Differential Binding of Autoantibodies to MOG Isoforms in Inflammatory Demyelinating Diseases

Objective To analyze serum immunoglobulin G (IgG) antibodies to major isoforms of myelin oligodendrocyte glycoprotein (MOG-alpha 1-3 and beta 1-3) in patients with inflammatory demyelinating diseases. Methods Retrospective case-control study using 378 serum samples from patients with multiple sclerosis (MS), patients with non-MS demyelinating disease, and healthy controls with MOG alpha-1-IgG positive (n = 202) or negative serostatus (n = 176). Samples were analyzed for their reactivity to human, mouse, and rat MOG isoforms with and without mutations in the extracellular MOG Ig domain (MOG-ecIgD), soluble MOG-ecIgD, and myelin from multiple species using live cell-based, tissue immunofluorescence assays and ELISA. Results The strongest IgG reactivities were directed against the longest MOG isoforms alpha-1 (the currently used standard test for MOG-IgG) and beta-1, whereas the other isoforms were less frequently recognized. Using principal component analysis, we identified 3 different binding patterns associated with non-MS disease: (1) isolated reactivity to MOG-alpha-1/beta-1 (n = 73), (2) binding to MOG-alpha-1/beta-1 and at least one other alpha, but no beta isoform (n = 64), and (3) reactivity to all 6 MOG isoforms (n = 65). The remaining samples were negative (n = 176) for MOG-IgG. These MOG isoform binding patterns were associated with a non-MS demyelinating disease, but there were no differences in clinical phenotypes or disease course. The 3 MOG isoform patterns had distinct immunologic characteristics such as differential binding to soluble MOG-ecIgD, sensitivity to MOG mutations, and binding to human MOG in ELISA. Conclusions The novel finding of differential MOG isoform binding patterns could inform future studies on the refinement of MOG-IgG assays and the pathophysiologic role of MOG-IgG.


Conclusions
The novel finding of differential MOG isoform binding patterns could inform future studies on the refinement of MOG-IgG assays and the pathophysiologic role of MOG-IgG.
Serum immunoglobulin G (IgG) autoantibodies against myelin oligodendrocyte glycoprotein (MOG-IgG) are associated with a spectrum of neurologic diseases including optic neuritis (ON), acute disseminated encephalomyelitis (ADEM), myelitis, seizures, encephalitis, and with brainstem and/or cerebellar involvement. [1][2][3][4][5][6][7][8][9] In addition, MOG-IgG appears to be supportive to discriminate these disorders from multiple sclerosis (MS) 10,11 as reflected by the first diagnostic recommendations for MOG-IgG-associated disorders (MOGADs). 1,6,12 Furthermore, MOGADs are not only characterized by clinical but also neuropathologic features. 13,14 Although most studies use live cell-based assays (CBAs) with the MOG alpha (α) 1 isoform for the measurement of MOG-IgG, previous results were often discrepant, because of different MOG expression vectors (full-length vs extracellular domain), cell lines, read-out systems (immunofluorescence [IF] vs flow cytometry), and other test variations, which aimed to increase specificity and eliminate nonspecific lowtiter positivity. 1,[15][16][17] Several studies attempted to define the molecular epitopes of MOG-IgG with the help of amino acid substitutions or deletions and discovered distinct binding patterns. [18][19][20][21] The most frequent epitopes were located in the loops between the β sheets of the extracellular Ig domain of human MOG (MOG-ecIgD). These findings were extended by other studies showing that only a subset of human MOG-IgG is also reactive to rodent MOG epitopes 18,19,22,23 and pathogenic in vitro or in vivo. 19,23,24 Earlier studies using MOG-ecIgD as an antigen for immunoassays have indicated a lower sensitivity compared with fulllength MOG (with the α1 isoform as the consensus sequence). 10,25 Although these results indicated binding differences to different MOG variants, no study so far has analyzed antibody responses to different MOG isoforms. Like most other human myelin genes, the MOG gene undergoes extensive alternative splicing and multiple different MOG isoforms have been described in primates, but not in rodents. [26][27][28][29] Although the extracellular Ig domain is present in all these isoforms, they show profound differences in the composition of the intracellular C terminus resulting in alpha (α) or beta (β) isoforms.
The aim of our study was to analyze the serum IgG antibody response to additional MOG isoforms (α2, α3, β1, β2, and β3; figure 1) in MOGα1-seropositive and MOGα1-seronegative patients and controls. Furthermore, we analyzed whether the use of additional MOG isoforms (either alone or in combinations) improves the specificity of MOG-IgG CBAs.

