Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 DOI10.1186/s40478-017-0495-8 RESEARCH Open Access Metabolomics reveals distinct, antibody- independent, molecular signatures of MS, AQP4-antibody and MOG-antibody disease Maciej Jurynczyk1,2, Fay Probert3* , Tianrong Yeo3,4, George Tackley1, Tim D. W. Claridge5, Ana Cavey1, Mark R. Woodhall1, Siddharth Arora6, Torsten Winkler7, Eric Schiffer7, Angela Vincent1, Gabriele DeLuca1, Nicola R. Sibson8, M. Isabel Leite1, Patrick Waters1, Daniel C. Anthony3* and Jacqueline Palace1* Abstract: Theoverlappingclinicalfeaturesofrelapsingremittingmultiplesclerosis(RRMS),aquaporin-4(AQP4)- antibody(Ab)neuromyelitisopticaspectrumdisorder(NMOSD),andmyelinoligodendrocyteglycoprotein(MOG)-Ab diseasemeanthatdetectionofdiseasespecificserumantibodiesisthegoldstandardindiagnostics.However, antibodylevelsarenotprognosticandmaybecomeundetectableaftertreatmentorduringremission.Therefore,there isstillaneedtodiscoverantibody-independentbiomarkers.Wesoughttodiscoverwhetherplasmametabolicprofiling couldprovidebiomarkersofthesethreediseasesandexploreifthemetabolicdifferencesareindependentofantibody titre.Plasmasamplesfrom108patients(34RRMS,54AQP4-AbNMOSD,and20MOG-Abdisease)wereanalysedby nuclearmagneticresonancespectroscopyfollowedbylipoproteinprofiling.Orthogonalpartial-leastsquares discriminatoryanalysis(OPLS-DA)wasusedtoidentifysignificantdifferencesintheplasmametaboliteconcentrations andproducemodels(mathematicalalgorithms)capableofidentifyingthesediseases.Inallinstances,themodelswere highlydiscriminatory,withadistinctmetabolitepatternidentifiedforeachdisease.Inaddition,OPLS-DAidentified AQP4-AbNMOSDpatientsampleswithlow/undetectableantibodylevelswithanaccuracyof92%.TheAQP4-Ab NMOSDmetabolicprofilewascharacterisedbydecreasedlevelsofscyllo-inositolandsmallhighdensitylipoprotein particlesalongwithanincreaseinlargelowdensitylipoproteinparticlesrelativetobothRRMSandMOG-Abdisease. RRMSplasmaexhibitedincreasedhistidineandglucose,alongwithdecreasedlactate,alanine,andlargehighdensity lipoproteinswhileMOG-Abdiseaseplasmawasdefinedbyincreasesinformateandleucinecoupledwithdecreased myo-inositol.Despiteoverlapinclinicalmeasuresinthesethreediseases,thedistinctplasmametabolicpatterns supporttheirdistinctserologicalprofilesandconfirmthattheseconditionsareindeeddifferentatamolecularlevel. Themetabolitesidentifiedprovideamolecularsignatureofeachconditionwhichisindependentofantibodytitreand EDSS,withpotentialusefordiseasemonitoringanddiagnosis. Keywords:Multiplesclerosis,Neuromyelitisoptica,Metabolomics,Biomarker,MOGantibodydisease *Correspondence:[email protected];[email protected]; [email protected] MaciejJurynczykandFayProbertcontributedequally. DanielCAnthonyandJacquelinePalacecontributedequally. 3DepartmentofPharmacology,UniversityofOxford,MansfieldRoad,Oxford OX13QT,UK 1NuffieldDepartmentofClinicalNeurosciences,JohnRadcliffeHospital, UniversityofOxford,Level3,WestWing,HeadleyWay,OxfordOX39DU,UK Fulllistofauthorinformationisavailableattheendofthearticle ©TheAuthor(s).2017OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.0 InternationalLicense(http://creativecommons.org/licenses/by/4.0/),whichpermitsunrestricteduse,distribution,and reproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinkto theCreativeCommonslicense,andindicateifchangesweremade.TheCreativeCommonsPublicDomainDedicationwaiver (http://creativecommons.org/publicdomain/zero/1.0/)appliestothedatamadeavailableinthisarticle,unlessotherwisestated. Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page2of13 Introduction several brain imaging studies have been able to distin- The field of central nervous system (CNS) inflammatory guish MS from AQP4-Ab NMOSD or MOG-Ab disease, demyelinating diseases (IDD) has undergone consider- the almost identical presentation of the latter two condi- able change with the discovery of antibodies against the tions means distinction using radiological features alone aquaporin-4 water channel (AQP4-Ab) in neuromyelitis is not possible [21, 22]. Thus, while the underlying optica spectrum disorders (NMOSD) [30, 31]. More re- mechanisms appear to be unique, the molecular pro- cently, antibodies against conformational epitopes of the cesses which lead to convergent, downstream histo- myelin oligodendrocyte glycoprotein (MOG) have been logical and radiological signs remain unknown. The reported in AQP4-Ab negative NMOSD [28, 35] as well absence of a biomarker for MS means that diagnosis is as in pediatric acute disseminated encephalomyelitis predicated on the exclusion of competing diagnoses and (ADEM) [40]. The