classifying the antibody-negative nmo syndromes

ARTICLE OPEN ACCESS

Classifying the antibody-negative NMOsyndromesClinical imaging and metabolomic modeling

Tianrong Yeo MRCP Fay Probert PhD Maciej Jurynczyk MD PhD Megan Sealey PhD Ana Cavey CNS

Timothy DW Claridge DPhil Mark Woodhall PhD Patrick Waters PhD Maria Isabel Leite MD DPhil

Daniel C Anthony PhDdagger and Jacqueline Palace FRCP DMDagger

Neurol Neuroimmunol Neuroinflamm 20196e626 doi101212NXI0000000000000626

Correspondence

Prof Palace

jacquelinepalacendcnoxacuk

or Prof Anthony

danielanthonypharmoxacuk

AbstractObjectiveTo determine whether unsupervised principal component analysis (PCA) of comprehensiveclinico-radiologic data can identify phenotypic subgroups within antibody-negative patientswith overlapping features of multiple sclerosis (MS) and neuromyelitis optica spectrum dis-orders (NMOSDs) and to validate the phenotypic classifications using high-resolution nuclearmagnetic resonance (NMR) plasma metabolomics with inference to underlying pathologies

MethodsForty-one antibody-negative patients were recruited from the Oxford NMO Service Thirty-sixclinico-radiologic parameters focusing on features known to distinguish NMOSD and MSwere collected to build an unbiased PCA model identifying phenotypic subgroups withinantibody-negative patients Metabolomics data from patients with relapsing-remitting MS(RRMS) (n = 34) and antibody-positive NMOSD (Ab-NMOSD) (aquaporin-4 antibody n =54 myelin oligodendrocyte glycoprotein antibody n = 20) were used to identify discriminatoryplasma metabolites separating RRMS and Ab-NMOSD

ResultsPCA of the 36 clinico-radiologic parameters revealed 3 phenotypic subgroups within antibody-negative patients an MS-like subgroup an NMOSD-like subgroup and a low brain lesionsubgroup Supervised multivariate analysis of metabolomics data from patients with RRMS andAb-NMOSD identified myoinositol and formate as the most discriminatory metabolites (bothhigher in RRMS) Within antibody-negative patients myoinositol and formate were signifi-cantly higher in the MS-like vs NMOSD-like subgroup myoinositol (mean [SD] 00023[00002] vs 00019 [00003] arbitrary units [AU] p = 0041) formate (00027 [00006] vs00019 [00006] AU p = 0010) (AU)

ConclusionsPCA identifies 3 phenotypic subgroups within antibody-negative patients and that the me-tabolite discriminators of RRMS and Ab-NMOSD suggest that these groupings have somepathogenic meaning Thus the identified clinico-radiologic discriminators may provide usefuldiagnostic clues when seeing antibody-negative patients in the clinic

These authors contributed equally to the manuscript

daggerCorresponding author metabolomics data and analyses

DaggerCorresponding author clinical data and analyses

From the Department of Pharmacology (TY FP MS DCA) University of Oxford UK Department of Neurology (TY) National Neuroscience Institute Singapore NuffieldDepartment of Clinical Neurosciences (MJ AC MW PW MIL JP) John Radcliffe Hospital University of Oxford UK Department of Chemistry (TDWC) Chemistry ResearchLaboratory University of Oxford UK

Go to NeurologyorgNN for full disclosures Funding information is provided at the end of the article

The Article Processing Charge was funded by Oxfordrsquos RCUK Open Access Block Grant

This is an open access article distributed under the terms of the Creative Commons Attribution License 40 (CC BY) which permits unrestricted use distribution and reproduction in anymedium provided the original work is properly cited

Copyright copy 2019 The Author(s) Published by Wolters Kluwer Health Inc on behalf of the American Academy of Neurology 1

In the multiple sclerosis (MS) or neuromyelitis optica spec-trum disorders (NMOSD) clinic one of the greatest di-agnostic challenges is differentiating antibody-negativepatients with NMOSD from those with opticospinal MS Thisconundrum was demonstrated when large diagnostic dis-agreement was shown even among experts in this field de-spite having the 2015 NMOSD diagnostic criteria in fact thecriteria were not consistently used1

It is clear that the use of discriminatory models on plasmametabolites or conventional MRI can distinguish patientswith relapsing-remitting MS (RRMS) from those withaquaporin-4 antibody (AQP4-Ab) NMOSD and RRMSfrommyelin oligodendrocyte glycoprotein antibody (MOG-Ab) disease remarkably accurately2ndash4 Thus we aim to usethese methods to tackle the diagnostic difficulties inantibody-negative patients who have features overlappingNMOSD and MS The primary methodologic barrier toidentifying discriminators of MS and primary antibody-mediated NMOSD is the lack of a gold standard diagnostic

tool to test accuracy against Therefore there is no publishedstudy to date to resolve this clinical dilemma Given that thetreatment of MS and antibody-mediated NMOSD is mark-edly different and many MS-specific therapies can worsenantibody-mediated NMOSD5ndash12 it is paramount that neu-rologists are able to identify individuals who have antibody-mediated pathology and those with MS pathology withinantibody-negative patients presenting with overlappingclinico-MRI features

In this study we aim to classify a group of difficult-to-diagnose antibody-negative patients into those whose un-derlying pathology are antibody-mediated and those who arelikely to have MS First we assess whether there are spon-taneous clusters of these patients based on their clinical andMRI features using principal component analysis (PCA)Next we explore whether these clusters appear to segregateinto plausible disease-specific groups If these spontaneousclusters appear to identify ldquoMS-likerdquo and ldquoNMOSD-likerdquocohorts we then apply the metabolomics discriminators of

GlossaryAb-NMOSD = antibody-positive NMOSD ANOVA = analysis of variance AQP4-Ab = aquaporin-4 antibody AU = arbitraryunits AUC = area under the curve CPMG = Carr-Purcell-Meiboom-Gill LBL = low brain lesion MOG-Ab = myelinoligodendrocyte glycoprotein antibodyMRS = magnetic resonance spectroscopy NMOSD = neuromyelitis optica spectrumdisordersOPLS-DA = orthogonal partial least square discriminant analysis PCA = principal component analysis ppm = partsper million RRMS = relapsing-remitting MS VIP = variable importance in projection

2 Neurology Neuroimmunology amp Neuroinflammation | Volume 6 Number 6 | November 2019 NeurologyorgNN

MS vs antibody-positive NMOSD (Ab-NMOSD) (obtainedby combining AQP4-Ab and MOG-Ab patients) to furthervalidate that these spontaneous clusters are likely to berepresenting underlying pathologic processes If the metabolicdifferentiators do support the spontaneous clinico-radiologicclusters one could use the most important differ-entiating clinico-MRI features when making diagnostic andtreatment decisions on antibody-negative patients inthe clinic