Patients and Controls
We performed a retrospective case-control study including serum samples from 378 patients with inflammatory demyelinating diseases and healthy controls (HCs) with known MOG-IgG serostatus recruited from the participating centers for this study (Innsbruck and Vienna, Austria; Verona, Italy; Barcelona, Spain; Dublin, Ireland; Oxford, UK; Zurich, Switzerland; Munich and Lübeck, Germany; and Lyon, France). 2,10,16,22 We included samples from children (age <18 years) and adults (≥18 years) with an aquaporin-4seronegative non-MS demyelinating disease typically associated with MOG-IgG (n = 214, thereof 191 seropositive MOGAD), MS (n = 64), 11 and HCs (n = 100). We also included MOG-IgG-positive MS (n = 8) and HC (n = 3) samples collected at different centers, and therefore, MOG-IgG seropositivity in MS and HCs is overrepresented compared with previous studies. Samples were sent to Innsbruck and reclassified there as MOG-IgG negative (n = 176; 47%; MOG-IgG(H + L) titer 0 to 1:80; MOG-IgG(Fc) negative) or MOG-IgG positive (n = 202; 53%; MOG-IgG(H + L) titer 1:160 to 1:20,480; MOG-IgG(Fc) positive) using a combination of 2 different live IF MOGα1 CBAs (CBA-IF) as recently described. 16 In addition, 106 follow-up samples of 35 individuals were analyzed with a median observation period of 4.1 years (range 0.8-12.1 years). Because a detailed clinical and radiologic description was not the primary scope of this study, we summarized clinical data of non-MS patients in predominant clinical phenotypes: opticospinal (isolated optic neuritis, myelitis, brainstem syndrome, or a combination of those, a proportion of these patients fulfilled the current diagnostic criteria for neuromyelitis optica spectrum disorder 30 ), cerebral (ADEM according to current diagnostic criteria, 31 Glossary ADEM = acute disseminated encephalomyelitis; CBA = cell-based assay; FACS = fluorescence-activated cell sorting; HCs = healthy controls; IF = immunofluorescence; MOG = myelin oligodendrocyte glycoprotein; MOGAD = MOG-IgG-associated disorders; MOG-ecIgD = extracellular MOG immunoglobulin domain; MOG-IgG = serum IgG antibodies against MOG; MS = multiple sclerosis; NMOSD = neuromyelitis optica spectrum disorder; ON = optic neuritis; PCA = principal component analysis. multiphasic disseminated encephalomyelitis, brainstem encephalitis, or encephalitis), or a combination of both (ADEM followed by optic neuritis and opticospinal with cerebral symptoms) and disease courses (monophasic vs recurrent). Clinical and demographic data according to the MOG-IgG serostatus are shown in table e-1, links.lww.com/NXI/A498. Neuropathologic investigations were available from 6 MOG-IgG-positive cases: 2 with biopsy and autopsy, 3 with biopsy, and 1 with autopsy. Four of these patients had previously been described as case reports and included in a series of patients to describe the pathology of CNS demyelination accompanied by MOG-IgG. 13 16). All patients or their caregivers and controls gave written informed consent. All samples from participating centers were anonymized before sending them to Innsbruck, Austria.

