Neuromyelitis optica MOG-IgG causes reversible lesions in mouse brain

Introduction Antibodies against myelin oligodendrocyte glycoprotein (MOG-IgG) are present in some neuromyelitis optica patients who lack antibodies against aquaporin-4 (AQP4-IgG). The effects of neuromyelitis optica MOG-IgG in the central nervous system have not been investigated in vivo. We microinjected MOG-IgG, obtained from patients with neuromyelitis optica, into mouse brains and compared the results with AQP4-IgG. Results MOG-IgG caused myelin changes and altered the expression of axonal proteins that are essential for action potential firing, but did not produce inflammation, axonal loss, neuronal or astrocyte death. These changes were independent of complement and recovered within two weeks. By contrast, AQP4-IgG produced complement-mediated myelin loss, neuronal and astrocyte death with limited recovery at two weeks. Conclusions These differences mirror the better outcomes for MOG-IgG compared with AQP4-IgG patients and raise the possibility that MOG-IgG contributes to pathology in some neuromyelitis optica patients.

A few NMO patients without AQP4-IgG have IgG against myelin oligodendrocyte glycoprotein (MOG-IgG), which recognize extracellular conformational domains of MOG [10][11][12][13]. MOG is expressed on the outer surface of CNS myelin sheaths and comprises about 0.05% of total myelin protein [14]. There is mounting evidence that MOG-IgG NMO has more favorable clinical outcome than AQP4-IgG NMO, with resolution of imaging abnormalities [10,11,15,16]. It is currently unclear whether MOG-IgG plays any role in NMO by causing lesions in the CNS in vivo. Here we compared the effects of MOG-IgG with those of AQP4-IgG in the intracerebral injection mouse model. We used total IgG from a normal subject (IgG CON ) and from NMO patients with AQP4-IgG (IgG AQP4 ) or MOG-IgG (IgG MOG ).

IgG and complement
NMO patients with MOG-IgG or AQP4-IgG were identified using live cell-based assays. Briefly, AQP4-IgG and MOG-IgG positivity was determined by visualization of binding to human embryonic kidney cells, transfected with the extracellular and trans-membrane domains of MOG or with full-length M23-AQP4. Details of the assays are given elsewhere [5,10,11,15]. IgG was purified using Protein G from sera or plasmas of five patients with AQP4-IgG NMO, five MOG-IgG NMO or one healthy volunteer. The effect of injecting IgG and C hu from healthy volunteers into mouse brain was extensively investigated in our earlier studies [5,17,18]. The purified, dialysed and pooled total IgG preparations (6 -38 mg/ml) are termed IgG AQP4 , IgG MOG and IgG CON . Clinical details of the 5 AQP4-IgG + [5] and 5 MOG-IgG + [11,15] patients are given elsewhere. To deplete MOG-IgG, the IgG MOG was adsorbed by incubation with MOG-HEK cells until MOG-IgG became undetectable (IgG MOG(AdsMOG-HEK) ). IgG MOG adsorbed against untransfected HEK cells (IgG MOG(AdsHEK) ) was used as control. A chimeric mouse-human recombinant monoclonal anti-mouse MOG antibody, MOG-IgG 2B7 , was produced as described [19]. Human recombinant monoclonal anti-AQP4 IgG 1 , termed AQP4-IgG 53 , was also generated [20]. A measles virus-specific antibody termed CON-IgG 2B4 was used as isotype control [20]. The source of C hu was fresh serum from healthy volunteers [5].