identification of these biomarkers, to- so, to date, reliable cell-based assays detecting antibodies gether with immunopathological studies, has led to their against AQP4 and MOG remain the gold standard for increasing recognition as distinct clinical entities separ- diagnosing and differentiating these three conditions. In atefrom multiple sclerosis(MS)[25,34,46,52,53].This spite of this, the most sensitive assays for these anti- has important prognostic and therapeutic implications, bodies are not widely available and can still fail to detect since it is now known that disability in AQP4-Ab the antibodies in patients with low antibody titre when NMOSD is wholly dependent on relapses and that MS- treated and/oroutsideofrelapses. specific treatments are not effective in reducing relapses Despite this invaluable role of AQP4 and MOG inthesepatients[29,41,55]. antibodies in diagnosis of CNS IDD, antibody titre does TherehasbeencontroversyastowhetherCNSIDDas- notappeartocorrelatewithdiseaseseverityorpredictre- sociated with MOG-Ab represents a distinct condition lapsesinthefewpublishedstudiesavailable[3,17,27,50]. separatefromMS.Earlyon,MOGwasproposedasacan- This is particularly true for AQP4 NMOSD, though in didateautoantigenforMSandMOGisstillroutinelyused MOG-Abdiseasethisislessclear.MOG-Abseemstorap- asanimmunogeninanimalmodelsofMS[1,6]including idly decline following monophasic ADEM and persist in thoseusedtoexplorethetreatmentmechanismofglatira- chronic CNS demyelinating conditions [44]. The consist- mer acetate and fingolimod, both of which are approved ent observation for both conditions, however, is that the drugs with proven efficacy in MS patients [10, 45]. In antibody titre can decrease with treatment or when the addition, the specificity of MOG-Abs remained a concern diseaseisinactive,makingthediagnosisinasmallnumber asMOG-Abswerefoundinpatientswithotherinflamma- of cases very problematic. Indeed, of the 54 AQP4-Ab torydiseasesandinhealthycontrols[24,51].Indeed,early NMOSDpatientsincludedinourstudy,24hadloworun- studiesrevealedthepresenceofMOG-AbinMSpatients, detectableantibodylevelsatthetimeofsamplecollection. however, these studies only detected antibodies against Therefore, there is still a need for the discovery of linearepitopes of MOG which werelater found to not be antibody-independent biomarkers which would provide clinically relevant [45]. Recent histopathology studies of additional information on the molecular mechanisms patients with fully conformational MOG-Ab showed fea- underlyingthesediseases. tures compatible with pattern II MS pathology, reflecting High-resolution 1H nuclear magnetic resonance humoral mediated mechanisms [42]. Observations of ab- (NMR) spectroscopy is a non-invasive tool, which, sentor verylowlevelsofconformationalMOG-AbinMS coupled with multivariate statistical analysis, identifies patients, and reports of imaging features distinct from metabolite patterns in biofluid samples. We have previ- MS, supports that MOG-Ab disease is a separate clinical ously shown that this technique is able to distinguish entity from both MS and AQP4-Ab NMOSD, although RRMS from healthy controls with 100% accuracy and pathological biomarkers have not been explored up to from secondary progressive MS (SPMS) with an accur- now[21,22,42,52]. acy of 87% using the metabolite profile alone [7]. Here, MS is believed to be due to an aberrant T-cell re- using a similar approach, we present a comparison of the sponse with B-cell mediated autoimmunity also playing plasma metabolic profiles of patients with RRMS, AQP4- a role [5], while autoantibodies are believed to be central Ab NMOSD, and MOG-Ab disease, in an effort to to the pathogenesis of AQP4-Ab NMOSD [18], and uncover the metabolic signatures representative of each MOG-Ab disease is now regarded as an antibody medi- disease state. Their different metabolic profiles provide ated condition. Despitetheseimmunopathologicaldiffer- further evidence that these three diseases are indeed dis- ences, clinical features overlap which can make clinical tinct,servingaspotentialdiagnosticbiomarkerswhichare distinction challenging [20]. RRMS, AQP4-Ab NMOSD, independent of antibody levels and EDSS. The discrimin- and MOG-Ab disease are all characterised by relapses atory metabolites identified also provide insight into the which involve similar topographical regions within the metabolic perturbations in each condition, allowing ex- CNS, interspersed with periods of remission. While plorationofunderlyingpathophysiologicalmechanisms. Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page3of13 Methods (pH 7.4). Samples were then centrifuged at 16,000 x g Patients for 3 min to remove any precipitate before transferring All patients included in the study were recruited from toa5-mm NMRtube. NMO and MS clinics in the John Radcliffe Hospital in Oxford.PatientswerediagnosedwithAQP4-AborMOG- NMRspectroscopy Ab disease if they had at least one clinically evident in- All NMR spectra were acquired using a 700-MHz Bru- flammatory demyelinating event and tested positive on ker AVII spectrometer operating at 16.4 T equipped cell-based assays for AQP4-Ab or MOG-Ab, respectively with a 1H (13C/15N) TCI cryoprobe. Sample temperature [52, 53]. No patient was double positive for AQP4 and was stable at 310 K. 1H NMR spectra were acquired MOG Abs. AQP4-Ab end point titre was determined by usinga1DNOESYpresaturationschemefor attenuation the reciprocal of the highest positive serum dilution by of the water resonance with a 2 s presaturation. A spin- cell-based assay. A fluorescence visual score of ≥1 (range echo Carr-Purcell-Meiboom-Gill (CPMG) sequence with 0–4)wasusedasthethresholdforpositivity.Patientswere a τ interval of 400 μs, 80 loops, 32 data collections, an diagnosed with RRMS if they fulfilled the revised 2010 acquisition time of 1.5 s, a relaxation delay of 2 s, and a McDonald criteria [43]. All patients gave their written fixed receiver gain was used to supress broad signals consent to participate in the study (Oxford Research Eth- arising from large molecular weight plasma components. ics Committee C Ref: 10/H0606/56 and 16/SC/0224). Pa- CPMG spectra provide a measurement of small molecu- tientdemographicsareshowninTable1. larweightmetabolitesandmobile side chains oflipopro- teins in the plasma sample and were used for all further Plasmasamplecollection analysis. Due to their large molecular weight, AQP4 and Blood was collected into vacutainer lithium-heparin MOG-IgG antibody resonances are not observed in the tubes (Becton Dickinson, product number 367375) and 1H CPMG spectra. 1H correlation spectroscopy (COSY) stored at room temperature for 30 mins before centrifu- spectra were acquired on at least one sample in each gation at 2200 x g for 10 mins. Plasma was immediately classification to aid in metabolite identification. For aliquotedandstored at −80°C. quality control, pooled plasma samples were spread throughouttheruntomonitortechnical variation. NMRsamplepreparation Plasma samplesweredefrosted atroom temperature and NMRdatapreprocessing centrifuged at 100,000 x g for 30 min at 4 °C. 150 μL of Resulting free induction decays (FIDs) were zero-filled the plasma supernatant was then diluted with 450 μL of by a factor of 2 and multiplied by an exponential func- 75 mM sodium phosphate buffer prepared in D O tion corresponding to 0.30 Hz line broadening prior to 2 Table1Patientinformation RRMS AQP4-AbNMOSD MOG-Ab Numberofpatients 34 54 20 Age,mean(range),y 41(18–60) 53(22–83)*† 39(16–70) Gender,No.female(%female) 25(74) 46(85) 9(45) EDSS,median(range) 4(0–7) 3(0–8) 2(0–8) Diseaseduration,median(range),months 89(1–301) 70(3–270) 16(1–420)*~ Timesincerelapse,median(range),months 23(0–133) 22(0–108) 8(1–39)*~ Onoralprednisolone,% 0% 85% 35% Onprednisolone,meandosemg 0 12 14 Onazathioprine,% 0 49 5 Onmethotrexate,% 0 11 5 Onmycophenolate,% 0 19 5 Oninterferonβ,% 15 0 5 Onglatiramer,% 12 0 0 Onfingolimod,% 12 0 0 Ondimethylfumarate,% 12 3 0 Onnatalizumab,% 15 0 0 TheKolmogorov-SmirnovtestwasusedtoidentifysignificantdifferencesofeachclasscomparedtoRRMS(*),AQP4-AbNMOSD(~),orMOG-Abdisease(†) Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page4of13 Fouriertransformation.Allspectrawerephased,baseline total. If these models perform better than models pro- corrected (using a 3rd degree polynomial), and chemical duced by random class assignments this ensures that the shiftsreferenced to the lactate-CH doubletresonance at separation observed between the groups is valid. A final 3 δ=1.33 ppm in Topspin 2.1 (Bruker, Germany). Spectra model is produced using all of the data and a completely were visually examined for errors in baseline correction, independent set of samples (n=10) is used to determine referencing, spectral distortion, or contamination and the predictive value of this model. A schematic of the then exported to ACD/Labs Spectrus Processor Aca- statistical approach can be found in Additional file 1: demic Edition12.01 (Advanced Chemistry Development, Fig. S1. For a more detailed explanation of the analysis Inc.). The regions of the spectra between 0.08–4.20 ppm