Immunoassays for the Detection of MOG-IgG
Binding to Different MOG Isoforms All live CBAs were performed using HEK293 cells transfected with the individual expression vectors encoding the different MOG isoforms described above. Because of the amount of analyses per sample, evaluation of antibody binding to the different isoforms and animal species was performed on the same day to reduce the number of freeze-thawing cycles. All CBAs (live CBA-IF, live CBA-fluorescence-activated cell sorting [FACS], and a commercial fixed CBA-IF from Euroimmun) and ELISAs were performed according to recently published detailed protocols. 16 We used a combination of 2 assays with secondary antibodies to human IgG(H + L) and confirmation by secondary antibodies to human IgG(Fc) to exclude isolated IgM reactivity. Final results were given in titer level (CBA-IF) or binding ratio (CBA-FACS). The binding of serum samples to human, mouse, and rat myelin was analyzed by a tissue-based IF assay as described before. 19 Competitive Binding Assay Using the Soluble Extracellular Domain of MOG We performed competitive binding assays by CBA-FACS 16 to investigate whether MOG-IgG recognizes soluble human MOG-ecIgD and to determine antibody affinities. Serum samples were diluted according to their FACS median binding ratio to standardize MOG-IgG amounts. Selection of samples was based on the MOG isoform binding pattern observed in live CBA-IF. Diluted serum samples were incubated with increasing amounts (0, 1.6, 3.2, and 9.6 μM) of soluble human MOG-ecIgD (provided by Euroimmun, Lübeck) 16 for 1 hour at room temperature and spun down at 10,000g for 5 minutes. After this preabsorption step, samples were analyzed by CBA-FACS. Nonlinear regression analysis was used to analyze competitive binding and calculate IC50 values.

Statistical Analysis
The primary hypothesis of this study was that there are differential disease-specific IgG antibody responses to MOG isoforms (α1, α2, α3, β1, β2, and β3). This hypothesis was tested for antibody titers using the Friedman nonparametric test, Spearman's rank nonparametric correlational analyses, and receiver operating curve analysis. Principal component analysis (PCA) was used to classify groups of log2-transformed MOG-IgG titers. This unsupervised, unbiased multivariate analysis approach was used to identify the set of variables (MOG isoforms) accounting for the greatest variation present in the data set. Loading plots were generated to visualize the combination of MOG isoforms responsible for clustering. The analyses of all secondary and other endpoints focused on estimates (common OR; positive and negative predictive values [PPV and NPV]; percentages, medians, and median differences) and 95% CI or 25th to 75th interquartile ranges. Groups were compared using Kruskal-Wallis or χ 2 tests. Statistical significance was defined as 2-sided p value < 0.05 (adjusted for multiple comparisons using Bonferroni's correction). Statistical analyses were performed using IBM SPSS software (IBM SPSS Statistics; Version 26.0. Armonk, NY: IBM Corp.) or GraphPad Prism 9 (GraphPad Software, La Jolla, CA).

Data Availability
The data set used and analyzed during the current study is included in the main text or the supplementary files or is available from the corresponding author on reasonable request. Expression plasmids for the different isoforms will be deposited with Addgene.

Results
Differential Antibody Binding to MOG Isoforms α1, α2, α3, β1, β2, and β3 Because previous studies could not consistently clarify which MOG epitopes are immunodominant, we used a different approach and compared the binding of human MOG-IgG with MOG isoforms α1, α2, α3, β1, β2, and β3. We analyzed serum samples from 378 individuals with MOG-IgG serostatus determined using MOGα1 live CBA (176 negative and 202 positive) and single transfections of HEK293 cells with the other MOG isoforms. Cell surface expression of isoforms was demonstrated by performing live CBA with serial dilutions of the humanized mouse monoclonal MOG antibody 8-18-C5 showing comparable detection endpoints (figure 2, A and B). Despite small differences in transfection efficiencies, MOG isoforms showed comparable specific surface binding of humanized MOG antibody 8-18-C5 (figures e-2 and e-3, links.lww.com/NXI/ A498). Therefore, all extracellular domains of the different MOG isoforms were determined equally available for antibody binding.
In a next step, we analyzed whether the individual MOG isoforms are useful for discriminating a non-MS demyelinating disease from MS or HC. As can be seen in figure  2C, low-titer antibodies against all MOG isoforms were also present in a few cases with MS and in HCs.