MethodsStudy participants and clinico-radiologic dataThe study workflow is outlined in figure 1

Antibody-negative cohort for PCAmodel building usingclinico-MRI featuresForty-one antibody-negative patients were recruited from theOxford national NMO service at the John Radcliffe Hospital

Figure 1 Outline of the study workflow

Ab-NMOSD = antibody-positive NMOSD AQP4-Ab = aquaporin-4 antibody AU = arbitrary units LBL = low brain lesion MOG-Ab = myelin oligodendrocyteglycoprotein antibody NMOSD = neuromyelitis optica spectrum disorder PCA = principal component analysis RRMS = relapsing-remittingMS VIP = variableimportance in projection

NeurologyorgNN Neurology Neuroimmunology amp Neuroinflammation | Volume 6 Number 6 | November 2019 3

from November 2013 to September 2015 All patients wereout of relapses and were referred by their primary neurologistsfor possible NMOSD and none had typical MS Serum in allpatients was negative on multiple occasions for both AQP4-Ab and MOG-Ab tested by cell-based assays as previouslydescribed1314

Clinico-radiologic data were obtained frommedical notes andreview of clinical MRIs supplemented by neuroradiologicreports Thirty-six predefined clinico-radiologic parameterswere collected focusing on features that have been describedto distinguish betweenMS andNMOSD (table e-1 linkslwwcomNXIA155)341516 These parameters were scored aspresent if a patient ever had that clinico-MRI feature Thisclinico-radiologic data set was used for unsupervised multi-variate PCA for unbiased pattern recognition to identifyphenotypic subgroups within the antibody-negative patients(see Statistical analyses)

Clinical cohort of patients with RRMS and Ab-NMOSDfor visualization of known diagnostic clusters withinthe PCA modelThe same 36 clinico-MRI parameters were collected from 45patients with established diagnosis (RRMS n = 15 AQP4-Abn = 15 MOG-Ab n = 15) randomly selected from the OxfordMSNMO research database These data were used as a pre-dictive set and inserted into the PCA model that was builtusing the clinico-MRI data from antibody-negative patientsallowing corroboration of phenotypic subgroups (if any) withknown diagnostic clusters

Reference cohort of patients with RRMS and Ab-NMOSD for plasma metabolomics discriminatoryanalysisPlasma metabolomics spectral data from an independentcohort of 108 patients with established diagnosis (RRMS n =34 AQP4-Ab n = 54 MOG-Ab n = 20) was used to builddiscriminatory models to identify metabolites separatingRRMS from Ab-NMOSD (ie AQP4-Ab combined withMOG-Ab patients) (see Statistical analyses)2 Sample col-lection protocols were identical and NMR metabolomicsexperiments were performed at the same time for both thereference cohort and antibody-negative cohort

Standard protocol approvals registrationsand patient consentsThis study was approved by the Oxford Research EthicsCommittee C (Ref 10H060656 and 16SC0224A) Allpatients gave their written consent to participate in the study

Plasma collection and NMR samplepreparation for metabolomics analysisBlood was collected into lithium-heparin tubes (BectonDickinson 367375) and left to stand at room temperature for30 minutes before centrifugation at 2200g for 10 minutesPlasma was immediately aliquoted and stored at minus80degC ForNMR experiments plasma was thawed at room temperaturefollowed by centrifugation at 100000g for 30 minutes at 4degC

One hundred fifty microliters of the plasma supernatant wasthen diluted with 450 μL of 75 mM sodium phosphate bufferprepared in D2O (pH 74) followed by centrifugation at16000g for 30 minutes before transferring to a 5-mm NMRtube

NMR spectroscopy and data processing formetabolomics analysisAll NMR experiments were performed using a 700-MHzBruker AVIII spectrometer Technical specifications of theNMR experiments and data processing have been previouslypublished2 Briefly 1D 1H NMR spectra were obtained usinga Carr-Purcell-Meiboom-Gill (CPMG) relaxation editingpulse sequence which retains resonances from small-molecular-weight metabolites and mobile side chains of lip-oproteins The CPMG spectra were preprocessed in Topspin21 (Bruker Germany) followed by visual inspection forerrors in baseline correction referencing spectral distortionor contamination Processed spectra were exported to ACDLabs Spectrus Processor Academic Edition 1201 (AdvancedChemistry Development Inc Toronto Canada) wherebyregions of the spectra between 080ndash420 parts per million(ppm) and 520ndash850 ppmwere split into 002-ppm-wide binsIntegral values of the spectral bins were computed and used asquantitative variables expressed in arbitrary units (AU) Me-tabolite assignment was performed by referencing to literaturevalues and the Human Metabolome Database17ndash21 Furtherconfirmation was achieved by inspection of the 2D spectra(presaturation correlation spectroscopy) spiking of knowncompounds and 1D total correlation spectroscopy spectra

Statistical analysesTo identify potential subgroups within the antibody-negativecohort using clinico-imaging data PCA was used SIMCAsoftware (MKS Data Analytics Solutions Umetrics Sweden)was used for PCA PCA is an unsupervised unbiased(ie without defining disease groups) multivariate analysisapproach to identify a set of variables (in this case clinico-MRI parameters) accounting for the greatest variation presentin the data set22 As the analysis is unsupervised clustering (ifany) is in no way influenced by the user but rather is whollydependent on the clinico-MRI data alone Furthermore thePCA approach allows the inclusion of correlated variableswhich reflects the actual real-life clinico-MRI (often corre-lated) data gathered by a neurologist when seeing a patientThis approach was used to analyze the 36 predefined clinico-radiologic parameters (binary data) to evaluate the degree ofclustering between the 41 antibody-negative patients basedon clinico-MRI features enabling clusters (if any) to beidentified Loading plots were generated to visualize theclinico-radiologic parameters responsible for clustering