methodsseetheelectronicAdditionalfile1. and 5.20–8.50 ppm were divided in to 0.02 ppm width ‘buckets’ and the absolute value of the integral of each Lipoproteinsubclassquantification spectral bucket was Pareto scaled. Resonances were To interrogate the lipoprotein subclasses in more detail assigned by reference to literature values [32, 47] and and obtain fully quantitative values the AXINON® lipo- the Human Metabolome Database [56–58] and further FIT® system (numares AG, Germany) was used. Thistest confirmed by inspection of the 2D spectra, spiking of systemdeconvolutesthebroadmethyllipoproteinreson- known compounds,and1D–TOCSYspectra. ance into its constituent parts allowing the direct meas- urement of the cholesterol content, number of particles, Statisticalanalysis andmeanparticlediameterofeachlipoproteinsubpopu- The bucket integrals were imported into R software (R lation. Lipoprotein groups measured include very low foundation for statistical computing, Vienna, Austria) density lipoprotein (VLDL), low density lipoprotein [48]. All multivariate analysis was carried out using in- (LDL), intermediate density lipoproteins (IDL), and high house R scripts and the ropls package [49]. Principal density lipoprotein (HDL) with each group further di- component analysis (PCA) was used to visualize the de- vided intolargeandsmall subpopulations. gree of separation between the disease classifications anddetectpotentialoutliers. Results OPLS-DA was employed to investigate differences in Patientsandpotentialconfounders the disease classifications using the scheme outlined in NMR spectra from 108 patients (34 RRMS, 54 AQP4- Fig. 1. The quality of classification was assessed by first Ab NMOSD, and 20 MOG-Ab disease) were included in correcting for unequal class sizes and then splitting the this study (Fig. 1) to determine whether we could gener- data into a training set (90%) used to build the model ate model algorithms that could accurately distinguish which is then tested on the remaining 10% of the data. between plasma samples from patients with RRMS, This process was repeated to produce 1000 models in AQP4-Ab NMOSD or MOG-Ab disease. Demographic and clinical data, including treatment regimes, for the patients in each group were collected (Table 1). No sig- nificant differences were observed in any of the parame- ters recorded between AQP4-Ab NMOSD samples with high titre (≥ 200) (n=30) and low titre/negative samples (<200)(n=24). Consistent with previous clinical reports, the majority of RRMS and AQP4-Ab patients were female (74% in the RRMS cohort and 85% in the AQP4-Ab NMOSD cohort) and the mean age of the AQP4-Ab NMOSD pa- tients was higher (53 years in AQP4-Ab NMOSD com- pared to 41 and 39 in RRMS and MOG-Ab disease respectively) as a result of older age of onset in this con- dition. No significant differences or correlations were observed in any of the NMR data as a result of differ- ences in the parameters described in Table 1, and multi- variate analysis was unable to discriminate between the NMR spectra based on any of the factors recorded. Fur- thermore, the inclusion of gender, age, disease duration, time since relapse, or medication as variables in the Fig.1Illustrationofthemultivariateanalysismethodology multivariate analysis did not improve the OPLS-DA employedinthisstudy models, confirming that the differences in the Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page5of13 parameters in Table 1 are not responsible for the dis- (Fig. 3). Each point in the scores plot represents an crimination between the three disease groups reported NMR spectrum from a single patient while the axis rep- below. It was noted that the majority of AQP4-Ab resents variation inthe metabolite profile;the metabolite NMOSD patients were on immunosuppression, whilst pattern of points lying close together in the plot are the majority of RRMS patients were on disease modify- similar. The distinct clusters observed for AQP4-Ab ing therapies (DMTs). Therefore, the effect of each NMOSD or RRMS suggests the presence of a distinct medication (listed in Table 1) on the metabolic profile metabolic profile for each condition. In order to confirm was investigated in greater detail. In all cases, multivari- that this separation had not occurred by chance, and ate analysis was unable to discriminate patients on ther- that the OPLS-DA model produced is predictive of apy versus those not on the therapy. For example, we RRMS/AQP4-Ab NMOSD status, a 10-fold cross valid- were unable to build models to discriminate AQP4-Ab ation scheme with repetition was employed