Identification of Distinct Non-MS-Associated MOG Isoform Binding Patterns
In a next step, we investigated combinations of MOG isoforms using unsupervised PCA to generate a PCA scores plot of all samples and isoforms (figure 3A). The variables driving clustering are clearly differentiated in PC1 (all isoforms with strongest contributions of α1 and β1; 75.3% of variance, Eigenvalue 28.7) and PC2 (α2, α3, β2, and β3; 21.2% of variance, Eigenvalue 8.1) together explaining 96.5% of the total variance in the data set. The biplot shown in figure 3A indicates the following major groups: negative for PC1 (seronegative for MOG-IgG), positive for PC1 but negative for PC2, and positive for both PC1 and PC2. This separation into different binding clusters is shown in figure 3 and table 1. The "negative" cluster (PC1 negative, n = 176) was associated with MOG-IgG-negative (MOGα1 titer 0-1:80) non-MS, MS, and HC samples. By contrast, the other binding patterns were predominantly associated with non-MS and dominated by either PC1 or PC2: (1) equal binding to α1 and β1 and no binding to α2, α3, β2, and β3 (pattern α1β1, n = 73), (2) equal binding to α1 and β1, weaker binding to α2 and α3, and no binding to β2 and β3 (pattern α1-3β1, n = 64), and (3) binding to all MOG isoforms (pattern α1-3β1-3, n = 65). Three representative CBA-IF stainings for samples assigned to these isoform binding patterns are shown in figure 4.
It is important that these binding patterns were associated with a non-MS demyelinating disease consistent with MOGAD when compared with the MOG-IgG-negative patients and controls (  (high-titer MOG-IgG positive) specifically associated with MOGAD 16 was higher in the α1-3β1-3 (74%) than in the α1-3β1 (58%) or α1β1 (59%) binding patterns.

Associations of Different MOG-IgG Binding Patterns With Non-MS Clinical Phenotype, Disease Course, and Neuropathology
Children with a non-MS disease course were more frequently associated with MOG isoform binding patterns α1β1 and α1-3β1, whereas adults were more frequently associated with α1-3β1-3 or seronegative MOG-IgG (table 1). The percentage of females was comparable between binding patterns. Furthermore, the percentage of patients with different clinical presentations of a non-MS demyelinating disease (cerebral, opticospinal, or mixed presentation) or monophasic or recurrent course at the last follow-up was comparable in all 3 MOG isoform binding patterns. Moreover, there was no association of IgG titers or seropositivity against the individual   Finally, we analyzed the neuropathologic features at disease/ relapse onset and follow-up in the 6 cases with available biopsies and/or autopsies (figure e-5, links.lww.com/NXI/ A498). One of these cases (perivenous demyelination) had α1β1 antibodies, 1 had the α1-3β1 binding pattern (perivenous demyelination, conversion to low-titer MOG-IgG at follow-up), and 4 had the α1-3β1-3 binding pattern (1 with perivenous and 3 with confluent demyelination, one of them with intrathecal MOG-IgG).

MOG-IgG Binding Patterns to Different Isoforms Are Associated With a Differential Recognition of MOG Epitopes
We performed a number of additional experiments (reactivity to several mutations in the extracellular Ig domain of MOGα1, competitive binding experiments with soluble MOG-ecIgD, reactivity in CBA-FACS and ELISA, reactivity to fixed MOG expressing cells, and reactivity to human, mouse, and rat MOG in brain tissue) to elucidate the molecular mechanisms underlying the 3 different MOG isoform binding patterns in a subset of samples. Figure 5A shows the differential binding to human MOGα1 mutants, as well as mouse and rat MOG observed in the 3 MOG-IgG binding patterns in relation to MOGα1. The P42S and E64K mutations strongly affected MOG-IgG binding in all 3 binding patterns, with the strongest reduction seen in the α1β1 pattern. The N31Q, A75S, R86Q, and H103A + S104E mutations had comparable effects in all binding patterns. Binding to mouse and rat MOG was reduced in all 3 isoform binding patterns.
Next, we analyzed whether the binding to different MOG isoforms can be inhibited by competition with increasing concentrations (0, 1.6, 3.2, and 9.6 μM) of soluble human MOG-ecIgD in a CBA-FACS competitive binding assay. As shown in figure 5B, addition of soluble MOG-ecIgD was able to reduce the binding to MOGα1 in the α1-3β1-3 pattern comparable with the monoclonal MOG antibody 8-18-C5, but not in pattern α1β1. Furthermore, pattern α1-3β1-3 also had similar IC50 and affinities (0-10 μM, figure  5C). By contrast, antibodies of pattern α1β1 did not seem to bind to the soluble MOG-ecIgD. Of interest, pattern α1-3β1 appears to hold a position between the 2. Because median MOG-IgG titers were comparable (1:640 for all 3 binding patterns), we can exclude an antibody concentration effect. Similar findings were obtained using an MOG-IgG ELISA in which α1-3β1-3 antibodies showed better binding than the other patterns ( figure 5D). By contrast, we observed no differences for reactivity in CBA-FACS, fixed CBA-IF, and antibody binding to human, mouse, or rat myelin.