To identify metabolic differences between RRMS and Ab-NMOSD using metabolomics spectral data orthogonal partialleast square discriminant analysis (OPLS-DA) statisticalmethods were used2 R software (R foundation for statisticalcomputing Vienna Austria) was used for OPLS-DA using


in-house R scripts and the ropls package23 OPLS-DA is anextension of PCA allowing supervised multivariate analysis toexplore variables (in this case metabolites) accounting for classdiscrimination between user-defined classes22 This approachwas used to investigate metabolic differences of patients withRRMS vs Ab-NMOSD (ie AQP4-Ab combined with MOG-Ab) from the reference cohort and to identify the key metab-olites driving the separation between them In brief after cor-rection for unequal class sizes the metabolomics data were splitinto a training set (90 of data) and a test set (10 of data)The training set was used to build the model on which the testset was applied to to determine the predictive accuracy of themodel Ten-fold cross-validation with 100 iterations was per-formed creating an ensemble of 1000 model accuracies Tovalidate the metabolic separation between the disease groupsthe mean accuracy of the ensemble of model accuracies wascompared with the mean accuracy of a separate ensemblecreated by random class assignments

Analysis of other clinicoimaging and metabolomics data wasperformed with STATA software (Release 14 StataCorp LPCollege Station TX) and R software Chi-square tests orFisher exact tests were used for categorical variables as ap-propriate whereas 2-sample t testone-way analysis of vari-ance (ANOVA) with Tukey Honestly Significant Difference(HSD) post hoc correction or Mann-Whitney UKruskal-Wallis tests were used for continuous variables as appropriateTwo-tailed p values of lt005 were considered statisticallysignificant

Data availabilityAnonymized data can be shared by request from any qualifiedinvestigator

ResultsPCA of clinico-radiologic data within theantibody-negative cohort identifies 3 distinctpatient subgroupsTo identify potential phenotypic subgroups within antibody-negative patients we performed unsupervised PCA of the 36specified clinico-radiologic parameters and generated a PCAscores plot (figure 2A) Each point in the plot represents all36 clinico-radiologic parameters from 1 patient pointscloser to one another are more clinically alike Spontaneousseparation of the antibody-negative cohort into 3 patientclusters (dashed blue circles) was observed on the PCA plot(figure 2A) This observation suggested a distinct clinicalprofile for each cluster and we sought to explore the reasonfor clustering

The variable loadings plot of the PCA was constructed toidentify the variables driving the clustering (figure 2B) Thevariables driving the top cluster are features characteristic ofMS324 whereas the ones defining the bottom right cluster aremore typical of NMOSD151625 The bottom left cluster is

characterized by no or low brain lesion load This allowed usto classify these 3 phenotypic clusters into an MS-like sub-group an NMOSD-like subgroup and a low brain lesion(LBL) subgroup (figure 2A) with the most principal variableslisted in the inset

To corroborate these phenotypic assignments with patientswith established diagnosis the 36 clinico-radiologic parame-ters were collected from patients in the clinical cohort ofknown RRMS and Ab-NMOSD Insertion of this data setconfirmed that most of the patients with RRMS clustered withthe MS-like subgroup whereas the majority of the patientswith AQP4-Ab NMOSD and MOG-Ab disease clustered tothe NMOSD-like subgroup (figure 2C) It is interesting tonote the clustering of patients with AQP4-Ab and MOG-Aband this is consistent with previous studies that have shownthat AQP4-Ab NMOSD and MOG-Ab disease in adults havelargely identical clinical presentations and cannot be distin-guished on conventional MRI426 Of note some patients withRRMS AQP4-Ab NMOSD and MOG-Ab disease clusteredwith the LBL subgroup highlighting that these diseases haveoverlapping clinico-radiologic features

Taking these observations in totality PCA of clinico-radiologicdata within the antibody-negative cohort identified 3 pheno-typically distinct subgroups an MS-like subgroup (n = 6) anNMOSD-like subgroup (n = 14) and an LBL subgroup (n =21) Table 1 shows the demographic and clinical data of theantibody-negative patients grouped by the 3 PCA-definedsubgroups and the proportions of patients having each of the36 clinico-radiologic parameters

Plasma myoinositol and formate discriminatebetween RRMS and Ab-NMOSD with highaccuracy within the reference cohortAlthough unbiased PCA of extensive clinico-radiologic data isable to identify distinct phenotypes within the antibody-negative cohort pathophysiologic relevance at a molecularlevel with respect to the reference diseases (ie MS pathologyvs antibody-mediated pathology) is lacking Thus to in-vestigate whether plasma metabolomics can identify meta-bolic biomarkers separating the antibody-negative phenotypicsubgroups with inference to their underlying pathologies weobtained discriminatory metabolic markers in the referencecohort of patients with known RRMS and Ab-NMOSD FirstOPLS-DA was used to build discriminatory models usingmetabolomics spectral data to distinguish between RRMS andAb-NMOSD within the reference cohort A representativeOPLS-DA scores plot was generated (figure 3A) Each pointin the plot represents all metabolomics data from 1 patientpoints closer to one another are more metabolically similar Aclear separation between RRMS and Ab-NMOSD was ob-served on the scores plot This separation was validated as themean accuracy (of the ensemble of accuracies) of the diseasegroups model was significantly greater than the mean accu-racy of the random class assignment model (mean [SD]807 [42] vs 523 [76] p lt 0001) (figure 3B) No


Figure 2 Identification of phenotypic subgroups within the antibody-negative cohort by PCA using clinico-radiologic data

(A) Spontaneous separation of antibody-negative patients into 3 distinct clusters using the 36 predefined clinico-radiologic parameters alone (dashed bluecircles) (B) Variable loadings plot of the clinico-radiologic parameters allows visualization of parameters responsible for patient clustering Each parameter isrepresented by a gray diamond The number beside eachdiamond corresponds to the number listed in table e-1 (linkslwwcomNXIA155) This enables the 3phenotypic clusters to be classified as an MS-like subgroup an NMOSD-like subgroup and an LBL subgroup (panel A inset) (C) Insertion of clinico-radiologicdata from the clinical cohort of patients with RRMS AQP4-Ab NMOSD and MOG-Ab disease into the PCA scores plot shows corroboration of the phenotypicsubgroups with known diagnostic clusters AQP4-Ab = aquaporin-4 antibody EDSS = Expanded Disability Status Scale IPND = International Panel for NMODiagnosis LBL = low brain lesion MOG-Ab = myelin oligodendrocyte glycoprotein antibody NMOSD = neuromyelitis optica spectrum disorders PCA =principal component analysis RRMS = relapsing-remitting MS


Table 1 Demographic and clinico-radiologic data within the antibody-negative cohort grouped according to the 3 PCA-defined subgroups

MS-like (n = 6) NMOSD-like (n = 14) LBL (n = 21)

Age at sampling median (range) y 542 (375ndash715) 386 (244ndash704) 457 (230ndash597)

Female no () 3 (500) 8 (571) 13 (619)

Duration of disease (disease onset to sampling) median (range) y 54 (13ndash174) 34 (00ndash175) 52 (02ndash206)