Mice
Experiments were performed at St. George's, University of London using CD1 mice 8 -12 30 -35 g, 8 -12w old. Protocols were approved by the British Home Office (Project Licence, PPL 70/7081). After administering 2,2,2tribromoethanol i.p., mice were mounted onto a stereotactic frame (Benchmark, Neurolab, St Louis, MO, USA). Four burrholes were made on the right side using a high speed drill (0.7 mm burr, Foredom, Bethel, CT, USA) at the following coordinates in millimetres from the bregma (lateral, anterior): (1, 0), (1, −1), (1, −2), (2, −1). Mice were allocated to the different experimental groups by a person unaware of the aim of the study. A 30 g needle attached to 50 ml gas-tight glass syringe (Hamilton, Reno, NV, USA) was inserted 3 mm deep to micro-infuse (1 μL/min) into the right hemisphere 16.8 μL IgG MOG , IgG AQP4 or IgG CON or 16.8 μL (20 μg) MOG-IgG 2B7 or AQP4-IgG 53 + 11.2 μL C hu (or normal saline) as described [5]. Rectal temperature was kept 37 -38°C with a heating lamp. After regaining the righting reflex, mice were returned to their cages, kept in 12 hour light/dark cycle and given water and normal chow ad libitum. Mice (5 per group) were killed at 24 hours, seven days or two weeks. Investigators were unaware of which antibody was injected.

Mouse brain histology and immunohistology
Mice were anaesthetized and perfused-fixed by injecting 4% formaldehyde through the left cardiac ventricle. Brains were removed, post-fixed in 4% formaldehyde overnight and processed into paraffin. Coronal tissue sections (7 μm thick) through the injection tract were stained with H + E, Luxol Fast Blue (LFB) [5] or immunostained.
Photomicrographs were taken using an Olympus BX-51 microscope.

Data analysis
Coded photomicrographs were analysed with ImageJ (v1.45S, NIH). Neurofilament immunoreactivity in the injected hemisphere was quantified as mean staining intensity minus background. AnkG and Caspr expression was the number of fluorescent spots/mm 2 in four photomicrographs, 90 μm × 67 μm, taken from the injected hemisphere 0.5 mm from the needle tract. After subtracting background, formatting images to 8-bit, adjusting threshold, the 'analyse particles' function of Image J was used. Spots < 0.01 μm 2 were excluded as noise.

Lesions induced by IgG MOG compared to IgG AQP4
IgG MOG + C hu caused brain edema at 24 hours, but by seven days and two weeks the brain appeared normal ( Figure 1A). Although IgG AQP4 + C hu also caused edema at 24 hours, at seven days there was marked leukocyte infiltration and by two weeks reactive gliosis ( Figure 1A). IgG MOG + C hu caused loss of Luxol Fast Blue (LFB) staining at 24 hours, but this had reversed by two weeks, while the IgG AQP4 + C huinjected tissue showed increased loss of LFB staining at seven days and only partially recovered at two weeks ( Figure 1B). The recruitment of inflammatory cells also differed markedly between the two preparations. IgG MOG + C hu did not produce inflammation while IgG AQP4 + C hu caused inflammation at 24 hours (perivascular neutrophils) and seven days (mostly macrophages) ( Figure 1C).
We immunostained for two astrocyte markers, AQP4 and GFAP. Loss of AQP4 and GFAP was seen in IgG AQP4 + C huinjected brains (at 24 hours and seven days) but IgG MOG + C hu did not reduce AQP4 and GFAP (Figure 2). At two weeks there was marked gliosis (increased AQP4 and GFAP) in brains injected with IgG AQP4 + C hu , compared to little gliosis in brains that received IgG MOG + C hu (Figure 2).

MOG-IgG binds mouse MOG and causes loss of LFB staining
To confirm that IgG MOG binds mouse myelin, it was applied to brain sections. IgG MOG bound the corpus callosum; binding co-localized with a commercial anti-MOG antibody ( Figure 3A). IgG MOG adsorbed by incubation with MOG-expressing human embryonic kidney (MOG-HEK) cells until MOG-IgG became undetectable Figure 1 Brain lesions caused by MOG-IgG and AQP4-IgG. Mice received IgG CON + C hu (purple), IgG MOG + C hu (green) or IgG AQP4 + C hu (blue), were killed at 24 hours (d1), seven days (d7) or two weeks (d14) and coronal brain sections were cut through the injection site. (IgG MOG(AdsMOG-HEK) ) did not bind the corpus callosum, unlike IgG MOG adsorbed against untransfected HEK cells (IgG MOG(AdsHEK) ) ( Figure 3B). To confirm that the MOG-IgG was responsible for the loss of LFB staining, the two adsorbed preparations were injected with C hu and mice were killed at seven days. Loss of LFB staining in the injected hemisphere was only found when IgG MOG(AdsHEK) + C hu was used ( Figure 3C).