Discussion
Our results show that the strongest MOG-IgG responses are directed against the longest MOG isoforms α1 and β1, whereas the shorter isoforms α2, α3, β2, and β3 are less  patients with MS and HCs. 16 This observation is also seen in our study with MOG-IgG-positive MS and HC cases found only in the low-titer positive group present in all 3 binding patterns. However, MOG-IgG seropositivity in MS and HCs in our study was overrepresented compared with previous studies. 1 The recognition of different MOG isoforms by MOG-IgG is not associated with relevant differences in clinical presentations, disease course, or neuropathology. Because our study has several limitations such as its retrospective design, the heterogenous observation period, and missing information on treatment, this needs further investigation in a prospective study. However, demographic and clinical characteristics of our study population are consistent with previous reports. 1,2,5,6,9,12,22 Our results indicate that α1β1 and α1-3β1 MOG-IgG could be more common in children and α1-3β1-3 MOG-IgG in adults. This observation is consistent with previous findings showing a dominant expression of the isoforms MOG α1 and β1 in the human brain, and especially in early development, only these 2 isoforms are detected. 28,29 We used a number of additional immunologic investigations to clarify our findings. Only MOG-IgG of the α1-3β1-3, but not the α1β1, binding pattern also bound to soluble MOG-ecIgD with high affinity (20 μM-10 nM), which might explain why immunoprecipitation assays found MOG-IgG less frequently than CBAs 25,34 and why attempts to isolate human MOG-IgG using affinity binding to MOG-ecIgD were only partly successful. 21,23 This result is surprising because all MOG-IgG presumably bind to epitopes in the extracellular Ig domain, and therefore, they should compete similarly with MOG-ecIgG. Additional studies are now needed to elucidate the underlying molecular mechanisms of this finding. Furthermore, the fact that only a subset of human MOG-IgG (those of MOG α1-3β1-3 binding pattern) was reactive in MOG ELISA may explain discrepancies found in the literature. 16,21,25,35 Finally, all 3 MOG isoform binding patterns show different sensitivities to mutations in the extracellular MOG Ig domain: whereas all patterns were sensitive to the P42S mutation associated with an immunodominant MOG epitope 18,21,23,32 and the E64K mutation in the C0D loop, [36][37][38] this effect was even more pronounced in the α1β1 binding pattern. Because the α1-3β1 pattern shares many features with both of the other 2 patterns, it possibly represents a mixture of the other patterns present in polyclonal human serum samples. Further work is now necessary to define the molecular targets of antibodies associated with the identified MOG isoform binding patterns. However, it is a limitation of our study that we were not able to perform these experiments with all samples because of limited availability of samples and reagents.
Our data indicate that these antibody binding patterns are associated with clear structural differences, and therefore, it seems very likely that they represent different antibody clones. The observed MOG isoform binding patterns could be explained by the differential recognition of MOG dimers or multimers caused by clustering of MOG isoforms. 37 This explanation is supported by a recent study demonstrating that the second hydrophobic domain of MOG (only present in α1 and β1) enhanced the recognition of the extracellular part of MOG by human MOG-IgG. 39 Moreover, MOG-IgG from most patients may require bivalent binding to MOG dimers, whereas a smaller subset of MOG-IgG shows monovalent binding to monomers such as the 8-18-C5 monoclonal antibody.
To conclude, our novel finding of differential MOG isoform binding patterns could explain previous discrepant reports and instruct future studies to improve and refine MOG-IgG antibody assays.