as described NMOSD plasma treated with steroids from those not in detail in the electronic Additional file 1. This treated with steroids. Importantly, RRMS and AQP4-Ab approach validates the observed separation by creating NMOSD patients not treated with steroids could still be an ensemble of OPLS-DA models from randomly se- distinguished from each other with high accuracy (Add- lected (size-matched) subsets of the data. Each model itional file 1: Fig. S2). This was the case for all medica- wasthentestedonanindependent subsetofthesamples tions listed in Table 1 including DMTs. Taken together, (excluded from the training set) in order to establish the these findings suggest that none of the demographic and accuracy of the ensemble of models. The same approach clinical parameters described in Table 1 had an appre- was applied to a random set of data, produced by ran- ciableeffectonthemetabolicprofile. dom class assignment of the NMR dataset, and the en- semble accuracies are compared. Figure 2b illustrates TheNMRmetabolitesignatureofRRMSplasmaisdistinct that the accuracy of the RRMS vs. APQ4-Ab NMOSD fromAQP4-AbNMOSDplasma ensemble is significantly greater than that achieved by Simultaneous measurement of multiple metabolites in random chance and, in a rigorous manner, validates the plasma using 1H–NMR spectroscopy followed by the separationobserved andconfirms that themetabolic sig- application of multivariate analysis (OPLS-DA) to gener- natures ofthesetwo diseasesaredistinct. ate predictive models (mathematical algorithms) was Investigation of the variables responsible for the separ- employed here to build models to discriminate between ation in the models revealed several metabolites with RRMS(n=29) and AQP4-Ab NMOSD plasma with titre significant perturbations. Scyllo-inositol, histidine, glu- ≥200 (n=25) at time of sampling. This subset was se- cose, anda subsetofsmalllipoproteinparticles weresig- lected as it represents a diagnostically robust group of nificantly increased in RRMS plasma whilst lactate, AQP4-Ab NMOSD patients suitable for model building. alanine, and a subset of large lipoprotein particles were InspectionoftheresultingOPLS-DAscoresplot(Fig.2a) significantly decreased relative to AQP4-Ab NMOSD revealed a significant difference in the NMR metabolic plasma. pattern of RRMS plasma compared to that of AQP4-Ab As the above analysis proved that the discrimination NMOSD patients with titre ≥200 at time of sampling between diseases did not occur by chance, it was a b Fig.2aOPLS-DAscoresplotofRRMS(black)andAQP4-AbNMOSDwithtitre≥200(red)NMRspectra.bOPLS-DAmodelvalidation.Theaccuracy oftheensembleof1000RRMSV.AQP4-AbNMOSDmodels,asdeterminedbyclassificationofanindependenttestset,issignificantlygreater thanthatofrandomdata.Kolmogorov-Smirnovtestp-values<0.001arerepresentedby*** Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page6of13 Fig.3Average1HCPMGspectraofRRMS(black),AQP4-AbNMOSD(red),andMOG-Ab(yellow)plasmasamples.Boxplotsillustratesignificant differencesintheNMRspectralintegralsforaselectionofmetabolitesselectedbytheOPLS-DAmodels.One-wayANOVAwithpost-hoc(Fisher’s LSD)p-valueslessthan0.05,0.01,and0.001arerepresentedby*,**,and***respectively valid to use the identified metabolites to predict dis- Ab NMOSD vs. RRMS. Of the 24 samples in this subset ease classification. In order to further investigate the only two (one with a titre of 100 and one with a titre of utility of the metabolites described and to produce a 50) were incorrectly identified as RRMS by the OPLS- single algorithm for testing additional plasma samples, DA model, resulting in an accuracy of 92%. This finding a single OPLS-DA model was produced using RRMS indicates that the algorithm can identify AQP4-Ab (n=29) and high titre (≥200) AQP4-Ab NMOSD (n= NMOSDindependent ofantibodytitre.Indeed, allofthe 25) plasma NMR spectra validated above. This model AQP4-Ab NMOSD patients investigated whose antibody was then tested with plasma samples from a ran- levels had become undetectable were correctly identified domly selected set of 10 entirely independent patients and no significant correlations or clustering on principle (5 with RRMS and 5 with AQP4-Ab NMOSD titres component analysis (Additional file 1: Fig. S3) were ob- ≥200), i.e. these naïve samples were not used in the served between any of the NMR metabolites measured model validation at any stage. This final model cor- and antibody titre. The same methods were used to in- rectly predicted the disease classes of this independ- vestigate the effect of EDSS on the metabolite profile. ent