MOG-IgG 2B7 causes loss of LFB staining largely independent of immune cells or complement activation
In case the amount of MOG-IgG in the patient preparations was insufficient to cause inflammatory cell infiltration, a large amount (20 μg) of the humanized anti-mouse MOG-IgG 2B7 was co-injected with C hu . At seven days, MOG-IgG 2B7 + C hu caused loss of LFB staining, but without inflammatory cell infiltration ( Figure 4A). At 24 hours after injecting MOG-IgG 2B7 + C hu there was faint C5b-9 immunoreactivity in white matter tracts suggesting slight complement activation, whereas injection of a monoclonal recombinant anti-AQP4 (AQP4-IgG 53 ) + C hu caused strong perivascular C5b-9 immunoreactivity ( Figure 4B). Moreover, intracerebral injection of MOG-IgG 2B7 without C hu produced loss of LFB staining at 24 hours similar to MOG-IgG 2B7 + C hu ( Figure 4C).

MOG-IgG causes reversible damage to myelinated axons
At two weeks there was marked neuronal loss in IgG AQP4 + C hu lesions compared to little neuronal loss in brains injected with IgG MOG + C hu ( Figure 5A). We investigated the effect of IgG MOG + C hu on myelin and axonal proteins including myelin basic protein (MBP), neurofilament, ankyrin G (AnkG) and contactin associated protein (Caspr) ( Figure 5B). MBP adheres adjacent cytoplasmic faces of myelin together, neurofilament provides structural support for axons, AnkG clusters voltage-gated Na + channels at nodes of Ranvier [21] and Caspr attaches paranodal myelin loops to the axons [22]. At 24 hours after IgG MOG + C hu injection, MBP expression appeared abnormal ( Figure 5C) and there was significant reduction in AnkG ( Figure 5D) and Caspr ( Figure 5E) immunoreactivities. At two weeks, the MOG-IgG + C huinduced changes in MBP, AnkG and Caspr had recovered and neurofilament expression was normal ( Figure 5F), indicating intact axons.

Discussion
Although there is growing interest in the potential pathogenicity of MOG antibodies in NMO, the effects of NMO MOG-IgG have not been explored in vivo.
Our results indicate that MOG-IgG directly damages myelin. The detrimental effects of MOG-IgG markedly differ from those of AQP4-IgG and are reversible (see Table 1).
AQP4-IgG lesions are characterized by astrocyte damage followed by leukocyte infiltration that entirely depend on complement activation [5]. We showed that recovery of myelin loss in AQP4-IgG lesions is limited, with gliosis and neuronal death. This finding may explain why clinical recovery after AQP4-IgG NMO attacks is often limited [7][8][9]. By contrast, MOG-IgG, as examined here, damages myelin and axons temporarily, with little complement activation, and no leukocyte infiltration. The myelin and axonal recovery and lack of neuronal death mirror the reported good outcomes of MOG-IgG NMO patients [10,11,15,16].
One study suggested that IgG MOG obtained from children with demyelination does not bind mouse MOG [23], but another study showed that human MOG-IgG binds mouse MOG [24]. Our IgG MOG samples obtained from adult NMO patients, and the anti-mouse MOGspecific monoclonal antibody, both recognized mouse MOG in frozen brain sections, and produced comparable LFB loss without inflammation. This discrepancy may be due to differences in MOG-IgG levels and specificity or differences in MOG glycosylation state, which plays a key role in MOG-IgG binding [24], between children and adults.
The effects of MOG-IgG on cultured oligodendrocytes have already been studied. MOG-IgG binds extracellular epitopes on MOG [23] and can cause crosslinking [25] and internalization [12] of MOG molecules and reversible retraction of oligodendrocyte processes [25]. At high concentration, MOG-IgG causes complement-mediated lysis of MOG-expressing cells [12]. Passive transfer of MOG-IgG antibodies exacerbates CNS damage in experimental autoimmune encephalomyelitis rodent models in which cellular immunity is the predominant pathogenic mechanism [26,27]. Using the intracerebral injection mouse model, we have shown unequivocally that NMO MOG-IgG directly damages myelin in vivo independent of preexisting cellular immunity and complement.
MOG-IgG changed MBP architecture and reduced expression of axonal proteins. Caspr and AnkG are required for the integrity of the nodes of Ranvier and normal action potential firing [21,22]. Mice that lack MBP have a characteristic motor dysfunction including tremor and seizures [28], mice that lack Caspr have severe motor paresis [22] whereas mice lacking cerebellar ankG develop progressive ataxia [21]. Therefore, the altered MBP expression and reduced Caspr and AnkG expression produced by MOG-IgG are predicted to produce a neurological deficit if the NMO lesion is in an eloquent region of the CNS. Unlike AQP4-IgG, MOG-IgG did not produce axonal disintegration or neuronal death. Given the 96% homology between mouse and human MOG [14], our findings raise the possibility that MOG-IgG may also cause similar reversible lesions in the human CNS.
MOG-IgG has been reported in other non-NMO diseases including multiple sclerosis, acute disseminated encephalomyelitis and even some normal subjects [29]. Does MOG-IgG from these non-NMO subjects also cause the same reversible CNS changes, as described here for NMO MOG-IgG? This question is difficult to answer at present because of the variety of assays used to detect MOG-IgG. For example, the assay used here, which employs C-terminal truncated rather than fulllength MOG, did not detect MOG-IgG in adult multiple sclerosis patients and normal individuals [11], which suggests that different assays detect different subpopulations of MOG-IgG. It is important to first standardize the assays before determining which subpopulations of MOG-IgG can cause CNS damage and in which diseases.
The mechanism of MOG-IgG-induced myelin damage in vivo is unknown. Our data show that MOG-IgGmediated myelin damage is a direct effect of MOG-IgG and that complement activation is not necessary. MOG-IgG binding may cause MOG conformational changes or internalization that disrupts the myelin structure and secondarily alters axonal protein expression. To explain the lack of complement involvement, we hypothesize that, after MOG-IgG binding, MOG might not aggregate (because of its low abundance) or MOG might become internalized (thus prohibiting C1q activation). The full recovery within two weeks of the MOG-IgG-induced LFB, MBP, ankG and Caspr changes suggests that MOG-IgG does not kill the oligodendrocytes, but causes a reversible damage.
Our findings raise the possibility that MOG-IgG contributes to pathology in some NMO patients. If MOG-IgG is pathogenic, antibody depletion (plasmapheresis) or suppression with steroids should be effective, as indeed appears to be the case [10,11,15,16]. Conversely, some of the newly proposed therapies for AQP4-IgG NMO, such as sivelestat for inhibiting neutrophils [17], or eculizumab for inhibiting complement [30], are less likely to be needed in MOG-IgG NMO. Examining lesions from MOG-IgG NMO patients may help elucidate the pathogenicity of MOG-IgG in the human CNS. Rectangles show sites of Sections. B. Mouse brain immunostaned for C5b-9 at 24 hours after injection of IgG CON + C hu , IgG MOG + C hu , or IgG AQP4 + C hu . Lu, lumen; wm, white matter. Weak (gray arrows) and strong (black arrows) immunoreactivity. C. Loss of LFB staining at 24 hours after injection of MOG-IgG 2B7 , MOG-IgG 2B7 + C hu , or isotype control (CON-IgG 2B4 ). 5 mice per group. Mean ± SEM. P < 0.05*, < 0.01**. Bar

Conclusions
MOG-IgG obtained from neuromyelitis optica patients causes myelin changes and alters the expression of axonal proteins when injected in mouse brain. These effects are not associated with inflammatory cell infiltration, are largely independent of complement and recover within two weeks. AQP4-IgG obtained from neuromyelitis optica patients causes complement-mediated myelin loss, inflammatory cell infiltration, neuronal and astrocyte death with limited recovery at two weeks. These findings raise the possibility that MOG-IgG contributes to pathology in some neuromyelitis optica patients.

Availability of supporting data
No supporting data.