend that there are also early, sub-lytic, and highly inflammatory astrocytic responses to NMO IgG that contribute to early granulocytic recruitment. In our model, granulocyte recruitment is an upstream
cause of injury and is triggered by astrocyte signaling, rather than astrocyte death. We observed clear granulocytic accumulation in NMO patient tissue that was not dependent on complement deposition, active demyelination, or tissue destruction, along with evidence of reactive astrocytes in these regions (Fig.
1). Building on these observations, we utilized an astroglial culture system to examine the rapid cellular and molecular events induced by stimulation with NMO IgG in the absence of exogenous complement. We found that such stimulation engaged a highly inflammatory and reactive astrocyte transcriptional program that included the upregulation of numerous genes encoding pro-granulocytic chemokines (Fig.
2). Further analysis of the transcriptional program initiated by stimulation with NMO IgG revealed that the NFκB signaling pathway was significantly upregulated. Confirming engagement of the NFκB signaling pathway, we observed phosphorylation of IκB-α and nuclear translocation of the NFκB transcription factor p65 in astrocytes following stimulation with NMO IgG. Treatment with a spectrum of NFκB inhibitors effectively blocked these responses (Fig.
3). Finally, we found that the potent pro-granulocytic chemokines CCL5, CCL2, CXCL1, and CXCL2 were released by cells following stimulation with NMO IgG, and that the release of all except CXCL2 was effectively blocked by inhibition of NFκB (Fig.
4), including inhibition using clinically relevant proteasome inhibitors (Fig.
5).
NFκB is a well-studied master regulator of autoimmunity that is crucial for both inflammation and immune tolerance. While NFκB activation occurs transiently in the course of a normal immune response, chronic activation of this signaling pathway in target tissues is associated with pathogenesis in many autoimmune diseases [
32]. Importantly, therapeutic targeting of the NFκB signaling pathway is clinically feasible and may provide a strategy for controlling the transition from normal immunity to autoimmunity. In our model, proteasome inhibition by MG132, bortezomib, and PR-957 effectively blocked the release of several pro-granulocytic chemokines (Figs.
3,
4, and
5). The efficacy of bortezomib is of interest due to its current therapeutic use in multiple animal models and in patients. Bortezomib treatment results in decreased inflammation in animal models of contact hypersensitivity [
33], allograft rejection [
34], and SLE [
35]. Bortezomib is an approved therapy for multiple myeloma patients where treatment inhibits NFκB and induces myeloma cell apoptosis [
36]. Bortezomib was also recently shown to halt autoantibody production and kill plasma cells in a murine model of myasthenia gravis [
37]. Of note, long-lived plasma cells, due to the metabolic demand required by a high rate of IgG production, are particularly sensitive to proteasome inhibition [
35] but are generally resistant to immunosuppressive drugs, including anti-CD20 antibodies such as rituximab that are currently used to functionally inhibit or deplete B cell populations [
38]. Considering the upregulation of B cell-related factors and other inflammatory drivers in astrocytes stimulated with NMO IgG (Fig.
2) and the published evidence for intrathecal production of IgG in NMO patients [
39,
40], therapeutic proteasome inhibition may provide the dual advantage of simultaneously targeting both the inflammatory astrocytic response and plasma cell survival—essentially blocking both the stimulator and the concomitant stimulation. Furthermore, the clinical efficacy of bortezomib coupled with immunomodulators such as lenalidomide or methylprednisolone in transplant and multiple myeloma patients suggests that combined drug strategies may confer significant therapeutic advantage in complex inflammatory diseases such as NMO.
One complication with bortezomib is the potential for peripheral neuropathy at therapeutic doses. Therefore, we expanded our study to include testing of PR-957 (ONX-914), an immunoproteasome inhibitor that has shown efficacy in an animal model of rheumatoid arthritis (RA) at far lower doses than are employed for the proteasome inhibitor carfilzomib [
21]. Similar results were found in mouse models of colitis, MS [
41], and SLE [
22], where PR-957 treatment significantly inhibited the production of pro-inflammatory cytokines and reduced the severity of disease symptoms. The immunoproteasome is mostly found in cells of the immune system but can be expressed in other cells, including astrocytes, upon exposure to various stressors [
42]. In a pro-inflammatory environment, virtually all newly synthesized 20S proteasomes incorporate inducible subunits associated with the immunoproteasome rather than constitutive subunits [
43]. Treatment with PR-957 in our system resulted in efficient blockade of NFκB p65 translocation in astrocytes and inhibited the release of CCL5, CCL2, and CXCL1 (Fig.
5). These results suggest that astrocytic inflammatory responses in patients with NMO may be targeted by specific inhibition of the immunoproteasome. Finally, evidence that immunoproteasome inhibition also suppresses the production of autoantibodies [
22] and selectively targets plasma cell survival [
21] further suggests that such a strategy may confer significant disease-modifying effects in patients with NMO. We propose that the use of drugs such as PR-957 in NMO patients may reduce or block disease pathogenesis via parallel pathways that reduce astrocyte reactivity, suppress CNS inflammation, block granulocyte recruitment, reduce production of NMO IgG, and deplete autoantibody-producing plasma cells that are resistant to current therapies such as rituximab [
38].
One limitation of our study is the absence of a suitable animal model in which to test our hypothesis. Existing mouse models of NMO rely on the concurrent initiation of active (via immunization) or passive (via transfer of myelin-specific T cells) EAE [
44‐
46], co-injection of NMO IgG and human complement factors into the CNS [
47‐
49], or injection of the proinflammatory cytokine IL-1β directly into the brain to drive granulocyte recruitment and complement factor production [
50]. The “success” of these models is due to inflammatory modulation of the blood–brain barrier and direct induction of complement-mediated pathology. Unfortunately, the use of any of these models would compromise the analysis of early astrocytic responses to NMO IgG and are therefore not suitable to test our hypothesis. Another area in which a suitable animal model is necessary but currently unavailable is the analysis of antibody-induced granulocytic recruitment in the absence of overt tissue destruction or complement deposition. A pathogenic role for neutrophils in CNS autoimmune disease was suggested by the occurrence of severe exacerbations in some MS and NMO patients given recombinant granulocyte-colony stimulating factor (G-CSF) which stimulates the function and proliferation of granulocytes [
51]. Although the presence of neutrophils in typical MS lesions is rare, studies in the EAE model indicate that neutrophils are found at high frequency in the CNS parenchyma during the preclinical phase, increase dramatically in the meninges both preclinically and at relapse, and may potentiate the formation of lesions by mediating the breakdown of the blood–brain or blood-cerebrospinal fluid barriers or by stimulating the maturation of local antigen-presenting cells [
52]. With regard to eosinophils, degranulated cells are found in both NMO meninges and early lesions [
30]. While historically recognized as endstage effectors in parasitic immunity and allergic diseases, it is quite likely that eosinophils directly contribute to tissue injury in NMO via release of cytokines, chemokines, lipid mediators, oxygen burst components, and cytotoxic granule cationic proteins [
53]. Elucidating and targeting the mechanisms that recruit granulocytes to early NMO lesions and discovering strategies to inhibit the initial damage triggered by these cells will require the creation of animal models that do not involve the tautological induction of complement-dependent injury or the initiation of inflammation that does not build from a reactive astrocyte response.
Competing interests
Dr. Lucchinetti may accrue revenue for a patent regarding AQP4-associated antibodies for the diagnosis of neuromyelitis optica; she receives royalties from the publication of Blue Books of Neurology: MS 3 (Saunders Elsevier, 2010); she has received personal compensation for consultation services from Biogen-Idec; and she receives research support from Novartis, Alexion, and Biogen (principal investigator). Dr. Howe receives research support from Alexion and Sanofi (principal investigator). The other authors report no conflicting financial or non-financial interests. The funders of the present work had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Authors’ contributions
MEW-C performed the experiments, analyzed the data, prepared the figures, and wrote the manuscript. YG performed the experiments, analyzed the data, and prepared the figures. RKJ and CBM provided technical support for the experiments. PDF-G provided statistical support and analyzed the data. CFL designed the experiments, analyzed the data, and wrote the manuscript. CLH designed the experiments, analyzed the data, prepared the figures, and wrote the manuscript. All authors read and approved the final manuscript.