test set with an accuracy of 100% (Fig. 4a), We were unable to build models to discriminate be- confirming that the OPLS-DA model is able to iden- tween high and low EDSS patients and no clustering in tify AQP4-Ab NMOSD and RRMS using the NMR the scores plots was observed (Additional file 1: Fig. S4). metabolite profile alone. No significant differences were observed between the metabolite concentrations of the high and low AQP4-Ab TheNMRmetabolicprofileidentifiesAQP4-AbNMOSD titre samples and so these cohorts were combined into a andRRMSinamannerindependentofantibodytitre single AQP4-Ab NMOSD (n=54) cohort for all further We next investigated whether the metabolite profile analysis. could identify AQP4-Ab NMOSD patient samples with titre <200 or that were undetectable at the time of sam- pling (n=24). All patients in this ‘low titre’ cohort had TheplasmaNMRmetabolicprofileofMOG-Abdiseaseis previously robust antibody titre results and a confirmed distinctfrombothRRMSandAQP4-AbNMOSD diagnosis of AQP4-Ab NMOSD. NMR spectra from OPLS-DA was able to discriminate between the plasma AQP4-Ab NMOSD plasma with [100≤titre <200] metabolic profiles of MOG-Ab disease and RRMS (Fig. 4b), [20≤titre <100], (Fig. 4c), and negative results (Fig. 5a) and of MOG-Ab disease and AQP4-Ab (Fig. 4d) at the time of sample collection were assessed NMOSD (Fig. 5b) with accuracies of 73±4% and 73± using the OPLS-DA model generated above for AQP4- 7% respectively. Once again, the accuracy of the Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page7of13 a b c d Fig.4PredictiveOPLS-DAmodelforthediscriminationofRRMSandAQP4-AbNMOSDplasma.OPLS-DAscoresplotsofRRMS(blackcircle)and AQP4-AbNMOSD(redcircle)andpredictedclassificationsofa)independenttestsetofRRMS(blacksquare)andAQP4-AbNMOSD(redsquare),b) lowtitre(≥100and<200)AQP4-AbNMOSD(yellowsquare)c)verylowtitre(<100)AQP4-AbNMOSD(bluesquare),andd)negativetitreAQP4- AbNMOSD(greensquare)plasmasamples ensemble of models was significantly greater than the information is not obtained from the MOG-Ab assay null distribution (random chance models) validating the employed. Nevertheless, no correlation was observed be- models and confirming that there are significant differ- tween the metabolic profile and the semi-quantitative ences in the metabolic profiles of these three diseases. fluorescence visualscore (Additionalfile1:Fig.S5). Interrogation of the discriminatory metabolites selected bythemodelsrevealedthatthe concentration offormate Lipoproteinprofilingrevealsperturbationstoplasma and leucine was increased in MOG-Ab plasma relative lipoproteinpopulationsinAQP4-AbNMOSDandRRMS to both AQP4-Ab NMOSD and RRMS whilst the con- Lipoproteins were identified as highly discriminatory; centration of myo-inositol was decreased. A summary of the removal of lipoprotein measurements resulted in a the significant plasma metabolite changes is given in marked decrease inmodel accuracy. However,the stand- Table 2; a unique set of metabolites was found to vary ard metabolomics NMR experiment (1H CPMG) is un- significantly for each disease class relative to both other able to categorize individual lipoprotein subpopulations, diseases illustrating that the plasma metabolite profiles measure lipoprotein particle number, size, or cholesterol of these three diseases are distinct. Quantitative titre concentration. As a result, we investigated the plasma Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page8of13 a b Fig.5StatisticallysignificantOPLS-DAmodelsfortheseparationofMOG-Ab(yellow)froma)RRMS(black)andb)AQP4-AbNMOSD(red).Model validitywasassessedbycomparingtheaccuracy,sensitivity,andspecificityoftheensemblewiththatofanulldistributionusingthe Kolmogorov-Smirnovtest(p-values<0.001areindicatedby***) lipoproteins with an NMR-based lipidomics platform to HDL particles was significantly lower. As a consequence, define which lipoprotein subpopulations in RRMS anincrease inthemean sizeofthe HDLparticles and an plasma were different from those in AQP4-Ab NMOSD increase in the concentration of cholesterol in the subset (Fig. 6). The number of large LDL and large HDL parti- of large HDL particles in AQP4-Ab NMOSD were also cles was significantly increased in AQP4-Ab NMOSD observed. Interestingly, no significant differences were when compared with RRMS while the number of small observed in the total number of HDL particles or total Table2SignificantdifferencesintheplasmametabolitesofAQP4-AbNMOSD,RRMS,andMOG-Abdisease AQP4-AbNMOSD RRMS MOG-Ab ConcentrationoflargeLDLparticles ↑ SizeofHDLparticles ↑ Glucose ↑ CholesterolconcentrationinlargeHDL(subclassA) ↑ LargeLDLparticles ↑ SmallHDLparticles ↓ Phosphocholine/lipoprotein ↓ Scyllo-inositol ↓ Lysine/creatinine/creatine ↑ Histidine ↑ LargeHDLparticles ↓ Lactate ↓ Unsaturatedlipid ↓ Alanine ↓ Formate ↑ Leucine ↑ Myo-inositol ↓ Increasesanddecreasesrelativetotheothertwodiseaseclassificationsareindicatedwith↑and↓respectively MetaboliteslistedwereidentifiedasdiscriminatorybyOPLS-DA Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page9of13 Fig.6LipoFIT®variableswithsignificantdifferencesbetweengroupsAQP4-AbNMOSD(red),RRMS(black),andMOG-Ab(yellow)followingone- wayANOVA.Post-hoc(Fisher’sLSD)p-valueslessthan0.05,0.01,and0.001arerepresentedby*,**,and***respectively.LLDL-p,largelowdensity lipoproteinparticleconcentration;LHDL-p,largehighdensitylipoproteinparticleconcentration;SHDL-p,smallhighdensitylipoproteinparticle concentration;HDL.A-c,largehighdensitylipoproteincholesterolconcentration HDL-cholesterol concentration between these diseases NMOSD has not been achieved by conventional MRI indicating that discrimination between these diseases [21]. Thus, our observations contribute to the growing would not be possible using a standard cholesterol body of evidence that suggests that MOG-Ab disease test panel. All values obtained from this lipoFIT® represents a distinct clinical and pathophysiological en- analysis matched the intensities obtained from the tity. Since the metabolic profiles of MOG-Ab disease CPMG NMR spectra further validating the significant andAQP4-AbNMOSDaredifferent,despitethembeing differences in the metabolites identified by the multi- predominantly humorally mediated conditions, their in- variate analysis. dividual separation from RRMS is not simply a result of an up-regulated humoral immune response. Further in- Discussion vestigation into the discriminatory metabolites could In this study, using 1H NMR spectroscopy, we demon- provide valuable information on the divergent patho- strate that RRMS, AQP4-Ab NMOSD, and MOG-Ab physiologicalprocessesunderpinningeachcondition and disease display distinct plasma metabolite patterns. In their associated metabolic perturbations. However, the the case of AQP4-Ab NMOSD, the discriminatory me- absence of direct histopathological correlates remains a tabolite pattern identified is independent of antibody limitation here as we cannot easily determine whether titre. The most significant differences in metabolite con- the magnitude of the responses is proportional to the centrations observed across the three conditions in- CNSdiseaseburden. cluded changes in plasma lipoprotein and amino acid Different distributions of lipoprotein populations were levels along with changes in scyllo-inositol and myo- observed in AQP4-Ab NMOSD and RRMS, small HDL inositol. In particular, MOG-Ab disease which overlaps was decreased and large LDL was increased in AQP4- with AQP4-Ab NMOSD both clinically and in treatment Ab NMOSD relative to both RRMS and MOG-ab dis- response, was associated with unique changes in the ease, while large HDL was decreased in RRMS relative levelsofformate,leucine,andmyo-inositol,allowingdis- to AQP4-Ab NMOSD. Total HDL and LDL particle tinction from both RRMS and AQP4-Ab NMOSD. This number did not change significantly across the diseases separation between MOG-Ab disease and AQP4-Ab suggesting that the lipoprotein subclasses have been Jurynczyketal.ActaNeuropathologicaCommunications (2017) 5:95 Page10of13 modified while overall lipoprotein particle numbers re- RRMS relative to MOG-Ab disease. Serum scyllo-inositol main the same. This is supported by the fact that no sig- concentration has been shown previously to be decreased nificantdifferenceswereobservedintotalcholesterol,LDL- in73%ofaheterogeneouspopulationofNMOSDpatients cholesterol,HDL-cholesterol,ortriglycerideconcentrations (both AQP4-Ab positive and negative) when compared and clarifies why a standard lipid panel is unable to dis- with RRMS [37]. This is consistent with our observation criminate between these three diseases. The lipoprotein that scyllo-inositol is decreased in AQP4-Ab NMOSD population in AQP4-Ab NMOSD plasma is skewed to- relative to both RRMS and MOG-Ab disease. Inositol wards larger particles while in RRMS plasma the lipopro- phosphate and myo-inositol are components of myelin, teinparticlesaresmaller.Ourpreviousmetabolomicsstudy and have roles in neural function and homeostasis in the showed that RRMS serum has increased phosphocholine CNS [13, 16, 36] and so the plasma myo-inositol and along with decreased β-hydroxybutyrate, and lipoprotein scyllo-inositol