Annualized relapse rate median (range)a 02 (01ndash07) 07 (02ndash17) 03 (01ndash11)

Interval between last attack to sampling median (range) y 54 (10ndash174) 18 (02ndash138) 32 (02ndash152)

Interval between disease onset to latest MRI brain median (range) y 53 (05ndash174) 30 (03ndash177) 45 (0003ndash140)

Interval between disease onset to latest MRI spine median (range) y 25 (03ndash174) 30 (06ndash177) 45 (02ndash173)

On immunosuppressant no () 0 (00) 8 (571) 6 (286)

Azathioprine mdash 5 (357) 3 (143)

Mycophenolate mofetil mdash 2 (143) 2 (95)

Methotrexate mdash 1 (71) 1 (48)

On prednisolone no () 1 (167) 7 (500) 5 (238)

On MS disease-modifying therapy no () 0 (00) 0 (00) 1 (48)b

The 36 clinico-radiologic variables used for PCA multivariate analysis

Any transverse myelitis no () 4 (667) 14 (1000) 16 (762)

LETM no () 1 (167) 12 (857) 5 (238)

T1 hypointensity with corresponding T2 hyperintensity in acute stage of cordlesion no ()

0 (00) 5 (357) 1 (48)

Cord lesion spanning cervical medullary junction no () 0 (00) 1 (71) 1 (48)

Predominant central cord involvement no () 2 (333) 13 (929) 4 (190)

Conus involvement no () 2 (333) 4 (286) 1 (48)

EDSS score ge6 at nadir of any attack no () 1 (167) 12 (857) 2 (95)

Any optic neuritis no () 2 (333) 11 (786) 9 (429)

Severe optic neuritis no () 0 (00) 6 (429) 6 (286)

Simultaneous bilateral optic neuritis no () 0 (00) 5 (357) 2 (95)

Simultaneous optic neuritis and transverse myelitis no () 0 (00) 5 (357) 0 (00)

Long segment optic neuritis no () 0 (00) 0 (00) 1 (48)

Optic chiasm involvement no () 0 (00) 0 (00) 0 (00)

Area postrema syndrome no () 0 (00) 2 (143) 0 (00)

No brain lesion no () 0 (00) 0 (00) 7 (333)

1ndash3 brain lesions no () 0 (00) 6 (429) 12 (571)

ge4 brain lesions no () 6 (1000) 8 (571) 2 (95)

Dawson fingers no () 6 (1000) 2 (143) 0 (00)

Lesion touching body of the lateral ventricle no () 6 (1000) 3 (214) 0 (00)

Inferior temporal lesion no () 2 (333) 1 (71) 0 (00)

Corpus callosum lesion no () 1 (167) 6 (429) 3 (143)

Diffuse splenial lesion no () 0 (00) 2 (143) 0 (00)

Fluffy infratentorial lesion no () 0 (00) 3 (214) 0 (00)

Continued


potential confounders were identified within this data set afterextensive investigation as reported previously2

Next to identify the most important metabolites driving theseparation between RRMS and Ab-NMOSD variable impor-tance in projection (VIP) scores were generated A VIP score isa measure of a variablersquos importance to the OPLS-DA modelthe higher the VIP score the greater the contribution a variablemakes to the model Ranking of VIP scores revealed thatmyoinositol and formate (both metabolites being higher inRRMS) were the 2 most important metabolites driving thisseparation (figure 3C) with a VIP score of 257 and 251 re-spectively Receiver operating characteristic analysis revealedhigh diagnostic accuracies as measured by the area under thecurve (AUC) ofmyoinositol (AUC0914 95CI 0862ndash0967)and formate (AUC 0907 95 CI 0849ndash0965) (figure 3D)

Myoinositol and formate levels aresignificantly higher in the MS-like subgroupcompared with the NMOSD-like subgroupwithin the antibody-negative cohortAs myoinositol and formate could accurately discriminatebetween RRMS and Ab-NMOSD we explored whether thesemetabolites are different between the MS-like and NMOSD-like clinico-radiologic subgroups within the antibody-negativecohort Myoinositol was significantly higher in the MS-likesubgroup compared with the NMOSD-like subgroup (mean[SD] 00023 [00002] vs 00019 [00003] AU p = 0041)

(figure 4A) Formate was also significantly elevated in theMS-like subgroup vs the NMOSD-like subgroup (00027 [00006]vs 00019 [00006] AU p = 0010) On one-way ANOVAformate was significantly different across the 3 subgroups[F(238) = 502 p = 0012] post hoc comparisons using theTukey HSD test showed formate to be higher in the MS-likesubgroup compared with the NMOSD-like subgroup (p =0013) as indeed compared with the LBL subgroup (00027[00006] vs 00020 [00005] AU p = 0017) (figure 4B)Taking successive discriminatory metabolites with cutoff VIPscores ge175 (before the second drop-off in VIP scores seefigure 3C) showed similar trends in separating the MS-likefrom NMOSD-like subgroups (figure 5) Next we exploredwhether the MS-like and NMOSD-like patients were meta-bolically similar to patients with RRMS and Ab-NMOSDrespectively Using metabolomics spectral data we were un-able to distinguish MS-like patients from patients with RRMSand NMOSD-like patients from patients with Ab-NMOSD(figure e-1 linkslwwcomNXIA154)

In summary the 2 most discriminatory metabolites obtainedfrom the OPLS-DAmodel of RRMS vs Ab-NMOSD were alsosignificantly different between the MS-like and NMOSD-likesubgroups (and in the same direction) within antibody-negative patients This suggests that theMS-like and NMOSD-like subgroups have different underlying pathologies akin totheir respective reference diseases (ie RRMS and antibody-mediated NMOSD)

Table 1 Demographic and clinico-radiologic data within the antibody-negative cohort grouped according to the 3 PCA-defined subgroups (continued)


Lesion adjacent to the 4th ventricle no () 1 (167) 5 (357) 0 (00)

Lesion adjacent to the 3rd ventricle no () 0 (00) 2 (143) 0 (00)

Periaqueductal lesion no () 0 (00) 2 (143) 0 (00)

Area postrema lesion no () 0 (00) 2 (143) 0 (00)

Hypothalamicthalamic lesion no () 0 (00) 1 (71) 0 (00)

Tumefactive lesion no () 0 (00) 3 (214) 0 (00)

Corticaljuxtacortical lesion no () 1 (167) 6 (429) 2 (95)

Juxtacortical S- or U-shaped lesion no () 0 (00) 2 (143) 0 (00)

Fulfill 2016 MAGNIMS dissemination in space criteria no () 4 (667) 10 (714) 5 (238)

Fulfill 2015 IPND seronegative NMOSD criteria no () 0 (00) 12 (857) 0 (00)

Disability progression independent of relapses no () 3 (500) 1 (71) 4 (190)

Unmatched CSF oligoclonal bands no () 4 (667) 713 (538) 918 (500)

Coexisting autoimmunity andor autoantibodies no () 2 (333) 3 (214) 5 (238)

Abbreviations EDSS = ExpandedDisability Status Scale IPND = International Panel for NMODiagnosis LBL = low brain lesion LETM = longitudinally extensivetransverse myelitis MAGNIMS = Magnetic resonance Imaging in Multiple Sclerosis NMOSD = neuromyelitis optica spectrum disorders PCA = principalcomponent analysisa Calculated with the onset attack included and restricted to patients with at least 1-year interval between the onset attack and samplingb Glatiramer acetate


Differences in myoinositol and formate levelsare not accounted for bypotential confoundersAs a higher proportion of patients in the NMOSD-like andLBL subgroups were on immunosuppressants and predniso-lone at the time of plasma sampling compared with the MS-like subgroup (table 1) it was explored whether theseaccounted for the differences in myoinositol and formatelevels By combining the NMOSD-like and LBL subgroupsmyoinositol and formate levels of patients on immunosup-pressants were compared with patients not on immunosup-pressants Similar analysis was performed for prednisoloneuse There were no statistically significant differences in bothmetabolites stratified by immunosuppressant or prednisoloneuse myoinositol by immunosuppressant use (on immuno-suppressant 00020 [00002] vs off immunosuppressant00021 [00004] AU p = 0384) myoinositol by prednisoloneuse (on prednisolone 00020 [00004] vs off prednisolone00021 [00003] AU p = 0224) formate by immunosup-pressant use (on immunosuppressant 00019 [00005] vs offimmunosuppressant 00020 [00005] AU p = 0714) andformate by prednisolone use (on prednisolone 00017[00005] vs off prednisolone 00020 [00005] AU p = 0111)

In fact within the NMOSD-like subgroup alone patients onimmunosuppressants had higher levels of myoinositol (onimmunosuppressant 00020 [00002] vs off immunosup-pressant 00018 [00005] AU p = 0370) and formate (onimmunosuppressant 00021 [00006] vs off immunosup-pressant 00016 [00002] AU p = 0143) and this would ifanything reduced the discriminatory power of the metabo-lites Similar analyses were performed for age sex diseaseduration and interval since last attack with no significantdifferencescorrelations in the levels of both metabolitesbased on these parameters (data not shown)

DiscussionOur findings confirmed that distinct phenotypic subgroupsexist within the antibody-negative cohort using advancedPCA pattern-recognition techniques coupled with extensiveclinico-radiologic data without a priori assumptions of theirclinical diagnosis We then applied the 2 metabolites that werethe most discriminatory between RRMS and Ab-NMOSDand confirmed that these same metabolites distinguishedbetween the antibody-negative subgroups that were MS-like

Figure 3 OPLS-DA score plot of metabolomics spectral data comparing RRMS with Ab-NMOSD from the reference cohort

(A) OPLS-DA scores plot shows good separation of patients with RRMS from patients with Ab-NMOSD based on metabolomics spectral data (B) Meanaccuracy of the disease groups model is significantly greater than that of the random class assignment model (mean [SD] 807 [42] vs 523 [76] p lt0001) (C) The top 2 discriminatory metabolites myoinositol and formate are identified by their high VIP scores (D) High AUC of both myoinositol andformate in distinguishing RRMS and Ab-NMOSD Ab-NMOSD = antibody-positive neuromyelitis optica spectrumdisorders AUC = area under the curve OPLS-DA = orthogonal partial least square discriminant analysis RRMS = relapsing-remitting MS VIP = variable importance in projection


and NMOSD-like This suggests that the clinico-radiologicseparation by PCA is pathophysiologically meaningful and wesuggest that in clinical practice the features shown in figure2A (inset) are pathologically relevant for classification Thishas the potential to help guide treatment decisions whenseeing antibody-negative patients in the clinic

Myoinositol is a component of the cell membrane and myelinand is involved in intracellular signaling in many CNS cells27

More importantly it has been recognized as a marker of as-trocyte activation and proliferation28 Low myoinositol levelshave been observed in AQP4-Ab NMOSD compared withMS after transverse myelitis using 1H magnetic resonancespectroscopy (MRS) of the spinal cord reflecting astrocyticnecrosis29 Conversely high myoinositol levels have beennoted in RRMS and clinically isolated syndrome comparedwith controls using 1H MRS of normal-appearing whitematter indicating astrocytosis and astrogliosis3031 UnlikeAQP4-Ab NMOSD MOG-Ab disease is not an astrocytop-athy and glial fibrillary acidic protein is not elevated in theCSF32 Although accurate quantification of astrocytes has notbeen performed in MOG-Ab disease in view of the smallnumber of cases with histopathology it is likely that extent ofgliosis as seen in MS (resulting from ongoing chronic neu-roinflammation) does not occur in MOG-Ab disease33 andthis may explain the reduced levels of myoinositol with re-spect to MS This needs further pathologic verification Ourfindings of higher myoinositol levels in RRMS and MS-likepatients compared with Ab-NMOSD and NMOSD-likepatients are in agreement with these observations Formatecauses mitochondrial damage by inhibiting cytochrome coxidase resulting in disruption of the electron transport chainand production of reactive oxygen species34 Formate-induced cytotoxicity has been demonstrated in rat

hippocampal cultures and in retinal (human and rat) cellcultures3536 Of interest methanol poisoning is mediated byformate producing optic nerve demyelination and sub-sequent progressive retinal axonal loss in humans3738 Asmitochondrial dysfunction has been implicated in MS path-ogenesis it is of interest to note the higher formate levels inpatients with MS39 How formate is involved in this process ifat all as a primary mediator or as part of an injurious cascadewill require further mechanistic studies

In view of the lack of accuracy of the McDonald criteria toseparate MS from NMOSD40ndash42 we have previouslyattempted to better delineate MS from Ab-NMOSD usingconventional MRI parameters34 Distinctive MRI brain fea-tures of MS include Dawson fingers inferior temporal lobelesion and lesion adjacent to the body of the lateralventricle34 which are also the variables driving the MS-likesubgroup in this current studyWe have previously shown thatblood-based metabolomics can accurately separate MS fromcontrols and from AQP4-Ab NMOSD and MOG-Abdisease243 The current study combines both approaches byusing metabolomics to give pathologic support to the spon-taneously separating clinico-radiologic phenotypes Of notethe clinico-MRI phenotypic classification identified the 2015seronegative NMOSD criteria as the most important dis-tinguishing NMOSD-like variable independently supportingthese criteria

Our study is limited by the small sample size due to the rarityof antibody-negative patients however we were still able toshow a remarkable similar pattern of discriminatory metab-olites in the MS-like against the NMOSD-like subgroups asseen in patients with RRMS against patients with Ab-NMOSD Our methodology is optimized to compare

Figure 4 Boxplots comparing myoinositol and formate levels between MS-like and NMOSD-like subgroups within theantibody-negative cohort

Both (A) myoinositol and (B) formate are significantly higher in the MS-like subgroup compared with the NMOSD-like subgroup On one-way ANOVA (B)formate was significantly different across the 3 subgroups and post hoc comparisons using the Tukey HSD test showed formate to be significantly higher intheMS-like subgroup comparedwith the NMOSD-like subgroup as well as to the LBL subgroup p values shown in (B) are fromone-way ANOVAwith post hocmultiple comparison corrections Boxplots of myoinositol and formate in patients with RRMS and Ab-NMOSD are constructed from the same data used togenerate the AUC graphs in figure 3D Ab-NMOSD= antibody-positiveNMOSD ANOVA = analysis of variance AU = arbitrary units AUC = area under the curveLBL = low brain lesion NMOSD = neuromyelitis optica spectrum disorders ppm = parts per million RRMS = relapsing-remitting MS


2 subsets and in the antibody-negative group there will bemultiple disorders hence we focused on the 2 phenotypicsubgroups which appeared to represent MS-like and anti-body-mediatedndashlike pathology The third phenotypic sub-group in our analysis contained patients with lower brainlesion load without any MS-like or NMOSD-like discrim-inators and pathologies among this subgroup will includeantibody-mediated pathologies MS other cell-mediated dis-orders such as CNS sarcoidosis and monophasic post-infectious conditions In view of the mixed conditions withinthe LBL subgroup we have kept it separate for analysisClinicopathologic classification within this LBL subgroup willbe particularly challenging However in patients with 1ndash3

brain lesions who have MS-like or NMOSD-like discrim-inators these clinico-radiologic discriminators are still po-tentially useful as illustrated by 43 of NMOSD-like patientshaving 1ndash3 brain lesions Future validation of our findings isneeded in an independent cohort of antibody-negativepatients

Our study demonstrates the strength of computationalmodeling of clinico-MRI features which cannot be done ina consistent and unbiased way by clinicians in the clinicalsetting given the huge amount of data available for each pa-tient We also demonstrate the use of metabolomics in sup-porting the results of such analysis We have selected

Figure 5 Boxplots of other discriminatory metabolites (VIP score ge175)

Other discriminatorymetabolites trend in the samedirectionwhen comparing theMS-likewithNMOSD-like subgroups aswith RRMS to Ab-NMOSD (A-I) Thistrend becomes less clear with lower VIP scores as shown by the last 3 metabolite bins in the panel (J) citrate (268ndash270 ppm VIP score 187) (K) mobilendashN(CH3)3free choline (320ndash322 ppm VIP score 185) and (L) argininelysineleucine (168ndash170 ppm VIP score 175) Ab-NMOSD = antibody-positive NMOSDAU = arbitrary units LBL = low brain lesion NMOSD = neuromyelitis optica spectrumdisorders ppm= parts permillion RRMS = relapsing-remittingMS VIP =variable importance in projection


a diagnostically challenging group of patients and have beenable to identify useful clinical and radiologic characteristicsthat support some individuals having likely MS and otherswith likely antibody-mediated pathology As the MRIparameters are not time restricted these observations aremore useful to apply in clinical practice Prospective work tostudy treatment responses and long-term outcome along withCSF metabolomics analysis and samples taken during relap-ses may further improve this classification especially inpatients within the LBL subgroup

Study fundingT Yeo is supported by the Ministry of Health Singaporethrough the National Medical Research Council ResearchTraining Fellowship (NMRCFellowship00382016)F Probert is supported by the MS Society M Jurynczyk issupported by the Medical Research Council Confidence inConcept Fund and received a research fellowship from thePolish Ministry of Science and Higher Education programmeMobilnosc Plus (1070MOBB20130)

DisclosureT Yeo F Probert M Jurynczyk M Sealey A Cavey TDWClaridge M Woodhall and DC Anthony report no dis-closures relevant to the manuscript P Waters and the Uni-versity of Oxford hold patents and receive royalties andrevenue for performing antibody assays in neurologic dis-eases MI Leite reported being involved in aquaporin-4testing receiving support from the National Health ServiceNational Specialised Commissioning Group for Neuro-myelitis Optica and the National Institute for Health ResearchOxford Biomedical Research Centre receiving speakinghonoraria from Biogen Idec and receiving travel grants fromNovartis J Palace is partly funded by highly specializedservices to run a national congenital myasthenia service anda neuromyelitis service She has received support for scientificmeetings and honorariums for advisory work from MerckSerono Biogen Idec Novartis Teva Chugai Pharma andBayer Schering Alexion Roche Genzyme MedImmuneEuroImmun MedDay Abide and ARGENX and grants Fulldisclosure form information provided by the authors isavailable with the full text of this article at NeurologyorgNN

Publication historyReceived by Neurology Neuroimmunology amp NeuroinflammationJune 28 2019 Accepted in final form August 13 2019

References1 Jurynczyk M Weinshenker B Akman-Demir G et al Status of diagnostic approaches

to AQP4-IgG seronegative NMO and NMOMS overlap syndromes J Neurol 2016263140ndash149

2 Jurynczyk M Probert F Yeo T et al Metabolomics reveals distinct antibody-independent molecular signatures of MS AQP4-antibody and MOG-antibody dis-ease Acta Neuropathol Commun 2017595

3 Matthews L Marasco R Jenkinson M et al Distinction of seropositive NMO spec-trum disorder and MS brain lesion distribution Neurology 2013801330ndash1337

4 Jurynczyk M Geraldes R Probert F et al Distinct brain imaging characteristics ofautoantibody-mediated CNS conditions and multiple sclerosis Brain 2017140617ndash627

Appendix Authors

Name Location Role Contribution

TianrongYeo MRCP

Department ofPharmacologyUniversity of OxfordUK

Author Designed andconceptualized thestudy analyzed thedata major role in theacquisition of dataand drafted themanuscript forintellectual content

Appendix (continued)


FayProbertPhD


Author Designed andconceptualized thestudy analyzed thedata major role in theacquisition of dataand revised themanuscript forintellectual content

MaciejJurynczykMD PhD

Nuffield Departmentof ClinicalNeurosciencesUniversity of OxfordUK

Author Major role in theacquisition of data

MeganSealeyPhD



Ana CaveyCNS



TimothyDWClaridgeDPhil

Department ofChemistry Universityof Oxford UK

Author Major role in theacquisition of data andrevised themanuscriptfor intellectual content

MarkWoodhallPhD



PatrickWatersPhD



MariaIsabelLeite MDDPhil



Daniel CAnthonyPhD


Author Designed andconceptualized studyinterpreted the dataand revised themanuscript forintellectual content

JacquelinePalaceFRCP DM


Author Designed andconceptualized thestudy interpreted thedata and revised themanuscript forintellectual content


5 Palace J Leite MI Nairne A Vincent A Interferon beta treatment in neuromyelitisoptica increase in relapses and aquaporin 4 antibody titers Arch Neurol 2010671016ndash1017

6 Kleiter I Hellwig K Berthele A et al Failure of natalizumab to prevent relapses inneuromyelitis optica Arch Neurol 201269239ndash245

7 Stellmann JP KrumbholzM Friede T et al Immunotherapies in neuromyelitis opticaspectrum disorder efficacy and predictors of response J Neurol Neurosurg Psychiatry201788639ndash647

8 Min JH Kim BJ Lee KH Development of extensive brain lesions following fingoli-mod (FTY720) treatment in a patient with neuromyelitis optica spectrum disorderMult Scler 201218113ndash115

9 Shimizu J Hatanaka Y Hasegawa M et al IFNbeta-1b may severely exacerbateJapanese optic-spinal MS in neuromyelitis optica spectrum Neurology 2010751423ndash1427

10 Azzopardi L Cox AL McCarthy CL Jones JL Coles AJ Alemtuzumab use in neu-romyelitis optica spectrum disorders a brief case series J Neurol 201626325ndash29

11 Wildemann B Jarius S Schwarz A et al Failure of alemtuzumab therapy to controlMOG encephalomyelitis Neurology 201789207ndash209

12 Yamout BI Beaini S Zeineddine MM Akkawi N Catastrophic relapses followinginitiation of dimethyl fumarate in two patients with neuromyelitis optica spectrumdisorder Mult Scler 2017231297ndash1300

13 Waters P Woodhall M OrsquoConnor KC et al MOG cell-based assay detects non-MSpatients with inflammatory neurologic disease Neurol Neuroimmunol Neuro-inflamm 20152e89 doi 101212NXI0000000000000089

14 Waters PJ McKeon A Leite MI et al Serologic diagnosis of NMO a multicentercomparison of aquaporin-4-IgG assays Neurology 201278665ndash671

15 Jurynczyk M Craner M Palace J Overlapping CNS inflammatory diseases differ-entiating features of NMO and MS J Neurol Neurosurg Psychiatry 20158620ndash25

16 Kim HJ Paul F Lana-Peixoto MA et al MRI characteristics of neuromyelitis opticaspectrum disorder an international update Neurology 2015841165ndash1173

17 Lenz EM Bright J Wilson ID Morgan SR Nash AF A 1HNMR-basedmetabonomicstudy of urine and plasma samples obtained from healthy human subjects J PharmBiomed Anal 2003331103ndash1115

18 Tang H Wang Y Nicholson JK Lindon JC Use of relaxation-edited one-dimensionaland two dimensional nuclear magnetic resonance spectroscopy to improve detectionof small metabolites in blood plasma Anal Biochem 2004325260ndash272

19 Wishart DS Jewison T Guo AC et al HMDB 30mdashthe humanmetabolome databasein 2013 Nucleic Acids Res 201341D801ndashD807

20 Wishart DS Knox C Guo AC et al HMDB a knowledgebase for the humanmetabolome Nucleic Acids Res 200937D603ndashD610

21 Wishart DS Tzur D Knox C et al HMDB the humanmetabolome database NucleicAcids Res 200735D521ndashD526

22 Worley B Powers R PCA as a practical indicator of OPLS-DA model reliability CurrMetabolomics 2016497ndash103

23 Thevenot EA Roux A Xu Y Ezan E Junot C Analysis of the human adult urinarymetabolome variations with age body mass index and gender by implementinga comprehensive workflow for univariate and OPLS statistical analyses J ProteomeRes 2015143322ndash3335

24 Arrambide G Tintore M Espejo C et al The value of oligoclonal bands in themultiple sclerosis diagnostic criteria Brain 20181411075ndash1084

25 Wingerchuk DM Banwell B Bennett JL et al International consensus diagnosticcriteria for neuromyelitis optica spectrum disorders Neurology 201585177ndash189

26 Hyun JW Woodhall MR Kim SH et al Longitudinal analysis of myelin oligoden-drocyte glycoprotein antibodies in CNS inflammatory diseases J Neurol NeurosurgPsychiatry 201788811ndash817

27 Rae CD A guide to the metabolic pathways and function of metabolites observed inhuman brain 1H magnetic resonance spectra Neurochem Res 2014391ndash36

28 Harris JL Choi IY Brooks WM Probing astrocyte metabolism in vivo protonmagnetic resonance spectroscopy in the injured and aging brain Front Aging Neu-rosci 20157202

29 Ciccarelli O Thomas DL De Vita E et al Low myo-inositol indicating astrocyticdamage in a case series of neuromyelitis optica Ann Neurol 201374301ndash305

30 Chard DT Griffin CM McLean MA et al Brain metabolite changes in cortical greyand normal-appearing white matter in clinically early relapsing-remitting multiplesclerosis Brain 20021252342ndash2352

31 Fernando KT McLean MA Chard DT et al Elevated white matter myo-inositol inclinically isolated syndromes suggestive of multiple sclerosis Brain 20041271361ndash1369

32 Kaneko K Sato DK Nakashima I et al Myelin injury without astrocytopathy inneuroinflammatory disorders with MOG antibodies J Neurol Neurosurg Psychiatry2016871257ndash1259

33 Shu Y Long Y Wang S et al Brain histopathological study and prognosis in MOGantibody-associated demyelinating pseudotumor Ann Clin Transl Neurol 20196392ndash396

34 Nicholls P The effect of formate on cytochrome aa3 and on electron transport in theintact respiratory chain Biochim Biophys Acta 197643013ndash29

35 Kapur BM Vandenbroucke AC Adamchik Y Lehotay DC Carlen PL Formic acida novel metabolite of chronic ethanol abuse causes neurotoxicity which is preventedby folic acid Alcohol Clin Exp Res 2007312114ndash2120

36 Treichel JL Henry MM Skumatz CM Eells JT Burke JM Formate the toxic me-tabolite of methanol in cultured ocular cells Neurotoxicology 200324825ndash834

37 Sharpe JA Hostovsky M Bilbao JM Rewcastle NB Methanol optic neuropathya histopathological study Neurology 1982321093ndash1100

38 Nurieva O Diblik P Kuthan P et al Progressive chronic retinal axonal loss followingacute methanol-induced optic neuropathy four-year prospective cohort study Am JOphthalmol 2018191100ndash115

39 Witte ME Mahad DJ Lassmann H van Horssen J Mitochondrial dysfunction con-tributes to neurodegeneration in multiple sclerosis Trends Mol Med 201420179ndash187

40 Pittock SJ Lennon VA Krecke K Wingerchuk DM Lucchinetti CF WeinshenkerBG Brain abnormalities in neuromyelitis optica Arch Neurol 200663390ndash396

41 Chan KH Tse CT Chung CP et al Brain involvement in neuromyelitis opticaspectrum disorders Arch Neurol 2011681432ndash1439

42 Asgari N Lillevang ST Skejoe HP Falah M Stenager E Kyvik KO A population-based study of neuromyelitis optica in Caucasians Neurology 2011761589ndash1595

43 Dickens AM Larkin JR Griffin JL et al A type 2 biomarker separates relapsing-remitting from secondary progressive multiple sclerosis Neurology 2014831492ndash1499


DOI 101212NXI000000000000062620196e626 Neurol Neuroimmunol Neuroinflamm

Tianrong Yeo Fay Probert Maciej Jurynczyk et al metabolomic modeling

Classifying the antibody-negative NMO syndromes Clinical imaging and

This information is current as of October 28 2019

ServicesUpdated Information amp

httpnnneurologyorgcontent66e626fullhtmlincluding high resolution figures can be found at

References httpnnneurologyorgcontent66e626fullhtmlref-list-1

This article cites 43 articles 5 of which you can access for free at

Subspecialty Collections

httpnnneurologyorgcgicollectionmultiple_sclerosisMultiple sclerosis

httpnnneurologyorgcgicollectiondevics_syndromeDevics syndromefollowing collection(s) This article along with others on similar topics appears in the

Permissions amp Licensing

httpnnneurologyorgmiscaboutxhtmlpermissionsits entirety can be found online atInformation about reproducing this article in parts (figurestables) or in

Reprints

httpnnneurologyorgmiscaddirxhtmlreprintsusInformation about ordering reprints can be found online

Academy of Neurology All rights reserved Online ISSN 2332-7812Copyright copy 2019 The Author(s) Published by Wolters Kluwer Health Inc on behalf of the AmericanPublished since April 2014 it is an open-access online-only continuous publication journal Copyright

is an official journal of the American Academy of NeurologyNeurol Neuroimmunol Neuroinflamm

In the multiple sclerosis (MS) or neuromyelitis optica spec-trum disorders (NMOSD) clinic one of the greatest di-agnostic challenges is differentiating antibody-negativepatients with NMOSD from those with opticospinal MS Thisconundrum was demonstrated when large diagnostic dis-agreement was shown even among experts in this field de-spite having the 2015 NMOSD diagnostic criteria in fact thecriteria were not consistently used1

It is clear that the use of discriminatory models on plasmametabolites or conventional MRI can distinguish patientswith relapsing-remitting MS (RRMS) from those withaquaporin-4 antibody (AQP4-Ab) NMOSD and RRMSfrommyelin oligodendrocyte glycoprotein antibody (MOG-Ab) disease remarkably accurately2ndash4 Thus we aim to usethese methods to tackle the diagnostic difficulties inantibody-negative patients who have features overlappingNMOSD and MS The primary methodologic barrier toidentifying discriminators of MS and primary antibody-mediated NMOSD is the lack of a gold standard diagnostic

tool to test accuracy against Therefore there is no publishedstudy to date to resolve this clinical dilemma Given that thetreatment of MS and antibody-mediated NMOSD is mark-edly different and many MS-specific therapies can worsenantibody-mediated NMOSD5ndash12 it is paramount that neu-rologists are able to identify individuals who have antibody-mediated pathology and those with MS pathology withinantibody-negative patients presenting with overlappingclinico-MRI features

In this study we aim to classify a group of difficult-to-diagnose antibody-negative patients into those whose un-derlying pathology are antibody-mediated and those who arelikely to have MS First we assess whether there are spon-taneous clusters of these patients based on their clinical andMRI features using principal component analysis (PCA)Next we explore whether these clusters appear to segregateinto plausible disease-specific groups If these spontaneousclusters appear to identify ldquoMS-likerdquo and ldquoNMOSD-likerdquocohorts we then apply the metabolomics discriminators of

GlossaryAb-NMOSD = antibody-positive NMOSD ANOVA = analysis of variance AQP4-Ab = aquaporin-4 antibody AU = arbitraryunits AUC = area under the curve CPMG = Carr-Purcell-Meiboom-Gill LBL = low brain lesion MOG-Ab = myelinoligodendrocyte glycoprotein antibodyMRS = magnetic resonance spectroscopy NMOSD = neuromyelitis optica spectrumdisordersOPLS-DA = orthogonal partial least square discriminant analysis PCA = principal component analysis ppm = partsper million RRMS = relapsing-remitting MS VIP = variable importance in projection



































Female no () 3 (500) 8 (571) 13 (619)














LETM no () 1 (167) 12 (857) 5 (238)


0 (00) 5 (357) 1 (48)





















Continued






















































Appendix Authors


TianrongYeo MRCP





FayProbertPhD






MeganSealeyPhD



Ana CaveyCNS






MarkWoodhallPhD



PatrickWatersPhD






Daniel CAnthonyPhD




























































Reprints





































Female no () 3 (500) 8 (571) 13 (619)














LETM no () 1 (167) 12 (857) 5 (238)


0 (00) 5 (357) 1 (48)





















Continued






















































Appendix Authors


TianrongYeo MRCP





FayProbertPhD






MeganSealeyPhD



Ana CaveyCNS






MarkWoodhallPhD



PatrickWatersPhD






Daniel CAnthonyPhD




























































Reprints
























































Appendix Authors


TianrongYeo MRCP





FayProbertPhD






MeganSealeyPhD



Ana CaveyCNS






MarkWoodhallPhD



PatrickWatersPhD






Daniel CAnthonyPhD




























































Reprints

























































Reprints




classifying the antibody-negative nmo syndromes

Documents