concentrations may reflect perturbations in triacylglycerol (−CH , −(CH -) , and –CH CH CO) con- inositol phosphate metabolism as a result of demyelin- 3 2 n 2 2 centrations with respect to SPMS patient serum [7]. In ation. Taken together, the set of metabolic perturbations addition, a recent report demonstrated that RRMS serum identified may reflect alterations in lipid transport, mem- LDL particles were smaller in RRMS compared to both branebreakdown,andenergymetabolism. SPMS and control samples [19], further supporting that The data presented demonstrates that, in the case of lipoprotein populations are perturbed in MS. Indeed, al- AQP4-Ab NMOSD, the metabolite profile is independ- teredlipidprofileshavebeenpreviouslylinkedwithdisease ent of antibody titre. The OPLS-DA model was able to activity and progression in MS patients [54] although data correctly identify AQP4-Ab NMOSD samples with anti- on lipid alterations in non-MS CNS inflammatory diseases body titres <200 at the time of sampling with an accur- arelacking. Inoneof thefew studiesinvestigating lipopro- acy of 92%, whilst high titre (≥ 200) AQP4-Ab NMOSD teins in AQP4-Ab NMOSD patients, higher serum Apoli- samples and RRMS samples were discriminated with poprotein B levels were observed when compared with 100% accuracy. This finding suggests that the metabolic RRMS [33]. To the best of our knowledge, ours is the first profile could be diagnostically useful in cases of study to compare plasma lipoprotein subclass data in suspected AQP4-Ab NMOSD where antibody level has RRMS, AQP4-Ab NMOSD, and MOG-Ab disease in decreased or become undetectable over time with treat- detail. ment. Patients in the acute setting are often treated em- The primary role of lipoprotein particles is the trans- pirically and the rarer diagnosis of NMOSD is usually port of lipid and other hydrophobic molecules around addressed downstream. Thus, metabolic profiling may the body. Thus, lipoproteins are involved in a wide array be particularly useful in cases where samples are taken of physiological processes including cell signaling, lipid after established immunosuppression and / or outside of homeostasis, and the acute phase response [8, 14, 23]. onset of relapse. Whilst this may only be the case in a Previous studies have suggested that HDL particles are small number of individuals, we believe it is particularly capable of crossing the blood brain barrier [12] and that noteworthyas,outofthe54AQP4-AbNMOSDpatients LDL is present in the parenchyma of early MS lesions (randomly sampled from the Oxford NMO clinic) in- [39]. Lipoprotein modifications also occur in response to cluded in this study, 7% (n=4) were seronegative at the inflammation [2, 38]. Therefore, it is plausible that the time of sampling. Nevertheless, the samples from these plasma lipoprotein perturbations observed here are the four patients were correctly identified as AQP4-Ab result of lipoprotein modifications in response to CNS NMOSD by the OPLS-DA model. We explored whether injury and the inflammatory response. Alternatively, this observation is also true of MOG-Ab disease. How- changes in lipoproteins may reflect altered energy me- ever, due to the lower incidence of this condition, only tabolism in these patients. Indeed, the plasma glucose two samples from patients who were seronegative at concentration was higher in AQP4-Ab NMOSD and lac- time of sampling were available for analysis. Whilst both tate decreased in RRMS plasma. Metabolomics analysis of these samples were correctly identified as MOG-Ab has previously revealed decreased lactate levels in cere- disease by the OPLS-DA model (data not shown), future bral spinal fluid of RRMS patients compared to AQP4- work on a larger cohort will confirm whether the Ab NMOSD although no significant difference in glu- metabolic profile is independent of MOG-Ab serosta- cose wasobservedinthisbiofluid[26]. tus at sampling time-point. Given that the current Myo-inositol was previously reported to be low within work suggests that the AQP4-Ab NMOSD metabolic spinal cord lesions in a group of 5 NMOSD patients (2 profile is independent of antibody titre, and that ob- AQP4-Ab positive and 3 negative) [4] and elevated white servational studies have reported that titres are not matter myo-insoitol has been linked with multiple scler- predictive of relapses, future work will explore the osis [11].Interestingly, the OPLS-DA model revealedthat role of the discriminatory metabolites as biomarkers myo-inositol was higher in both AQP4-Ab NMOSD and of disease activity.
Description: